Applied Evolutionary Economics andEconomic Geography

Applied EvolutionaryEconomics andEconomic Geography

Edited by

Koen Frenken

Utrecht University, The Netherlands

Edward ElgarCheltenham, UK • Northampton, MA, USA

© Koen Frenken 2007

All rights reserved. No part of this publication may be reproduced, stored in aretrieval system or transmitted in any form or by any means, electronic,mechanical or photocopying, recording, or otherwise without the priorpermission of the publisher.

Published byEdward Elgar Publishing LimitedGlensanda HouseMontpellier ParadeCheltenhamGlos GL50 1UAUK

Edward Elgar Publishing, Inc.William Pratt House9 Dewey CourtNorthamptonMassachusetts 01060USA

A catalogue record for this bookis available from the British Library

Library of Congress Cataloguing in Publication Data

European Meeting on Applied Evolutionary Economics (4th : 2005 : Utrecht,Netherlands)

Applied evolutionary economics and economic geography / edited by KoenFrenken.

p. cm.Summary: ‘The volume Applied Evolutionary Economics and Economic

Geography is the fourth book published by Edward Elgar on applied evolutionaryeconomics and stems from the fourth European Meeting on AppliedEvolutionary Economics (EMAEE) held in Utrecht, 19–21 May, 2005. ... Thepresent volume Applied Evolutionary Economics and Economic Geography aimsto advance empirical methodologies in evolutionary economics, this time with aspecial emphasis on geography’–from the Preface.

Includes bibliographical references and index.1. Evolutionary economics–Congresses. 2. Economic geography–Congresses.

I. Frenken, Koen, 1972– II. Title.HB97.3E84 2005330.1–dc22 2006017913

ISBN 978 1 84542 845 7

Printed and bound in Great Britain by MPG Books Ltd, Bodmin, Cornwall

Contents

List of figures viiList of tables ixList of boxes xiList of contributors xiiPreface xiv

1. Introduction: applications of evolutionaryeconomic geography 1Ron A. Boschma and Koen Frenken

PART I ENTREPRENEURSHIP

2. The Cambridge high-tech cluster: an evolutionary perspective 27Elizabeth Garnsey and Paul Heffernan

3. Sophia-Antipolis as a ‘reverse’ science park: from exogenous toendogenous development 48Michel Quéré

PART II INDUSTRIAL DYNAMICS

4. The evolution of geographic structure in new industries 69Steven Klepper

5. Constructing entrepreneurial opportunity: environmentalmovements and the transformation of regional regulatory regimes 93Brandon Lee and Wesley Sine

6. Absorptive capacity and foreign spillovers: a stochasticfrontier approach 121Jojo Jacob and Bart Los

PART III NETWORK ANALYSIS

7. Informational complexity and the flow of knowledgeacross social boundaries 147Olav Sorenson, Jan W. Rivkin and Lee Fleming

v

8. Networks and heterogeneous performance of cluster firms 161Elisa Giuliani

9. Social networks and the economics of networks 180Daniel Birke

PART IV SPATIAL SYSTEMS

10. Diversity, stability and regional growth in theUnited States, 1975–2002 203Jürgen Essletzbichler

11. Inter-regional knowledge flows in Europe: aneconometric analysis 230Mario A. Maggioni and T. Erika Uberti

12. Explaining the territorial adoption ofnew technologies: a spatial econometric approach 256Andrea Bonaccorsi, Lucia Piscitello and Cristina Rossi

PART V PLANNING

13. Evolutionary urban transportation planning?An exploration 279Luca Bertolini

Index 311

vi Contents

Figures

1.1 Evolutionary economic geography applied atdifferent levels of aggregation 4

2.1 High-tech firms in Cambridgeshire, 1960–2004 312.2 Changing sectoral growth patterns among

Cambridge high-tech firms 332.3 Firm turnover (turbulence) in the Cambridge

high-tech sector 342.4 Comparison of survival rates for cohorts of

Cambridge high-tech firms 352.5 Companies with founders from Cambridge

University engineering departments 362.6 New firms started by founders and employees

of Acorn Computers 372.7 Industrial inkjet printing spin-outs originating

in Cambridge 382.8 Biotech firms originating in 12 Cambridge

University departments 403.1 Cumulative number of organisations and employment 493.2 Cumulative number of organisations

and employment (ICT) 523.3 Cumulative number of organisations

and employment (life sciences) 533.4 Size distribution of ICT firms 593.5 The geography of ICT activities 614.1 Entry, exit and number of producers in the television

industry, 1946–1989 714.2 Percentage of television producers in New York, Chicago

and Los Angeles, 1946–1989 724.3 Entry, exit and number of producers in the

automobile industry, 1895–1966 754.4 Percentage of automobile producers in the

Detroit area, 1895–1941 764.5 Entry, exit and number of producers in the

tire industry, 1901–1980 814.6 Percentage of tire producers in Ohio, 1906–1980 81

vii

6.1 Labour productivity growth decomposition 1257.1 Landscape without interdependence 1497.2 Landscape with maximal interdependence 1508.1 Types of networks: (a) BI network in CP;

(b) KN network in CP; (c) BI network in BVC;(d) KN network in BVC; (e) BI network in CV;(f) KN network in CV 167

9.1 Number of subscribers in the UK 1849.2 Development of subscriber market shares 1859.3 Distribution of distances between nodes 1909.4 Interaction network of students 192

10.1 Volatility, diversity, growth 21610.2 Relationship between volatility and growth/diversity 21811.1 Co-patent network, 1998–2002, including Oberbayern,

Darmstadt, Düsseldorf and Île de France 24411.2 Co-patent network, 1998–2002, excluding Oberbayern,

Darmstadt, Düsseldorf and Île de France 24512.1 Distribution of ICT adoption and added value per inhabitant

across Italian provinces: the ‘Three Italies’ 26312.2 ICT adoption and added value per employee,

univariate LISA 26413.1 Changes in the built-up area and infrastructure

in the Amsterdam region, 1967–2001 29913.2a Coping with uncertainty in planning 30713.2b Coping with irreducible uncertainty in

planning (or ‘chaos’) 307

viii Figures

Tables

1.1 Two types of regional innovation policy 155.1 Summary statistics and correlations for state-avoided

cost analysis 1105.2 GEE model predicting state-avoided costs 1116.1a Summary statistics of 17 5-digit ISIC industries: levels 1326.lb Summary statistics of 17 5-digit ISIC industries: average

annual growth rates 1346.2 Stochastic frontier estimates for 17 5-digit industries 1366.3 Decomposition of productivity growth, 1985–1996 1406A.1 Industrial classification 1447.1 Rare events logit models of the likelihood of a focal patent

receiving a citation from a future patent 1568.1 Firm characteristics by cluster 1688.2 Collection of key variables 1698.3 Descriptive statistics and correlation matrix 1738.4 Probit estimations with marginal effects 1749.1 Gender and nationality of respondents 1879.2 Frequencies for choice criteria 1889.3 Do you know which operator your friends/family/

partner use? 1889.4 Local network density 1909.5 Mobile phone operators and nationality 1939.6 Determinants of choosing the same operator 1949.7 Predicted probabilities of using the same operator 1949.8 Friendship determinants 1959.9 Predicted probabilities of calling each other 196

10.1 Correlates of volatility 21310.2 BEA area rankings by stability, diversity and growth 21710.3 Basic statistics of dependent and independent variables 21910.4 Correlation coefficients between dependent and

independent variables 22010.5 Determinants of volatility 22111.1 Pearson and Spearman correlations between knowledge

flow variables 23911.2 QAP correlation between knowledge flow variables 240

ix

11.3 Network analysis indices of knowledge flow structures 24211.4 Gravity equation for knowledge flows 24712.1 ICT adoption: descriptive statistics 26112.2 ICT adoption in macro areas 26112.3 Ranking of Italian provinces by ICT adoption and

per capita income in 2001 26212.4 Spatial dependence tests for the dependent variable 26412.5 Specification of dependent and independent variables 26512.6 Correlation matrix 26612.7 Standard OLS models 26712.8 Spatial lag and spatial error models 26913.1 Overview of different domains of change

in the Amsterdam urban region, 1946–1999 285

x Tables

Boxes

13.1 The late 1960s and early 1970s: a transport and land use policy transition dissected 292

13.2 The late 1980s and early 1990s: a land use policy transition dissected 294

xi

Contributors

Luca Bertolini Amsterdam Institute for Metropolitan and InternationalDevelopment Studies (AMIDSt), University of Amsterdam, TheNetherlands.

Daniel Birke Nottingham University Business School, University ofNottingham, UK.

Andrea Bonaccorsi Facoltà di Ingegneria, Università di Pisa, Italy.

Ron A. Boschma Urban and Regional Research Centre Utrecht (URU),Utrecht University, The Netherlands.

Jürgen Essletzbichler Department of Geography, University CollegeLondon, UK.

Lee Fleming Harvard Business School, Cambridge, MA, USA.

Koen Frenken Urban and Regional Research Centre Utrecht (URU),Utrecht University, The Netherlands.

Elizabeth Garnsey Centre for Technology Management, Institute forManufacturing, University of Cambridge, UK.

Elisa Giuliani Science Policy Research Unit (SPRU), University ofSussex, Brighton, UK.

Paul Heffernan Centre for Technology Management, Institute forManufacturing, University of Cambridge, UK.

Jojo Jacob Eindhoven Centre for Innovation Studies (ECIS), EindhovenUniversity of Technology, The Netherlands.

Steven Klepper Department of Social and Decision Sciences, Carnegie-Mellon University, Pittsburgh, PA, USA.

Brandon Lee School of Industrial and Labor Relations, CornellUniversity, Ithaca, NY, USA.

Bart Los University of Groningen, The Netherlands.

xii

Mario A. Maggioni Department of International Economics,Development and Institutions (DISEIS) and Faculty of Political Science,Catholic University, Milan, Italy.

Lucia Piscitello Dipartmento di Ingegneria Gestionale, Politecnico diMilano, Italy.

Michel Quéré Centre National de la Recherche Scientifique – Groupe deRecherche en Droit, Economie, Gestion (GREDEG), France.

Jan W. Rivkin Harvard Business School, Cambridge, MA, USA.

Cristina Rossi Scuola Superiore Sant’Anna, Pisa, Italy.

Wesley Sine Johnson Graduate School of Management, CornellUniversity, Ithaca, NY, USA.

Olav Sorenson Rotman School of Management, Toronto, Canada.

T. Erika Uberti Department of International Economics, Developmentand Institutions (DISEIS) and Faculty of Political Science, CatholicUniversity, Milan, Italy.

Contributors xiii

Preface

The volume Applied Evolutionary Economics and Economic Geography is thefourth book published by Edward Elgar on applied evolutionary econom-ics, and stems from the fourth European Meeting on Applied EvolutionaryEconomics (EMAEE) held in Utrecht, 19–21 May 2005. The first volumeedited by Paolo Saviotti (Applied Evolutionary Economics, 2003) was thefounding volume. The second volume edited by John Foster and WernerHölzl (Applied Evolutionary Economics and Complex Systems, 2004) dealtwith empirical applications of complex systems theory. The third volumeedited by Andreas Pyka and Horst Hanusch (Applied EvolutionaryEconomics and the Knowledge Based Economy, 2006) discussed the role ofknowledge in the modern economy from an evolutionary perspective.

As with previous volumes, this one also aims to advance empiricalmethodologies in evolutionary economics, but with a special emphasis ongeography. We have seen a growing, mutual interest between economic geog-raphy and evolutionary economics in recent times. In particular, after the2001 Schumpeter Prize winning work by Steven Klepper on the spatial evo-lution of industries and the innovative applications of social network analy-sis by Stefano Breschi, Francisco Lissoni, Elisa Giuliani and others,evolutionary scholars have become interested in the geography of industriesand networks. These two meso levels of analysis will be central in the book.New developments on the interface between geography and evolutionaryeconomics also take place at the micro level, in particular, the geography ofentrepreneurship, and at the macro level, in particular, the dynamics of spatialsystems as reflected in the uneven growth of cities, regions and nations. Themicro and macro levels will supplement the meso level in the book. The bookthus consists of a four-layer structure as follows. Part I: Entrepreneurship(micro); Part II: Industrial Dynamics (meso); Part III: Network Analysis(meso); and Part IV: Spatial Systems (macro). The book opens with an intro-ductory chapter and ends with a policy chapter on planning.

Many people have contributed to this book. First, I would like to thankthe authors who have all written excellent chapters. Their efforts show thevalue added of an evolutionary approach in economic geography. Second, Iwould like to thank the 125 participants of the fourth EMAEE held inUtrecht. The conference theme being ‘geography, innovation and networks’,the event was unique in bringing together economists and geographers to

xiv

exchange ideas and advance our common understanding of economic geog-raphy. Third, I would like to thank the members of the scientific committeefor their review work and the members of the local organising committee forpreparing the conference. Special thanks go to Anet for her excellent plan-ning of the conference, to Jesse for the IT support, and to Siebren for his per-sonal assistance. Finally, I gratefully acknowledge the financial support ofthe Urban and Regional research centre Utrecht (URU), the NetherlandsGraduate School of Housing and Urban Research (NETHUR), the RoyalNetherlands Academy of Arts and Sciences (KNAW), the NetherlandsOrganisation for Scientific Research (NWO), and the Utrecht municipality.

Koen FrenkenUtrecht, March 2006

Preface xv

1. Introduction: applications ofevolutionary economic geographyRon A. Boschma and Koen Frenken

1. INTRODUCTION

Economic geography is the field of study that deals with the uneven dis-tribution of economic activities in space. Two conflicting theories arecurrently influential in the field: institutional economic geography and the‘new’ economic geography. Institutional economic geography is domi-nated by scholars with a geography background and is akin to institutionaleconomics (Hodgson, 1998). At the risk of oversimplification, institu-tional economic geography argues that the uneven distribution of wealthacross territories is primarily related to differences in institutions (Whitley,1992; Gertler, 1995; Martin, 2000). The new economic geography has beendeveloped by neoclassical economists (Krugman, 1991; Fujita et al., 1999;Brakman et al., 2001), who view uneven distributions of economic activ-ity as the outcome of universal processes of agglomeration driven bymobile production factors. Recent debates between geographers and econ-omists have been fierce and with little progress (for example, Martin, 1999;Amin and Thrift, 2000; Overman, 2004). The lack of cross-fertilisationbetween the two disciplines can be understood from two incommensur-abilities between institutional and neoclassical economics (Boschma andFrenken, 2006).

First, institutional economic geography and new economic geographydiffer in methodology. Institutional economic geographers tend to dismiss apriori the use of formal modelling. Instead, they apply inductive, often, case-study research, emphasising the local specificity of ‘real places’. By contrast,the new economic geography approaches the matter deductively using formalmodels based on ‘neutral space’, representative agents and equilibriumanalysis. Proponents of the latter approach do not value, or even reject alto-gether, case-study research. Second, the two theories differ in core assump-tions regarding economic behaviour. The new economic geography aims toexplain geographical patterns in economic activity from utility-maximisingactions of individual agents. By contrast, institutional scholars start from the

1

premise that economic behaviour is best understood as being rule guided.Agents are bounded rational and rely heavily on the institutional framework,which guides their decisions and actions. Institutions are embedded in geo-graphically localised practices, which implies that localities (‘real places’) arethe relevant unit of analysis. Institutions play no role in neoclassical models,or only in a loose and implicit sense. They are not regarded as essential toeconomic explanations, and their study should therefore be ‘best left to thesociologists’, as Krugman once put it (Martin, 1999: 75).

Evolutionary economic geography can be considered a third approach ineconomic geography. Evolutionary economists argue that ‘the explanationto why something exists intimately rests on how it became what it is’ (Dosi,1997: 1531). Rather than focusing on universal mobility processes underly-ing agglomeration (neoclassical) or the uniqueness of institutions inspecific territories (institutional), an evolutionary economic geographyviews the economy as an evolutionary process that unfolds in space andtime. In doing so, it focuses on the path-dependent dynamics underlyinguneven economic development in space (Martin and Sunley, 2006). In par-ticular, it analyses the geography of firm dynamics (such as the geographyof entrepreneurship, innovation and extinction) and the rise and fall oftechnologies, industries, networks and institutions in different localities. Inthis view, uneven economic development requires an understanding of theSchumpeterian process of creative destruction at different levels of spatialaggregation (cities, regions, nations, continents).

Even though evolutionary economics goes back at least to the seminalcontribution by Nelson and Winter (1982), evolutionary approaches to eco-nomic geography are fairly recent (Arthur, 1994; Swann and Prevezer, 1996;Boschma, 1997; Rigby and Essletzbichler, 1997; Storper, 1997; Boschma andLambooy, 1999; Antonelli, 2000; Caniëls, 2000; Klepper, 2001; Maggioni,2002; Breschi and Lissoni, 2003; Bottazzi et al., 2004; Brenner, 2004; Werkerand Athreye, 2004; Boschma and Wenting, 2005; Essletzbichler and Rigby,2005; Martin and Sunley, 2006). The difference between evolutionary eco-nomic geography and both new and institutional economic geography canbe summarised as follows (Boschma and Frenken, 2006). An evolutionaryapproach to economic geography is different from new economic geographyin that it attempts to go beyond the heroic assumptions about economicagents and the reduction of geography to transportation costs. At thesame time, evolutionary economic geography also differs from insti-tutional economic geography in that an evolutionary approach explainsterritorial differences not primarily by referring to different institutions,but from differences in the history of firms and industries residing in aterritory. An evolutionary analysis may well take into account the role ofinstitutions though, but in a co-evolutionary perspective (Nelson, 1995).

2 Introduction

Methodologically, evolutionary economic geography differs from bothinstitutional and new economic geography in that it combines all researchmethodologies: case-study research, surveys, econometrics, theoreticalmodelling exercises and policy evaluation can, in principle, all be based onevolutionary theorising.

The present volume, Applied Evolutionary Economics and EconomicGeography, aims to further develop an evolutionary economic geography.It does so by bringing together a selected group of excellent scholarscoming from business studies, economics, geography, planning and organ-isational sociology. All contributors share an interest in explaining theuneven distribution of economic activities in space and the historicalprocesses that have produced these patterns. The heterogeneity in back-grounds was overcome by a common understanding of the evolutionarynature of spatial processes. The end result is a volume of 13 chapters onvarious topics organised under the headings of entrepreneurship, industrialdynamics, network analysis, spatial systems and planning. The volume alsoreflects the variety of research methodologies characterising applied evo-lutionary economics, including case-study research (Garnsey andHeffernan, Chapter 2; Quéré, Chapter 3; Lee and Sine, Chapter 5;Bertolini, Chapter 13), duration models (Klepper, Chapter 4), data envel-opment analysis (Jacob and Los, Chapter 6), complexity theory (Sorensonet al., Chapter 7), social network analysis (Sorenson et al., Chapter 7;Giuliani, Chapter 8; Birke, Chapter 9; Maggioni and Uberti, Chapter 11),spatial econometrics (Essletzbichler, Chapter 10; Bonaccorsi et al., Chapter12) and gravity modelling (Maggioni and Uberti, Chapter 11).

2. EVOLUTIONARY ECONOMIC GEOGRAPHY:MICRO, MESO AND MACRO APPLICATIONS

Boschma and Frenken (2006) argued that applications of evolutionary eco-nomic geography primarily fall under four categories: firm, industry,network and spatial systems. Their scheme also underlies the structure ofthe book with the various chapters being organised under one of these fourheadings. Following Figure 1.1, the categories follow from aggregatingfirms to their relevant meso levels of the industry in which they competeand the networks in which they exchange commodities and share knowl-edge. Aggregating in turn the meso levels to the macro level, one obtainsthe macro level of spatial systems. Following this scheme, localities inspatial systems, be it cities, regions or countries, can be characterised bytheir sector composition and their position in spatial networks, and struc-tural changes herein over time (Castells, 1996).

Introduction 3

4 Introduction

Source: Adapted from: Boschma and Frenken (2006, p. 293).

Figure 1.1 Evolutionary economic geography applied at different levels ofaggregation

Macro

Meso

Micro

Spatial systems(Chapters 10–12)

Industrial dynamics(Chapters 4–6)

Network analysis(Chapters 7–9)

Entrepreneurship(Chapters 2–3)

Policy(Chapter 13)

Introduction(Chapter 1)

Entrepreneurship

We consider evolutionary economic geography to involve a synthesis ofevolutionary economics and economic geography. Following evolutionaryeconomics, our starting point is the firm, which competes on the basis ofits routines and core competences that are built up over time (Nelson andWinter, 1982). Organisational routines and core competences consist for alarge part of learning-by-doing and tacit knowledge, which are hard tocodify and difficult to imitate by other firms (Teece et al., 1997; Maskell,2001). Consequently, organisations are heterogeneous in their routines, andpersistently so (Klepper, Chapter 4; Giuliani, Chapter 8). Models can thusno longer rely on assuming a ‘representative agent’, but have to account forheterogeneous firms. This variety provides the fuel for selection processes,which causes some firms to prosper and grow and others to decline and pos-sibly exit. From this evolutionary process of firm dynamics based on com-petition, innovation and selection, an emergent spatial pattern of economicactivity arises. This evolving economic landscape, as reflected by spatialheterogeneity in firms’ routines, can be understood as the joint outcome ofgeographical proximity (enhancing innovation and imitation) on the onehand, and spatial differences in selection conditions on the other (Boschmaand Lambooy, 1999; Essletzbichler and Rigby, 2005).

In the context of economic geography, firm location, or more generally,the locational behaviour of firms, is the central explanandum (Stam, 2003).Demographically, the evolutionary economic process unfolding in spaceand time is driven by entry of new firms, exit of incumbent firms and re-location of incumbent firms. Through this process, new routines are beingdiffused in space. From an evolutionary perspective, one does not analysenew firm location solely as the outcome of rational decisions directed byprice differentials, as in neoclassical theory, or in terms of comparing insti-tutional frameworks in different areas, as in institutional theory. Rather,one is interested in the history of the founder and key employees of a newventure to account for routines transferred from a previous activity,and how that affects their survival. And, to understand uneven rates ofregional entrepreneurship and entrepreneurial success, one is interested inthe spatial distribution of resources required to start up a new business. Asentrepreneurs require resources (capital, labour, networks, knowledge) tostart new ventures, and resources tend to concentrate in space, as in urbanareas (Hoover and Vernon, 1959) or specialised clusters (Porter, 2000), theprobability of starting a new venture can also be made dependent onterritorial conditions. This is not to say that price differentials (the neo-classical view) and place-specific institutions (the institutional view) do notmatter. Rather, prices and institutions only condition the range of possible

Introduction 5

economic behaviours and their locations, while the actual behaviours aredetermined by the path-dependent history of actors involved in particularterritorial settings (Boschma and Frenken, 2006).

The core concept of path dependence can also be fruitfully applied tofirm location. Location decisions by firms are heavily constrained by thepast. For example, many firms just start at locations where the founder lives,due to bounded rationality, or because the founder is socially embedded inlocal networks, and it is well known that most spin-offs locate near theparent firm (Cooper and Dunkelberg, 1987; Klepper, 2001). In either case,previous decisions taken in the past determine the location decision of anew firm. Path dependence also affects the probability of relocation asfirms are expected to display a considerable degree of locational inertia.The probability of relocation decreases over time as a firm develops a stableset of relations with suppliers and customers and sunk costs accumulatein situ (Stam, 2003). Of course, even though path dependence constrainsrelocation of the firm, one can expect the firm to outsource parts of the pro-duction to low-wage locations, in particular, activities that rely less on theorganisational and core competences built up in situ over time (see Vernon,1969). The probability and economic success of off-shoring, however,depends on a firm’s capability to transfer its routines to different localities(Kogut and Zander, 1993).

Research has paid special attention to the geography of high-tech entrepre-neurship (Hall and Markusen, 1985; De Jong, 1987; Aydalot and Keeble, 1988;Saxenian, 1994; Stuart and Sorenson, 2003). New high-tech firms are com-monly thought to fuel employment growth and regional economic develop-ment. In the present volume, we focus on two exceptional European regions thathave been successful in fostering high-tech entrepreneurship in information andcommunication technology (ICT). The two cases concern Cambridge, UK andSophia-Antipolis near Nice. The development of Cambridge as a high-techregion can be understood as resulting from an endogenous evolutionaryprocess of entrepreneurs setting up business and hereby improving the con-ditions for new ventures to occur (Garnsey and Heffernan, Chapter 2). Theendogenous process encompassed the founding of companies by members ofthe university, spin-offs, the rise of local suppliers and the emergence ofspecialist labour markets. This process, however, has not been entirely ‘auto-matic’. Once congestion became problematic and university regulations wereperceived unfavourable for entrepreneurship, collective action resulted in insti-tutional reform. Thus, the history of the Cambridge region illustrates both theendogenous nature of entrepreneurship and the co-evolutionary process ofentrepreneurship, regional development and institutional change. Anotherexample of successful regional development is the science park of Sophia-Antipolis. However, its development was far from endogenous. Rather the

6 Introduction

process was triggered by the presence of a few large companies, a favourableliving environment and a visionary man (Quéré, Chapter 3). Interestingly, theprocess transformed from being triggered by external factors into a moreendogenous process from the early 1990s onwards. The endogenous nature ofthe more recent history is evidenced by the fact that even though some largerfirms left the park in the early 1990s to go to larger agglomerations such as Parisand London, employees decided not to leave the region, but to start their ownventures instead. In this particular case, it is the employee rather than the firmthat shows locational inertia. Thus, the two cases of Cambridge and Sophia-Antipolis are different yet equally successful in the creation of new high-techfirms (see also, Garnsey and Longhi, 2004).

Industrial Dynamics

Starting from the firm, the first meso level of aggregation that is specificallyimportant in evolutionary economic geography is the industry level. In thiscontext, the main phenomenon to be explained is the process of spatial con-centration or de-concentration of an industry over time. Arthur (1994)developed two simple evolutionary models of spatial concentration byspin-off and by agglomeration economies (see also, Boschma and Frenken,2003). In the spin-off model an industry comes into being as a Polya processof firms giving birth to firms giving birth to firms and so on. This processis known to have played an important role in the rapid growth and spatialconcentration of several industries, including the concentration of the USautomobile industry in the Detroit area (Klepper, 2001), the ICT sector inSilicon Valley (Saxenian, 1994) and the biotechnology sector in Cambridge,UK (Keeble et al., 1999).

Klepper (2001, 2002) extended the spin-off model in an industry life-cycle model, which synthesises five assumptions: routines are hetero-geneous; spin-offs inherit the routines of parent firms; more successfulfirms grow faster; larger firms produce more spin-offs; and worse-perform-ing firms are forced to exit due to competition. The first four mechanismsensure that the region that hosts early, experienced and successful entrantswill come to dominate the industry. In contrast to Arthur’s spin-off model,this truly concerns a process of inheritance in which the experience ofparent firms is inherited by spin-offs with a positive impact on their survivalrates. The fifth mechanism of cost competition at the sector level asym-metrically affects regions, causing the region hosting the less successfulfirms to decline, leaving the region hosting the successful companies todominate the industry. Typically, cost competition becomes fierce only afteran industry has developed for a number of years, that is, after productstandardisation has taken place and innovation shifts to process innovation

Introduction 7

in line with the product life-cycle hypothesis (Abernathy and Utterback,1978). The result is a shakeout forcing many firms to exit the industry,which strongly affects the spatial distribution of the industry since routinesare heterogeneous and unevenly spread. The predictions of the model canbe tested econometrically in a relatively straightforward way using durationmodels (Klepper, 2001, 2002).

Arthur’s (1994) second model of agglomeration economies assumes thatnew firms start up rather than spin off from incumbent firms. The locationchoice of a new firm can therefore not be ‘automatically’ determined by thelocation of the parent company: the location of the firm becomes a choicedecision. Arthur assumes that each firm has a locational preference for oneparticular region. While Arthur is far from explicit on this matter, this het-erogeneity in preferences can stem from bounded rationality yet may alsobe given an empirical meaning: start-ups typically locate their business inthe region where the founder lives and/or held previous employment.Agglomeration economies arising from spatial concentration of firmsoperating in the same industry, cause the industry to concentrate in onesingle region even though the individual firms have different individualpreferences. The reason is that once one region has attracted slightly moreentrants than other regions, a critical threshold is passed, and suddenly allfirms will opt for this one region: a case of spatial lock-in.

In an empirical context, the outcomes of the spin-off model are not easilydistinguishable from the outcomes of the agglomeration economies model.We have, indeed, two different explanations for the same phenomenon ofspatial concentration of an industry. As spin-off dynamics and agglomer-ation economies may well contribute to spatial concentration simultane-ously, the challenge for empirical research is to disentangle both processesso as to assess their presence and importance. One out of the few studies thathave attempted to do so is Klepper’s (2001) study of the US automobileindustry. In his econometric analysis, he included a dummy for being locatedin the Detroit area. The dummy showed no positive effect on the survival offirms, which suggests that agglomeration economies were not present. Theuse of a Detroit control variable, however, can be questioned, since a subsetof firms within the Detroit area may have benefited from each other’s pres-ence through local networks (Giuliani, Chapter 8) or firms may havebenefited from knowledge spillovers over a longer distance (Jacob and Los,Chapter 6). Despite this shortcoming, the result by Klepper (2001) stronglysuggests that the concentration of the US automobile industry in Detroitcan be attributed mainly to the self-reinforcing dynamics of successful firmscreating successful spin-offs, and so on.

A study by Boschma and Wenting (2005) on the spatial evolution ofthe British automobile sector came to similar conclusions regarding the

8 Introduction

self-reinforcing nature of spin-off dynamics, which, in the British case, ledto a concentration in the Birmingham–Coventry area. However, Boschmaand Wenting also accounted for the presence of related industries (such ascoach and cycle making) in a region as a potential source of agglomerationeconomies, which was shown to have a positive effect on the survival rateof firms. Thus, the local presence of related industries appeared to bebeneficial due to, for example, knowledge spillovers and skilled labour, yetthe local presence of a high number of firms operating in the same indus-try turned out to be harmful due to increased competition, lowering thesurvival chances of new entrants. Another recent elaboration on Klepper’smodel is by Cantner et al. (2005), whose methodology using instrumentalvariable estimation allows for post-entry innovation. In doing so, the sur-vival probabilities are not only dependent on initial conditions of entrants,but also on the research and development activities they undertake duringtheir lifetime. These contributions suggest that survival analysis is a promis-ing research methodology in evolutionary economic geography.

Importantly, in an evolutionary context, spatial concentration (or itsabsence) is not only an outcome of a process of industrial evolution, butalso affects an industry’s further evolution. This recursive relationship iscentral in another empirical tradition in industrial dynamics known as‘organisational ecology’ or ‘firm demography’ (Hannan et al., 1995; Carrolland Hannan, 2000; Stuart and Sorenson, 2003; van Wissen, 2004). First,geographical concentration of industrial activities can generate positivefeedbacks on entry rather than performance. This means that an industrycan become concentrated through a self-reinforcing process of entry trig-gering more entry. Second, geographical concentration of firms increasesthe level of competition and makes entry less likely. This negative feedbackset limits to spatial concentration. Typically, positive feedbacks operate atthe start of an industry life cycle, while negative feedback takes over aftera certain threshold of spatial concentration is passed. Interestingly, the twoprocesses causing positive and negative feedbacks may well operate atdifferent spatial scales depending on the type of industry (Jacob and Los,Chapter 6). In industries where demand is local and knowledge spilloversmore global, one expects negative feedbacks to operate at a lower spatiallevel than positive feedbacks, resulting in a more even spatial distribution(Hannan et al., 1995). However, in markets where competition is global, butknowledge spillovers rather local, the reverse may well be the case.

Institutions also affect the spatial evolution of industries. From an evo-lutionary perspective, the question is not so much whether particular insti-tutions triggered the development of a particular industry in a certainregion, but rather how institutions have co-evolved with the emergence ofa new sector (Nelson, 1995). The co-evolutionary perspective is important

Introduction 9

because it acknowledges that innovations leading to new sectors oftenrequire the restructuring of old institutions and the establishment of newinstitutions (Freeman and Perez, 1988). Examples of the co-evolution ofnew sectors and institutions are the rise of the synthetic dye industry in thesecond half of the nineteenth century in Germany (Murmann, 2003) andthe evolution of the UK retail banking industry from the 1840s to the 1990s(Consoli, 2005). In their study of the spatial diffusion of the renewableenergy technology, Lee and Sine (Chapter 5) also emphasise the differentialinstitutional changes occurring in different American states.

Network Analysis

Networks provide another unit of analysis. Unlike the competitive natureof industrial dynamics, network relationships are less competitive and of amore complementary nature. One important aspect of networks in evo-lutionary economic geography is that these act as vehicles for knowledgespillovers. A key research question is then to determine whether knowledgediffusion and innovation is more a matter of being in the right place, in theright network, or in both (Boschma and Ter Wal, 2006). Social networkanalysis provides a rich toolbox for the analysis of the structure and evo-lution of networks (Wasserman and Faust, 1994; Carrington et al., 2005).What is more, there is a lot of interest in theorising about networks andnetwork formation starting from the pioneering work by Granovetter(1973) and Burt (1982) to more recent, but already classic contributions ofWatts and Strogatz (1998) and Barabasi and Albert (1999).

In evolutionary economics, interest in networks stems primarily fromthe increasing importance of networks among high-technology firms(Hagedoorn, 1993; Powell et al., 1996), while geographical studies haveshown the role of networking in clusters (Uzzi, 1996; Maskell andMalmberg, 1999). The central question has been whether agents profit fromsimply being co-located or whether network relationships are required tocarry these knowledge flows. A related question is whether geographicalproximity facilitates the formation of network links. An innovative studyby Breschi and Lissoni (2003) found that, using co-inventor data to indi-cate social networks and patent citations to indicate knowledge flows, geo-graphical localisation of knowledge spillovers can be largely attributed tosocial networks and labour mobility. This study shows considerableprogress over the study by Jaffe et al. (1993), who treated geographicalspace as a black box. The Breschi–Lissoni study suggests that geographicalproximity is neither a necessary nor a sufficient condition for knowledgespillovers to occur. Rather, knowledge diffuses through social networks,which are dense between proximate actors, but also span across the globe.

10 Introduction

Network analysis between firms in specialised clusters is another field inwhich social network analysis can be fruitfully applied. Using survey data,Giuliani (2005) has been able to map the business and knowledge networksamong wine producers in three different clusters. She found that the the dis-tribution of connectivity is much more skewed in knowledge than in busi-ness networks, which suggests that only a few central firms profit fromknowledge spillovers. This hypothesis has been put to the test in a follow-up study presented in this volume (Giuliani, Chapter 8), in which it isshown that a firm’s centrality in knowledge networks is indeed positivelyaffecting innovative performance, even after controlling for heterogeneityin internal competencies. A recent study by Boschma and Ter Wal (2006)on a footwear district in southern Italy tends to suggest that the absorptivecapacity of firms is indirectly related to their innovative performance,through having non-local instead of local relationships. That is, the higherthe absorptive capacity of a district firm, the better it is connected to organ-isations outside the district, which, in turn, impacts positively on their inno-vative performance. These studies show that social network analysis is apowerful tool in analysing the geography and structure of knowledge net-works and the effect of a firm’s network position in these networks on itsperformance. In a similar fashion, the concept of regional innovationsystems (Cooke et al., 1998) can be operationalised empirically more sys-tematically by mapping the various network relations of actors that arepart of the regional system with other actors within and outside theregional system.

Evolutionary theorising has also argued that, due to bounded rational-ity, consumers also rely on personal networks. As a result, certain decisionsby central actors can propagate through the network, leading many con-sumers to opt for the same product (Cowan et al., 1997; Plouraboue et al.,1998; Solomon et al., 2000). The strength of these networks effects, and thegeographical nature of such personal networks, can also be explored empir-ically using social network analysis. A nice example of such an approach isthe study by Birke (Chapter 9), who conducted a survey among studentsasking them about their personal networks and their choice of mobile tele-phone operator so as to analyse the effect of personal networks on thechoice of operator.

Hitherto, the use of social network analysis in evolutionary economics hasbeen almost exclusively static. A future challenge is to understand the spatialevolution of networks. This requires longitudinal data and methods toanalyse the dynamics of networks over time. An influential theoretical modelof network dynamics is the model by Barabasi and Albert (1999). In thismodel, a network grows as new nodes connect to a network. Nodes areassumed to attach themselves to other nodes with a probability proportional

Introduction 11

to the latter’s connectivity. This principle is known as ‘preferential attach-ment’, which means that a new node prefers to link with a well-connectednode so as to profit from its connectivity. Well-connected nodes will thentend to become even more connected, while peripheral nodes in the networkwill tend to remain peripheral. The resulting distribution of connectivity willbe extremely skewed (scale free). Which of the nodes becomes the centralnode is path dependent, and thus unpredictable, although early entrants willhave a much higher probability of becoming central than later entrants. Thestochastic logic underlying the Barabasi–Albert model of network formationhas also been applied to the spatial evolution of networks where new nodescan occur anywhere in space, and connections between nodes are madedependent on both geographical space (negatively) and preferential attach-ment (positively). The resulting topology and spatial organisation of anetwork can then be understood as a purely stochastic and myopic sequence(Andersson et al., 2003, 2006) that may generate hub-and-spokes networks,as observed in infrastructure networks (for example, Guimerà and Amaral,2004; Barrat et al., 2005). Empirical research in this field, however, has stillbeen rather limited.

Spatial Systems

Aggregating sectors and networks to the macro level of spatial systems, oneobtains a model of the growth of localities (cities, regions, countries), asdepending on their sectoral composition and global network position, andthe structural changes herein occurring over time. The sectoral logic under-lying the evolution of spatial systems is better known as the process ofstructural change (Freeman and Perez, 1988; Boschma, 1997, 2004). Citiesand regions that are capable of generating new sectors with new productlife cycles will experience growth, while cities and regions that are lockedinto earlier specialisations with mature life cycles will experience decline.Importantly, there is no automatic economic or political mechanism toensure that cities or regions will successfully renew themselves. Rather, oneexpects localities in most instances to experience decline after periods ofgrowth due to vested interests, institutional rigidities and sunk costs asso-ciated with previous specialisations (Grabher, 1993). There are, however,still very few systematic evolutionary studies on convergence and diver-gence at different spatial scales (for example, Pumain and Moriconi-Ebrard, 1997; Caniëls, 2000). This can be partly understood from thedemanding data requirements for systematic analysis of long-term dynam-ics, especially if one is interested in analyses at subnational levels.

A particularly popularly topic in economic geography concerns the roleof variety in regional growth. Economic theory has long been focused on

12 Introduction

explaining economic growth by a combination of growth in inputs andefficiency improvements (Solow, 1957). The underlying qualitative nature ofeconomic development, in terms of the variety of sectors or the variety oftechnologies, has been addressed only rarely. One can distinguish three typesof relationships between variety and economic development (Frenken et al.,2005, 2006). The first approach centres on variety, knowledge spillovers andgrowth, which has become a central theme in what is called ‘new growththeory’. It has been argued that, apart from spillovers occurring betweenfirms within a sector, spillovers also occur between sectors, which are com-monly referred to as ‘Jacobs externalities’, after Jacobs (1969). A secondway to relate variety to regional economic development is to view variety asa portfolio strategy to protect a region from external shocks in demand(Essletzbichler, Chapter 10). In this context, one also speaks of regionaldiversification analogous to corporate diversification as a risk-spreadingstrategy. A third type of relationship between variety and economic devel-opment concerns the long-term effect of variety on the economic system.An economy that does not increase the variety of sectors over time, willsuffer from structural unemployment, and will ultimately stagnate. In thisview, the development of new technologies and sectors in an economy isrequired to absorb labour that has become redundant in existing sectors(Pasinetti, 1981, 1993; Saviotti and Pyka, 2004). This process underlyinglong-term growth has major geographical implications, when new sectorsemerge in other areas than the ones where old sectors are located. Thiswould imply that labour becomes redundant primarily in areas where theold sectors are concentrated, while new employment is primarily created innew areas. This imbalance may be counteracted by labour migration fromold to new areas and by firm migration in the opposite direction.

Although many empirical studies have analysed the effects of variety onregional growth in the past decade or so, some methodological issues inempirical research remain. First, the measurement of variety is not trivial.For example, one would like to distinguish between related variety under-lying spillovers and unrelated variety underlying the portfolio effects(Frenken et al., 2006; Essletzbichler, Chapter 10). Second, explainingregional phenomena requires a careful econometric specification so as toallow different effects to take place at different spatial levels of aggregation.For example, the rate of regional growth or the rate of regional informationtechnology (IT) adoption can be made dependent on the rate of growth inneighbouring regions through the use of spatial autocorrelation econo-metrics (Essletzbichler, Chapter 10; Bonaccorsi et al., Chapter 12).

The network perspective also lends itself for aggregation to the macrolevel. By aggregating networks between firms to the locations of these firms,one obtains inter-city and inter-regional networks. The underlying concept

Introduction 13

of ‘network cities’ has become very common among geographers (Pred,1977; Hohenberg and Lees, 1995; Castells, 1996). The central idea under-lying the concept of network cities holds that connectivity contributes bothto urban economic growth and to urban inequalities. Examples of empiri-cal studies that map urban networks include networks based on the tiesbetween headquarters and subsidiaries of multinational organisations(Taylor, 2001; Alderson and Beckfield, 2004), on transportation networks(Matsumoto, 2004) or IT infrastructure (Moss and Townsend, 2000). Inthese views, cities can develop a more central network position by attract-ing corporate headquarters or functioning as transportation or IT hubs.The concept of inter-city networks can also be applied to inter-regional net-works, as the contribution by Maggioni and Uberti (Chapter 11) shows.Regions acting as central hubs in the development and diffusion of knowl-edge will be more central in these networks, while other regions will staymore peripheral. Network position is thus expected to affect regionalgrowth, as central hubs will receive more, and more relevant, knowledgespillovers. Using Tinbergen’s (1962) gravity model from international tradetheory, one can also analyse to what extent geographical distance affects thestrength of knowledge flows between any two regions. This question hasalso been taken up by Maggioni and Uberti (Chapter 11).

As for the study of firm networks, the dynamic analysis of urban andregional networks is still in its infancy. Understanding the structure of anetwork at one moment in time requires an understanding of the evo-lutionary process that has given rise to such structures. An interestingresearch avenue is to analyse the determinants of changes in network struc-tures in a spatial system. For example, does the accession of EasternEuropean countries reorganise the hierarchy in the European city system?And, historically, can we relate the rise and fall of cities to their changingpositions in global knowledge networks around emerging technologies andinfrastructures (Pumain, 1997)?

3. POLICY

The contributions in the present volume focus on understanding spatialphenomena from an evolutionary perspective. General policy implicationsare often hard to draw, if only because evolutionary theorising leaves roomfor ‘small events’ to have long-lasting effects. Some may even go a stepfurther to suggest that evolutionary analysis often shows the limited poten-tial of policy makers to truly influence long-term geographical patterns ofeconomic growth. For example, Klepper’s (Chapter 4) conclusion that theUS automobile industry became concentrated in Detroit for accidental

14 Introduction

reasons, suggests that efforts to attract new industries to a particular city orregion have a low probability of success. What matters most is to have com-petent entrepreneurs, the presence and actions of which are hard toinfluence by policy. Similarly, the success story of Sophia Antipolis (Quéré,Chapter 3) suggests that its success is unique and difficult to copy. Theprocess of regional development was set in motion by external factors suchas climate, the presence of multinationals, the international airport, andone visionary man. And, in the case of Cambridge, regional developmentwas fuelled by its excellent university as well as by the benefits of theGreater London area at just one hour from Cambridge (Garnsey andHeffernan, Chapter 2).

Even if policy implications of evolutionary economics are inherentlydifficult to derive, a growing number of evolutionary economists are tryingto draw some policy implications (Perez and Soete, 1988; Metcalfe, 1995;Foray, 1997; Nelson, 1999; Lambooy and Boschma, 2001; Chang, 2003).The point of departure is that the focus on static efficiency in neoclassicaleconomics is to be replaced by dynamic efficiency (Nelson and Winter,1982). In other words, one is not only interested in the allocation of scarceresources present today, but also in the opportunities to create newresources in the future.

In the context of economic geography, the question becomes how todesign policies that promote dynamic efficiency at urban and regionallevels. Boschma (2005) distinguished between two types of regional policy:evolutionary and revolutionary (Table 1.1). Evolutionary regional policytakes the specific local context and industrial structure as the starting point.It is a fine-tuning policy that aims to strengthen the connectivity betweenthe elements of the regional system. In these circumstances, local policymakers have few degrees of freedom, yet are more likely to be successful aslong as their actions are localised, that is, focused on reproducing andstrengthening the existing structures. In other words, the local environment

Introduction 15

Table 1.1 Two types of regional innovation policy

Evolutionary type of policy Revolutionary type of policy

Location-specific policy Generic policyFine-tuning Restructuring of institutional frameworkStrengthening existing connectivity Stimulating new connectionsBenefiting from specialisation Stimulating diversityFew degrees of freedom More degrees of freedomLess uncertainty More uncertainty

Source: Adapted from Boschma (2005).

determines to a large extent available options and probable outcomes ofregional policy.

The goal of a revolutionary regional policy, by contrast, is the restruc-turing of the social and institutional framework by constructing newregional systems, increasing diversity and a high degree of openness regard-ing the inflow of labour, capital and knowledge. In these circumstances,local policy makers have more degrees of freedom, but at the cost of ahigher degree of uncertainty regarding the actual outcome of regionalpolicy making and its success. Since path dependence is less relevant, it isless meaningful to account for the location-specific context as a startingpoint for regional policy. Radically new trajectories of industrial develop-ment build on generic conditions, because the existing actors and insti-tutional environment are unlikely to provide the specific stimuli. The caseof Sophia Antipolis seems to be a good example of such a development.

The paradox of regional policy holds that it can be very effective andsuccessful in conserving economic activity by means of evolutionary poli-cies, yet it has difficulty triggering, or even opposes new economic activitynecessary for long-term development. Note, however, that evolutionaryand revolutionary policies are not mutually exclusive. One can pursue fine-tuning policies in existing sectors while improving the generic conditionsfor revolutionary change to take place. However, such a two-goal policyrequires careful policy making, because policies designed for one goal mayin practice hamper the achievement of the other one. A way to combineboth objectives is to enhance the creation of new industrial trajectories, beit new technologies or new sectors, by means of building upon the existingcompetence base of firms, employers and employees in the region. Radicalinnovations often stem from the (quite unexpected) recombination of exist-ing technologies in entirely new ways (Levinthal, 1998). A famous examplehas been the rise of an environmental sector after the decline of the miningindustry in the Ruhr area. A broad engineering base in the Emilia Romagnaregion provided a fertile ground for the emergence of a broad range ofindustries such as ceramics, food packaging, robotics, car manufacturingand agricultural machinery during the post-war period (Boschma, 2004).Another example is the birth of the automobile industry in theCoventry–Birmingham area in England, which was partly determined bythe strong presence of the bicycle and carriage industry (Boschma andWenting, 2005). This policy captures the importance of creating ‘relatedvariety´ in a region, which broadens a region’s sectoral base, while foster-ing knowledge spillovers between the sectors (Frenken et al., 2005, 2006).

Another domain of policy, which is of crucial importance for urban andregional economic growth, is infrastructure provision. The growth ofagglomerations is limited by the capacity and quality of its infrastructure

16 Introduction

networks. For this reason, successful regional policy always requiresa complementary transportation infrastructure policy. Again, SophiaAntipolis serves as a successful example (Quéré, Chapter 3), whileCambridge suffered precisely from a mismatch between its economicdevelopment and infrastructure provision (Garnsey and Heffernan,Chapter 2). Adopting an evolutionary approach to transportation plan-ning in the agglomeration of Amsterdam, Bertolini (Chapter 13) attemptsto derive some general guidelines for planning. Given the inherent andirreducible uncertainty about the future regional development and land-use claims, urban transportation systems should be capable of resilience,that is, still function properly in the face of change. At the same time, ifnecessary, the system must also be responsive to change, that is, it must beadaptable. In transport systems, resilience is best shown by the networkmorphology and multi-modality, while adaptability is foremost a propertyof the policy system. The link between the two is important: in the case ofAmsterdam, the resilience of the transport network morphology has beena condition for the adaptability of land use and mobility managementpolicies, because it allowed a choice at all times between substantiallydifferent policy courses.

4. DISCUSSION

Using the micro–meso–macro scheme in Figure 1.1 as a framework, wehave discussed various applications of evolutionary economics in the fieldof economic geography. The common denominator in these approaches isto view spatial structures as the outcome of historical processes, and asconditioning but not fully determining economic behaviour. The explicithistorical nature of evolutionary analysis, however, poses demandingrequirements for empirical research. One needs to collect time-series dataof evolving populations, be it from technologies, sectors, networks, cities orregions, and to apply appropriate methodologies to analyse the data col-lected. The contributions by Klepper (Chapter 4), Jacob and Los (Chapter6) and Essletzbichler (Chapter 10) are fine examples of the use of econo-metric techniques applied to time-series data. However, other methodol-ogies are also available to fruitfully apply evolutionary economics. Forexample, case-study research, combining written and oral sources, canprovide an understanding of long-term planning processes (Bertolini,Chapter 13) and the multi-faceted process of regional development(Garnsey and Heffernan, Chapter 2; Quéré, Chapter 3). Static analysis,although dealing with snapshots of an otherwise evolving process, can alsobe approached from an evolutionary perspective, for example, by deriving

Introduction 17

hypotheses on expected inequalities in network positions (Giuliani,Chapter 8) or rates of technology adoption (Bonaccorsi et al., Chapter 12).Nevertheless, such phenomena could be understood better if time-seriesdata were available.

Apart from data limitations and methodological challenges ahead, thereare still a number of conceptual weaknesses that hamper the application ofevolutionary economics to economic geography: for example, the conceptof routines still needs to be refined (Becker, 2004), and their role in thedevelopment of multi-locational organisations is still quite unclear (Stam,2003, 2006); and the evolutionary theory of the firm has little to say aboutmultinational organisations, exceptions aside (Kogut and Zander, 1993;Cantwell and Iammarino, 2003). Another key concept in evolutionary eco-nomics is path dependence. Yet, its fruitful application in economic geog-raphy is still surrounded by a number of unsolved issues (Martin andSunley, 2006). Finally, as Breschi and Lissoni (2001) have argued at length,the concept of knowledge spillovers is, both conceptually and empirically,still ill-defined. Despite the growing number of studies on knowledgespillovers, the mechanisms underlying such spillovers are still poorly under-stood as well as to what extent these mechanisms are sector and/or regionspecific. Furthermore, the importance of knowledge spillovers may bespecific for the geographical distance over which they occur. The moreimportant information flows typically stem from more distant locations, ageographical principle that might reflect the strength of weak ties(Granovetter, 1973). However, research that takes into account globalspillovers is still scarce (Jaffe and Trajtenberg, 1999). In this light, the con-tributions by Jacob and Los (Chapter 6) and Maggioni and Uberti(Chapter 11) are especially important.

The ‘big question’ regarding the unequal distribution of wealth amongnations needs to be addressed more often and more systematically. An evo-lutionary economic geography may provide a new understanding ofcore–periphery patterns at different spatial scales as evolutionary outcomesof path-dependent dynamics. Such an approach would combine theSchumpeterian analysis of structure change with the spatial process ofagglomeration and global networking. However, evolutionary growththeory (as does growth theory more generally) still lacks an explicit spatialstructure. A challenge ahead is to transform evolutionary growth theoryinto a theory explaining the evolution of uneven distribution of economicactivities in space.

In all, recent research, including the chapters in this volume, has shownthe value added of an evolutionary approach in economic geography. Anevolutionary economic geography aims to improve our theoretical andempirical understanding of the economy as an evolutionary process that

18 Introduction

unfolds in space and time. Starting from the seminal contribution by Nelsonand Winter (1982) and its theoretical elaborations in subsequent works(Dosi et al., 1988; Dopfer, 2005), a number of frameworks are being devel-oped that specifically deal with geographical issues, including locationtheory and entrepreneurship, the spatial evolution of sectors, the geogra-phy of social networks, the evolution of spatial systems, and urban andregional planning. Methodologically, a variety of approaches are beingpursued ranging from case-study research and social network analysis toduration models and spatial econometrics. Theoretically coherent andmethodologically open, an evolutionary perspective is helpful in under-standing the specific histories of firms and regions using a framework thatis less restrictive than the neoclassical paradigm, yet more generally applic-able than the institutionalist approach. It is time to take geography seri-ously in applied evolutionary economics.

BIBLIOGRAPHY

Abernathy W.J. and Utterback, J. (1978), ‘Patterns of industrial innovation’,Technology Review, 50: 41–7.

Alderson, A.S. and Beckfield, J. (2004), ‘Power and position in the world citysystem’, American Journal of Sociology, 109: 811–51.

Amin, A. and Thrift, N. (2000), ‘What kind of economic theory for what kind ofeconomic geography?’, Antipode, 32(1): 4–9.

Andersson, C., Frenken, K. and Hellervik, A. (2006), ‘A complex network approachto urban growth’, Environment and Planning A, 38: 1941–64.

Andersson, C., Hellervik, A., Lindgren, K., Hagson, A. and Tornberg, J. (2003),‘Urban economy as a scale-free network’, Physical Review E, 68(3): 036124 Part 2.

Antonelli, C. (2000), ‘Collective knowledge communication and innovation: the evi-dence of technological districts’, Regional Studies, 34(6): 535–47.

Arthur, W.B. (1994), Increasing Returns and Path Dependence in the Economy, AnnArbor, MI: University of Michigan Press.

Aydalot, P. and Keeble, D. (eds) (1988), High Technology Industry and InnovativeEnvironments: The European Experience, London and New York: Routledge.

Barabasi, A.L. and Albert, R. (1999), ‘Emergence of scaling in random networks’,Science, 286(5439): 509–12.

Barrat, A., Barthelemy, M. and Vespignani, A. (2005), ‘The effects of spatial con-straints on the evolution of weighted complex networks’, Journal of StatisticalMechanics, Art. No. P05003.

Becker, M.C. (2004), ‘Organizational routines. A review of the literature’, Industrialand Corporate Change, 13(4): 643–77.

Boschma, R.A. (1997), ‘New industries and windows of locational opportunity. Along-term analysis of Belgium’, Erdkunde, 51: 12–22.

Boschma, R.A. (2004), ‘Competitiveness of regions from an evolutionary perspec-tive’, Regional Studies, 38(9): 1001–14.

Boschma, R.A. (2005), ‘Rethinking regional innovation policy. The making andbreaking of regional history’, in G. Fuchs and P. Shapira (eds), Rethinking

Introduction 19

Regional Innovation and Change: Path Dependency or Regional Breakthrough?,Dordrecht: Springer, pp. 249–71.

Boschma, R.A. and Frenken, K. (2003), ‘Evolutionary economics and industrylocation’, Review of Regional Research, 23: 183–200.

Boschma, R.A. and Frenken, K. (2006), ‘Why is economic geography not an evo-lutionary science? Towards an evolutionary economic geography’, Journal ofEconomic Geography, 6(3): 273–302.

Boschma, R.A. and Lambooy, J.G. (1999), ‘Evolutionary economics and economicgeography’, Journal of Evolutionary Economics, 9: 411–29.

Boschma, R.A. and Ter Wal, A.L.J. (2006), ‘Knowledge networks and innovativeperformance in an industrial district. The case of a footwear district in the southof Italy’, Industry and Innovation, forthcoming.

Boschma, R.A. and Wenting, R. (2005), ‘The spatial evolution of the British auto-mobile industry’, Papers in Evolutionary Economic Geography 05.04, http://econ.geog.uu.nl.

Bottazzi, G., Dosi, G., Fagiolo, G. and Secchi, A. (2004), ‘Sectoral and geo-graphical specificities in the spatial structure of economic activities’, Laboratoryof Economics and Management (LEM) Working Paper 2004–21, www.lem.sssup.it/.

Brakman, S., Garretsen, H. and Marrewijk, C. van (2001), An Introduction toGeographical Economics, Cambridge: Cambridge University Press.

Brenner, T. (2004), Local Industrial Clusters. Existence, Emergence and Evolution,London and New York: Routledge.

Breschi, S. and Lissoni, F. (2001), ‘Knowledge spillovers and localinnovation systems: a critical survey’, Industrial and Corporate Change, 10(4):975–1005.

Breschi, S. and Lissoni, F. (2003), ‘Mobility and social networks: localised knowl-edge spillovers revisited’, Centre of Research on Innovation andInternationalization (CESPRI) Working Paper 142, www.cespri.unibocconi.it/.

Burt, R.S. (1982), Toward a Structural Theory of Action: Network Models of SocialStructure, Perception, and Action, New York: Academic Press.

Caniëls, M. (2000), Knowledge Spillovers and Economic Growth. Regional GrowthDifferentials across Europe, Cheltenham, UK and Northampton, MA, USA:Edward Elgar.

Cantner, U. Dreßler, K. and Krüger, J.J. (2005), ‘Knowledge compensation inthe German automobile industry’, Jenaer Schriften zur Wirtschaftswissenschaft,11/ 2005.

Cantwell, J.A. and Iammarino, S. (2003), Multinational Corporations and EuropeanRegional Systems of Innovation, London: Routledge.

Carrington, P.J., Scott, J. and Wasserman, S. (eds) (2005), Models and Methods inSocial Network Analysis, Cambridge: Cambridge University Press.

Carroll, G.R. and Hannan, M.T. (2000), The Demography of Corporations andIndustries, Princeton, NJ: Princeton University Press.

Castells, M. (1996), The Rise of the Network Society, Oxford: Blackwell.Chang, H.-J. (2003), ‘Kicking away the ladder: infant industry promotion in his-

torical perspective’, Oxford Development Studies, 31(1): 21–32.Consoli, D. (2005), ‘The dynamics of technological change in UK retail banking

services. An evolutionary perspective’, Research Policy, 34: 461–80.Cooke, P., Uranga, M.G. and Extebarria, G. (1998), ‘Regional innovation systems:

an evolutionary perspective’, Environment and Planning A, 30: 1563–84.

20 Introduction

Cooper, A.C. and Dunkelberg, W.C. (1987), ‘Entrepreneurial research: old ques-tions, new answers and methodological issues’, American Journal of SmallBusiness, 11: 11–23.

Cowan, R., Cowan, W. and Swann, G.M.P. (1997), ‘A model of demand with inter-actions among consumers’, International Journal of Industrial Organization,15(6), 711–32.

De Jong, M.W. (1987), New Economic Activities and Regional Dynamics,Netherlands Geographical Studies no. 38, Amsterdam: Royal NetherlandsGeographical Society.

Dopfer, K. (ed.) (2005), Evolutionary Principles of Economics, Cambridge:Cambridge University Press.

Dosi, G. (1997), ‘Opportunities, incentives and the collective patterns of techno-logical change’, Economic Journal, 107(444): 1530–47.

Dosi, G., Freeman, C., Nelson, R., Silverberg, G. and Soete, L. (eds) (1988),Technical Change and Economic Theory, London: Pinter.

Essletzbichler, J. and Rigby, D.L. (2005), ‘Competition, variety and the geographyof technology evolution’, Tijdschrift voor Economische en Sociale Geografie,96(1): 48–62.

Foray, D. (1997), ‘The dynamic implications of increasing returns: Technologicalchange and path dependent inefficiency’, International Journal of IndustrialOrganization, 15(6): 733–52.

Freeman, C. and Perez, C. (1988), ‘Structural crisis of adjustment, business cyclesand investment behaviour’, in Dosi et al. (eds), pp. 38–66.

Frenken, K., van Oort, F.G., Verburg, T. and Boschma, R.A. (2005), ‘Variety andregional economic growth in the Netherlands’, Papers in Evolutionary EconomicGeography 05.02, http://econ.geog.uu.nl.

Frenken, K., van Oort, F.G. and Verburg, T. (2006), ‘Related variety, unrelatedvariety and regional economic growth’, Regional Studies, forthcoming.

Fujita, M., Krugman, P. and Venables, A.J. (1999), The Spatial Economy. Cities,Regions and International Trade, Cambridge, MA: MIT Press.

Garnsey, E. and Longhi, C. (2004), ‘Complex processes and innovative places: theevolution of high tech Cambridge and Sophia-Antipolis’, International Journalof Technology Management, 28(3–6): 336–55.

Gertler, M.S. (1995), ‘ “Being there”: proximity, organization, and culture in thedevelopment and adoption of advanced manufacturing technologies’, EconomicGeography, 71: 1–26.

Giuliani, E. (2005), ‘The structure of cluster knowledge networks: uneven andselective, not pervasive and collective’, Danish Research Unit for IndustrialDynamics (DRUID) Working Paper 2005-11, www.druid.dk.

Grabher, G. (1993), ‘The weakness of strong ties: the lock-in of regional develop-ment in the Ruhr area’, in G. Grabher (ed.), The Embedded Firm: On the Socio-Economics of Interfirm Relations, London and New York: Routledge, pp. 255–78.

Granovetter, M. (1973), ‘The strength of weak ties’, American Journal of Sociology,78(6): 1360–80.

Guimerà, R. and Amaral, L.A.N. (2004), ‘Modelling the world-wide airportnetwork’, European Physical Journal B, 38(2): 381–5.

Hagedoorn, J. (1993), ‘Understanding the rationale of strategic technology’,Strategic Management Journal, 14(5): 371–85.

Hall, P.G. and Markusen, A.R. (eds) (1985), Silicon Landscapes, Boston, MA:Allen & Unwin.

Introduction 21

Hannan, M.T., Carroll, G.R., Dundon, E.A. and Torres, J.C. (1995), ‘Organizationalevolution in a multinational context: entries of automobile manufacturers inBelgium, Britain, France, Germany, and Italy’, American Sociological Review,60(4): 509–28.

Hodgson, G.M. (1998), ‘The approach of institutional economics’, Journal ofEconomic Literature, 36(1): 166–92.

Hohenberg, P.M. and Lees, L.H. (1995), The Making of Urban Europe 1000–1994,Cambridge, MA: Harvard University Press.

Hoover, E.M. and Vernon, R. (1959), Anatomy of a Metropolis: The ChangingDistribution of People and Jobs in New York Metropolitan Region, Cambridge,MA: Harvard University Press.

Jacobs, J. (1969), The Economy of Cities, New York: Vintage Books.Jaffe, A.B. and Trajtenberg, M. (1999), ‘International knowledge flows: evidence

from patent citations’, Economics of Innovation and New Technology, 8: 105–36.Jaffe, A.B., Trajtenberg, M. and Henderson, R. (1993), ‘Geographic localization of

knowledge spillovers as evidenced by patent citations’, Quarterly Journal ofEconomics, 108(3): 577–98.

Keeble D., Lawson, C., Moore, B. and Wilkinson, F. (1999), ‘Collective learningprocesses, networking and “institutional thickness” in the Cambridge region’,Regional Studies, 33(4): 319–32.

Klepper, S. (2001), ‘The evolution of the U.S. automobile industry and Detroit asits capital’, Paper presented at 9th Congress of the International Joseph A.Schumpeter Society, Gainesville, FL, March.

Klepper, S. (2002), ‘The capabilities of new firms and the evolution of the U.S. auto-mobile industry’, Industrial and Corporate Change, 11(4): 645–66.

Kogut, B. and Zander, U. (1993), ‘Knowledge of the firm and the evolutionarytheory of the multinational corporation’, Journal of International BusinessStudies, 24: 625–46.

Krugman, P.R. (1991), ‘Increasing returns and economic geography’, Journal ofPolitical Economy, 99(3): 483–99.

Lambooy, J.G. and Boschma, R.A. (2001), ‘Evolutionary economics and regionalpolicy’, Annals of Regional Science, 35: 113–31.

Levinthal, D.A. (1998), ‘The slow pace of rapid technological change: Gradualismand punctuation in technological change’, Industrial and Corporate Change, 7:217–47.

Maggioni, M.A. (2002), Clustering Dynamics and the Location of High-Tech-Firms,Springer: Heidelberg.

Martin, R. (1999), ‘The new “geographical turn” in economics: some criticalreflections’, Cambridge Journal of Economics, 23(1): 65–91.

Martin, R. (2000), ‘Institutional approaches in economic geography’, in E.Sheppard and T.J. Barnes (eds), A Companion to Economic Geography, Oxfordand Malden, MA: Blackwell, pp. 77–94.

Martin, R. and Sunley, P. (2006), ‘Path dependence and regional economic evolution’,Papers in Evolutionary Economic Geography 06.06, http://econ. geog. uu.nl.

Maskell, P. (2001), ‘The firm in economic geography’, Economic Geography, 77(4):329–44.

Maskell, P. and Malmberg, A. (1999), ‘Localised learning and industrial competi-tiveness’, Cambridge Journal of Economics, 23(2): 167–86.

Matsumoto, H. (2004), ‘International urban systems and air passenger and cargoflows: some calculations’, Journal of Air Transport Management, 10(4): 239–47.

22 Introduction

Metcalfe, J.S. (1995), ‘Technology systems and technology policy in an evolution-ary framework’, Cambridge Journal of Economics, 19: 25–46.

Moss, M.L. and Townsend, A.M. (2000), ‘The Internet backbone and the Americanmetropolis’, The Information Society, 16: 35–47.

Murmann, J.P. (2003), Knowledge and Competitive Advantage: The Co-evolution ofFirms, Technology, and National Institutions, Cambridge: Cambridge UniversityPress.

Nelson, R.R. (1995), ‘Co-evolution of industry structure, technology and support-ing institutions, and the making of comparative advantage’, International Journalof the Economics of Business, 2(2): 171–84.

Nelson, R.R. (1999), ‘The sources of industrial leadership: a perspective on indus-trial policy’, De Economist, 147: 1–18.

Nelson, R.R. and Winter, S.G. (1982), An Evolutionary Theory of Economic Change,Cambridge, MA and London: Belknap Press.

Overman, H.G. (2004), ‘Can we learn anything from economic geography proper?’,Journal of Economic Geography, 4: 501–16.

Pasinetti, L.L. (1981), Structural Change and Economic Growth, Cambridge:Cambridge University Press.

Pasinetti, L.L. (1993), Structural Economic Dynamics, Cambridge: CambridgeUniversity Press.

Perez, C. and Soete, L. (1988), ‘Catching up in technology: entry barriers andwindows of opportunity’, in Dosi et al. (eds), pp. 458–77.

Phelps, N.A. (2004), ‘Clusters, dispersion and the spaces in between. For an eco-nomic geography of the banal’, Urban Studies, 41(5–6): 971–89.

Plouraboue, F., Steyer, A. and Zimmerman, J.-B. (1998), ‘Learning induced criti-cality in consumers’ adoption pattern: a neural network approach’, Economics ofInnovation and New Technology, 6: 73–90.

Porter, M.E. (2000), ‘Location, competition, and economic development: localclusters in a global economy’, Economic Development Quarterly, 14: 15–34.

Powell, W.W., Koput, K.W. and SmithDoerr, L. (1996), ‘Interorganizational col-laboration and the locus of innovation: networks of learning in biotechnology’,Administrative Science Quarterly, 41(1): 116–45.

Pred, A. (1977), City-Systems in Advanced Economies, London: Hutchinson.Pumain, D. and Moriconi-Ebrard, F. (1997), ‘City size distributions and metropo-

lisation’, Geojournal, 43(4): 307–14.Rigby, D.L. and Essletzbichler, J. (1997), ‘Evolution, process variety, and regional

trajectories of technological change in US manufacturing’, Economic Geography,73(3): 269–84.

Saviotti, P.P. and Pyka, A. (2004), ‘Economic development by the creation of newsectors’, Journal of Evolutionary Economics, 14(1): 1–35.

Saxenian, A. (1994), Regional Advantage, Cambridge, MA: Harvard UniversityPress.

Solomon, S., Gerard, W., de Arcangelis, L., Jan, N. and Stauffer, D. (2000), ‘Socialpercolation models’, Physica A, 277: 239–47.

Solow, R.M. (1957), ‘Technical change and the aggregate production function’,Review of Economics and Statistics, 39: 312–20.

Stam, E. (2003), ‘Why butterflies don’t leave: locational evolution of evolving enter-prises’, Dissertation, Utrecht University.

Stam, E. (2006), ‘A process model of locational change: implications for an evo-lutionary economic geography, in A. Pyka and H. Hanusch (eds), Applied

Introduction 23

Evolutionary Economics and the Knowledge-Based Economy, Cheltenham UKand Northampton, MA, USA: Edward Elgar, pp. 143–65.

Storper, M. (1997), The Regional World: Territorial Development in a GlobalEconomy, London: Guilford Press.

Stuart, T. and Sorenson, O. (2003), ‘The geography of opportunity: spatial hetero-geneity in founding rates and the performance of biotechnology firms’, ResearchPolicy, 32(2): 229–53.

Swann, P. and Prevezer, M. (1996), ‘A comparison of the dynamics of industrialclustering in computing and biotechnology’, Research Policy, 25: 1139–57.

Taylor, P.J. (2001), ‘Specification of the world city network’, Geographical Analysis,33: 181–94.

Teece, D., Pisano, G. and Shuen, A. (1997), ‘Dynamic capabilities and strategicmanagement’, Strategic Management Journal, 18(7): 509–33.

Tinbergen, J. (1962), Shaping the World Economy: Suggestions for an InternationalEconomic Policy, New York: 20th Century Fund.

Uzzi, B. (1996), ‘The sources and consequences of embeddedness for the economicperformance of organizations: the network effect’, American Sociological Review,61(4): 674–98.

Vernon, R. (1969), ‘International investment and international trade in the productlife-cycle’, Quarterly Journal of Economics, 80: 190–207.

van Wissen, L. (2004), ‘A spatial interpretation of the density dependence model inindustrial demography’, Small Business Economics, 22(3–4): 253–64.

Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods andApplications, Cambridge: Cambridge University Press.

Watts, D.J. and Strogatz, S.H. (1998), ‘Collective dynamics of “small-world” net-works’, Nature, 393: 440–42.

Werker, C. and Athreye, S. (2004), ‘Marshall’s disciples: knowledge and innovationdriving regional economic development and growth’, Journal of EvolutionaryEconomics, 14: 505–23.

Whitley, R. (1992), Business Systems in East Asia: Firms, Markets and Societies,London: Sage.

24 Introduction

PART I

Entrepreneurship

2. The Cambridge high-tech cluster:an evolutionary perspectiveElizabeth Garnsey and Paul Heffernan

1. INTRODUCTION

New knowledge-based firms emerge and grow around centres of learningand research. A process akin to ecological succession occurs as resources inthe local science base are converted into and attract business activity, givingrise to a richer, more diverse economic habitat. To gain a better understand-ing of these developments we need to examine how processes of changeoperate over time. Cambridge provides an exemplar of endogenous form-ation of a high-tech cluster through spin-off, agglomeration and institutionaladaptation. The importance of the strong science base at Cambridge is uni-versally acknowledged, but the evolutionary micro processes through whichits influence was exerted require further elucidation. Positive externalitiesproviding incentives to firms to cluster in an area are not necessarily presentat the time of the emergence of a new cluster of activity. Explanations interms of measurable externalities beg the question of the evolutionaryprocesses which gave rise to them. In other high-tech centres (Simmie et al.,2004), spillover effects have resulted from government spending on infra-structure, from large company investments, from metropolitan structuresand defence spending on information technology (IT). These influences wereabsent in the case of Cambridge.

After identifying the conceptual building blocks, we then examine indicesof the growth of high-tech clusters in the Cambridge area. Underlying theseaggregate trends are self-reinforcing mechanisms involving business spin-outs and networks of knowledge diffusion. Case studies illustrating theseprocesses, in the clustering of scientific instrumentation, information andcommunication technology (ICT), technical design consultancies andbiotechnology, further develop the thesis. Global linkages of Cambridgefirms and the local process of institutional transformation in the region,resulting from collective action of local firms, reveal how global and localconnections interact.

27

2. CLUSTERING AS AN ENDOGENOUS PROCESS

While traditional approaches in economic geography explained the loca-tion of industry in terms of exogenous factors such as demand or inter-national trade effects, more recently endogenous factors have beenidentified as drivers of local clustering. In particular, spillover effects aresaid to shape location decisions where firms derive benefit from local invest-ments, the cost of which they do not themselves bear (Breschi and Lissoni,2001). Value chain considerations shaping location decisions of firms aresometimes subsumed under spillover and sometimes treated as an alterna-tive externality affecting location decisions. From an evolutionary perspec-tive, the most striking feature of Cambridge as a high-tech centre is the wayin which participants transformed the area, creating the very externalitiesthat are assumed to be attributes of the area in leading accounts.

Although local firms undoubtedly benefited from the university infra-structure, particularly ventures incubated within university departments,the concept of knowledge spillovers does not fully explain spin-out activ-ity. University scientists can create their own firms to capture returns fromtheir intellectual capital; they do not have to allow commercial firms to reapthe gains from their intellectual capital, as is implied by the free-rider con-notations of spillover (Zucker et al., 1998). Patents offer means of attract-ing capital through the prospects they offer for appropriating returns.However, the formal and informal relationships that these scientistsdevelop with commercial organisations extend well beyond the market re-lations identified by Zucker et al. Through the operation of their networks,academic entrepreneurs have been among those creating a more favourablelocal business environment.

The other endogenous determinant of clustering recognised in the liter-ature, relates to local supply chain benefits. This approach sees clusters asgeographic concentrations of interconnected companies and their special-ised suppliers, service providers and firms in related industries, togetherwith associated institutions (Porter, 1990). Supply chain benefits stem fromco-locating with suppliers and customers, with a consequent lowering oftransport and coordination costs, while co-location may also provideinformation about new input technologies and potential products(Krugman, 1991). Co-locating with customers provides information abouttheir market needs, and may offer economies of scale to suppliers. Butsupply chain effects are not spillovers in that they do not offer free-ridergains (Zucker et al., 1998). Firms must invest time and effort in internalis-ing supply chain ‘externalities’ through deliberate measures in order tobenefit from labour availability, to reach price agreements, to establishpartnerships or make procurement arrangements. Firms in competition

28 Entrepreneurship

with each other will only co-locate if they can achieve such benefits andovercome any disadvantages in having nearby competitors. Informationon supply chain arrangements that work well is diffused over time on acumulative basis.

The customers of Cambridge high-tech companies are internationalrather than local, and relatively few of them have local suppliers in theirspecialised production chain. However, they do benefit from wider valuechain externalities (Krugman, 1991). We shall see that high-tech firms inthe Cambridge area currently make use of value chain complements or sub-stitutes for the firms’ internal activities, in particular by outsourcing to locallegal and business services. These have been attracted to the area by thepresence of high-tech firms, illustrating the endogenous dimension of thetransformation of the area. For knowledge-intensive firms, access tospecialised labour is a key feature of local value chain advantages. However,it is not only the supply of new graduates that has become an asset of thelocality, but a labour market of experienced specialised professionals,another outcome of cumulative processes.

Explanations of local clustering in terms of positive externalities havelargely adopted cross-sectional methodologies to identify the determi-nants of clustering (Breschi and Lissoni, 2001). This method is unable toexplain why clusters emerged in some places and not in others before suchtime as these externalities became local attributes. In contrast, Arthur(1994) developed two evolutionary models to explain spatial clustering,showing how chance occurrences may lead to a set of cumulative factorsfavouring the growth of a specific location, a finding anticipated byMaruyama’s work on positive feedback (Maruyama, 1963). The first is adynamic model in which firms sequentially choose locations on the basisof agglomeration economies, including skilled labour markets, specialisedsuppliers and knowledge spillovers. Arthur showed that local concen-trations of firms in a given industry are neither fully determined by geog-raphy – for example, local resource endowments, transport potential andfirm requirements – nor entirely a matter of historical chance, as assumedin competing theories of industrial location. Instead, both geographicattractiveness and ‘accidental historical order of choice’ come into theequation, with various outcomes following from upper limits to agglom-eration benefits.

A second model by Arthur (1994) is based on industry formationthrough spin-off (or spin-out1). Clustering can be explained by a sequentialprocess in which the probability of a region producing spin-offs at timet � 1 is dependent on the relative number of firms that located at time t. Bydrawing randomly, at each time t, one firm that produces a spin-off, anevolving spatial distribution of firms in an industry is simulated. Typically,

The Cambridge high-tech cluster 29

the resulting spatial distribution is highly skewed, because some regionswill by chance have a relatively high number of spin-offs early on and,subsequently, these will produce further spin-offs. The outcome of the spin-off model is dependent on the assumption that spin-offs locate in the sameregion as the parent company, since without this assumption one wouldobtain a random location model, but the assumption that spin-offs locatenear the parent is empirically robust (Cooper and Dunkelberg, 1987;Klepper, 2001). The departure of members of one firm to found another,on a friendly or hostile basis, was recognised in the 1970s by Cooper in theSan Francisco Bay area where spin-outs from Shockley and Fairchild werecelebrated (Cooper, 1971) as well as in a study of clustering of the US auto-mobile industry in Detroit (Klepper, 2002).

Arthur’s approach is illuminating in exploring the interplay of chanceand cumulative process manifest in path dependence (Boschma andFrenken, 2003). In particular it highlights the mechanisms of spin-offs andof agglomeration economies which are likely to operate simultaneously.The contribution of these models lies in demonstrating the principle thateven for regions equivalent in terms of institutions and endowments, theincidence of specialist clusters will be highly uneven because it occursthrough chance events that set in motion a self-amplifying process in whichsuccess breeds success.

Arthur’s abstract models are designed to exclude a variety of othercausal factors influencing the emergence of local activity, among which areculture, institutions and endowments. In the following analysis, we observethat the endowment represented by Cambridge University provided criti-cal conditions for the emergence of local industry through clustering. Thisendowment can be viewed either as systematic or as a chance initial con-dition, depending on whether the geography of the university’s location orthe historical accident of its location is emphasised. Exogenous factorswere also at work in the form of international demand for high-techproducts and services to which Cambridge companies responded. Butas Arthur’s model predicts, endogenous spin-off and agglomerationeconomies have played the major role in the emergence of high-tech indus-try in Cambridge.

In this chapter, clustering is related above all to the inter-generationalspin-out process in Cambridge. The firms in question are connectedlocally by mobile people and knowledge to a greater extent than by supplyrelations, and operate in value chains that have global reach. We showthat the cumulative, path-dependent nature of local value creation andappropriation processes is the source of their locational impact, to anextent ignored by the static methodologies of literature that neglectevolutionary processes.

30 Entrepreneurship

3. THE EMERGENCE OF THE CAMBRIDGE HIGH-TECH CLUSTER

Cambridgeshire is located northeast of London; the county extends 1300square miles and had a population of about 700 000 at the turn of the mil-lennium. Local firms are active in IT, advanced electronics and engineering,materials, instrumentation, biotechnology and research services, developingthe emerging and early diffusing technologies characteristic of high-techsectors (Butchart, 1986). The development of high-tech industry in theCambridge region has been rather exceptional, also labelled the ‘Cambridgephenomenon’ (Garnsey and Cannon-Brookes, 1993; Heffernan and Garnsey,2002; Keeble, 1989; Keeble et al., 1999). From about 50 firms in the mid-1960s,by 1985 there were over 300 firms and 16 000 jobs in the Cambridge high-techsector. A consultants’ report had identified extensive spin-out activity (Segalet al., 1985). By the end of the century there were more than 1200 technology-related firms (depending on definitions and area); these firms employed36 000 people, approximately 10 per cent of the total Cambridgeshire work-force (Figure 2.12). By 2000, the combined turnover of technology-basedenterprises was over £3.5 billion sterling (Heffernan and Garnsey, 2002).3

The recession of the early 1990s reduced employment in the larger, olderfirms in electronics and instrumentation, but new firms and specialisationsarose, restoring technology-employment totals. Manufacturing declined as

The Cambridge high-tech cluster 31

Source: 1960–82 figures based on Garnsey and Cannon-Brookes (1993); 1984 and 1986figures interpolated; data since 1988 derived from Cambridgeshire County CouncilEmployment database, as described in Heffernan and Garnsey (2002), extended to 2004 andexcluding Peterborough. The data differ from those used by PACEC (2003), which examinedhigh-tech activity over the Greater Cambridge Partnership area, a wider area thanCambridgeshire.

Figure 2.1 High-tech firms in Cambridgeshire, 1960–2004

0

200

400

600

800

1000

1200

1400

1960

1964

1968

1972

1976

1980

1984

1988

1992

1996

2000

2004

Firms

a proportion of high-tech activity as service activities increased, as in theUK economy at large. Figure 2.2 shows the changing distribution of firmsand employment across a selection of high-tech sectors (Butchart, 1986).Software firms were the largest single group, reflecting both extensive ITexpertise and low barriers to entry. Instrumentation, and research anddevelopment (R&D) were also important sectors, the latter being thelargest employer. Biotechnology-related firms accounted for 15 per cent offirms and 29 per cent of employment by 2004. Absolute numbers ofCambridge high-tech manufacturing firms were on the increase until the2004 survey (in contrast with figures for the same sectors in the UK).

High-tech employment in the area in and around Cambridge4 fell byabout 5 per cent between 2002 and 2004, but the number of firms remainedunchanged in the face of the technology slump and collapse in market valu-ations that followed the boom of the 1990s.

The Cambridge phenomenon has grown mainly through the entry ofnew enterprises, most of these remaining small. The mean size of firms in2004 was 33, less than that in 1988 (38), though median size increasedslightly from 6.5 to 7. It must be noted, however, that a skewed firm-size dis-tribution is universal and applies over time as well as nationally and region-ally; only a small proportion of any cohort of firms grow to become majoremployers over time (Storey, 1994). It requires a large pool of start-ups togenerate a few major companies. Firms of small size dominate in high-techCambridge to a lesser extent than is found across the UK as a whole as seenwhen relevant sectors are matched (Heffernan and Garnsey, 2002). Thecontinued expansion of high-tech activity depended, however, on the entryof new firms exceeding exit rates. Between 1989 and 2004 the mean numberof new high-tech firms registered by the local authorities in Cambridgeeach year was 67, but about 47 firms closed or moved away (Figure 2.3)

Local capability to sustain new firms is shown by survival rates. Figure 2.4shows that survival rates for Cambridge technology-based firms were con-sistently higher than the regional and national averages, and compared wellwith survival rates reported in other studies (see Kirchoff, 1994; Slatter, 1992;Storey, 1994). This graph suggests that Cambridge firms had superior capa-bilities compared to the UK average.

4. CASE STUDIES OF CAMBRIDGE CLUSTERS

Scientific Instrumentation

The earliest Cambridge high-tech cluster was of scientific instrumentationfirms, a response to the rise of the science departments of Cambridge

32 Entrepreneurship

33

Sou

rce:

Cal

cula

ted

from

Cou

nty

Cou

ncil

reco

rds.

Fig

ure

2.2

Cha

ngin

g se

ctor

al g

row

th p

atte

rns

amon

g C

ambr

idge

hig

h-te

ch fi

rms

Num

ber

of fi

rms

050100

150

200

250

300

350

1988

1990

1992

1994

1996

1998

2000

2002

2004

Em

ploy

men

t

0

1 00

0

2 00

0

3 00

0

4 00

0

5 00

0

6 00

0

7 00

0

8 00

0

9 00

0

10 0

00

1988

1990

1992

1994

1996

1998

2000

2002

2004

Tel

ecom

mun

icat

ions

Ele

ctric

al e

ngin

eerin

g

IT h

ardw

are

IT s

ervi

ces

IT s

oftw

are

Res

earc

h an

dde

velo

pmen

t

Bio

tech

Inst

rum

enta

tion

University in the nineteenth century. By the 1980s this cluster includedCambridge Instruments, UniCam and other Pye (Philips) sites employingseveral thousand in the area. Large numbers of managers and techniciansin more recently founded firms gained experience in these companies.However, venerable firms such as Cambridge Scientific Instruments,founded by Charles Darwin’s son Horace, like many other UK manufac-turing firms, proved unable to adapt to rapid technological change andinternational competition (Koepp, 2002). They were acquired and down-sized in a series of retrenchments and new, much smaller firms were formedby former managers and technicians. By the late 1990s all the older instru-mentation companies had been acquired and de-merged. The instrumen-tation sector now includes about 100 firms in the Cambridge region withvery diverse products. Some newer ones have responded to internationalopportunities by addressing expanding markets for automating biotech-nology research. The presence of biotech networks in the area has been afactor in the recognition of lab automation opportunities and in access torelevant competence in the life sciences. But overall, the capability of thearea in instrumentation is less than could be expected from the concentra-tion of science laboratories.5 There was a failure of established firms tomaintain a leading role in the area such as Hewlett Packard achieved in

34 Entrepreneurship

Source: Calculated from County Council records.

Figure 2.3 Firm turnover (turbulence) in the Cambridge high-tech sector

–150

–100

–50

0

50

100

150

200

250

1990 1992 1994 1996 1998 2000 2002 2004

New firms

Lost firms

Net effect

Silicon Valley (Saxenian, 1994) or to be renewed through government–uni-versity manufacturing modernisation initiatives such as were undertakenelsewhere in the US (Best and Forrant, 2000).

ICT

The information technology sector has shown the highest rates of newfirm formation, reflecting demand for IT products and services and lowbarriers to entry. There has been no official audit of companies foundedby members of the University of Cambridge. Only firms in which the uni-versity took out equity are registered, a very small proportion of the totaloriginating in the university. New evidence shows that over a hundredcompanies were founded by members of the university computer sciencedepartment, about 80 per cent located in the UK and about 50 per centin the Cambridge area.6 IT companies have also originated from otherdepartments, above all the engineering department, which has been thesource of a diverse set of new technologies and firms mainly located inthe area (Figure 2.5). As with most populations of new firms, only a smallproportion of departmental spin-outs have grown to above average size,

The Cambridge high-tech cluster 35

Note: ONS rates are based on value-added tax (VAT) registrations and deregistrationswhich excludes some small firms.

Source: National and regional data obtained from office of National Statistics (ONS).

Figure 2.4 Comparison of survival rates for cohorts of Cambridgehigh-tech firms (with regional and national averages)

40%

50%

60%

70%

80%

90%

100%

110%

0 1 2 3 4 5 6 7 8 9 10

Years after start-up

% F

irms

rem

aini

ng

1989/90 cohort

1991/92 cohort

1993/94 cohort

1995/96 cohort

1997/98 cohort

UK East 1995

UK 1995

but several firms founded by former students have become leading com-panies in their sector, including, for example, Autonomy.

Several hardware pioneers did not survive the maturation of the micro-computer industry. But Acorn Computers, founded in 1979, illustrates theinfluence that an originating firm can have on its progeny. The founders ofAcorn Computers, Hermann Hauser and Chris Curry, drew on expertisefrom the University Computer Lab to develop an innovative microcom-puter, the production of which was subcontracted outside the area. Acorn’sexplosive growth was stemmed by a sudden slump in consumer demand in1984 and overcommitment to suppliers. The company only survivedthrough acquisition by Olivetti. Acorn’s proprietary operating system cameunder competition from the new industry standard, MS-DOS. UnderOlivetti’s ownership, Acorn attempted to pioneer the Network Computerin an international strategic partnership that fell through, the NetworkComputer being a product before its time. Acorn was wound up in 1999 soas to realise for Olivetti the value of its shares in ARM, and to create thenew spin-out, Element-14. Acorn had funded the development of ARM’stechnology before spin-out and was its first customer. Acorn Computers

36 Entrepreneurship

Figure 2.5 Companies with founders from Cambridge Universityengineering departments

Failed

Acquired

Univ IP/ equity

Still in existancein 2002

2000

1990

1980

1970 Delcam 1968

HRM Software 1986

PCME 1991

Richard Marshall Ltd 1996

Spark Ltd 1998

Jumpleads 2000

Splashpower 2002

Shearline Precision Engineering 1973TopExpress 1979 Acq

Adder Technology 1984Synoptics 1985

Cambustion1987JMEC 1989

Asymptote 1989

Biorobotics 1992 Acq

Entropic 1995 AcqIonotec1995

Sintefex Audio Lab 1997

Transversal 1998

CAM 3D 1999

ChyGwyn 1999

Vulcan Machines 2000

Blue Technologies 2001

Cirocco 2001Alphamosaic 2001

Cool Analgesia 2001

Granta Design 1993

Softsound1995 Acq

CMIL 1998

Cambridge FlowSolutions 1999

CRISP 1999

Cambridge Semiconductors 2000

CEDAR Audio 1989

CCL 1960

Qudos 1986 failed Acq

Vivamer 2002 Chem Eng

Chem Eng

CambridgeUniversity

EngineeringDepartment

was a learning organisation for the whole area, providing experience tomany local entrepreneurs and managers. The range and depth of compe-tence developed at Acorn made it possible for former members to startlarge numbers of local spin-outs (Figure 2.6). A further 20 firms werefounded by employees leaving ARM to start a business.7

Technical Design Consultancies

A cluster unique to Cambridge is made up of technical design consultan-cies working with leading international clients. These have become reposi-tories of a wide range of competence, technical and managerial. Theyengage in prototype and small-scale production as well as advisory consul-tancy. The pioneer, Cambridge Consultants Ltd (CCL), founded by formermembers of the chemical engineering department in 1960, attempted toestablish a manufacturing unit in their early days. Lack of competence inmanufacturing led to a cash flow crisis and acquisition by A.D. Little.Thereafter the company specialised in technical services, a sector in whichfirms can minimise the sunk costs that often undermine UK ventures.Figure 2.7, below, shows the spin-outs from CCL. When employees wishedto use CCL’s intellectual property (IP) to develop a product, CCL helped

The Cambridge high-tech cluster 37

Figure 2.6 New firms started by founders and employees of AcornComputers

1985ABC, 1988

1990

1995

2000

Pre 1980

2005

Orbis1978

Ubisense (Ubiquitous Systems), 2002Level 5 Networks (Cambridge Internetworking), 2002

Cambridge Broadband, 2000

Adaptive Broadband, 1998

(Cambridge Network Ltd, 1998)

Amadeus Capital Partners, 1997

NetChannel Ltd, 1996

NetProducts Ltd, 1996

IPV (Telemedia Systems), 1995

Advanced Displays

Electronic Share Information Ltd, 1993

SynGenix, 1992Vocalis,

1992

EO Inc. (with AT&T)

E*Trade UK, 1998

IXI Ltd

Harlequin Ltd, 1986

Qudos Technology Ltd, 1985IQ Bio, 1981

Real VNC, 2002

Clearswift (Net – Tel Computer Systems),1982GIS, 1985

ARM, 1990

STNC, 1993ANT, 1993

Xemplar Education, 1996

nCipher, 1996

Element 14, 1999Commtag, 2000

Pogo Mobile Solutions, 2002

Icora Semiconductor, 2003

ATM 1993, VIRATA

ACORNFounders

them with seed capital, taking out equity in such ventures (Auton andBiddle, 2001). Former CCL employees later spun out other design consul-tancies, via another consultancy with a local office. One of these, TheTechnology Partnership (TTP), has been the source of several further spin-outs. TTP’s sales were under £20 million in 2001, but with its combinedspin-outs, sales were over £80 million.8

Industrial Ink Jet Printing

Largely spinning out of CCL is a cluster of Cambridge firms that haveachieved global expansion on the basis of a common set of technologies.Figure 2.7 depicts the genealogy of seven local industrial ink jet printingcompanies (Imaje is located in France and Willett in Corby). By around2000 the ink jet printing firms employed over 3000 directly and theircustom provided further jobs in PCB and precision engineering firms inthe regions. They were dominant in international markets for non-impactproduct identification, a smaller market than the major market for smallbusiness and desktop ink jet printing. Their revenues were estimated at

38 Entrepreneurship

Figure 2.7 Industrial inkjet printing spin-outs originating in Cambridge

Domino Printing Sciences

Cambridge Consultants

Limited

2000

1990

1980

1970

Imaje

Inca Digital Printers

XaarBiodot

Xennia

Elmjet

Linx

Willett

£500 million in 2000.9 The production chain is international; ink jet print-ing (IJP) firms source jewels from Switzerland, pumps from the US andprecision components from many other sources, but they also share somelocal suppliers. The firms do not formally collaborate, or regularly supplyeach other, but interviewees report frequent informal interaction and link-ages resulting from their common origins and the mobility of staff. IJPcustomer companies in the area have helped local PCB and precision engi-neering suppliers to upgrade their performance; these contractors werethen able to help other customers in the area to upgrade their productsand production process. Local IJP clients remain an important source ofcustom for local suppliers, but subassemblies came to be sourced inter-nationally as the industry matured. The role of leading firms in produc-tion networks is confirmed by the experience of Domino PrintingSciences, which grew to over 1000 employees and had a particular impacton suppliers in eastern England such as Hansatech in King’s Lynn. Theindustrial IJP firms continue to provide a local labour market skilled inrelevant competences. When ink jet technologies were adopted by newentrants developing advanced materials such as light emitting polymers(Cambridge Display Technology and Plastic Logic) they were able to hireprofessional staff with experience in the local IJP cluster, demonstratingthe role of job mobility in the diffusion of competence in the area.However, at technician and operator levels, IJP firms found it difficult torecruit skilled personnel at competitive rates, and it was partly for thisreason that Videojet (the acquirer of Elmjet) moved out of the Cambridgearea, while Xaar’s manufacturing function was relocated as a result of amerger.

Biotechnology

The biotech cluster clearly illustrates the interaction of endogenous andexogenous influences. There are about 110 biotech companies in theCambridge area, 40 or so indigenous spin-outs from university depart-ments and local research institutes, the others being local start-ups or exter-nal entrants. Because biotech has extensive applications, these companiesembody a wide range of competence in diverse life sciences, instrumen-tation, chemistry and computing. Twelve different university departmentswere the source of the biotech companies recognised by the university asspin-outs because they embody university IP and equity. The nautilus formof Figure 2.8 illustrates the increasing incidence of biotech spin-out activ-ity over time. Among the 42 firms shown in Figure 2.8, there were 20 in bio-pharm therapeutics and three in biopharm diagnostics, all of these engagedin R&D activities. Only one firm, Cambridge Antibody Technologies, was

The Cambridge high-tech cluster 39

40 Entrepreneurship

Notes:1. Celltech, 1980 15. Cambridge Drug 29. Cambridge Biotechnology,2. Affinity Chromatography, Discovery, 1997 2001

1987 16. Kudos, 1997 30. Akubio, 20013. Cantab Pharmaceuticals, 17. AdproTech, 1997 31. Purely Proteins, 2002

1989 18. Abcam, 1998 32. Genepta, 20024. Cambridge Antibody 19. Sense Proteomic, 33. Smart Holograms, 2002

Technology, 1990 1998 34. Daniolabs, 20025. Cambridge Sensors, 20. Paradigm 35. Blue Gnome, 2003

1991 Therapeutics, 36. Vivamer, 20036 BioRobotics, 1993 1998 37. Diagnostics for7. Microbial, 1994 21. Solexa, 1998 the Real World, 20038. Hexagen, 1996 22. Cambridge 38. Ionscope, 20039. Cambridge Microbial 39. Ampika, 2003

Combinatorial, Technologies, 1999 40. Cambridge Lab on Chip,1996 23. Clinical & 2003

10. Metris Therapeutics, Biomedical 41. Protein Logic, 20031996 Computing, 42. Novexin, 2004

11. Cambridge 1999Biotransforms, 24. De Novo 1997 Pharmaceuticals,

12. RiboTargets, 1997 199913. Biotica Technology, 25. Astex

1997 Technology, 199914. Cambridge 26. Diversys, 2000

Bioclinical, 27. Avidis, 20001997 28. Cool Analgesia, 2001

Source: Research Services, Cambridge University.

Figure 2.8 Biotech firms originating in 12 Cambridge Universitydepartments

1985

1990

1995

2000

1980

2005

1

2 34

56

8

9

10

11

12

1314

1516

171819202122

23

24

25

26

27

28

29

30

31

32

33

34

3536

37

40 41 423938

7

in integrated drug discovery and production. The remaining spin-outs weresupporting drug discovery through bioinformatics (10) and specialistservices, devices and supplies (8). Ownership changes have been consider-able, with over a quarter of the biotech spin-out ventures being acquired ormerged. Local biopharm ventures together with attracted firms aresufficiently numerous to provide a local labour market in commercial lifesciences. As the cluster has matured, the professional networks formed withcompanies elsewhere and around local career structures have becomeincreasingly significant.10

The strength of life sciences in the university and location of the MedicalResearch Council and other research institutes in the area ensured that asmarket applications for biotech emerged, Cambridge would be a leadingcentre. This was confirmed when a number of biotech incubators wereestablished in the late 1990s and the Human Genome project was sitedlocally. Spin-outs were attracted from big pharmaceuticals and from otheruniversities. Case histories reflect complex alliances and reconfigurations.For example, Enzymatics, originally financed by British Sugar, was closedwhen its funding was terminated, but its IP became the endowment of alocal spin-out, Chiroscience, which later merged with Celltech, sold to aBelgian pharmaceutical company in 2004.

5. INTERNATIONALISATION ANDINSTITUTIONAL ADAPTATION

The case studies all show that the emergence of high-tech clusters hasbeen largely an endogenous process driven by spin-outs and emergingagglomeration advantages. Two accompanying processes that have furthernurtured the clustering processes deserve further attention. First, theincreasing international linkages of Cambridge firms and the transfor-mation of institutions such that they have become more supportive for thefurther development of high-tech clusters.

Internationalisation

New firms in the area struggled to obtain investment, reflecting the short-term focus of capital markets and rates of return that are higher in other morecartelised, less innovative activity in the UK economy. The introduction ofstandardised credit rating by British banks did not favour innovative busi-nesses. Until the late 1990s, venture capital in the area consisted in three fundsinvesting in about five ventures each among all Cambridge high-tech com-panies. Financial conditions improved when high returns on US tech-based

The Cambridge high-tech cluster 41

investments changed the investment climate at the turn of the century. By thelate 1990s, several hundred million pounds worth of venture capital wasunder management in the area through new venture capital funds seeking toinvest in technology-based firms in Cambridge and beyond. Further seed-capital initiatives by the new government in favour of the commercialisationof science-based technology programmes were undertaken. A few high-profile Cambridge firms realised the financial advantage of their boomingshare prices. Cambridge companies survived the Internet boom and crashbetter than elsewhere because of the diversity of applications and markets.

The attraction of international firms to the area has been a relatively newdevelopment (Garnsey and Longhi, 2004). But the ability of the area toattract inward investment had earlier been demonstrated by internationalacquisition of local enterprises. A 1993 study showed that the most promis-ing Cambridge high-tech enterprises were prone to acquisition, reducingthe pool of companies with prospects for independent growth (Shah andGarnsey, 1993). This pattern has continued with the acquisition of a rangeof promising young Cambridge companies by US corporations in particu-lar. There is case-study evidence (for example, Acorn, CIS, Smallworld andBioRobotics) that acquisition reduces innovative potential by subjectingthe acquired unit to corporate strategic priorities. A counter-effect is thatinward investment through acquisition stimulates further spin-out activityin the area.

Although high-tech firms in Cambridge are linked locally by commonorigins of their members and local expertise, their production networksare more international than local. Cambridge high-tech firms are highlyexport oriented; even the technical design consultancies report that themajority of their clients are international. Export performance was out-standing even before the high-tech boom. The need to establish relationswith customers overseas from start-up stage makes considerable demandson young Cambridge companies; those engaged in ‘micro-global’ effortscannot take the easier route of building initial relationships with domesticcustomer and supplier companies which is open to many US start-ups. Theneed for new firms to build the foundations for growth from the outsetthrough a structure of alliances was the lesson drawn by local firms fromearlier experience in the area. New business models developed byCambridge firms involve licensing and establishing close linkages withmanufacturing partner firms globally. Alliance capabilities are needed toreach multiple markets and are facilitated by international networks devel-oped by local entrepreneurs. Cambridge enterprises that include ARMand CDT have demonstrated that licensing can secure substantial returnsprovided that the technologies are well protected and have multiple marketapplications.

42 Entrepreneurship

Institutional Adaptation

The Cambridge University long operated as a community of scholarsmade up of colleges and departments rather than as a conventionallymanaged organisation; the central administration was minimal until after2000 and did not have the means or inclination to manage technologytransfer centrally. From 1986, British universities had rights to IP in workfunded by the research councils. The Cambridge University was unusualin vesting this entitlement to inventors on its staff. In laissez-faire mode,no active support for technology transfer was provided in the early period.But the administration of the university did not prevent faculty membersfrom developing commercial applications or starting new businesses solong as they carried out their teaching and research duties. In professionalcommunities moral pressures can be brought to bear on individuals whofail to conform to norms and expectations. Cambridge academics areembedded in a network of obligations, collegial and departmental in a uni-versity with a strong research and teaching ethos. Entrepreneurial aca-demics who undertook business activities over and above their academicduties and used the revenues to fund research students and departmentalresearch were able to alter a climate of opinion that had earlier been hostileto business activity by academics. The transformation of Cambridge fromintellectual retreat to high-tech centre took place initially as enterprisingIT experts associated with the university detected and responded togrowing economic demand for information technology. By the 1970s,expertise in science and technology developed at Cambridge Universitywas diffusing from research into business activity. By 1995, the SciencePark housed over 70 firms and 3500 jobs. The Innovation Centrefounded by St John’s College became the nodal centre of the CambridgePhenomenon.11

Local and central government were unsupportive of business expansionin Cambridge. But the authorities were unable to stem the expansion intechnology enterprise. After years of spontaneous expansion, Cambridgehigh-tech companies faced problems they could not solve on an individualbasis. Inadequate public transport and shortages of housing and skilledtechnical labour had become significant constraints on growth and had notbeen addressed by government. Informal linkages were rapidly mobilisedto create a formal network to provide a voice for the high-tech community.Among the aims of the ‘Cambridge Network’ was to improve access to USmarkets and funding for Cambridge high-tech firms, drawing on contactsforged by entrepreneurs who had strong US connections. Business net-works in the Cambridge area extended their reach into the policy arena.Since 1997, Cambridge entrepreneurs have taken part in government policy

The Cambridge high-tech cluster 43

making. Local government officials and businesses formed the GreaterCambridge Partnership, recognising that a larger territory than the city andits villages was the appropriate unit for sustainable development. WhetherCambridge could or should be the hub of a new high-tech region remainedcontroversial; strong anti-growth feeling and concern about the polaris-ation of the labour market coexisted with advocacy of economic potential.Working groups were set up to encourage related expansion in the regionand beyond. Local recognition of the finance gap for firms developingemerging technologies and the extent to which US technology enterprisehas benefited from much higher levels of government support, for example,through Defense Advanced Research Agency (DARPA) and SmallBusiness Innovation Research and (SBIR) programmes, resulted in lobby-ing for a bill, sponsored by a Cambridge Member of Parliament, providingfor more grant and procurement support on the US model to high-tech ven-tures in the UK.

6. CONCLUSION

In summary, endogenous developments in Cambridge encompass thefounding of companies by current and former members of the university,clustering stimulated by serial spin-outs from originator firms, the rise oflocal suppliers and, especially significant, the emergence of specialistlabour markets. These developments depended on demand for high-techoutput and exerted attraction effects through business services drawn to thearea, through the implantation of international subsidiaries, inward invest-ment via acquisition and the attraction of venture capital funds. Togetherthese processes, endogenous and exogenous, contributed to the develop-ment of local competence and capabilities resulting in the formation andsuccess of many new firms. While the process of transformation has manyfeatures in common with ecological succession in the natural world, thisnew economic habitat developed its own identity and local consciousnessat a higher-order level. The creation of a favourable selection environmentwas the unintended outcome of many local decisions, but over time a senseof purpose was created in the local high-tech community whose championsset out deliberately to attract resources to the area. The need for such actionpoints to the limitations of spontaneous expansion unsupported byregional or technology policy.

It is difficult to isolate the effects of knowledge networks around astrong science base in places where many other influences have been atwork. Co-determinants elsewhere include earlier industrial experience(Oxford), government spending on infrastructure (Sophia-Antipolis),

44 Entrepreneurship

large company effects (Siemens in Munich, Eriksson in Stockholm),metropolitan influences (London, Paris) and defence spending and pro-curement (Silicon Valley, Route 128). But in the Cambridge area there wasminimal impact from exogenous investment influences such as have con-tributed to local spillover effects for technology-based firms elsewhere.Thus Cambridge provides unique evidence of the economic potential of aknowledge-based centre as the engine of expansion of innovative industry.

NOTES

1. The term ‘spin-out’ is sometimes used to convey the ‘third mission’ of the universities,the transfer and exploitation of intellectual property (Research Services, University ofCambridge). In this usage the term ‘university start-up company’ is used to refer to busi-nesses founded by students and faculty and former or current members the university,but not making use of university intellectual property (IP) or lacking a university equitystake. We use the term ‘spin-out’ in the inclusive sense long used in the research litera-ture and ‘official spin-out’ company for those started with university IP or equity.

2. Total employment 351 170, based on the 1997 Employment census, adjusted by theCambridge County Council Research Group. The number of firms and levels of employ-ment reported here are derived from an adjusted dataset, refined from that used by theCounty Council which employs a less restrictive definition of ‘high-tech’ and countsoperating units rather than firms. Alternative definitions of high-tech activity are usedby PACEC (2003), based on the wider Greater Cambridge Partnership area, and by theCambridge investment consultancy, Library House, for their commercial database. Forcontinuity and comparability we used SIC categories rather than non-standardised cat-egories devised for high-tech activities.

3. The methodology for this calculation is described in Heffernan and Garnsey (2002).4. Defined here as South Cambridgeshire and the City of Cambridge. The data presented

in Figures 2.3 and 2.4 were derived from the Cambridgeshire County CouncilEmployment database, as described in Heffernan and Garnsey (2002). For the purposesof this chapter, the data have been extended to 2004, but the area covered has beenrestricted to the City of Cambridge and the adjacent South Cambridgeshire area, whichhave a concentration of firms originating in the university.

5. For example, while manual DNA sequencing originated in Cambridge, automated genesequencing equipment essential to the Human Genome project was developed byAdvanced Biosystems in California.

6. Cambridge Ring Newsletter. University of Cambridge Computer Science Department,2004.

7. Robin Saxby, Chairman of ARM, Diebold Conference, LSE, 28 April 2004.8. TTP records kindly made available by the company.9. Calculations by Alan Barrell, former CEO of Domino Printing Sciences and Willetts.

10. Caspar and Karamanoz (2003) argue that non-university ties are more important thanuniversity ties to the Cambridge biotech cluster from a study of recent spin-outs, but donot examine changes over time in the cluster and its origins.

11. By 1969 Cambridge was already a centre of applied science; 25 per cent of the researchand technical staff of the university were involved in applied research supported byoutside funding (Garnsey and Cannon-Brookes, 1993). Cambridge University scientiststook part in international networks, as a result of which they were the first to recognisethe benefits to academic science departments of having companies sited locally to helpcommercialise scientific inventions, provide research collaboration and jobs for graduates.Social networks in the university and city encouraged innovation, with social interaction

The Cambridge high-tech cluster 45

across professional groups through college and community connections. Individual man-agers of banks’ local branches became part of an emerging business community and wereable to provide overdrafts on a discretionary basis. Individuals among college adminis-trators championed the Science Park and Innovation Centre (Garnsey and Longhi, 2004).

REFERENCES

Arthur, W.B. (1994), Increasing Returns and Path Dependence in the Economy, AnnArbor, MI: University of Michigan Press.

Auton, P. and Biddle, H. (2001), ‘Successful spin-outs, by design’, Arthur D. LittlePrism, Issue 1.

Best, M.H. and Forrant, R. (2000), ‘Regional industrial modernization programmes:two cases from Massachusetts’, European Planning Studies, 8(2): 211–23.

Boschma, R.A. and Frenken, K. (2003), ‘Evolutionary economics and industrylocation’, Review of Regional Research, 23: 183–200.

Breschi, S. and Lissoni, F. (2001), ‘Knowledge spillovers and local innovationsystems: a critical survey’, Industrial and Corporate Change, 10(4): 975–1005.

Butchart, R.L. (1987), ‘A new UK definition of the high technology industries’,Economic Trends, No. 400, February 1987, Crown Copyright.

Caspar, S. and Karamanoz, A. (2003), ‘Commercializing science in Europe: theCambridge biotechnology cluster’ European Planning Studies, 11(7): 805–21.

Cooper, A. (1971), ‘Spin-offs and technical entrepreneurship’, IEEE Transactionson Engineering Management, 18(1): 2–6.

Cooper, A.C. and Dunkelberg, W.C. (1987), ‘Entrepreneurial research: old ques-tions, new answers and methodological issues’, American Journal of SmallBusiness, 11: 11–23.

Garnsey, E. and Cannon-Brooke, A. (1993), ‘The Cambridge phenomenon revis-ited: aggregate change among Cambridge high technology firms since 1985’,Entrepreneurship and Regional Development, 5: 179–207.

Garnsey, E. and Longhi, C. (2004), ‘Complex processes and innovative places: theevolution of high tech Cambridge and Sophia-Antipolis’, International Journalof Technology Management, 28(3–6): 336–55.

Heffernan, P. and Garnsey, E. (2002), ‘Technology and knowledge based businessin the Cambridge Area: a review of the evidence’, Centre for TechnologyManagement Working Paper 2002/01, University of Cambridge.

Keeble, D. (1989), ‘High-technology industry and regional development in Britain: thecase of the Cambridge phenomenon’, Environment and Planning C, 7(2), 153–72.

Keeble, D., Lawson, C., Moore, B. and Wilkinson, F. (1999), ‘Collective learningprocesses, networking and “institutional thickness” in the Cambridge region’,Regional Studies, 33(4): 319–32.

Kirchoff, B.A. (1994), Entrepreneurship and Dynamic Capitalism: The Economics ofBusiness Firm Formation and Growth, Westport, CT: Praeger.

Klepper, S. (2001), ‘Employee startups in high-tech industries’, Industrial andCorporate Change, 10(3): 639–74.

Klepper, S. (2002), ‘The evolution of the U.S. automobile industry and Detroit asits capital’, Paper presented at 9th Congress of the International Joseph A.Schumpeter Society, Gainesville, FL, March.

Koepp, R. (2002), Clusters of Creativity: Enduring Lessons on Innovation andEntrepreneurship from Silicon Valley and Europe’s Silicon Fen, Chichester: Wiley.

46 Entrepreneurship

Krugman, P. (1991), ‘Increasing returns and economic geography’, Journal ofPolitical Economy, 99(3): 483–99.

Maruyama, M. (1963), ‘The second cybernetics: deviation-amplifying, mutualcausal processes’, American Scientist, 51: 164–79.

PACEC (2003), ‘The Cambridge Phenomenon – Fulfilling the Potential’, Report forthe Greater Cambridge Partnership, Public and Corporate EconomicConsultants, Cambridge.

Porter, M. (1990), ‘The competitive advantage of nations’, Harvard BusinessReview, March–April: 73–93.

Saxenian, A. (1994), Regional Advantage: Culture and Competition in Silicon Valleyand Route 128, Cambridge, MA and London: Harvard University Press.

Segal Quince Wicksteed Ltd. (1985), The Cambridge Phenomenon: The Growth ofHigh Technology Industry in a University Town, 2nd edn, 1990, Cambridge, UK.

Shah, S. and Garnsey, E. (1993), ‘The acquisition of high tech firms: evidence fromCambridge’, Judge Institute Research Papers in Management, University ofCambridge.

Simmie, J., Siino, C., Zuliani, J., Jalabert, G. and Strambach, S. (2004), ‘Local inno-vation system governance and performance: a comparative analysis ofOxfordshire, Stuttgart and Toulouse’, International Journal of TechnologyManagement, 28 (3–6): 534–59.

Slatter, S. (1992), Gambling on Growth, Chichester: Wiley.Storey, D. (1994), Understanding the Small Business Sector, London: Routledge.Zucker, L., Darby, M. and Brewer, M. (1998), ‘Intellectual human capital and the birth

of U.S. biotechnology enterprises’, American Economic Review, 88(1): 290–306.

The Cambridge high-tech cluster 47

3. Sophia-Antipolis as a ‘reverse’science park: from exogenous toendogenous developmentMichel Quéré

1. INTRODUCTION

The Sophia-Antipolis science park is often presented in the media as aEuropean model of science park development. There are obvious reasonsfor that, especially because of the historical background of the experiment.The park started from scratch in the 1970s and reached an impressive stageof accumulation whereby more than 25 000 jobs are now in existence onsite. We shall argue that the park constituted a unique experiment due tothe fact it has to be considered as a ‘reverse’ science park as the universityand research institutions joined the park only at a later stage. A relatedfeature of the park holds that for a long time its development has drawn onexogenous sources. Only recently have some developments become trulyendogenous, rendering the success of the park more complicated to assess.

Section 2 will discuss the historical patterns characterising the Sophia-Antipolis project. From these background conditions, Section 3 explores amore specific issue, which is the capability of that project to transform itselfinto a real science park project, encouraging and benefiting from localentrepreneurial initiatives. Section 4 deals with a general discussion aboutthat transformation, with a specific insistence on governance issues andmore in particular with a discussion about how innovation opportunitiesprogressively emerge locally. Section 5 focuses on innovative behavioursarising from local interactive learning in information and communicationtechnology (ICT) activities, in accordance with either the type of firmsinvolved in these processes or internal–external interactions. Section 5 alsodeals in more detail with the policy implications of that transition towardsa science park functioning in ICT activities. Concluding remarks about thesustainability of the Sophia-Antipolis science park and local policy makingwill be provided in the final section.

48

2. SOPHIA-ANTIPOLIS: MAIN HISTORICALCHARACTERISTICS

The usual definition of a science park requires the existence of comple-mentary resources in a similar location, namely academic resources includ-ing training and research capabilities, and high-tech start-ups that eitherresult from or interact with the local research environment. These back-ground conditions are usually complemented by any kind of governancestructure favouring the connecting dimension among those components.What characterises Europe with regard to the governance and regulation ofscience parks is the wide variety of mechanisms and the diversity of gover-nance structures aimed at ensuring the development of experiments(Gaffard and Quéré, 1996; Quéré, 1998). The governance structures varyconsiderably among European countries, from university or even privatestructures in Anglo-Saxon types of regulation to public-owned offices aswell as private/public specific organisations in countries such as France,Italy and/or Germany.

The Sophia-Antipolis science park is an original example for discussingtypes of governance mechanisms appropriate to encouraging innovation andbusiness opportunities from local learning and interactions among variouseconomic actors. Partly this results from the genesis of the project, which isquite recent (35 years ago), and partly from the different developmentalstages faced by the project. Figure 3.1 depicts the quantitative accumulationfrom the mid-1970s and it demonstrates the unique character of the project.

Sophia-Antipolis as a ‘reverse’ science park 49

Source: Sophia Antipolis Enterprise Management – Sophia Antipolis Côte d’Azur(SAEM–SACA), Sophia-Antipolis.

Figure 3.1 Cumulative number of organisations and employment

0

200

400

600

800

1000

1200

1400

1600

1800

2000

1975

1977

1979

1981

1983

1985

1987

1989

1991

1993

1995

1997

1999

Years

Le

ga

l ent

itie

s

0

5000

10000

15000

20000

25000E

mpl

oym

ent

Cumulative number of legal entities Cumulative employment

Initially, Sophia-Antipolis was planned to be a city of science, cultureand wisdom. The genesis of the science park was a purely private initiative,instigated by Pierre Laffitte. Laffitte was a member of the board of theENSMP (École Nationale Supérieure des Mines de Paris), one of the so-called French ‘Grandes Écoles’, and what he had in mind was to establisha sort of community of scientists in the South of France benefiting fromthe Sun Belt effect. He targeted one of the few landscapes still availablelocally, which was a forest near the village of Valbonne, seven kilometresnorth of the city of Antibes. The enterprise started in 1969 and the firstbuildings were erected in 1972. However, the project collapsed financially;the infrastructure costs were extremely high and the private organisationinvolved was unable to survive the mismatch between costs and benefitswithin the first years of implementation.

The initial project could have come to a halt at this point, since no realaccumulation process had taken place when the private initiative was with-drawn. However, this has not been the case due to a transformation processfrom a private to a public governance structure. The local public authori-ties (the Conseil Général des Alpes Maritimes: CGAM) have retained theproject with the aim of complementing the local economy with other typesof activity that were thought to be compatible with the dominant charac-ter of tourism. The local environment (the Côte d’Azur) traditionallydepends on tourism activities and it was a central concern for the publicauthorities to diversify without damaging that dominant economic infra-structure. High-tech activities were thought to be acceptable in the sensethat their relative image and perception in the public would not endangertourist flows.

From 1977, the Sophia-Antipolis experiment shifted from a purelyprivate to a public initiative in which the CGAM has played a critical role.The latter empowers the local Chamber of Commerce to manage the exper-iment. This change of governance goes hand in hand with a change in boththe concept and the content of the project: from a ‘City of Science, Cultureand Wisdom’ to the concept of an ‘International Industrial Park’. Localpublic authorities favoured the location of research and development(R&D) units of international firms, that is, they aimed to attract externalresources that would diversify the economy of the Côte d’Azur. A deliber-ate choice not to accept the location of manufacturing activity was madeand the selective character of attracting R&D units was due to the searchfor compatibility with tourism. From that time, a change in public supportwas also important. Not only did the public authorities take over from theinitial private initiative in order to ensure the physical infrastructures, butthey also developed an active advertising strategy in order to promote thelocation at the international level and in the United States in particular. As

50 Entrepreneurship

a result, a change in the scale of the experiment occured and a progressiveaccumulation of external resources assured the sustainability of the project(Longhi and Quéré, 1997; Quéré, 1997).

The shift in governance has ensured the quantitative success of theproject. In that period, from 1977 to the end of the 1980s, the park exhib-ited a spectacular accumulation process of external activities (see Figure3.1). This particular (second) stage of development can be characterised bythe following structural patterns:

● an influential resilience from the private (individual) initiative, even iflocal authorities have moved away from the initial concept to a ‘stan-dard’ international industrial park;

● a spectacular process of accumulation of external resources, mainlyR&D units from large international firms;

● a site attractiveness mainly due to public infrastructures and utilitiesused by tourism activities on the Côte d’Azur (international airport,accommodation facilities, congress facilities and so on)

● a heterogeneous accumulation process, as far as sectoral patterns areconcerned and, as a consequence, a very low level of local connec-tions and interactive learning on site; and

● an important dependence of local units from exogenous (inter-nationally wide) firm decision making.

These structural characteristics induced an obvious fragility in the sus-tainability of the experiment. However, this stage can also be thought ofas a necessary step to ensure and secure a local mass effect and the emer-gence of local interactions among the components of the park. Thealmost random accumulation process resulted in the progressive special-isation into two major activity areas: ICTs and life sciences. ICTs includesoftware, telecom and image engineering activities but also electric, elec-tronic and micro-electronic activities, automation, and signal and systemengineering. This gave rise to a number of R&D activities that are quitelarge and diversified, resulting in potential opportunities for interactions.Moreover, this potential has been enhanced by the location of publicresearch institutes working in the same knowledge domains. Figure 3.2shows the accumulation process in the ICT activities during the 1990s. Itespecially shows how ICT activities have faced a transition phase andescaped from a relative quantitative stagnation to increase again from themid-1990s onwards.

By 2000, ICTs accounted for about 75 per cent of ‘high-tech’ localemployment. They originated from either large domestic companies such asAir-France, Thalès and Télémécanique/Schneider or large foreign ones such

Sophia-Antipolis as a ‘reverse’ science park 51

as Amadeus, Accenture and DEC/Compaq. Both types of firms locatedtheir R&D activity in the park, which complemented the accumulation ofpublic research resources locally. If ENSMP was already among the firstactors locating in Sophia-Antipolis, research units from CNRS (CentreNational de la Recherche Scientifique: National Centre for ScientificResearch), INRIA (Institut National de Recherche en Informatique etAutomatique) and the University of Nice-Sophia-Antipolis finally formeda critical mass that favoured the existence of business opportunities throughendogenous interactive learning.

The situation in life science activities is different, because no similar masseffect has been reached. No such eclectic accumulation as in ICT activitieshas been obtained and there are no more than 50 local firms involved inthose activities. During its development, the park benefited from R&Dunits from firms such as Dow Chemical, Dow Corning, Wellcome,Allergan, Rohm & Haas, Aventis and others. These units were acting essen-tially in activities such as fine chemistry, pharmaceuticals and dermatology.In that respect, business opportunities stemming from the diversity ofcapabilities that have been accumulated locally are not easy to develop.Most of these local units are internally oriented, as they often have anexclusive customer located at the international level. As such, they are notvery interactive with other actors in the park, since they function mostly asan internal service provider. Some of the units are more research oriented,such as Cird/Galderma and Cordis/Zeneca, but their local relationships arestill very limited. Thus, within-park opportunities through interactivelearning are not well developed. Figure 3.3 reports the quantitative accu-mulation in those sectors and demonstrates the lack of a significant masseffect in life sciences.

52 Entrepreneurship

Figure 3.2 Cumulative number of organisations and employment (ICT)

0

50

100

150

200

250

300

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Years

Lega

l ent

ities

0

2

4

6

8

10000

12000

Em

ploy

men

t

Cumulative number of legal entities Cumulative employment

000

000

000

000

In sum, by the turn of the 1990s, the first stage of the experiment hadresulted in an obvious quantitative success as about 700 firms involvingabout 10 000 employees were located in that area which was, to reiterate,just a forest 20 years earlier. As such, this characteristic made the Sophia-Antipolis park quite unique in Europe, as no similar pattern of R&D accu-mulation in such a short time period has been found elsewhere.

A second stage of the experiment began at the turn of the 1990s. At thattime, there was a cyclical crisis in ICT activities, which provoked anincreased instability and volatility in local R&D units. This endangered thesustainability of the project, and from 1990 to 1995, no real improvementin employment accumulation was possible. The quantitative observationexpressed a transition phase in the local accumulation regime. On the onehand, no serious candidate to locate in Sophia-Antipolis was available,probably due to increasing competitive pressure from other Europeancountries (see Section 3); on the other, some of the R&D units that were atthe source of the success left the park, under pressure from external (inter-national) headquarters and their related decision making on how to organ-ise R&D and business units in Europe. Curiously, this business-cycle crisishas benefited the park as it has induced a change in the accumulationregime, from an exogenously driven process to a more endogenous processof entrepreneurial initiatives. The local process of accumulation was againsensitive from 1995 but is now much more the result of an internally drivenprocess than one of externally driven accumulation of resources. Since themid-1980s, the Sophia-Antipolis park has become a European ‘model’ of ascience park. This is somewhat misleading. Sophia-Antipolis should beproperly thought of as a reverse science park, as the local university was

Sophia-Antipolis as a ‘reverse’ science park 53

Figure 3.3 Cumulative number of organisations and employment (lifesciences)

0

10

20

30

40

50

60

1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000

Years

Lega

l ent

ities

0

500

1000

1500

2000

2500

Em

ploy

men

t

Cumulative number of legal entities Cumulative employment

one of the last institutions to join the process of local development. Thelocation of PhD training from the university system has occurred only sincethe mid-1980s, but has been extremely powerful in favouring the previoustransition from an exogenous-towards an endogenous-driven process ofgrowth. Beyond the university system, the progressive accumulation ofresearch institutes, especially in ICT sectors, has also favoured the trans-ition towards a more traditional type of science park development. Fromthe mid-1990s, the major structural patterns characterising the economicworking of the park can be characterised by:

● a larger influential role of the university of Nice-Sophia-Antipolis inthat transition;

● a modest but steady improvement in the accumulation of academicand public research resources, beyond the local university;

● a substitution of large R&D units by more endogenous entrepre-neurial initiatives (especially in ICT activities); and

● a more interactive process of localised learning allowing for exploringinnovative business opportunities on a large scale from the local envi-ronment (the emergence of a local entrepreneurial climate and culture).

From this rough historical description of the Sophia-Antipolis park, onecan provide some insights about the local process of growth as well as aboutthe governance mechanisms that allow for such a successful experiment.The passage of time has progressively revealed two major sets of activities(ICT and life sciences). However, the Sophia-Antipolis experiment isexhibiting quite a large economic bio-diversity, which makes the experi-ment unique and distinctive. Furthermore, due to exogenous forces, thepark could reach a critical mass in ICT activities, which appears to be a nec-essary condition for innovative opportunities to arise. However, such amass effect cannot in itself be a sufficient condition for endogenous devel-opment to be sustainable. It has to be complemented by further efforts,especially from local public authorities, including physical infrastructures,governance mechanisms and innovation policy efforts. This is the issue dis-cussed in the next section.

3. THE GOVERNANCE OF A SCIENCE PARK:LESSONS FROM THE SOPHIA-ANTIPOLISEXPERIMENT

The governance of the Sophia-Antipolis science park has become a centralissue in order to guarantee the sustainability of the project in the future. To

54 Entrepreneurship

understand the context of policy making, we first need to understand thestructural factors that have influenced the evolution and governance of thepark. Among these factors are the Côte d’Azur Sun Belt effects, which areas various as the existence of an attractive international airport, local infra-structures facilities and the sunny climate. But these factors also encompassmore dedicated efforts such as an active advertising policy developed bylocal policy makers when they decided to follow up the project. The maincyclical factor was the need from extra-European firms to locate some oftheir R&D facilities in Europe. Basically, the European market was per-ceived to be so complicated for legal and governmental reasons that firmsneeded to benefit from some local R&D facilities, adjusting their productportfolios to the decentralised and complicated ‘local’ legislations that char-acterised Europe at that time. As such, Sophia-Antipolis was a good candi-date to adjust correspondingly with intra-European customers. In thatrespect, the quality of physical infrastructures on the Côte d’Azur that wasinherited from tourism activities has been a positive factor that has madethat location competitive relative to other potential choices within Europe.

Second, the mass effect in ICT activities has also been enhanced by aca-demics and public research organisations. Due to the influential role ofPierre Laffitte, some training and research activities have been located in thatarea from the starting phase. However, training and research infrastructurehas not experienced a similar growth trend as private R&D units.Nevertheless, the infrastructure has progressively supported quite asignificant volume of employment currently involving about 2000 individu-als and 5000 students on site. Part of this improvement is due to nationaldecision making, which deliberately decided to locate some research unitsfrom the French CNRS or from other public research institutes includingthe aforementioned ENSMP, INRIA and ADEME (Agence del’Environnement et de Maîtrise de l’Energie). Part of this improvement is dueto an increasing interest from the University of Nice-Sophia-Antipolis in thepark from the mid-1980s onwards, when the university decided to locate PhDtraining related to ICT and life science activities in the park.

Third, the mass effect resulting from an active international advertisingstrategy and the accumulation of academic resources has provoked asignificant change in the local growth regime. Both factors changed thedriving force of local growth from an exogenous-driven process to a moreendogenous one. The period of that transition regime corresponds to theearly 1990s to the mid-1990s. At that time, a crisis occurred locally, pri-marily due to a cyclical downturn, leading to a decreasing attractiveness ofthe Sophia-Antipolis park for R&D units of large international firms andeven to the departure of some of them from the park. The reason for thestructural shock was on the one hand an influential effect of the European

Sophia-Antipolis as a ‘reverse’ science park 55

Union (EU) integration process and, on the other, an improvement inknowledge about the EU market for local R&D units. In short, there wasa lack of relative competitiveness with the Sophia-Antipolis park, as far astraditional comparative cost advantages were concerned. Furthermore, the‘crisis’ has also been the source of a transition in the local growth regime.Most of the individuals concerned with the intra-EU mobility process werenot convinced and decided to stay on the French Côte d’Azur for personalreasons. As such, they tried to engage in individual business by develop-ing entrepreneurial initiatives locally. As a consequence, the transitionenhanced local interactions and favoured the development of a local labourmarket for high-skilled capabilities. This improvement in the labour marketfor highly qualified personnel is to be thought of as a consequence of thesuccess of the park in its capability to shift towards a more endogenousgrowth development, that is, to behave as a real science park.

This evolution has also been favoured by the progressive accumulationof academics, public research personnel and PhD students on site. A stockof around 2000 individuals acting in academic and public research insti-tutes added to a volume of 5000 students is part of the transition as it hasbeen extremely useful in supporting the previous entrepreneurial initiatives.Some institutions have taken a particularly active role in the process; forexample, INRIA has been especially active in launching entrepreneurialinitiatives from internal research projects. Moreover, other institutions suchas ETSI (European Telecom Standard Institute) have had a specificinfluence in attracting European firms in telecom activities and in con-tributing in a local distinctive European capability in that sector. Both insti-tutions have been particularly helpful in favouring the local sustainabilityof the transition towards a more endogenous process of accumulation.

Fourth, the most spectacular dimension of the transition lies in the birthof a huge number of high-tech very small firms, the results of complicatedcombinations of individuals from R&D units, PhD students and publicresearch fellows. These combinations emerge locally almost as a reactivestrategy to either not leaving the Côte d’Azur, or being unable to find a suit-able position in the public research system. However, new business oppor-tunities from local interactive learning have been quantitatively spectacular.Moreover, it is not simply a business-cycle effect of the ‘dotcom’ revolutionthat occurred around 2000; it is a real process of transformation of the parkthat leads us to question its ability to behave as a real science park and todiscuss the transition towards an actual local innovation system. As a con-sequence, the governance of the project is at stake as the initial advertisingpolicy that ensured its success has also been questioned. What kind of policyinitiative is needed to ensure the success of the transformation process isnow becoming a real issue.

56 Entrepreneurship

Fifth, and as a consequence of the previous remark, the transition resultsin a new effect: a much higher volatility of local firms. As far as ICT activ-ities are concerned, it is possible to identify a group of about 300 firmslocally. However, when that group is traced back from the start of the 1990s,more than double that number of ICT firms have been located in the park.We can identify a population of 345 firms that disappeared from Sophia-Antipolis within that period. This is a sign of structural weakness, es-pecially because ICT firms are still very small in size and do not depend ona local market base. Thus, a local mismatch between R&D units from largeinternational firms and small and very small high-tech firms is still aproblem, as far as a discussion about the existence (or not) of a real localsystem of innovation is concerned.

Therefore, it would be interesting to discuss further the ability of firmsto participate in the transformation of the park, especially because thevariety of strategies to deal with and accompany that transformation ishigh. The ability of the territory to behave as a science park where inno-vation can occur from interactive learning among the local components canbe questioned from different viewpoints. In what follows, the implicationsof the change towards a more endogenous accumulation process will beconsidered in more detail.

4. SOPHIA-ANTIPOLIS AS A SCIENCE PARK:CRITICAL GOVERNANCE ISSUES

To obtain a deeper understanding of the park, it is useful to distinguishbetween three main categories of firms that are developing innovativebehaviour locally: (i) R&D units of large international firms; (ii) small andmedium-sized enterprises (SMEs); and (iii) spin-offs/starts-ups. These cat-egories are quite different and, as such, call for different kinds of gover-nance mechanisms as well as different kinds of public policy.

R&D Units of Large International Firms

From the mid-1970s, the park development has been based on the matchbetween supply and demand. The supply was the availability of squaremetres and/or offices in an attractive area surrounding the Côte d’Azur; thedemand was the need of large international firms to adjust their productportfolio to specific requirements from European countries. As we havealready indicated, the location of R&D units in Sophia-Antipolis has pro-gressively been challenged by other European locations. At the turn of the1990s, some R&D units located in Sophia-Antipolis considered alternative

Sophia-Antipolis as a ‘reverse’ science park 57

locations in Europe – Ireland or Scotland for clear advantages in terms ofhigh-qualified labour costs; southern Germany or northern Italy when theywanted to improve interactions with their main or representative Europeancustomers; Paris or London when their EU location was dealing moreextensively with financial and/or administrative matters.

As a consequence, the relative advantages of Sophia-Antipolis progres-sively diminished and those R&D units either relocated elsewhere inEurope or downsized their local units in terms of employment. Note,however, that these firms had little involvement in local learning with otheractors in the park, be it other firms or public research institutions. R&Dunits were mainly dedicated to intra-firm activity and used as an exclusiveprovider of internal services for other units of the company. As such, theirembeddedness in the local economic environment was weak. Moreover,subcontracting activities were mostly non-local, too. Thus, these R&Dunits were fully dependent on external decision making that was also inter-nationally driven and unaffected by local policy making. Therefore, eventhough these R&D units of international firms ensured the quantitativesuccess of the Sophia-Antipolis park, they were not fundamental to localinteractive learning. Nevertheless, two caveats apply. First, the externalthreat to relocate stemming from internal firm hierarchy had a positiveeffect on local interactions. Proving the existence of complementaryresources locally was perceived as the best means of avoiding a departurefrom the park. Therefore, those R&D units developed localised interac-tions and learning as a sort of reactive strategy in order to enhance the localattractiveness and to avoid moving from Sophia-Antipolis. Second, asignificant number of individuals concerned with the relocation processdecided not to leave the South of France but to engage in new local inde-pendent businesses. They refused the related mobility to other parts ofEurope and developed entrepreneurial initiatives locally. Moreover, someof the local R&D units (such as Lucent on site, but also Alcatel, IBM andTexas Instruments in the immediate vicinity) encouraged the transitionthemselves and favoured the establishment of intra-entrepreneurialsupport (such as in-house incubators). The transition to a more endoge-nous type of development is thus partly the result of a reactive strategy ofR&D units of large firms that were not fully convinced about the wisdomof leaving the site. Yet, this reactive strategy was not pervasive enough toprevent some firms from relocating their R&D units.

Small and Medium-sized Enterprises

A main weakness of the Sophia-Antipolis park lies in the lack of techno-logical SMEs. There are very few of them, as the Côte d’Azur has never

58 Entrepreneurship

been an industrial area. Most of the firms that are emerging locally do notreach the critical size seen as necessary for ‘real’ SMEs. Figure 3.4 illustratesthis argument by considering the size patterns of the mass effect reached inICTs. Most of the ICT firms located in Sophia-Antipolis have fewer than50 employees.

Obviously, this characteristic makes these SMEs quite vulnerable locally.There are basically no local markets as the area does not benefit from anindustrial infrastructure. As a consequence, the SMEs have difficulty devel-oping and growing locally. They essentially depend on a very small numberof customers and that factor makes their development unlikely to be locallysustainable in the long term. As a consequence, improving their growth sus-tainability means either that they expand in geographical terms and have todevelop towards a market base that is no longer local (mainly Bordeaux,Toulouse, Montpellier and Marseille), or that they merge or make con-tractual agreements with larger firms that can be perceived as a guaranteeto secure their development (especially in financial terms and in related firminvestment).

Local Spin-offs and Start-ups

From the discussion about the two previous firm types, it is obvious thatmany firms created locally are partly the result of the crisis when R&Dunits from large firms decided to create local spin-offs. The latter are thesource for the existence of technological SMEs that exploit market nichesfrom their Sophia-Antipolis location in diverse activities such as techno-logical expertise and computer services (system design, software, networkengineering and so on). A strategic positioning of these SMEs is part of thedotcom revolution. They are not fully part of the impressive economicdevelopment of e-business applications, but they provide the technological

Sophia-Antipolis as a ‘reverse’ science park 59

Figure 3.4 Size distribution of ICT firms

0

20

40

60

80

100

120

140

160

180

0 to 10 11 to 50 51 to 100 101 to 300 301 to 500 501 and more

Employment

Lega

l ent

ities

logistics to develop innovation in those activities. As such, these SMEs aremore technology oriented and appear critical to ensure the technologicalimprovement and related logistics to the so-called ‘dotcom revolution’.Therefore, these firms have not suffered so much from the recent dotcomcrisis and have succeeded in surviving and even improving their technolog-ical expertise and capabilities despite the cyclical downturn. The existenceof these SMEs has also been favoured by a general improvement inscience–industry relationships within the French context. In a countrywhere the academic environment has been extremely isolated from indus-try, a real change has occurred during the last decades. Each university andpublic research institute is now trying to engage and develop connectionswith firms. This national transformation has obviously benefited the entre-preneurial climate in Sophia-Antipolis. The academic and public researchsystems have become more receptive to the idea of engaging with privateactors and developing professional relationships with the latter (contrac-tual terms, patents and licence agreements, and so on). The improvementin the professional character of negotiations related to the whole range ofscience–industry relationships has mutually benefited local interactivelearning and the development of business initiatives stemming from com-plementarities between private and public expertise, essentially, in the fieldof ICT activities (Quéré and Ravix, 1997). Another general trend is thedevelopment of intermediation through venture capital initiatives and theavailability of a diversified intermediation supply. This concerns not onlyventure capital as traditionally defined, but also other intermediaries. Thelatter can be purely private (such as business angels), purely public (such asacademic incubators established by the Law on Innovation from 1999), orsemi-public. These changes have triggered many more interactive relationswithin the park and, more generally, the emergence of a local entrepre-neurial climate and endogenous business opportunities.

5. THE GOVERNANCE OF THETRANSFORMATION TOWARDS A SCIENCEPARK MODEL: POLICY IMPLICATIONS

In order to behave as a ‘real’ science park, the Sophia-Antipolis experi-ment has to encourage a transformation from exogenous to endogenousdevelopment. The necessary condition for such a transformation is theexistence of a significant critical mass effect. It is obvious that the latterhas been reached only in ICT activities. This is why the following discus-sion is centred on ICT and the way interactions occur and take place in abusiness environment, which is obviously larger than the frontiers of the

60 Entrepreneurship

Sophia-Antipolis park itself. Figure 3.5 makes the point in a more syn-thetic manner. It offers a more comprehensive framework on how inno-vative behaviours through local interactive learning take place in a largermarketplace.

The figure represents the distribution of ICT activities in a ‘type andspace’ framework. Of course, this is just a rough approximation of a‘stylised fact’ representation of those ICT activities. But it helps the dis-cussion as the matrix between the type of activity and the importance oflocal versus non-local relationships provides a synthetic view about thedominant characteristics of innovative firms’ behaviours through interac-tive learning in the Sophia-Antipolis environment.

Most of the telecom providers (operators) located in the park are notextremely active locally. Even France Telecom, which could be thought ofas having a localised advantage, cannot be considered as a central actor inlocal innovative behaviours. These firms are essentially interested inabsorbing positive benefits from local capabilities but in a unilateral way.The main concern of this category of firms is external, that is, non-local,and the park is an exploratory potential that can be helpful in order toreflect on the future of markets and services as well as to benefit fromattractive human resources. However, their active participation in localisedinnovation processes is weak. They benefit from an image effect and theycan also offer higher earnings to specific human resources that are revealedas useful to them from the working of the park. But this is a sort of pre-dating strategy (both conscious and unconscious) that does not fullybenefit the overall entrepreneurial climate.

Sophia-Antipolis as a ‘reverse’ science park 61

Figure 3.5 The geography of ICT activities

‘Hardware’activities

Operators Equipment providers

Local interactions

Applications development

Global interactions

‘Software’activities

Telecom equipment providers, however, have a pivotal role in the eco-nomic working of the park. On the one hand, they are participating in anindustrial context which implies that they will be identified and act asglobal players. On the other, they are apparently more active in terms oflocal interactive learning than operators. In that respect, telecom equip-ment providers consider Sophia-Antipolis to be a ‘technological platform’where they can test technological opportunities thanks to local technolog-ical start-ups. Moreover, the location of specific institutes like ETSI andpublic research institutions is essential to them in order to secure innovativeinvestment. In that respect, local collaborations are a means of exploringand defining new market opportunities. The latter include networking andconnectivity interfaces, technical compatibility between different equip-ment and standards and so on. Such an experimental process is basicallythe result of the local accumulation of external resources that creates adiversified productive environment where diverse and various capabilitiesare possibly complementing each other. Some equipment suppliers haveeven developed in-house incubators in order to benefit from that variety ofcapabilities existing in the park. This gives to the park, at least for specifictypes of activities, a status of European experimental project enhancing thereputation of Sophia-Antipolis as one of the leading locations in Europeas far as particular telecom activities are concerned.

Application and service providers represent the category of firms forwhich interactive learning is the most developed and possibly mostbeneficial. It is possible to distinguish two types of firms. First, SMEs thatsucceeded in taking an active part in the e-economy, that is, those thatbenefited from the dotcom revolution and, as such, enhance the Sophia-Antipolis image effect. However, success stories have been relatively fewand probably did not contribute significantly to the local economy. In fact,many of these firms disappeared from the local environment when acquiredby large external firms. As such, local entrepreneurship then benefits therest of the world economy instead of benefiting the local economy. This islargely the case for firms such as Respublica, Lybertysurf, Echointeractiveor Odisei. ‘High-tech’ SMEs can also be obliged to connect more denselywith their main customers and suppliers, which leads local entrepreneurialinitiatives to relocate in other locations in order to secure their market andapplication bases.

The second category of firms is science and technology-based SMEsthat develop technological support and services on the application side ofthe so-called ‘new economy’. Here, the interesting aspect is that these firmsare the result of local interactive learning among different components,including public research institutes, PhD student projects, spin-offs andso on. They are the most important result of the transition phase in the

62 Entrepreneurship

accumulation process from an exogenous towards an endogenous type ofdevelopment. A byproduct of these ventures is the higher local sustain-ability as they develop distinctive capabilities that would not be possible toreproduce easily from another location. Thus, the localised character oftheir innovativeness is the best guarantee for the sustainability of theSophia-Antipolis park. Any departure from Sophia-Antipolis shouldresult in high transition costs and in a risk for firms interested in benefitingfrom SME expertise. As a consequence, the second category of firms is theone that best expresses the transformation of Sophia-Antipolis towards anactual science park. Finally, it is important to stress that the number ofsuch firms (about 20) is relatively small in compared with the whole set ofresources acting locally in ICT activities.

On the whole, the gap between the success in the media of Sophia-Antipolis as a science park and the actual effect of innovation based oninteractive learning and resulting in new high-tech start-ups is quite impor-tant. It stressed that Sophia-Antipolis can be considered as a real sciencepark but only for a subset of ICT activities for which all the necessary com-ponents for a science park have reached a sufficient mass effect to favoursuch endogenous initiatives. This observation brings us back to the parkgovernance issue and to discuss in more detail what kind of policy makingis appropriate to encourage and accelerate the transformation of thescience park.

What is particularly clear from the development of the park is that policymaking has almost exclusively been concerned with the provision of phys-ical infrastructures. This is of course a peculiar characteristic due to the his-torical background of the project (which started from scratch in the 1970s).The proactive character of policy making has basically been to advertisethe project properly at the international level. However, such a policy hasquite exclusively been based on promoting the local physical infrastruc-tures. As a consequence, there is no real policy that has been concerned withthe internal understanding of localised and interactive learning within thedifferent components of the science park. Endogenous development hasmainly been thought of as a natural process that should derive automati-cally from the accumulation mass effect. Obviously, it is not and the previ-ous transformation depicted from ICT activities questions seriously theability of policy making to encourage the economic sustainability of thatexperiment.

Moreover, ensuring the accumulation of resources stemming from endoge-nous development is a difficult issue. On the one hand, it still has to face theintrinsic unstable character of international players (telecom operators andsome equipment providers as well) for which external decision making isstill the dominant regulation for their location in Sophia-Antipolis. As such,

Sophia-Antipolis as a ‘reverse’ science park 63

these firms have a low level of sensitivity to any kind of local public decisionmaking. On the other hand, endogenous SMEs are difficult to secure in thepark as the economic climate is traditionally not a marketplace for thesehigh-tech firms. With tourism still dominating the local economy, policy isstill relatively unconcerned with high-tech activities.

With regard to local policy making, three main issues can be addressed inrelation to the sustainability of the economic transformation of Sophia-Antipolis and its transition towards a real science park. First, the role ofphysical infrastructures is not to be denied. If the park is now a leading placein Europe for some ICT activities, it is basically to be found in the choice oflocal policy makers to invest in a high-broadband telecom network based ona fibreoptic technology within the park by the 1980s. This choice has had aconsiderable impact on the site attractiveness from that time and has beencrucial for ICT firms located in Europe to test for high volumes of infor-mation exchange and the related development of new applications. In otherwords, the quality of local infrastructures is still a necessary, if not asufficient, condition for the economic sustainability of the park. However,what type of infrastructures are appropriate for the future of the park isprobably something that should currently be elaborated within the frame-work of the park and in deeper connection with private actors. This isapparently not the case as the connections between private and public de-cision making are weak. The connectivity between private and public deci-sion making within the framework of the park to define the physicalinfrastructures appropriate to its future development is a challenging issuefor the sustainability of the park.

Second, within the last decade, different forms of collective coordinationamong local private and public actors have emerged. Specific associationsand thematic clubs have developed, most of them in a spontaneous way inorder to favour local coordination and collective innovative projects on alocal base. Some of them are in fact the result of a reactive strategy to dealwith the risk of departure from the park. It was a means of bargaining moreeffectively with external decision making to justify the need to stay on site.This mechanism has been complemented by another trend, which is theemergence of a local market for some local SMEs within the park.Technology and business consulting firms, intellectual property rights andfinance-oriented firms are parts of a logistical infrastructure to high-techSMEs. Such consulting infrastructure is progressively transforming theSophia-Antipolis park into a real marketplace. Both effects are part of atransformation process from an exogenous to an endogenous mode ofdevelopment. However, both mechanisms are also disconnected frompublic policy concern. There is, however, a need for a more effective way todeal with the interplay between private and public characteristics of that

64 Entrepreneurship

transformation. Policy makers can no longer consider interactive inno-vative learning as the result of a spontaneous order. They have to thinkabout proper investment favouring the coordination of all the actors thatare part of that shift towards a more endogenous type of development.

Third, for a long time the Sophia-Antipolis park was perceived as an iso-lated ‘island’ in the Côte d’Azur environment. This is less and less true asthe connectivity of firms located in the park to the environment is increas-ing. The improvement of localised innovative learning in ICT activities alsoimplies complementing R&D with non-R&D activities (prototyping,designing and testing applications and so on). The latter require other capa-bilities and relationships with various sorts of markets. As far as local capa-bilities are available, this process can be expanded locally through all sortsof connections to other firms in the surrounding environment. To mentionone example, one challenge is to connect the dominant local economicactivity (tourism) to the park in a more developed perspective. Tourism isan interesting application area where ICT activities can be applied invarious ways to improve innovation in that sector. The distinctive capabil-ities accumulated in the park in areas such as wireless and Global Systemfor Mobile Communications (GSM) technologies can certainly be com-bined in original ways for tourism applications. However, this againrequires from policy making a forecasting ability to deal properly with suchchallenges.

Altogether, the critical issues show the difficulty of developing policiessupporting the transition of the science park from an exogenously drivento an endogenously evolving economy. To most local policy makers,however, the dominant perception is still that the endogenous character oflocal development should be a natural result of the accumulation process.However, this is obviously not the case as endogenous development can behampered by local constraints and global trends (Garnsey and Heffernan,this volume).

6. CONCLUDING REMARKS

The Sophia-Antipolis experiment is unique. The main factor explaining thesuccess of the project lies in the general advantages exhibited by the FrenchRiviera that were well suited to large multinational companies for locatingin Sophia-Antipolis either European administrative centres or R&D units.At that first stage of its development, Sophia-Antipolis was exogenouslydriven and the experiment exhibited a significant comparative advantageregarding other potential locations in Europe. However, the economicworking of Sophia-Antipolis was vulnerable, because innovative activities

Sophia-Antipolis as a ‘reverse’ science park 65

were not based on local relations. Nevertheless, the park has benefited froma large accumulation of external (international) resources that has been anecessary condition to ensure its quantitative success and to create therequired conditions to shift to real science park dynamics. The latteroccurred 20 years after the starting phase of the experiment and the situ-ation seems to be improving in recent years. However, the transformationis still modest in quantitative terms in comparison with the overall accu-mulated activities in the location and does not really benefit from an activeand appropriate support from local policy making. The transition, essen-tially limited to ICT activities, is to be thought of as a sort of spontaneouschange, mostly due to incentives from individuals not to leave the enjoyableCôte d’Azur environment, which is essentially the result of behaviours fromprivate actors. As a consequence, the transition is still very fragile andshould be central for policy making (Quéré, 1999). The latter should con-centrate on understanding the transition and defining more precise inter-vention to encourage its sustainability. This is still thought to be necessaryin order to make the park viable in the medium term.

REFERENCES

Gaffard, J.L. and Quéré, M. (1996), ‘The diversity of European regions and the con-ditions for a sustainable economic growth’, in X. Vence Deza and J.S. Metcalfe(eds), Wealth from Diversity, Dordrecht: Kluwer Academic.

Longhi, C. and Quéré, M. (1997), ‘The Sophia-Antipolis project or the uncertaincreation of an innovative milieu’, in R. Ratti, A. Bramanti and R. Gordon (eds),The Dynamics of Innovative Regions, Aldershot: Ashgate, pp. 219–36.

Quéré, M. (1997), ‘Sophia-Antipolis as a local system of innovation’, Economia &Lavoro, 3–4: 259–72.

Quéré, M. (1998) (ed.), Les Technopoles en Europe, Enjeux et Atouts de la Diversité,Paris: AFT/DATAR.

Quéré, M. (1999), ‘Innovation, growth, and co-ordination through institutions: adiscussion about “innovation systems” ’, in O. Fabel, F. Farina and L.F. Punzo(eds), European Economies in Transition, London: Macmillan, pp. 131–47.

Quéré, M. and Ravix, J.L. (1997), ‘Production de connaissance et institutions inno-vatrices: le chercheur-entrepreneur’, Revue d’Économie Industrielle, 79: 213–32

66 Entrepreneurship

PART II

Industrial Dynamics

4. The evolution of geographicstructure in new industriesSteven Klepper*

1. INTRODUCTION

In recent years there has been a resurgence of interest among economistsin questions related to geography. In part spurred by the clustering of eco-nomic activity in Silicon Valley, attention has focused on why certain indus-tries agglomerate narrowly in one or a few regions. Ellison and Glaeser(1997) developed an index to measure geographic clustering in industries.It calibrates the extent to which clustering in an industry exceeds whatwould be expected merely on the basis of the chance location of a limitednumber of plants of unequal size. Their findings suggest that some degreeof agglomeration is the norm, but the kind of extreme clustering present inSilicon Valley is the exception. While clustering in some industries can beexplained by the uneven geographic distribution of a key input, instancesof clustering like in Silicon Valley seem to reflect a deeper process at work.Exactly what that process is and why its effects vary across industries hasbeen the object of much theorizing in recent years. Testing of the new the-ories of geography, though, has lagged behind. The object of this chapteris to review recent evidence and theorizing on the evolution of a selectgroup of new industries to probe the determinants of the geographic struc-ture of industries.

Modern theories of geography feature the influence of agglomerationeconomies on the location of producers. Such economies can derive fromthe sharing of inputs whose production involves increasing returns, labormarket pooling that facilitates a better match between the needs of firmsand the skills of workers, and spillovers of knowledge that are mediated bydistance (see Marshall, 1920). Other mechanisms, such as firms locatingcloser to demanders to economize on transportation costs, can alsoinduce agglomeration (Krugman, 1991). All of these benefits impart a self-reinforcing character to agglomerations. The more firms in an area then thegreater agglomeration benefits, and the greater such benefits then the morefirms will be drawn to an area and the better firms in the area will perform.

69

Congestion costs, in the form of higher land prices and compensating wagedifferentials, ultimately limit the extent of agglomerations. Until that limitis reached, though, all firms located in agglomerated areas benefit from theexternalities resulting from their collective presence.

If agglomeration economies are influential, it might be expected thatindustries would agglomerate around regions where successful early entrantslocated. Such regions would initially produce more output, employ morelabor, and be subject to more innovation, all of which would contribute toagglomeration economies that would attract subsequent entrants andenhance the performance of firms located there. There has been little empir-ical investigation of the evolution of the geographic structure of new indus-tries, but Klepper (2003) argues that this is not the way either the automobileor television receiver industries evolved. Both industries were initially char-acterized by a large number of producers and then experienced sharp shake-outs and evolved to be tight oligopolies. The automobile industry becamefamously agglomerated around Detroit, MI even though production in theDetroit area was initially negligible and early entry provided a decided com-petitive advantage (Klepper, 2001). Television producers were initiallyheavily concentrated in just three cities: New York, Chicago and LosAngeles. Even though early entry was also advantageous in TVs (Klepper,2002a), over time New York and Los Angeles lost all their producers andChicago did not grow, causing the industry to become more dispersed overtime.

Klepper (2003) advanced a hypothesis based on the ideas of organiza-tional birth and heredity to explain the evolution of the geographic struc-ture of both industries. Subsequently Buenstorf and Klepper (2005a,2005b) explored the evolution of the geographic structure of the pneumatictire industry, which was also famously agglomerated around a single city,Akron, Ohio. Similar to autos and television receivers, initially many firmsproduced tires and then the industry experienced a sharp shakeout andevolved to be a tight oligopoly. Unlike automobiles, the industry was con-centrated around Akron from its outset, and over time the agglomeration ofthe industry there grew. Buenstorf and Klepper (2005a, 2005b) investigatedthe extent to which agglomeration economies influenced the location andperformance of tire firms. They concluded that it was not primarily agglom-eration economies but similar forces to those operating in the automobileindustry that caused the industry to become so heavily agglomerated.

We review the evolution of the market and geographic structure of theautomobile, television receiver and tire industries in order to gain insightsinto the primary forces governing the agglomeration of industries.1 Webegin with television receivers, which is the easiest to understand andprovides a useful backdrop for the automobile and tire industries. Next we

70 Industrial dynamics

consider the evolution of the automobile industry, followed by the evolutionof the tire industry. We conclude with observations about the importance oforganizational heredity and birth in shaping industry agglomeration.

2. TELEVISION RECEIVERS

The annual number of entrants, exits, and producers of television receiversin the United States over the 1946–89 period based on listings in TelevisionFactbook (Warren Publishing Company) is presented in Figure 4.1. A totalof 177 firms entered the industry, most of them by 1951. Experimental tele-vision systems were introduced prior to the Second World War, but the wardelayed the start of the industry until 1946. RCA and DuMont were thefirst firms to begin producing in 1946. Many firms followed soon after,reflecting the rapid growth in the sales of television receivers. Entry peakedin 1948 at 54 firms, and by 1955 entry was negligible. The number of firmsrose from 1946 to 1949, reaching a peak of 105 in 1949, and then fellsharply. International competition, initially from Japan, began in the late1960s when roughly 30 US-based producers were left in the industry. Atthat point RCA and Zenith were the top two US producers of televisionreceivers, accounting for 39 per cent of US sales of black and white TVsand 48 per cent of the sales of color TVs, and the four-firm concentration

The evolution of geographic structure in new industries 71

Source: See Klepper (2003).

Figure 4.1 Entry, exit and number of producers in the television industry,1946–1989

0

20

40

60

80

100

120

1940 1950 1960 1970 1980 1990 2000

Producers EntrantsExits

ratios in black and white and color TVs were 61 and 65.5 per cent, respec-tively. International competition mounted over time and the number ofUS-based firms fell steadily. By 1989 only three US producers were left inthe industry, all of which were destined to exit within a short period.

Klepper (2003) analysed the location of the TV producers, which washeavily concentrated in three US cities: New York, Chicago and LosAngeles. Although these three cities accounted for only 15 per cent of theUS population, 73 per cent of television producers entered in the threecities, with 44 per cent entering in New York, 15 per cent in Chicago and 14per cent in Los Angeles. Figure 4.2 presents the annual percentage of tele-vision producers based in each of these three cities from 1946 to 1989. NewYork initially contained over 50 per cent of the television producers, butover time this percentage declined sharply. By 1970 New York’s share haddeclined to 20 per cent, and by the end of the 1970s no firm was based inNew York. Firms were slower to enter in Los Angeles, but by the mid-1950s, 20 per cent of the producers were located there. Subsequently theshare of producers in Los Angeles fell sharply, and by the mid-1970s nofirm was based in Los Angeles either. Chicago accounted for around 25 percent of television producers in the initial years of the industry. It main-tained its share through about 1980, after which its share increased sharplyas the number of firms dwindled from eight to three. Thus, at the start ofthe industry, television producers were heavily concentrated in three cities,but from the mid-1950s until 1980 the collective share of producers in thethree cities declined from 70 to 25 per cent.

72 Industrial dynamics

Source: See Klepper (2003).

Figure 4.2 Percentage of television producers in New York, Chicago andLos Angeles, 1946–1989

0%

20%

40%

60%

80%

100%

120%

1940 1950 1960 1970 1980 1990 2000

% NY producers

% Chicagoproducers% LA producers

The evolution of the location of the television producers was greatlyinfluenced by the location of firms in the radio industry. Entry, exit and thelocation of radio producers was reconstructed from annual listings of radioproducers in Thomas’ Register of American Manufacturers (ThomasPublishing Company; see also Klepper and Simons, 2000; Klepper, 2003).At the start of the television industry, 266 US firms produced radios. Theywere heavily concentrated in New York, Chicago and Los Angeles, whichaccounted for 33, 15 and 7 per cent, respectively, of all radio producers in1945–48. Of the 177 television entrants, 58 or approximately one-thirddiversified from the radio industry, and nearly all began producing tele-vision receivers where they produced radios. Thus, it is not surprising thatamong the 58 diversifiers, 55 per cent of them located in New York,Chicago and Los Angeles, which mirrors the fraction of radio producers inthe three cities. Perhaps more surprising is that among the remaining 119entrants, 82 per cent also located in these same three cities, especially inNew York and Los Angeles, which accounted for 53 and 18 per cent, respec-tively, of these entrants (Klepper, 2003).

Klepper and Simons (2000) demonstrated that the radio producers thatdiversified into the television industry tended to be the largest and mostexperienced ones, and they tended to enter earlier than other entrants intothe TV industry. They also found that the diversifiers from the radio indus-try had much lower hazards of exit at all ages than non-radio diversifiers,and among the radio diversifiers, the larger ones had much lower hazardsat all ages. Indeed, 13 of the top 14 television producers over the history ofthe industry were diversifiers from the radio industry, and four of the topfive television producers were among the top five radio producers as of1940 (the other radio producer among the top five in 1940 was among thetop 10 TV producers).

The location of the leading radio producers was the dominant forceshaping the location of TV producers in the long run and the evolution ofthe geographic structure of the industry. Only one of the top radio pro-ducers was located in New York, and it accounted for only 11 per cent ofthe sales of the top radio producers as of 1940, and Los Angeles had noleading radio producer as of 1940. With the leading radio producers ulti-mately dominating the television industry, New York and Los Angeles weredestined to experience a sharp decline in their share of television produc-ers as the industry proceeded through its shakeout. Chicago had five of thetop 16 radio producers that jointly accounted for 38 per cent of the sales ofthe leading radio producers as of 1940. Correspondingly, Chicago hadthree of the top 10 television producers and maintained its share of tele-vision producers over time. The other leading radio producers were scat-tered throughout the Northeast and Midwest. Many of these firms,

The evolution of geographic structure in new industries 73

including RCA, Philco and GE, became leading television producers. Theysurvived much longer than other firms, and as a result the base location oftelevision producers became increasingly dispersed throughout theNortheast and Midwest as the industry evolved. In a statistical analysis offirm hazard rates, New York and Los Angeles firms had higher hazards ofexit than firms located elsewhere, but once the background and time ofentry of firms was controlled, there were no significant differences in thehazard rates of firms by region (Klepper, 2003).

International competition further contributed to a geographic dispersalof television production. While US firms maintained their base locations,they increasingly moved their operations into low-wage countries in orderto counter foreign competition (Levy, 1981, pp. 261–78). But this did nothead off their demise. They were behind the technological frontier, es-pecially regarding the use of semiconductor technology. They lost much oftheir market share to Japanese firms that had pioneered the use of semi-conductor components in radios and that were consistently ahead of theUS firms in the use of semiconductor components in television receivers(La France, 1985).

Television receivers illustrate a few themes that are pertinent to auto-mobiles and tires. First, firms in related industries are important seeds forfirms in a new industry, and their location is an important determinant ofwhere entrants into the new industry locate. Second, the pre-entry experi-ence of firms has a profound effect on their ability to compete. In televisions,experience in radios was so significant that no new firm was successful in theindustry over the long term. Consequently, over the long run the base lo-cation of the leading radio firms was the dominant influence on the locationof television producers. Third, agglomeration economies were not a majorfactor shaping the base location of television producers. Two of the threeregions where firms were concentrated declined over time, and regionaldifferences in firm performance were largely due to differences in their pre-entry experience rather than any influence of the regions themselves. Finally,as the number of US firms declined, the leading firms increasingly movedtheir production into lower-wage areas, further dispersing production.

3. AUTOMOBILES

The annual number of US entrants, exits and producers of automobilesfrom the start of the industry in 1895 through 1966 based on a list compiledby Smith (1968) is presented in Figure 4.3. Only firms that produced a non-negligible number of automobiles are included, which encompassed 725firms through 1966 (see Klepper, 2002a). In contrast to televisions, entry

74 Industrial dynamics

was initially low, reflecting the limited demand for automobiles when theywere introduced. Subsequently entry grew, peaking at 87 firms in 1907.Entry remained high for the next three years and then declined to anaverage of 15 firms per year for the next 12 years, after which it became neg-ligible. The number of firms peaked at 272 in 1909, after which it fellsteadily despite average annual output growth of over 20 per cent duringthe next 15 years. By 1941 only nine firms were left in the industry. As of1911, the top two firms in the industry, Ford and General Motors,accounted for 38 per cent of the output of automobiles. They increasedtheir joint market share to over 60 per cent by the 1920s and with Chrysler,which emerged out of two early entrants in the 1920s, they jointlyaccounted for over 80 per cent of the output of the industry by the 1930s.

The industry became famously agglomerated around Detroit, MI, butinitially no firm produced a non-negligible number of automobiles in theDetroit area. Figure 4.4 reports the annual percentage of producers basedin the Detroit area from the start of the industry through 1941, when onlynine firms were left in the industry.2 No producer was located in the Detroitarea until 1901, when Olds Motor Works began production in Detroit andLansing, MI. Olds was the first great firm in the industry. After Olds’ entrythe percentage of automobile producers in the Detroit area steadily roseinto the 1910s, when it peaked at over 20 per cent. It then fell back a littlebut rose again after 1920, exceeding 50 per cent by 1941. The firms based

The evolution of geographic structure in new industries 75

Source: See Klepper (2003).

Figure 4.3 Entry, exit and number of producers in the automobileindustry, 1895–1966

0

50

100

150

200

250

300

1880 1900 1920 1940 1960 1980

Producers

Entrants

Exits

in the Detroit area were extraordinarily successful. By the mid-1910s theyproduced 13 of the 15 leading makes of automobiles, and over 60 per centof automobiles were produced in Michigan, nearly all in the Detroit area.Although 69 producers entered the industry in 1895 to 1900 before any pro-ducer entered in the Detroit area, Detroit nonetheless became the capitalof the US automobile industry by the mid-1910s, and it maintained itshegemony for many years thereafter (Klepper, 2001).

Entry was far more dispersed geographically than in televisions.Michigan had more entrants than any other state, but it accounted for only18.6 per cent of the 725 entrants through 1966, followed by New York with13.5 per cent and Ohio with 12.3 per cent. Thus, the top three statesaccounted collectively for 44.7 per cent of the entrants whereas the topthree cities in televisions accounted for 73 per cent of the entrants. Thislargely reflects that the leading seeding industries for automobiles were con-siderably more dispersed geographically than the radio industry. Klepper(2001) identified firms that diversified into autos or were founded by anindividual who headed a firm in another industry based on the listings inSmith (1968) and the brief histories of automobile firms in Kimes (1996).The industry from which the greatest number of these two types of entrantscame was carriages & wagons. In a statistical analysis of the location ofautomobile entrants, Klepper (2003) found that states with more carriage& wagon production not only had more entrants originating from the

76 Industrial dynamics

Source: See Klepper (2003).

Figure 4.4 Percentage of automobile producers in the Detroit area,1895–1941

0

10

20

30

40

50

60

70

1880 1900 1920 1940 1960

% Detroit producers

carriage & wagon industry but also more of other types of entrants as well.Unlike radio producers, which were concentrated in three cities, the car-riage & wagon industry was dispersed throughout the Northeast andMidwest, with the top three states accounting for only 32 per cent of all car-riage & wagon producers. Moreover, among US states Michigan was ninthin terms of carriage & wagon producers and fourth in terms of value of car-riage & wagon production. The second most important seeding industryfor automobiles was bicycles, which was also dispersed throughout theNortheast and Midwest, and few bicycle firms were located in Michigan.Consequently, the geographic dispersion of entrants and the slow start ofthe industry around Detroit were predictable.

Similar to televisions, automobile entrants that diversified from otherindustries, particularly carriages & wagons, bicycles and engines, had lowerhazards of exit at all ages, as did new firms founded by individuals whoheaded firms in these industries (Klepper, 2001). But diversifiers were farless important in automobiles than televisions. Whereas 33 per cent of theentrants into the television industry diversified from the radio industry,only 16.6 per cent of entrants into the automobile industry were diversifiersfrom any industry (Klepper, 2003). In large part this reflects the novel tech-nological challenges faced by automobile firms. Automobiles soon requiredprecision manufacturing to produce interchangeable parts, manufacturingwas done on an unprecedented scale, and technological progress was farmore rapid than had occurred in carriages & wagons and other relatedindustries. Consequently, experience in related industries was much lessvaluable in automobiles than in televisions.

This opened up opportunities for new firms, especially firms with one ormore founders that previously worked for an incumbent automobile firm,which are called ‘spin-offs’. Klepper (2001) identified the spin-off entrantsand the firms their founders previously worked for, dubbed their parents,based on the brief firm histories in Kimes (1996). Approximately 20 percent of entrants into automobiles were spin-offs, most of which werefounded by top managers or heads of incumbent firms. At their peak in1916, spin-offs accounted for 11 of the 15 leading makes of automobiles.Nearly all of these spin-offs descended from the leading automobile pro-ducers in the sense that their founders had worked for one of these firms(Klepper, 2001, 2005). Statistical analyses indicated that the annual likeli-hood of a firm having employees leave to start spin-offs was greater forbetter-performing firms, and on average better-performing firms hadbetter-performing spin-offs (Klepper, 2001). One explanation for these pat-terns is that leading incumbent firms provided a superior venue for employ-ees to learn about organizational best practices, especially top employees.Top firms were also magnets for talented individuals, which is another

The evolution of geographic structure in new industries 77

possible reason why their spin-offs performed so well. Spin-offs formed forvarious reasons. Among the top firms, many of them arose from internaldisputes about strategy and technology, reflecting control struggles thatwere common in the early years of the industry (Klepper, 2005).

In large part because of the influence of Olds Motor Works, spin-offsplayed a key role in the concentration of the industry around Detroit. OldsMotor Works had been a successful producer of steam and gasolineengines before it entered the automobile industry. Its manufacturing andmarketing experience enabled it to become the first firm to sell over 1000automobiles in a year, selling more than 5000 by 1904. Olds subcontractedall of its parts, which involved orders of unprecedented size, to variouslocal firms, providing its subcontractees with invaluable experience. Two ofthese firms were instrumental in the formation and success of Cadillac andFord Motor Co., both of which were located in Detroit Another one ofOlds’ subcontractors initially financed Buick, which was located in Flint,MI near Detroit. Buick was the cornerstone of the later merger that formedGeneral Motors. This same contractor also co-founded another successfulfirm, Maxwell-Briscoe, which later evolved into Chrysler.

Olds Motor Works, Cadillac, Ford Motor Co. and Buick/GeneralMotors were among the most successful early automobile producers, andthey collectively unleashed a spin-off juggernaut that propelled Detroit tobecome the automobile capital of the United States. They were the mostprolific parents in the industry, reflecting the greater rate of spin-offsamong the better firms. Olds had more descendants than any other firm inthe industry, and in total 41 firms descended from Olds, Cadillac, Ford andBuick/General Motors. These firms mainly located in the Detroit area,reflecting the general tendency for spin-offs to locate close to their parents(Klepper, 2001). Together Olds, Cadillac, Ford and Buick accounted for 11of the 13 spin-offs that produced leading makes of automobiles after 1903,with each firm spawning at least two of these spin-offs. Consequently, bythe mid-1910s nearly all the leading makes of automobiles were made byfirms based in the Detroit area. With the leading makes accounting forover 80 per cent of the output of the industry, Detroit firms totally domi-nated the industry. Indeed, what distinguished Detroit was primarily itsspin-offs, which accounted for 48 per cent of the entrants in Detroit versusonly 15 per cent of the entrants elsewhere. Moreover, spin-offs in Detroitgreatly outperformed spin-offs elsewhere, whereas the rest of the entrantsin Detroit performed comparably to their counterparts elsewhere. In a stat-istical analysis, the superior performance of firms in Detroit was confinedto its spin-offs, and their superior performance in turn was largely attrib-utable to their superior heritage rather than being located in Detroit(Klepper, 2001).

78 Industrial dynamics

The leading firms remained based in Detroit, but over time they con-ducted more of their business outside of Detroit as they established branchassembly plants throughout the United States. It was much cheaper to shipparts rather than a finished car. Consequently, if a firm had sufficient outputto accommodate multiple plants of minimum efficient size then it madesense to build branch assembly plants closer to the market. Ford, the largestproducer in the 1910s, was the first to build branch assembly plants in the1910s. It was followed by General Motors in the 1920s and later Chryslerand two of the other large automobile firms in the 1930s (Rubenstein, 1992).While this caused auto production to become more dispersed over time,Michigan still accounted for over 40 per cent of the output of automobilesas of 1931, and the leading firms remained based in the Detroit area.

Some of the lessons that emerge from automobiles are similar to TVs.Like TVs, firms in related industries, such as Olds, were important seeds forthe new industry. Also like TVs, there was enormous heterogeneity inentrants in terms of their pre-entry experience that persistently affectedtheir performance. The key difference between autos and TVs was thatspin-offs were competitive with, if not superior to, diversifiers. Thisreflected both the limited relevance of prior industries to autos and poss-ibly the distinctive opportunities within firms, particularly the leadingfirms, for high-level employees to learn valuable tacit organizational knowl-edge that they could apply to their own firms.

With better firms having higher spin-off rates and better-performingspin-offs, the spin-off process effectively led to a build-up of firms and activ-ity around the leading firms in the industry. This was especially potent inautos because of the location of four of the most successful early firms inone narrow region, fueling a great agglomeration of activity there. The fourfirms were connected through Olds, which was the catalyst for the agglom-eration of the industry around Detroit. But the other three were importantparents of spin-offs, and their creation near the industry leader addedanother random element to the agglomeration process that could helpexplain why agglomerations as extreme as autos are rare. Indeed, whileDetroit was part of the manufacturing belt dating back to the 1860s, it washardly the most likely place for the automobile industry to agglomerate. Itsdevelopment was largely attributable to the influence of Olds Motor Worksand the inherent randomness in the location of any one firm.

Agglomeration economies from locating near other producers do notappear to have been a major factor in causing the industry to agglomeratearound Detroit. Indeed, the establishment of branch assembly plantsthroughout the US by the leading firms is indicative of the potential dis-advantages of the leading firms clustering in one area. It is notable that ittook the industry leaders to exploit the advantages of assembling cars

The evolution of geographic structure in new industries 79

outside of Detroit. This is indicative of the difficulty of imitating theleaders of the industry from afar, no doubt in part due to the tacit knowl-edge the leaders possessed. It is also suggestive of why the leading firms hadmore and better spin-offs – their high-level employees had access to valu-able tacit knowledge about how to structure their own firms.

The evolution of the market structure of the automobile industry mayhave influenced the evolution of its geographic structure, though notdirectly. The TV industry also experienced a shakeout and evolved to be anoligopoly, yet its geographic structure evolved in an opposite direction toautos. Moreover, the hegemony of Detroit was established before theshakeout in the automobile industry began. It is possible, though, that theeventual drying up of entry that characterized the industry after 1925 elim-inated a force that could have unseated the leaders and conceivably reducedthe concentration of the industry around Detroit.

4. TIRES

The annual number of entrants, exits and producers of tires in the UnitedStates over the 1905–80 period based on listings in Thomas’ Register ofAmerican Manufacturers is presented in Figure 4.5. With the initial demandfor automobiles limited, the demand for tires was initially modest and entrystarted out low. Subsequently it grew for many years, peaking in the early1920s before falling off sharply and becoming negligible by 1930. A total of533 firms entered the industry through 1930, after which no significant firmentered. The number of firms peaked in 1922 at 278 and then went througha long shakeout despite robust output growth interrupted only by the GreatDepression. Only 51 firms were left in the industry in 1940, and by 1970only 24 firms were still in the industry. The industry evolved to be a tightoligopoly dominated by Goodyear, Goodrich, Firestone and U.S. Rubber(Uniroyal). Together these four firms accounted for over 53 per cent of theoutput of the industry in 1926, which they increased to over 70 per cent by1933 and then subsequently maintained (Klepper, 2002a).

Similar to automobiles, the tire industry became heavily concentratedaround a single city, Akron, OH, located in the northeastern part of Ohionear Cleveland. Figure 4.6 reports the annual percentage of 1930 and earlierentrants that were located in Ohio from 1906 to 1980. For the first 25 yearsor so Ohio generally accounted for between 20 and 30 per cent of all pro-ducers, but after 1930 the percentage of firms in Ohio rose steadily and by1959 it exceeded 50 per cent. Firms in Ohio, especially around Akron, weredistinctly successful, and by 1935 over 65 per cent of the output of tires wasproduced in Ohio (Buenstorf and Klepper, 2005b). Much of this output was

80 Industrial dynamics

produced by Goodyear, Goodrich and Firestone, all of which were based inAkron. But firms in northeastern Ohio also dominated the next cadre offirms. As of 1920, six of the next 20 largest firms were located in Akron, andfour others were located nearby in northeastern Ohio (ibid.).

The evolution of geographic structure in new industries 81

Source: See Buenstorf and Klepper (2005b).

Figure 4.5 Entry, exit and number of producers in the tire industry,1901–1980

0

50

100

150

200

250

300

1880 1900 1920 1940 1960 1980 2000

Producers

Entrants

Exits

Source: See Klepper and Buenstorf (2005b).

Figure 4.6 Percentage of tire producers in Ohio, 1906–1980

0

10

20

30

40

50

60

70

1900 1920 1940 1960 1980 2000

% Ohio producers

Like automobiles, Ohio had more tire entrants than any other state, butit accounted for only 24 per cent of all the entrants through 1930, followedby New York with 15 per cent, New Jersey with 14 per cent, Pennsylvaniawith 8 per cent and Illinois with 7 per cent. Klepper (2002) and Buenstorfand Klepper (2005b) identified the entrants into tires that diversified fromanother industry, which in most cases was the rubber industry. Similar toTVs and autos, in a statistical analysis Buenstorf and Klepper (2005a)found that states with more rubber producers had more tire entrants thatwere diversifiers and also more of other types of entrants, and within Ohiocounties with more rubber producers had more diversifying entrants.Similar to Michigan and autos, Ohio was not the leading state in terms ofrubber producers, but was fifth in 1890 with 3.5 per cent of US rubber pro-ducers. While diversifiers had lower hazards of exit on average than othertypes of entrants, diversifiers accounted for only 15.6 per cent of allentrants, similar to automobiles. In part, this reflects that automobile tiresrepresented a considerable break from prior rubber products. Bicycle tiresdid not readily scale to automobiles, tire manufacturing was much morecomplex than other rubber products, and tires were subject to much moretechnological change than other rubber products. Thus, like automobilesopportunities existed for regions that were not well stocked with firms inrelated industries.

Within Ohio, the most important rubber producer at the start of the tireindustry was BF Goodrich, which was located in Akron, where the (limitednumber of) rubber producers in Ohio were concentrated. Goodrich was aleading bicycle tire producer and successful producer of other rubber prod-ucts, and like Olds Motor Works it was an important catalyst for the indus-try in Akron. It produced the first pneumatic automobile tire in 1896 andimmediately became one of the leading producers of tires. Goodrich wasinfluential in four other early tire firms locating and prospering in Akron –Diamond Rubber, which merged with Goodrich in 1912, Kelly-Springfield,Firestone and Goodyear. Diamond was a 1894 rubber spin-off fromGoodrich. Goodrich produced Kelly-Springfield’s initial carriage tirebased on a patented design before Kelly-Springfield initiated the produc-tion of automobile tires in Akron in 1899. Goodrich also initially producedtires for Firestone after its entry in Akron in 1900 and then suppliedFirestone with prepared rubber and fabric when it began producing its owntires in 1903. Finally, Goodyear was founded in 1898 by the son of one ofthe original financiers of Goodrich that subsequently also operated arubber firm (Buenstorf and Klepper, 2005b).

With five leading firms located in Akron early on, the stage was setfor spin-offs to play a key role in the further development of the industryaround Akron. Buenstorf and Klepper (2005a) traced the origin of the

82 Industrial dynamics

126 firms that entered in the state of Ohio through 1930. Like Detroit, spin-offs accounted for a disproportionate share of the entrants that originatedfrom the Akron area – 58 per cent of the 36 entrants that originated inSummit County (the home of Akron) were spin-offs versus 35 per cent ofthe other entrants originating elsewhere in Ohio. Most of them wereformed by high-level employees, similar to the automobile industry.Furthermore, the bulk of the spin-offs that originated in Ohio entered ineither the same or a contiguous county to where their employer was located(ibid.). A statistical analysis of the rate at which employees left Ohio firmsto form spin-offs revealed that the highest spin-off rate among Ohio pro-ducers occurred in the leading Akron firms, followed by the next tier ofleading producers in Ohio (ibid.). In an analysis of firm performance(Buenstorf and Klepper, 2005b), firms located in the Akron area had lowerhazards than firms located in the rest of Ohio and outside of Ohio. Similarto Detroit, the distinctive performance of the firms in the Akron area wasconfined to the spin-offs located there. Furthermore, among all spin-offs inOhio, those that descended from the leading firms or the second tier ofleading producers had lower hazard rates, suggesting that the superior per-formance of the Akron spin-offs was largely attributable to their heritage.

Buenstorf and Klepper (2005a) traced where entrants in Ohio origi-nated, which for diversifiers was where they previously produced, for spin-offs where their parents were located, and for start-ups where their founderspreviously worked. Not only did spin-offs tend to locate in or close to theircounty of origin, but so did diversifiers and other start-ups. In a statisticalanalysis of the county where entrants located given their county of origin,Buenstorf and Klepper (2005a) found that the number of tire producersand the population of an entrant’s county of origin did not positivelyinfluence its likelihood of entering there. However, these same characteris-tics influenced whether an entrant located in a distant county. Figueiredoet al. (2002) found similar patterns for modern Portuguese entrepreneurialstart-ups. One interpretation of these findings is that entrants have valuableknowledge about their home region, such as where to find labor, input sup-pliers, transportation and even sources of knowledge spillovers, but theylack this knowledge about other regions. Consequently, even if their homeregion is not well stocked with firms in their industry and other industriesand local markets for labor, inputs and so on are thin, they still know whereto secure their needs. Without this knowledge about other regions, theywould be better off locating in regions with more firms in their industry andin other industries because such regions would have better developed localmarkets to supply their needs.

While the entrants in Ohio tended to locate near their geographic roots,when they established branch plants they tended to locate these away from

The evolution of geographic structure in new industries 83

their base location, similar to autos. In the 1920s, the leading tire produc-ers established branch manufacturing plants throughout the United Statesto save on transportation and labor costs (Jeszeck, 1982). This intensifiedafter 1935 due to increasing militancy on the part of the union represent-ing tire workers (ibid.), causing the share of tire production in Ohio todecline. Similar to the automobile industry, it was the leading firms thatwere in the vanguard of exploiting the advantages of more remote areas.Their willingness to set up plants outside of Ohio is suggestive of thelimited advantages of locating in Akron. Consistent with this, Akron wasnot a major draw for either start-ups or spin-offs that originated elsewhere,and a number of spin-offs that originated in Akron did not locate there(Buenstorf and Klepper, 2005a).

The lessons from the tire industry closely parallel those from autos. Firmsfrom the rubber industry, the most closely related industry, were importantseeds for tire entrants, but spin-offs were also significant competitors. Therewas great heterogeneity among entrants in terms of their pre-entry experi-ence that persistently affected their performance. One firm was a key cata-lyst for activity around Akron, both through its effects on other early Akronproducers and through the spin-offs that it and the other successful Akronproducers disproportionately spawned. As a consequence, the industrybecame extremely agglomerated around an unlikely region, reflecting boththe randomness in the location of any one firm and the unlikely combin-ation of early firms in one narrow region that was critical to the extremeagglomeration of the industry there. Agglomeration economies did notappear to play a major role in the agglomeration of the industry aroundAkron, and branching by the leaders eventually reduced the agglomerationof the industry there. The evolution of the market structure of the industry,particularly the eventual drying up of entry after 1930, may have con-tributed to the geographic concentration of the industry, but this concen-tration was established well before the industry underwent a shakeout.

5. OBSERVATIONS

Various themes emerge from the study of the three industries regarding theevolution of the geographic structure of new industries.

The Location of Firms in Related Industries Influences Where EntrantsLocate

In all three industries, the location of firms in related industries influencedthe location of entrants into the new industry. This was most apparent in

84 Industrial dynamics

televisions, where both diversifiers and other entrants concentrated in thethree cities where the radio firms were clustered. In both autos and tires,regions with more firms in related industries also had more diversifiers andother entrants. But firms in related industries were more dispersed in autosand tires than TVs, so entrants were more dispersed in these two industriesthan TVs. The radio industry may also have had more influence on the lo-cation of TV producers than any related industry had on autos and tiresbecause of the greater overlap between radios and TVs than any producthad with either autos or tires. This was reflected in the much higher frac-tion of entrants that were diversifiers (from the radio industry) in TVs thanautos and tires.

The influence of related industries on entry into TVs, autos and tires sug-gests two points. First, firms need competence to compete in a new indus-try, and one source of that competence is experience in related industries.Indeed, the fact that diversifiers in all three industries had lower hazards ofexit on average than other entrants suggests that experience in a relatedindustry was an important source of competence in all three industries.Second, diversifying entrants do not venture far geographically from theirroots, which also appears to have been the case for spin-offs and other start-ups. Buenstorf and Klepper’s (2005a) findings concerning the location ofOhio tire entrants suggests that entrants locate close to their roots to exploitvaluable local knowledge they possess based on their pre-entry experience.Consequently, an important determinant of regional entry into a newindustry is the stock of local firms that could provide the competenceneeded to succeed in the new industry.

Incumbents Can Also be Important Sources of Competence

Just as firms in related industries appear to be an important source of com-petence for a new industry, in autos and tires incumbent firms also appearto have been an important source of such competence, especially theleading incumbents. The leading firms had higher rates of spin-offs, and onaverage their spin-offs were better performers than spin-offs from lesserfirms. Moreover, their spin-offs were certainly competitive with if not su-perior performers to diversifiers from related industries, suggesting that theleading incumbent firms were also an important source of competence forentrants. The superior performance of spin-offs from the leading firmscould reflect that these organizations had more to pass down to offspring.Alternatively, it could reflect that better firms attracted better managerialtalent and more talented individuals founded superior firms.

Judging from the dominance of the industry by diversifiers from the radioindustry, spin-offs were not competitive in TVs. Two factors may have been

The evolution of geographic structure in new industries 85

at work. Radios and TVs overlapped considerably in terms of technologyand marketing whereas autos and tires represented a greater break frompast products. Consequently, diversifiers from radios may have had a greateradvantage in TVs than any kind of diversifier had in autos and tires, limit-ing the opportunities for new firms in TVs relative to autos and tires.Second, demand initially grew much faster in TVs than autos and tires,which may have limited opportunities for later entrants of all kinds, includ-ing spin-offs. Klepper (2002a) developed a model of shakeouts in whichearlier entrants have a head start in building up a market for their products,which enables them to apply their research and development over a largeroutput, providing them with a competitive advantage. Spin-offs naturallyenter later because they require a gestation period, in the form of employ-ees gaining experience in incumbent firms. Consequently, they will be at agreater disadvantage in markets in which demand initially grows rapidly, asoccurred in TVs relative to autos and tires. Consistent with this, entrybecame negligible within 10 years of the start of the TV industry whereas itcontinued much longer in autos and tires.

The Spin-off Process Can Induce Agglomerations around Successful Firms

In both autos and tires, better firms had higher spin-off rates. Spin-offs (andother entrants) did not venture far from their geographic roots, so entry wasgreater around successful firms. The spin-offs of successful firms also per-formed better than other spin-offs and were competitive with, if not su-perior to, diversifiers from related industries in autos and tires.Consequently, over time activity built up around successful early produc-ers, especially in Detroit and Akron, where successful early auto and tireproducers were concentrated.

Entry in Detroit and Akron was disproportionately composed of spin-offs and it was spin-offs in both regions that performed distinctly well, sug-gesting that the spin-off process alone can give rise to agglomerations. Onthe other hand, it is conceivable that the agglomerations in both Detroitand Akron were driven by agglomeration economies, which could give riseto the same patterns of entry being concentrated in agglomerated areas andfirms in agglomerated areas performing better than firms elsewhere. But ifagglomeration economies were the primary cause of the agglomerations inautos and tires, then entrants of all kinds should have been attracted toDetroit and Akron. Moreover, entrants of all types should have performedbetter in Detroit and Akron than their counterparts elsewhere. Yet in bothareas entry was disproportionately composed of spin-offs and only spin-offs performed better than their counterparts elsewhere. Furthermore,nearly all the spin-offs in Detroit and Akron had parents located there and

86 Industrial dynamics

so were not drawn from other regions. Judging from Buenstorf andKlepper’s (2005a) findings for tires, spin-offs (and other types of entrants)that originated from agglomerated regions were also no more likely tolocate in their home region than spin-offs that originated elsewhere. Thus,agglomeration economies do not appear to have played a major role in fos-tering the agglomerations in either autos or tires.

The TV industry is also instructive about the power of spin-offs versusagglomeration economies to generate agglomerations. Agglomerationeconomies would have been expected to operate as strongly in TVs as autosand tires. Yet despite the extraordinary concentration of entrants in threenarrow areas, the TV industry became less agglomerated over time. Whatappears to have been missing was a spin-off process that generated firmsthat were competitive with the leading diversifiers. This further suggeststhat the key to the agglomerations of the auto and tire industries was thespin-off process and not agglomeration economies.

The Spin-off Process Can Magnify an Early Cluster of Leading Firms intoan Extraordinary Agglomeration

Within the first 10 years of the auto and tire industries, most of the leadingproducers in autos and tires were present in Detroit and Akron. The settingwas ripe for the spin-off process to magnify the initial cluster of leadingfirms in each region, and this is precisely what occurred. Consequently, overtime the percentage of firms and activity around Detroit and Akronincreased and both regions evolved to account for over 60 per cent of activ-ity in their industries. While establishment-level data are not available tocompute the Ellison–Glaeser (1997) index of geographic concentration,conservative estimates would put both industries in the tail of manufactur-ing industries in terms of geographic concentration.

Having so many early leaders in one narrow region is surely an uncom-mon event, which would explain Ellison and Glaeser’s (1997) finding thatagglomerations as extreme as autos and tires are rare. On the other hand,even if the early leaders of an industry were not as clustered as in autosand tires, the spin-off process would still be expected to cause activity tobuild around successful early firms. Activity would still agglomerate, but itwould be dispersed across more regions than in autos and tires. This couldexplain Ellison and Glaeser’s finding that activity in most manufacturingindustries agglomerates to some degree, but generally much less so than inautos or tires.

In both autos and tires, the early leaders did not all locate in Detroitand Akron by chance. Rather, one firm in each region – Olds Motor Worksin autos and BF Goodrich in tires – played a key role in other successful

The evolution of geographic structure in new industries 87

producers being located nearby. Both firms were fertile sources of spin-offs,but their influence was broader. Olds provided valuable experience to itslocal subcontractors, some of whom later entered or financed the venturesof others, while Goodrich supplied inputs and sometimes initially manu-factured tires for local tire firms. In both industries, input markets were notyet well developed, and so expertise accumulated in one firm redounded tothe benefit of others nearby. This is a form of spillover, but it is of somewhatdifferent variety than the externalities featured in modern theories ofagglomeration economies. It is restricted to a small number of connectedfirms in a region and is consistent with the findings of Breschi and Lissoni(2002) concerning how technological knowledge is transmitted across firms.

The importance of a single firm helps explain how the agglomerations inautos and tires got established in an unexpected place. Neither Detroit norAkron was particularly distinguished in terms of activity in related indus-tries or anything else that would have fueled an agglomeration there. On theother hand, both regions were located in the so-called manufacturing beltand certainly had a healthy amount of activity in related industries. Aschance would have it, both ended up with a diversifier from a related indus-try that became the first outstanding performer in its industry. This was theseed for the agglomerations that emerged in both regions. With a single firmhaving so much influence on the agglomeration process, chance can play abig role in whether and where an agglomeration gets established. Again, theTV industry is instructive about the circumstances in which chance canhave such a big influence on the location of an industry. Without a strongspin-off process in TVs, no firm had the kind of influence on the locationof the TV industry as either Olds or Goodrich. Consequently, firms endedup congregating where firms in the radio industry were concentrated, lim-iting the possibility of activity clustering in an unexpected area.

Agglomerations and Shakeouts Are Not Directly Related

Clearly, the forces underlying shakeouts do not lead to agglomerations, asthe TV industry illustrates. Moreover, by the time the shakeouts began inautos and tires, Detroit and Akron were already well established as thecenters of activity in their respective industries. Thus, the shakeouts inautos and tires do not seem to have played a critical role in the agglomer-ation of either industry. But characteristic of shakeouts is the drying up ofentry, as occurred in all three industries after their shakeouts began. Thisremoved a force that potentially could have undermined the agglomer-ations that formed in autos and tires. Therefore, indirectly the agglomer-ations in autos and tires might have been promoted by the shakeouts bothindustries experienced.

88 Industrial dynamics

Ultimately, though, the forces underlying the shakeouts in both autos andtires appear to have caused both industries to become less agglomerated.Both industries were characterized by scale economies at the plant level,which no doubt led firms initially to enter with a single plant. Judging fromthe actions of the leading firms, though, the economies were not so over-whelming as to preclude the establishment of branch plants by the largestfirms in the industry. Thus, as both industries consolidated and the leaderstook over an increasing share of the industry’s output, the leading firmsestablished branch plants, which they generally located away from their baselocations to save on transportation and labor costs. With the leading firmsgenerally based in Detroit and Akron, this eventually caused the agglomer-ations in autos and tires to decline over time. The same forces also led theTV industry to become more dispersed over time as firms moved some oftheir operations offshore to take advantage of lower labor costs.

Dumais et al. (2002) found that in manufacturing industries branchplants generally are de-agglomerating in the sense that their location anduse causes employment to move away from agglomerated areas over time.They found that the main force sustaining agglomerations was the greaterlongevity of plants in agglomerated areas. Again, the findings for autos,tires and TVs are instructive about the forces possibly at work. Judgingfrom Buenstorf and Klepper’s (2005a) findings for tires, firms did notchoose their initial locations to minimize the costs of production but toexploit local knowledge they had accumulated through their pre-entryexperience. Thus, when branch plants were established, it was natural tolocate them away from where the firms were based, which was generally inagglomerated areas. Furthermore, agglomerations themselves can raise thecost of production such as by land prices being bid up, necessitating thepayment of compensating wage differentials. In tires, the agglomeration ofactivity in Akron also no doubt facilitated the union organization ofworkers, and militancy on the part of the union contributed to firms settingup branch plants elsewhere. The greater longevity of plants in agglomer-ated areas could result from a spin-off process comparable to the one thatoperated in autos and tires. Firms in agglomerated areas would be betterperformers and thus their plants would be longer lived.

Three Industries Do Not Make a General Theory

Generalizing based on three industries is certainly dangerous, but there isenough evidence from other industries to suggest that the forces at work inauto, tires and TVs are operative in other industries as well. Sorenson andAudia (2000) found that in the footwear industry, entry was more likely inagglomerated areas even though firms had higher hazards of exit in these

The evolution of geographic structure in new industries 89

areas. They interpreted this as a reflection of the natural tendency of entryto concentrate near incumbents even in the absence of agglomerationeconomies, consistent with the spin-off process in autos and tires. GordonMoore of Intel fame, along with his co-author, implicated spin-offs in thesemiconductor industry as the primary basis for the agglomeration ofactivity in Silicon Valley (Moore and Davis, 2004). The analog to Olds andGoodrich was Fairchild, whose offspring were so numerous that they weredubbed ‘Fairchildren’. Moore and Davis (2004) have a particularly inter-esting discussion of why working in an incumbent semiconductor firmprovided distinctive organizational knowledge that enabled high-level man-agers to form their own successful spin-offs.

A few studies analyse the spin-off process in specific industries withoutlinking it explicitly to geography, and their findings are also consistent withthose for autos and tires. In both the disk drive (Franco and Filson, 2000;Agarwal et al., 2004) and laser industries (Klepper and Sleeper, 2005), spin-offs performed distinctively well. In both industries more successful firmshad higher spin-off rates, which also appears to have been the case amongsemiconductor firms located in Silicon Valley (Brittain and Freeman, 1986).In disk drives, the spin-offs of more successful firms were also better per-formers, which appears to have been associated with an (involuntary) trans-fer of technology and marketing expertise from parents to their spin-offs.

Questions Abound about the Evolution of Industry Geographic Structure

The interpretation of the evolution of the geographic structure of the auto,tires and TV industries raises many questions. Firms are assumed to differfrom the outset in terms of their competence based on their pre-entry ex-perience. But what exactly does the pre-entry experience of firms providethem and how does this influence their performance not just initially, butfor many years after entry? Spin-offs played a key role in the agglomerationof both autos and tires. Why do spin-offs occur, why are they more preva-lent among the leading firms, and what drives the correlation between theperformance of spin-offs and their parents? Under what conditions do suc-cessful early entrants galvanize other firms to form and prosper nearby?More generally, what are the mechanisms that influence the transmission ofknowledge across firms in the same industry and between suppliers andproducers, and how is this mediated by geographic distance?

These are just some of the questions raised by the evolution of the threeindustries. Hopefully, further examination of the way the geographic struc-ture of new industries evolves will shed light on these questions and on thefundamental drivers of agglomerations and the geographic structure ofindustries.

90 Industrial dynamics

NOTES

* Helpful comments were provided by an anonymous referee. Support is gratefullyacknowledged from the Economics Program of the National Science Foundation, GrantNo. SES-0111429.

1. The review of the evolution of the market structure of the three industries is primarilybased on Klepper (2002a). The review of the evolution of the geographic structure of thethree industries is primarily based on Klepper (2001, 2002b, 2003, 2005) for automobiles,Klepper (2003) for TVs, and Buenstorf and Klepper (2005a, 2005b) for tires.

2. Firms established branches and moved within a 100-mile radius of Detroit. Accordingly,the market area around Detroit was defined to correspond to this 100-mile radius(Klepper, 2001).

REFERENCES

Agarwal, Rajshree, Raj Echambadi, April M. Franco and M.B. Sarkar (2004),‘Knowledge transfer through inheritance: spin-out generation, development andsurvival’, Academy of Management Journal, 47, 501–22.

Breschi, Stefano and Francisco Lissoni (2002), ‘Mobility and social networks:localised knowledge spillovers revisited’, mimeo, Bocconi University, Milan.

Brittain, Jack W. and John Freeman (1986), ‘Entrepreneurship in the semiconduc-tor industry’, mimeo, University of California, Berkley, CA.

Buenstorf, Guido and Steven Klepper (2005), ‘Regional birth potential, agglomer-ation economies, and home bias in the location of domestic entrants’, mimeo,Max Planck Institute, Jena.

Buenstorf, Guido and Steven Klepper (2005b), ‘Heritage and agglomeration: theAkron tire cluster revisited’, mimeo, Max Planck Institute, Jena.

Dumais, Guy, Glenn Ellison and Edward L. Glaeser (2002), ‘Geographic concen-tration as a dynamic process’, Review of Economics and Statistics, 84, 193–204.

Ellison, Glenn and Edward L. Glaeser (1997), ‘Geographic concentration in U.S.manufacturing industries: a dartboard approach’, Journal of Political Economy,105, 889–927.

Figueiredo, Octavio, Paulo Guimaraes and Douglas Woodward (2002), ‘Home-field advantage: location decisions of Portugese entrepreneurs’, Journal of UrbanEconomics, 52, 341–61.

Franco, April M. and Darren Filson (2000), ‘Knowledge diffusion throughemployee mobility’, Federal Reserve Bank of Minneapolis, Staff Report 272.

Jeszeck, Charles A. (1982), ‘Plant dispersion and collective bargaining in the rubbertire industry’, PhD dissertation, University of California, Berkeley.

Kimes, Beverly R. (1996), Standard Catalog of American Cars, 1805–1942, 3rd edn,Iola, WI: Krause Publications.

Klepper, Steven (2001), ‘The evolution of the U.S. automobile industry and Detroitas its capital’, mimeo, Carnegie Mellon University, Pittsburgh, PA.

Klepper, Steven (2002a), ‘Firm survival and the evolution of oligopoly’, RANDJournal of Economics, 33, 37–61.

Klepper, Steven (2002b), ‘The capabilities of new firms and the evolution of theU.S. automobile industry’, Industrial and Corporate Change, 11, 645–66.

Klepper, Steven (2003), ‘The geography of organizational knowledge’, mimeo,Carnegie Mellon University, Pittsburgh, PA.

The evolution of geographic structure in new industries 91

Klepper, Steven (2005), ‘The organizing and financing of innovative companies in theevolution of the U.S. automobile industry’, in Naomi Lamoreaux and KennethSokoloff (eds), The Financing of Innovation, Cambridge, MA: forthcoming.

Klepper, Steven and Kenneth L. Simons (2000), ‘Dominance by birthright: entry ofprior radio producers and competitive ramifications in the U.S. television receiverindustry’, Strategic Management Journal, 21, 997–1016.

Klepper, Steven and Sally Sleeper (2005), ‘Entry by spinoffs’, Management Science,51, 1291–306.

Krugman, Paul (1991), ‘Increasing returns and economic geography’, Journal ofPolitical Economy, 99, 483–99.

La France, Vincent A. (1985), ‘The United States television receiver industry’, PhDdissertation, Pennsylvania State University, University Park, PA.

Levy, Jonathan D. (1981), ‘Diffusion of technology and patterns of internationaltrade: the case of television receivers’, PhD dissertation, Yale University, NewHaven, CT.

Marshall, Alfred (1920), Principles of Economics, London: Macmillan.Moore, Gordon and Kevin Davis (2004), ‘Learning the Silicon Valley way’, in

Timothy Bresnahan and Alfonso Gambardella (eds), Building High-tech Clusters:Silicon Valley and Beyond, Cambridge: Cambridge University Press, pp. 7–39.

Rubenstein, James M. (1992), The Changing US Auto Industry, London: Routledge.Smith, Philip H. (1968), Wheels within Wheels, New York: Funk & Wagnalls.Sorenson, Olav and Pino G. Audia (2000), ‘The social structure of entrepreneurial

activity: geographic concentration of footwear production in the United States,1940–1989’, American Journal of Sociology, 106, 424–61.

Thomas Publishing Company (1905–1980), Thomas’ Register of AmericanManufacturers, New York: Thomas Publishing Company.

Warren Publishing Company, Television Factbook, vols 5–59, Washington, DC:Warren Publishing Company.

92 Industrial Dynamics

5. Constructing entrepreneurialopportunity: environmentalmovements and the transformationof regional regulatory regimesBrandon Lee and Wesley Sine*

1. INTRODUCTION

Past research on the geographic distribution of economic activity hasfocused primarily on specifying the conditions that sustain economic clus-ters rather than explaining regional variance in the conditions (that is,opportunities) that facilitate the emergence of new economic forms.Therefore, to understand when and where entrepreneurial opportunityexists, ‘theory must explain how information and resources for entrepre-neurial activity come to be disproportionately massed in some places andat some times’ (Romanelli and Schoonhoven, 2001: 41). Economic geogra-phers have begun to address this question by focusing attention on insti-tutional geography – those social, political and cultural–contextualelements that ‘enable, constrain, and refract economic development in spa-tially differentiated ways’ (Martin, 2000: 79). In this chapter, we advance theinstitutional geography agenda by investigating how differences in regionalcollective action affect state regulatory regimes and hence, opportunitiesfor economic activity.

Recent developments in organization theory that integrate social move-ment theory into accounts of the emergence, change and decline of insti-tutions (Davis and Thompson, 1994; Fligstein, 1996; Rao et al., 2000; Daviset al., 2005) provide a fruitful avenue for understanding variation in the cre-ation and existence of regional entrepreneurial opportunity. Research inboth entrepreneurship and economic geography has largely neglected therole that social movements play in generating new opportunities for entre-preneurs. While a substantial body of work has outlined how the charac-teristics of regions impact economic and entrepreneurial activity (Weber,1909; Marshall, 1922; Harris, 1954; Arrow, 1962; Piore and Sabel, 1984;Romer, 1986; Jaffe et al., 1993; Zucker et al., 1998; Almeida and Kogut,

93

1999; Sorenson and Audia, 2000; Stuart and Sorenson, 2003), there is littleresearch on how regional collective action influences geographic variationin the success of new industries and technologies. Increasingly, scholarshave shifted attention away from the characteristics and abilities of the loneentrepreneur (McClelland, 1961) and toward the demand side of entrepre-neurship (Thornton, 1999), with work centering on the impact of exoge-nous shocks or changes in the environment that stimulate opportunities fornew types of economic activity and organizational forms (Schumpeter,1934; Thornton, 1999; Sine and David, 2003). These environmental shocksor jolts provide a catalyst for action. Jolts disrupt the routines and practicesof individual organizations (Meyer, 1982) and can precipitate institutionaldecline for entire industries and organizational fields (Tushman andAnderson, 1986; Greve, 1995; Hoffman, 1999). Technological change andinnovation, deregulation, demographic shifts and changing consumer pref-erences can alter the relative value of information and resources in such away as to catalyse new types of economic activity. While these exogenousshocks are often necessary for the creation of new opportunities, they arenot sufficient in and of themselves. Extant research, however, reveals littleabout how these broader changes create and eliminate entrepreneurialopportunities and therefore this topic merits greater scholarly attention(Eckhardt and Shane, 2003).

To address this gap, we draw upon research that suggests the creation ofnew forms (and opportunities for new forms, as we argue in this chapter) isnot only a political process (Stinchcombe, 1965), but also a collective one(Fligstein, 1996; Rao et al., 2000; David et al., 2005; Davis et al., 2005).Taking this approach, we link exogenous change to the creation of entre-preneurial opportunity by examining how social actors engage in framingprocesses and the mobilization of resources (McAdam, 1996). By so doing,we clarify the mechanisms responsible for translating broader technolog-ical, economic and demographic disruptions into concrete entrepreneurialopportunities.

Work in organization theory that employs the social movement frame-work generally does so in the context of industry participants seeking tochange existing intra-industry arrangements or extra-industry constraintson their industry (for example, Fligstein, 1996; Davis and McAdam, 2000;Swaminathan and Wade, 2001). However, often, free-rider problems mayimpede collective action by industry participants and trade associationsmay not form until a later point in the industry life cycle (Rao, 2004).Taking a different approach, we suggest that broader social movementsthemselves, while potent motors for social change, also have the capacity togenerate new markets and possibilities for entrepreneurial activity. In sodoing, we answer calls to re-examine the linkages between organizational

94 Industrial dynamics

dynamics and broader societal changes (Stinchcombe, 1965; Perrow, 1986,2002; Friedland and Alford, 1991; Stern and Barley, 1996; Scott, 2001;Lounsbury and Ventresca, 2002). While the outcomes of social movementsare understudied (Giugni, 1998, 1999), scholarly consideration of theirinfluence on markets and entrepreneurial activity is even more scant (butsee Schneiberg, 2002; Lounsbury et al., 2003).

To address this shortcoming, we examine how social movementsinfluenced entrepreneurial opportunity creation in the US electrical powerindustry from 1978 to 1992. Prior to 1978, it was largely impossible forsmall, independent power producers using renewable energy technologies(RETs) to sell electricity to the US power grid. In 1978, federal legislationrequired incumbent utilities to purchase power from independent powerproducers. However, the federal government delegated all decision makingand enforcement of the particulars of these regulations to state govern-ments, opening the door for wide variation in the degree to which state-levelregulation supported independent power producers. In this chapter, weuse historical and quantitative analysis to examine the effect of stateenvironmental movement membership on regional regulatory environ-ments for RET entrepreneurs. By examining how social movements shapethe creation of regulatory environments across US states, we account forgeographic variation in entrepreneurial opportunity (Romanelli andSchoonhoven, 2001).

In the next section, we outline how social movement organizations con-struct entrepreneurial opportunity in nascent industries. We then providea brief history of the US electric power industry. This is followed by ahistorical narrative focused on how environmental social movementsinfluenced state level incentive structures for entrepreneurs interested indeveloping ventures using alternative energy technologies. We then explainthe data and methods employed and conclude with the analysis and a dis-cussion of the results.

2. THEORY

Social Movements and Entrepreneurial Opportunity

Increased scholarly attention has focused on the role that collective actionplays in the construction, change and devolution/destruction of institu-tions, technology regimes and new organizational forms (Garud and Vande Ven 1989; Davis and Thompson, 1994; Fligstein, 1996; Rao et al., 2000;Davis et al., 2005; Hargrave and Van de Ven, forthcoming). Because newforms, industries and technologies never emerge in empty social space,

Constructing entrepreneurial opportunity 95

processes of resource assembly, legitimation and integration require collec-tive action (Sine and David, 2003). Change and innovation within a field ormarket generally comes from peripheral players that engage in collectiveaction but do not have a stake in, or benefit from existing institutionalconfigurations (Leblebici et al., 1991). Collective challenges to existinginstitutions or technologies have been increasingly theorized through asocial movement lens to explain how collective efforts contest existing insti-tutions and construct new institutions and organizational forms thatembody shared interests and goals.

For example, in their account of the emergence of the organic foodindustry, Lee and Lounsbury (2005) show how local coalitions of organicfarmers collectively acted to distinguish organic from conventionally grownfood by developing and refining a process theory of how food should becultivated. By collectively organizing and embodying that theory in stan-dards and certification procedures, organic farmers successfully created anew category of products termed ‘organic’. This collective work created asolid jurisdictional boundary around the organic food category thatinfused these products with value over and above similar products that werenot labeled organic, allowing organic farmers and retailers to charge morefor their products.

In a similar vein, Schneiberg (2002) demonstrates how fire mutualsemerged as a new form to solve coordination problems that were not ade-quately addressed by either markets or hierarchies. These mutuals repre-sented a means by which property owners and agriculturalists resistedeconomic centralization and drew upon anti-company politics and agrar-ian protest to forcefully articulate and instantiate an alternative model oforganizing and managing risk in a new organizational form.

Finally, David et al. (2005) show how in the early 1900s institutional entre-preneurs publicly identified particular problems associated with the emer-gence of the corporate form (Chandler, 1990). The dramatic changes in sizeand scope of firms resulted in a new set of challenges for both owners andmanagers that institutional entrepreneurs such as Arthur Little, Edwin Boozand James McKinsey argued were best addressed by external consultantswell versed in contemporary social and natural science. These institutionalentrepreneurs and the firms and industry associations they formed workedtogether to create and proselytize problem–solution models that proposedusing external ‘consultants’ with established categories of expertise such aspsychology or chemistry for solving organizational problems. Due to the col-lective efforts of these early pioneers, within less than a decade, managementconsulting became an established part of the business landscape.

Drawing upon these studies, we suggest that localized collective actionefforts can lead to the construction of regional entrepreneurial opportunities

96 Industrial dynamics

by framing problems and solutions and constructing and utilizing mobiliz-ing structures (McAdam, 1996). We define framing as a collective attributionprocess through which people can voice their grievances, specify the under-lying logic of those grievances, and articulate and promulgate a solution (seeSnow et al., 1986; Snow and Benford, 1988; Benford and Snow, 2000).Mobilizing structures are those formal and informal vehicles by whichpeople engage in collective action (McCarthy, 1996). We argue that formalsocial movement organizations employ frames and mobilizing structures tocreate legal environments that facilitate particular types of entrepreneurialactivities.

Legal and Regulative Structure

Weber (1978) explicated the state’s role in fostering and shaping markets:‘Law can . . . function in such a manner that, in sociological terms, the pre-vailing norms controlling the operation of the coercive apparatus have sucha structure as to induce, in their turn, the emergence of certain economicrelations (667, emphasis added). Law and state regulation, through instru-mental as well as normative and cultural–cognitive means, influences thecreation and subsequent expansion of new markets (Dobbin, 1994;Fligstein, 2001; Fligstein and Stone Sweet, 2002). In their study of the rail-road industry in Massachusetts, Dobbin and Dowd (1997) found that ma-terial support from the state increased the number of foundings of railroadfirms. Similarly, Lomi (1995) shows how direct policy efforts codified in lawled to the diffusion of cooperative banks to rural areas in Italy whereprivate stock banks were not willing to do business. Wade et al. (1998)demonstrate how state governments can curb particular economic prac-tices. They find that during the Prohibition period in the United States,regulations at the state level were effective at barring in-state breweries, yetthese regulations unintentionally spurred brewery foundings in adjacentstates where these regulations did not exist. Over time, however, as thenumber of states with prohibition laws increased there was a decrease inbrewery foundings in all states.

These accounts, particularly the work by Wade and colleagues, demon-strate that policy governing economic activity has direct effects on organ-izational founding dynamics and that regulation can vary by region.However, we have little understanding of the origins and impetus for thistype of regulation and what accounts for the variation in its adoption.Attempting to specify the mechanisms of legal structure diffusion, recentwork (Schneiburg and Bartley, 2001; Strang and Bradburn, 2001; Ingramand Rao, 2004; Schneiberg and Soule, 2005) offers a more nuanced under-standing of the origins, antecedents and factors that account for regulation

Constructing entrepreneurial opportunity 97

within a market. Schneiberg and Bartley (2001), for example, find that inthe early US fire insurance industry, more robust regulation was likely to beenacted when marginal groups such as farmers and small businesses chal-lenged big business in the political arena.

Legal and regulatory structure within a region or locale can also serve asa mechanism that generates variation in agglomeration processes. Localinstitutions can structure access to resources and provide cognitive modelsthat are important in the emergence of new firms (Suchman et al., 2001).While it is generally accepted that institutional forces work at a level thattranscends regional boundaries, ‘some of the most important ones operatemore narrowly, within particular geographically or functionally boundedorganizational communities’ (Suchman et al., 2001: 357). For example, ina recent study of liquidity events and their effect on initial public offerings,Stuart and Sorenson (2003) find that urban areas in states that are lax intheir enforcement of non-compete clauses experience greater new ventureformation. Further, they find that enforcement of those non-competeclauses lessens the effects of liquidity events on founding rates. This studysupports Suchman et al.’s (2001) claim that, ‘Among the various insti-tutional structures that might regulate the flow of resources and models tonew firms, law occupies a particularly important place’ (362).

In the following section, we draw upon historical and archival docu-mentation to develop an analytical narrative that demonstrates the role thatenvironmental social movements played in contesting existing methods ofproducing electricity. Through their collective efforts, they facilitated theconstruction of a regulatory environment that provided the necessaryincentives for entrepreneurs to begin selling renewable energy on the grid.

3. CONTEXT: THE US POWER INDUSTRY

Until the late 1970s, electric utilities depended almost exclusively on a com-bination of oil, coal, large hydroelectric facilities, and to a lesser extentnatural gas and nuclear technology to generate power. Other than a fewsmall experimental facilities, utilities did not generate power using non-hydro renewable sources (US Department of Energy, 2001). This is not tosay that there was no interest in renewable technologies. For example, thefirst wind turbine that produced electricity was built in 1888, and since thattime a long line of fringe alternative power enthusiasts have promoted theirrespective technologies.

However, prior to 1978, the electric utility industry consisted of verticallyintegrated utilities that generated and distributed electricity. Utilitieslargely rejected RETs because they were viewed as expensive and risky

98 Industrial dynamics

when compared to the highly developed traditional generation technolo-gies. Utilities controlled both power generation and distribution in partic-ular regions. Renewable energy technologists could thereby only accessregional wholesale and retail energy markets with the cooperation of thelocal utility. These utilities often resisted interconnection with independentgenerators because interconnection would likely increase their costs andutilities were not eager to distribute power generated by potential competi-tors. Since utilities could lock out these potential competitors by refusinginterconnection, offering below market prices for independently producedpower, and charging higher-than-average prices for back-up power (Hirsh,1999: 81–3), there were no foundings of independent power generationfacilities that sold electricity to the grid prior to 1978.

This situation changed dramatically when in 1973 a Saudi oil embargoon the United States caused oil prices to more than double, reaching $25 abarrel in 1973. Further disruptions to oil supplies in 1978 pushed oil pricesto $50 a barrel, five times the 1972 price of oil. Utilities attempted to reducetheir reliance on oil by converting to expensive solid-fuel plants but despitethose efforts, electric prices remained high. These price increases motivatedpolicy makers to search for other ways to generate electricity that woulddecrease the country’s dependence on foreign oil, thereby providing insti-tutional entrepreneurs fertile opportunities to promote new technologicalagendas to these same policy makers (Sine and David, 2003).

Environmental activists brought these agendas to national attention byquestioning existing energy policies and practices. As early as the late 1960s,growing awareness of the harmful environmental effects of coal- and oil-burning power plants and fear of nuclear power increased the hostility ofmany environmental groups toward electric utilities (Fenn, 1984: 51–2).Large, established environmental groups such as the Sierra Club, theAudubon Society, the Union of Concerned Scientists and others began toactively promote an energy conservation agenda that included increaseduse of renewable energy and more efficient use of energy from all sources(McCloskey, 1992; McLaughlin and Khawaja, 2000). Advocates of RETsargued that while these new technologies were relatively underdevelopedcompared to coal- and oil-based technologies, they had several qualitiesthat made them potentially better sources of power than conventionalmeans. First, unlike coal, oil and gas, the process of generating power withRETs does not produce air or water pollution, making its environmentalfootprint smaller than that of large-scale hydroelectric plants. Moreover,unlike coal, the production of renewable energy does not require largemines or, as in the case of oil, run the risk of spills. Second, RETs are localsources of energy and thereby promote local jobs. Finally, given techno-logical progress, RETs had the future potential of being priced similarly to

Constructing entrepreneurial opportunity 99

energy produced by traditional sources. However, like most claims aboutfuture technology progress, this last point was highly uncertain.1

In reaction to the energy crisis and pressure from pro-conservation andrenewable energy advocates, the National Energy Act (NEA) was passed in1978 and section 210 of this law, the Public Utility Regulatory Policies Act(PURPA), allowed entrepreneurs to construct qualifying non-utility facili-ties free from the constraints of traditional utility regulation. Under section210, utilities were required to interconnect with qualifying non-utilitypower plants (that is, qualifying facilities) and purchase power from quali-fying facilities at utilities’ cost of generation (which came to be known inthe industry as ‘avoided cost’). Independent power plants qualified undersection 210 if they used alternative energy resources such as wind, solar,biomass, garbage, wood, sewage sludge and other lower-grade fuels, or usedcogeneration technology. Section 210 provided the legal structure at thenational level that allowed alternative energy entrepreneurs to interconnectwith and sell electricity to utilities. However, the Federal Energy RegulatoryCommission (FERC) left the interpretation and enforcement of section 210to state governments.

Social Movements and the US Power Industry

Social movements ‘deinstitutionalize existing beliefs, norms and valuesembodied in extant forms and establish new forms that instantiate newbeliefs, norms and values’ (Rao et al., 2000: 238). For social movements tobe successful in these deinstitutionalization and institutionalization pro-jects, movements must at the outset engage in the production and mainte-nance of meaning for constituents, antagonists and bystanders (Benfordand Snow, 2000: 613). Producing and maintaining collective meaning (thatis, collective action frames) includes three important processes: diagnostic,prognostic and motivational framing. Diagnostic framing involves creatinga shared understanding of a problematic condition that is in need ofchange and ascribes culpability to someone or something for the problem.Prognostic framing denotes the generation of a solution to the problem,and motivational framing refers to the construction of the rationale forengaging in collective action (Benford and Snow, 2000). Beginning in theearly 1970s, environmental movement organizations engaged directly indiagnostic, prognostic and motivational framing to advocate the use ofrenewable energy.

Diagnostic framingSuchman (1995) argued that the recognition and articulation of a problemfor which there is no adequate solution is the first step required for creating

100 Industrial dynamics

change in taken-for-granted ways of doing something (Scott, 2001). Theprocess of developing a description of, and detailed evidence about, aproblem, its cause and its negative consequences, focuses the attention ofthe public and powerful actors on unsolved difficulties (Edelman et al.,1999; Sine and David, 2003). Unanswered problems provide a social lo-cation for new solutions and simultaneously delegitimate existing insti-tutional arrangements that are not adequately addressing or solving theproblems (Cohen et al., 1972). Educating relevant actors and the publicabout problems catalyses solution generation processes as relevant actorsengage in sensemaking and collective problem solving (Mezias andScarselletta, 1994; Weick, 1995). The public articulation and elaboration ofproblems creates opportunities for institutional entrepreneurs to persuas-ively argue for the implementation of new practices and attracts pools ofalternative solutions.

Prior to the 1970s, the US environmental movement took little notice ofenergy policy. However, after the publication of Rachel Carson’s SilentSpring (1962), members of the environmental community engaged inframing coherent, consistent and salient critiques of the nation’s depen-dence on fossil fuels and nuclear power. The core idea emanating from thesecritiques was that the growing level of energy consumption was not sus-tainable. The most extreme advocates believed that ‘modern technologicalsystems were not simply malfunctioning or manifesting inefficiencies; theywere in some profound sense unable to preserve the environment, and thesocial and political structures that had produced those technologies couldnot make sufficient improvements to prevent a serious ecological disaster’(Laird, 2001: 122). Relatively unchallenged prior to the 1973 oil embargo,in the mid-1970s the energy sector found itself subject to intense scrutiny,with its technologies, resources and underlying values being challenged(Laird, 2001).

One of the earliest and most effective critics of US energy policy wasenvironmentalist Amory Lovins. In a strategic and creative way, Lovinseffectively reframed the energy problem. Instead of assuming an increaseddemand for energy, Lovins argued that people demand delivered servicessuch as ‘comfortable rooms, light, vehicular motion, food, tables, and otherreal things’ (Lovins, 1976: 78). Lovins’s perspective suggested that theproblem ‘is not simply where to get more energy, of any kind, from anysource, at any price, but rather how to supply just the amount, type andsource of energy that will provide each desired service at least cost’ (Lovinsand Lovins, 1974: 26). In sum, Lovins argued that the real energy problemlies in the mismatch of energy supply to end-use needs in scale and quality.By articulating the energy problem as a mismatch rather than a problem ofsupply, Lovins invoked the metaphor of someone who is unable to fill the

Constructing entrepreneurial opportunity 101

bathtub because the water keeps running out. Instead of thinking of theproblem as having too small a water heater, the problem lies in not havinga plug (Lovins and Lovins, 1974; Lovins, 1976).

The energy problem gained greater salience as Lovins ascribed blame forenergy shortages to shortsighted pricing policies. Instead of basing energydecisions on current average costs, as is common in the industry, Lovinsadvocated the use of long-run marginal costs, which would include capitalcosts associated with the construction of new centralized power plants.This was important because in many cases the cost of new capacity wassubstantially larger than past power facilities. He effectively argued that thelegal infrastructure of the time essentially subsidized the true costs of non-renewable energy by not accounting for externalities such as associatedenvironmental degradation and not taking into account future costs ofadditional capacity.

Prognostic framingPrognostic framing, or the generation of solutions, is closely associated withdiagnostic framing (the articulation of problems) (Gerhards and Rucht,1992). Lovins’s critique of US energy policy and his reframing of the energyproblem naturally dovetailed with the solutions he proposed. To remedy theproblems he described, he proposed to do more with less energy through theuse of energy-efficient technologies, allowing exactly the same output ofgoods and services. Another key characteristic of the solution includedtechnologies that relied on renewable energy flows that matched user needsin both scale and quality (Lovins, 1976). By scale, Lovins suggested thatlarge energy production facilities be utilized for large uses and small energyproduction technologies be used for small sources, such as a passive solardesign to heat a home. By quality, he meant that expensive forms of energy,such as electricity, should be reserved for applications where electricity isappropriate and indispensable, such as for smelting, subways and otherkinds of mechanical work, and not be used to perform small jobs such asdomestic space heating. Implicit in his solution was the decentralization ofenergy systems to redress the supply/demand mismatch.

Lovins’s creative reframing of the energy ‘problem’ and its associatedsolutions had an immediate and profound impact on the energy policydebate. He was attacked repeatedly for his position, yet responded articu-lately and rapidly to each criticism. One author noted,

He presented a thorough critique of conventional thinking about energy policyand the most creative and sophisticated arguments in favor of solar energy todate. His analysis provided a rationale for a group of growing importance in thesolar movement – those who based their attachment to solar energy on beliefs

102 Industrial dynamics

about the nature of environmental problems and the relationship of energy tech-nologies to them. (Laird, 2001: 126)

However, his work did not exclude more mainstream supporters, because itwas grounded in economic efficiency, suggesting that his approach wouldbe cheaper than the existing one. Furthermore, while Lovins’s approachwas consistent with ecological values, he argued repeatedly that it did notrequire them, suggesting that his approach could coexist with more con-ventional values like economic rationality (Laird, 2001).

Lovins’s diagnostic and prognostic framing of energy issues gained res-onance because the frame was consistent with both environmental andcost–benefit logics and therefore empirically credible (Benford and Snow,2000). Consequently, this new ‘energy frame’ gained acceptance from envi-ronmental groups that in turn promulgated and reinforced it through art-icles, studies and newsletters. For example, the president of the Sierra Club,in a letter to its members said, ‘there is, however, a positive choice that canbe made – a “soft path” alternative energy plan, one based on conservationand renewable sources. If adopted, it could solve our most pressing energyproblems faster, cheaper and more cleanly than Carter’s plan’ (Snyder,1979: 4).

The Union of Concerned Scientists (UCS) completed a study in 1980that critiqued both the current system and the industry’s nuclear alterna-tive and set out to provide an answer to the question, ‘if not nuclear power,what?’ (UCS, 1980: xvii). In the introduction to their study they state:

It is now abundantly clear that the world has entered a period of chronic energyshortages that will continue until mankind has learned to harness energy fromrenewable sources. . . . In the face of these mounting difficulties, a long-termstrategy emphasizing major improvements in energy productivity (with an atten-dant reduction in energy growth) and a reliance on renewable energy derivedfrom the sun[2] emerges as the clearest and most sensible solution to the greatchallenge we and future generations face. . . . We believe that the strategy out-lined in this study can help ease the burden of transition to a benign and sus-tainable energy future for the United States. (UCS, 1980: 21)

For the UCS, renewable energy provided a sensible solution to compet-ing solutions such as nuclear power:

On the other hand, we believe that most, if not all, of the major environmentaland societal risks posed by a large-scale breeder future could be avoided in awell-planned solar energy economy.[3] A solar energy future would in some waysbe simpler and in other ways more complex than a breeder future. While theindividual solar technologies are, in themselves, much less complicated thanbreeder reactors, the design and integration of a comprehensive solar energy

Constructing entrepreneurial opportunity 103

system would involve a greater degree of sophistication than a system basedentirely on breeders. However, the resulting system would be more flexible, lessdangerous, better adjusted to the diverse energy needs of this complex industrialstate, and thus a more stable basis for the nation’s continued prosperity than anysystem based on nuclear fission. (Ibid.)

Studies conducted by Friends of the Earth, the Audubon Society and theSierra Club all provided a solid basis for critiquing existing technology andpower generating sources and at the same time, advocated the use of alter-native energy. For example, the UCS strongly advocated for wind, suggest-ing that it was the most viable, safe, benign and easily commercialized of allalternative energy technologies:

Wind represents a large and nondepletable energy resource that can be utilizedwith minimal impact on the environment, producing no air and thermal pol-lution and requiring no water in its utilization. The simplicity of wind technol-ogy will allow for rapid deployment in comparison to many other energytechnologies. Finally, the economic prospects of wind systems are quite promis-ing. There are barriers as well to the widespread use of wind power, but noneshould prove insurmountable. (Ibid.: 145)

Through effective framing, these social movement organizations pro-vided meaning and momentum to efforts to change the structure of elec-tricity production in the United States. They served as sounding boards,amplifying the importance and urgency of fostering the development ofalternative energy sources. The development and subsequent promulgationof these frames became the basis for the mobilization of resources insupport of RETs in Congress and among the rank-and-file membership ofenvironmental groups.

Motivational framing and resource mobilizationMobilizing structures are forms of organization available to social move-ment actors that include informal friendships, neighborhood and work set-tings that facilitate and structure collective action (Tilly, 1978), and formalorganizations that embody and attempt to achieve the preferences andgoals of a broader social movement (McCarthy and Zald, 1977). For thepurposes of this chapter, we focus on formal social movement organiz-ations (SMOs) as the key mobilizing structures used to recruit members,obtain and distribute material resources, disseminate information, provideprotocols of action for members to follow, and facilitate a common cogni-tive framework among their members. For example, the Sierra Club servedas a principal mobilizing structure for collective efforts to advocate RETsand develop supportive regulative structure. The elaborate organizational

104 Industrial dynamics

structure of the Sierra Club (chapters in all 50 states) coupled with theflexibility to accommodate more informal and proactive activities in pro-moting RETs made it a potent organization for altering political land-scapes. An article in a 1970 newsletter described the relationship betweenformal organizational structure and grassroots action in mobilizationefforts:

Perhaps the most important aspect of chapters and groups is that they are theClub’s grassroots interface with the public. Some hold monthly membershipmeetings open to the public, with speakers and film programs. Others haveorganized speaker bureaus and presented programs to community groups. Manychapters publish excellent newsletters, which carry articles on local conservationcampaigns, as well as scheduled activities. A few have opened community con-servation centers, either with their own resources or in cooperation with otherenvironmental organizations. Very often, it is at the group level that the nuts andbolts work of the conservation campaigns is carried out – providing manpowerto gather data, prepare lawsuits, testify at public hearings, influence publicofficials, and generate the enthusiasm and dedication that will carry to state andnational levels, attracting support for our cause. . . . It is often at the group levelthat a conservation issue was first recognized, and then brought to the attentionof the chapter and to the Board of Directors, possibly to become one of theclub’s priority campaigns nationally. The process is reversed, though, whenappeals go out from the Club’s national offices to chapter and group leaders to‘activate telephone chains and get wires into Washington!’ The key element inthis complex structure, besides funds, public support and staff, is an informed,active membership. (Billings, 1971: 21)

A bridging concept between mobilizing structures and framing processesis motivational framing (Benford and Snow, 2000), which provides therationale for engaging in collective action. Important aspects of this typeof framing include articulating the severity of the problem, the urgency foraction, and the efficacy and propriety of a specific action (ibid.: 617). In thecase of renewable energy, environmental groups provided a rationale for itsmembers to support and advocate alternative energy sources by issuing acall to action, outlining the specific, urgent action to be taken, and empha-sizing the consequences of inaction. For example, the Audubon Society, anorganization whose primary focus had little to do with energy policy, pro-vided a strong rationale for its members to engage in political action:

It is not enough to be convinced that the solar/conservation approach is the mosteconomic and environmentally benign energy strategy. If organizations likeAudubon and the many environmentally oriented individuals who make up themembership of such organizations are unable to communicate to their neighborsand governmental leaders both the merits and the urgency of the solar/efficiencyapproach, then the goals of this Plan and others like it will not be met. In thatcase, the ‘energy growth’ point of view of organizations like the Edison Electric

Constructing entrepreneurial opportunity 105

Institute will dominate public policy debates, and the possibility of achievingeconomic growth without energy growth will be lost in the next decade.(National Audubon Society, 1984: 52–3)

This example illustrates the Audubon Society’s call to action anddescribes the kind of political action to be taken (communicating withneighbors and government leaders), the need for urgency (goals must bemet), and the effects of inaction (the Edison Electric Institute dominatingpublic policy debates). Other types of activities prescribed by this environ-mental organization included increasing public awareness, scrutinizing stateand local energy plans, lobbying at all levels of government, involving thecommunity in studying local energy problems, and calling upon experts atuniversities to analyse the benefits of conservation for taxpayers (ibid.: 105).

The Sierra Club constructed a similar mobilization frame and leveragedits formal and informal organizational structure to mobilize its members topromote regulatory change:

The threat posed by President Carter’s proposals is so great, and the need foraction by every member so urgent that the board of directors has called for themobilization of the Club’s full resources for this Emergency Energy Campaign.Only a massive outpouring of grass-roots concern can transform the presentpolitical climate, encouraging Congress to drop the damaging proposals andenact more rational energy alternatives. Intensive organizing efforts have alreadybeen set in motion, and letter writing and media contacts have begun. All SierraClub chapter and group leaders will be receiving regular updates as this cam-paign speeds along. (Snyder, 1979: 5)

It is time for face-to-face mobilization. Conservationists must start meeting withtheir elected officials and candidates to tell them what they, as voters, expect andwant. We need to stand up at rallies and ask, politely, but persistently, whyCongress ‘solved’ the energy crisis by putting most of the money into the leastpromising and most expensive technologies, the synthetic fuels. . . . If, in the nextsix months, every Sierra Club member would just once personally attend andparticipate in a political event, it would make a world of difference. (Coan andPope, 1980: 11–13, 47)

Environmental memberships readily accepted the mobilization framespromulgated by the leadership of environmental organizations and theircall for political activism. In fact, comprehensive surveys of Sierra Clubmembers reveal that their commitment reflected the directives, goals, andaspirations of the club’s leadership. A survey in 1971 revealed significantcommitment among the organization’s rank-and-file membership manifestin their political activism: 60 per cent said that they sent their views on con-servation matters to government officials at least once in the past year, with15 per cent reporting nine or more communications; 40 per cent of the

106 Industrial dynamics

members attended one or more political meetings or rallies in the last yearand 27 per cent reported devoting time to one or more political causes(Coombs, 1972).

Following the oil crisis of 1973, and the Three Mile Island incident of1979, energy policy was at the fore for both the rank-and-file members andleaders of the Sierra Club. A second survey in 1979 revealed a membershiphighly committed to changing energy policy. The opening sentence of thestory that introduced the results of the survey read:

A new broadened Sierra Club motto might be in order. It might read: . . . To pre-serve, explore and enjoy the nation’s wilderness, parks, forests and natural areas,to clean the nation’s air and water, to protect wildlife, promote the developmentof alternative, renewable energy resources and conserve energy to further this goal.(Emphasis added, Utrup, 1979: 16).

This second survey found that one-third of the members felt that ‘energyissues’ needed more effort on the part of the Club. Energy was also the thirdmost popular issue that Club members showed a willingness to spend moretime working on: 75 per cent of members expressed a strong interest insolar power, with the remainder expressing some or little interest (ibid.).

These surveys demonstrate that the logics advocated by the Sierra Clubleadership that framed conventional energy technology as an environmentalproblem and RET as a solution were accepted and enacted by the member-ship. In this section, we have provided qualitative evidence that suggests en-vironmental movement organizations effectively framed energy problems andsolutions, provided a rationale for action to their members, and marshaledresources through mobilizing structures to create favorable regulatory cli-mates for independent power firms using RETs. Given the intensity and scopeof activities engaged in by these social movement organizations, we hypothe-size that higher levels of social movement activity in a state will increase thelikelihood of a favorable regulatory environment for firms using RET.

4. DATA AND METHODS

We test our hypothesis by examining the effects of Sierra Club membershipon avoided cost rates. We draw on panel data that capture US state energyregulation from 1980 to 1992.

Dependent Variable

Our dependent variable is the average size of a state’s avoided costs in agiven year. Although the federal government loosely defined avoided cost

Constructing entrepreneurial opportunity 107

(the cost avoided by a utility for not generating the same amount of power),it was left up to the states to establish a precise definition. Consequently,the formula for constructing avoided cost varied from state to state. In somestates, it was based on the cost of fuel of the most likely alternative whileother states took into account capital costs. Calculations that took intoaccount capital costs tended to be controversial because they variedtremendously depending on the type of technology used (that is, nucleartechnology was more expensive than coal technology). Other sources ofambiguity included the choice of fuel type to use as the baseline to calcu-late the cost of generation. In other words, were public utility commissionsto define avoided costs by the most expensive type of fuel used in a utility’sportfolio or its least expensive? Further, should the public utility commis-sions assume high or low interest rates? Because of these ambiguities, therewere substantial differences in how states calculated avoided costs and con-sequently, their amounts. Our data on avoided costs come from threesources: Solar Law Reporter (1981), Energy User News (1982–85), andAvoided Cost Quarterly (1986–92).

Independent Variable

We obtained state-level Sierra Club membership data. During the period ofour study, the Sierra Club was one of the three largest environmentalorganizations in the United States (McCloskey, 1992). We chose the SierraClub membership data due to the organization’s size and status within theenvironmental community. We also acquired yearly state-level membershipdata from the UCS and found their membership distribution across USstates to be highly correlated with that of the Sierra Club’s (0.88).Membership data from the Audubon Society and several other large en-vironmental organizations were not available because such information wasnot retained during the late 1970s and 1980s.

Control Variables

Our model controlled for the following state characteristics: per capitagross state product, change in gross state product, state population, con-gressional voting record on energy issues,4 and net electricity imports. Wealso controlled for the availability of land with constant streams of high-quality wind because wind turbines were the most common RET used byindependent power entrepreneurs during our study period. Because theregulatory climate may be affected by the extent to which regulators arewilling to monitor compliance and punish firms for not following formalregulations or informal norms, we also controlled for the activism of state

108 Industrial dynamics

utility commissions. To assess commission activism, we measured thenumber of rate cases per utility, comprehensive audits per utility, compre-hensive audits of private utilities, and total audits per utility. Audits providea good measure of regulatory oversight because the purpose of audits isto verify the information given to the commission by utility companies.More audits conducted by a public utility commission are indicative ofan involved and active state commission. Rate cases examine a utility’sfinancial information to verify the justification for current or proposedrates. Utilities view audits and rate cases as highly disruptive and expensive.Excessive audits and rate cases can be viewed as either punitive measuresor as a way of ensuring compliance with state regulatory policies. Thesedata come from the National Association of Regulatory UtilityCommissioners (NARUC) annual utility surveys.

Model Specification and Estimation

We measure the impact of environmental social movement organizationson the regulatory climate using avoided costs as the dependent variable.This variable indicates the extent to which a state’s regulatory climate issupportive of alternative energy. We use panel data for each state-yearbetween 1980 and 1992. The analysis begins in 1980 because this is theearliest date that states defined avoided costs.

We employ a generalized estimating equation to predict the size of theavoided cost and corrected for common biases in the analysis of longitud-inal data introduced by autocorrelation among residuals by assuming anunstructured error correlation. We used the STATA XTGEE commandwith the unstructured correlation option. An advantage of this model isthat it does not assume that panels are uncorrelated. We used robust vari-ance estimators in our analyses, reducing problems associated with hetero-skedasticity and misspecification of the error structure (Allison, 1999). Wealso estimated the models using the ar 1 option, but found no significantdifference in the results. We used a variance inflation factor test to ensurethat our models were not significantly affected by multicollinearity andfound that the factor scores in both models presented were less than 5, sug-gesting an acceptable level of multicollinearity (Chatterjee and Price, 1991).

5. RESULTS

Summary statistics are provided in Table 5.1. In Table 5.2, we present theresults of the generalized estimating equation predicting supportive regu-latory climate. Several control variables significantly predict the amount of

Constructing entrepreneurial opportunity 109

110

Tab

le 5

.1S

umm

ary

stat

isti

cs a

nd c

orre

lati

ons

for

stat

e-av

oide

d co

st a

naly

sis

Var

iabl

eM

ean

S,D

.1

23

45

67

89

1011

12

1.A

void

ed c

osts

(ce

nts/

kWh)

2.79

1.26

0.22

2.G

SP p

er c

apit

a0.

020.

010.

000.

053.

Cla

ss 3

& 4

win

d av

aila

bilit

y49

333.

9479

787.

24�

0.11

�0.

280.

014.

Stat

e po

pula

tion

(ln

)4

692

763

507

669

50.

35�

0.20

�0.

16�

0.29

5.C

hang

e in

Gro

ss S

tate

7.

344.

970.

030.

16�

0.12

�0.

11�

0.01

Pro

duct

(G

SP)

6.N

et e

lect

rici

ty im

port

s 0.

542.

130.

130.

040.

02�

0.01

0.19

0.03

(mill

ion

kWh)

7.R

ate

case

s pe

r ut

ility

0.17

0.25

�0.

040.

140.

25�

0.11

�0.

230.

05�

0.01

8.C

ompr

ehen

sive

aud

its

per

0.06

0.16

0.11

0.19

0.41

�0.

06�

0.20

�0.

030.

090.

35ut

ility

9.C

ompr

ehen

sive

aud

its

1.19

3.22

0.22

0.02

0.00

�0.

030.

17�

0.04

0.45

�0.

100.

45pr

ivat

e ut

iliti

es10

.A

udit

s pe

r ut

ility

0.50

2.44

0.04

0.01

�0.

03�

0.10

0.08

0.04

�0.

010.

43�

0.01

�0.

0111

.C

ongr

essi

onal

vot

ing

49.6

520

.71

0.25

0.36

0.01

�0.

200.

120.

110.

220.

150.

070.

060.

13re

cord

/10

12.

Env

iron

men

tal

9.07

23.0

30.

440.

010.

17�

0.31

0.84

0.08

0.25

�0.

05�

0.01

0.22

0.11

0.24

mem

bers

hip/

1000

Not

e:A

ll co

rrel

atio

ns g

reat

er t

han

0.11

are

sig

.at

0.05

leve

l.

the avoided cost in a given state. The availability of windy land is significantand negatively correlated with state-avoided costs. This is unusual becauseone would expect that state governments would create legal structures thattake advantage of local natural resources. Instead, we find the opposite tobe true: regulators are less likely to try and use incentives to motivate windpower entrepreneurship in locations with substantial endowments ofwindy land. This suggests that state governments use incentives to com-pensate for shortages of natural resources in areas where constituents valuewind power. We also find that states with smaller populations and states

Constructing entrepreneurial opportunity 111

Table 5.2 GEE model predicting state-avoided costs

Variable Avoided cost

Model 1 Model 2

Control variablesGSP per capita �27.135 �24.485

(47.371) (48.746)Class 3 & 4 wind availability �0.000* �0.000�

(0.000) (0.000)State population (ln) � 0.429* �0.928**

(0.174) (0.287)Change in GSP �0.004 0.004

(0.017) (0.020)Net electricity imports �0.098 �0.111

(0.078) (0.082)Rate cases per utility 0.248 0.080

(0.588) (0.640)Comprehensive audits per utility �0.549 0.334

(1.429) (1.615)Comprehensive audits – private utilities 0.054 0.024

(0.057) (0.067)Audits per utility �0.007 �0.017

(0.044) (0.046)Congressional voting record/10 0.021** 0.021**

(0.007) (0.007)Independent variableEnvironmental membership/1000 0.634*

(0.255)

Constant 9.164** 11.367**(2.684) (3.074)

Chi squared statistic 24.45 34.58

Note: �p � 0.10; *p � 0.05; **p � 0.01. Standard errors are in parentheses.

with congressional representatives who have a history of voting in favor ofenvironmentally friendly energy policies are also more likely to have higheravoided costs.

The results in Table 5.2 support our hypothesis; environmental groupmembership positively and significantly predicts avoided costs. An increasein Sierra Club membership of 1000 increases the size of a state’s avoidedcosts by 0.63 cents per kWh.

6. DISCUSSION AND CONCLUSION

In this chapter, we have shown how regional differences in social movementorganization membership impacted the creation of supportive regulatoryenvironments for renewable energy entrepreneurs. Prior to 1978, electricutilities dominated the production and distribution of electricity in the US.These organizations enjoyed regional monopolies and local legal structuresthat had evolved over four decades to accommodate the needs and interestsof this organizational form. Thus, state legal structures were highly attunedto incumbent utilities and did not support other organizational forms suchas independent power. For example, the retail price utilities charged cus-tomers for electricity was regulated and determined by public utilities com-missions on a cost-plus basis which was defined as a utility’s cost ofgeneration plus a ‘fair’ profit. However, states did not have a procedure fordetermining the wholesale price of renewable electricity generated by inde-pendent power plants, leaving it up to utilities to determine whether and atwhat price to purchase independent power. Moreover, the existing regu-latory structure supported coal, natural gas and oil production through avariety of tax incentives and permitting systems. A similar supportive regu-latory infrastructure did not exist for independent power companies usingRETs. Transforming these regional regulatory environments to support anew organizational form (independent power generators using RET) was anecessary condition for entrepreneurial activity in this sector.

Environmental movement organizations paved the way for the emergenceof independent power using RETs by framing energy generation as not onlyan economic decision, but also a moral one. Environmental movementorganizations promulgated this frame via their regional chapters. Thesechapters mobilized local members to promote renewable energy and directlyengaged local constituents via educational programs, public relations anddirect lobbying of state governments. By the early 1980s, over half of themore than 100000 members of the Sierra Club were dedicating some timeeach week to supporting this agenda. In regions where the Sierra Club hada sizeable membership, political landscapes that favored incumbent utilities

112 Industrial dynamics

and their non-renewable technologies were transformed to support inde-pendent power producers using RETs. Our data demonstrate that much ofthe impact of national environmental organizations was quintessentiallylocal. Although the public relations arms of the national organizations pro-mulgated these new frames via the media nationwide, states’ support forthese new organizational forms varied substantially. States with large andactive Sierra Club or UCS chapters were significantly more likely to defineavoided costs in such a way as to create opportunities for entrepreneursusing renewable energy technologies.

Based on these findings, this chapter contributes to existing theory infour ways. First, unlike past work on industry emergence that treats regu-latory structures as fixed and immutable environmental attributes (Russo,2001), we view regulatory structures as socially constructed and dynamic(Edelman, 1990). We apply a co-evolutionary perspective on technicalchange, markets and institutions (Nelson, 1995). Regulation tends to fosterinertia and path dependency and generally reflects and supports the inter-ests and goals of dominant organizational forms (Fligstein, 1996, 2001). Assuch, dominant organizational forms defend and seek to preserve thesefavorable regulations. Consequently, such regulatory structures are difficultand expensive to change and because no single firm possesses the necessaryresource base and scope to precipitate such substantive regulatory change,collective action is necessary. Regions with a strong presence of collectiveactors supportive of new categories of economic activity are more likely tosee regulatory change that favors these new sectors. Building on researchfrom the institutional tradition (Schneiberg and Bartley, 2001; Scott, 2001),our study shows that regulatory environments that supported renewableenergy were more likely in regions with large environmental organizationmemberships such as the Sierra Club and the UCS. It is important to notethat environmental groups did not endorse renewable energy because of itsapparent economic efficiency. Nor was the emergence of RETs an instanceof an inevitable outcome of technological change. Rather, environmentalgroups advocated RETs because they had a more benign effect on the en-vironment – a critical externality that the regulatory environment at thattime did not take into account. These social movement organizationsworked to create a set of rules and incentives that supported technical solu-tions aligned with ideologically-driven preferences for renewable energyand conservation.

Second, our work advances an institutional approach to economic geog-raphy using an evolutionary perspective (Thrift and Olds, 1996; Crang,1997; Martin, 2000; Sine and Lee, 2006). Storper and Walker note thatsocio-institutional structure is an ‘essential underpinning of efficient capi-talist production’ (1989: 5). Institutional theorists likewise contend that

Constructing entrepreneurial opportunity 113

economic activity is socially and institutionally situated and embedded insocial and political structures (DiMaggio and Powell, 1983; Granovetterand Swedberg, 1992; Scott, 2001). The primary intent of an institutional-ist approach to economic geography is to understand ‘to what extent andthrough which means are the processes of geographically uneven capitalisteconomic development shaped and mediated by institutional structures’(Martin, 2000: 79). However, anecdotal evidence and theorizing have notspecified how supportive regulatory structure emerges ‘at the regional andlocal level and what precise role it plays in regional economic development’(ibid.: 88). By bringing the theoretical lens of social movements to bear onthe question of spatial variation of economic opportunity, we provide newanalytical leverage to these questions. Our research suggests the impor-tance of considering how pre-existing regional differences in non-economicsectors can shape economic outcomes. In the case study presented in thischapter, social movements promulgated particular norms and values alsoconstructed the cultural and regulatory infrastructure for entrepreneursusing RET.

Third, we contribute to social movement literature by emphasizing theimportant local effects that movements have on regional economies. Whilepast research on social movements has demonstrated how ideas diffuseacross regions, there has been little research on how movements changelocal environments to foster new economic activity (but, see Schneiberg andSoule, 2005). For example, research has demonstrated how social move-ments such as the temperance movement grew and eventually influencedthe ability of organizations to sell alcohol (Sandell, 2001), but we knowlittle about how these activities influenced the rise of other sectors such ascarbonated soda at the local level. Giugni (1999) has called for more empir-ical research that examines the relationship between social movements andsocial change by considering how the outcomes of social movements varyacross different contexts. By emphasizing regional differences in socialmovement activity and its impact on the transformation of regulatory en-vironments, this research provides evidence of the causal link betweenregional social movement activity and social and economic change.

Finally, this case study demonstrates how existing mobilizing structuresof social movement organizations can be leveraged to effectively advancenew social agendas. In less than 10 years, environmental organizationseffectively mobilized their members to help spread an obscure set of ideasabout energy generation technologies throughout the US and embed it inboth national and regional regulatory structures. It is precisely the abilityof environmental organizations to redirect and broaden mobilizing effortsto include advocating RET through its organizational structure and pre-existing repertoires of action that led to these dramatic regulatory changes

114 Industrial dynamics

in such a short period of time. We find that years of promoting the protec-tion of the environment led to the unanticipated consequence of a con-stituency ready to support renewable energy generation technologies. Oncethe national organization made the connection between environmentalvalues and methods of energy generation, local chapters went to work cre-ating fecund environments for this type of commercial enterprise.

NOTES

* This research was supported by the Initiative for Future Agriculture and Food SystemsGrant no. 2001-52104-11484 from the USDA Cooperative State Research, Education,and Extension Service.

1. The uncertainty of the economic rationale of RETs in the 1980s is further illustrated bythe fact that during our study the cost of power generation using RETs was alwayssignificantly higher than that of conventional technology, and many experts predicted thatthis would not change for decades.

2. Because the sun drives global climatic systems, energy derived from the sun includes solarthermal, photovoltaic, wind, biomass, ocean thermal energy conversion, tidal power, wavepower, ocean current, and salinity gradient energy systems (UCS, 1980).

3. A solar economy refers to an economic system in which all energy is generated usingrenewable energy sources.

4. These data were obtained from the League of Conservation Voters which annually tracksthe voting records of individual members of Congress on environmental-related legis-lation. We narrowed the counts to legislation strictly related to energy production.

REFERENCES

Allison, P. (1999), Multiple Regression: A Primer, Thousand Oaks, CA: Sage.Almeida, P. and B. Kogut (1999), ‘Localization of knowledge and the mobility of

engineers in regional networks’, Management Science, 45: 905–17.Arrow, K.J. (1962), ‘The economic implications of learning by doing’, Review of

Economic Studies, 29: 155–73.Benford, R.D. and D.A. Snow (2000), ‘Framing processes and social movements:

an overview and assessment’, Annual Review of Sociology, 26: 611–39.Billings, L. (1971), ‘Sierra Club 1970: “New chapters, groups, growth and change” ’,

Sierra Club Bulletin, 20–22.Chandler, A.D. (1990), Scale and Scope: The Dynamics of Industrial Capitalism,

Cambridge, MA: Belknap Press of Harvard University Press.Chatterjee, S. and B. Price (1991), Regression Analysis by Example, New York: John

Wiley & Sons.Coan, G. and C. Pope (1980), ‘Energy 1979 – what happened and why’, Sierra,

January/February, 11–13, 47.Cohen, M.D., J.G. March and J.P. Olsen (1972), ‘A garbage can model of organ-

izational choice’, Administrative Science Quarterly, 17: 1–25.Coombs, D. (1972), ‘The club looks at itself ’, Sierra Club Bulletin, July/August, 35–9.Crang, P. (1997), ‘Cultural turns and the reconstitution of the economic’, in R. Lee

and J. Wills (eds), Geographies of Economies, London: Arnold, pp. 3–15.

Constructing entrepreneurial opportunity 115

David, R.J., W.D. Sine and H.A. Haveman (2005), ‘Exploiting institutional change:entrepreneurship in the legitimation of new organizational forms’, Paper pre-sented at the Administrative Sciences Association of Canada Annual Meetings,Toronto, 28–31 May.

Davis, G.F. and D. McAdam (2000), ‘Corporations, classes, and social movements’,in B.M. Staw and R.I. Sutton (eds), Research in Organizational Behavior, Vol. 22,New York: Elsevier, pp. 193–236.

Davis, G.F., D. McAdam, W.R Scott and M.N. Zald (2005), Social Movements andOrganization Theory, Cambridge and New York: Cambridge University Press.

Davis, G.F. and T.A. Thompson (1994), ‘A social movement perspective on corpor-ate control’, Administrative Science Quarterly, 39: 141–73.

DiMaggio, P.J. and W.W. Powell (1983), ‘The iron cage revisited: institutional iso-morphism and collective rationality in organizational fields’, AmericanSociological Review, 48: 147–60.

Dobbin, F. (1994), Forging Industrial Policy: The United States, Britain, and Francein the Railway Age, Cambridge and New York: Cambridge University Press.

Dobbin, F. and T.J. Dowd (1997), ‘How policy shapes competition: early railroadfoundings in Massachusetts’, Administrative Science Quarterly, 42: 501–29.

Eckhardt, J. and S. Shane (2003), ‘Opportunities and entrepreneurship’, Journal ofManagement, 29: 333–49.

Edelman, L.B. (1990), ‘Legal environments and organizational governance: theexpansion of due process in the American workplace’, American Journal ofSociology, 95: 1401–40.

Edelman, L.B., C. Uggen and H. Erlanger (1999), ‘The endogeneity of legal regu-lation: grievance procedures as rational myth’, American Journal of Sociology,105: 406–54.

Fenn, S. (1984), America’s Electric Utilities: Under Siege and in Transition, NewYork: Praeger.

Fligstein, N. (1996), ‘Markets as politics: a political–cultural approach to marketinstitutions’, American Sociological Review, 61: 656–73.

Fligstein, N. (2001), The Architecture of Markets: An Economic Sociology of Twenty-first Century Capitalist Societies, Princeton, NJ: Princeton University Press.

Fligstein, N. and A. Stone Sweet (2002), ‘Constructing polities and markets: aninstitutionalist account of European integration’, American Journal of Sociology,107: 1206–45.

Friedland, R. and R. Alford (1991), ‘Bringing society back in: symbols, practices,and institutional contradictions’, in W.W. Powell and P.J. DiMaggio (eds), TheNew Institutionalism in Organizational Analysis, Chicago: University of ChicagoPress, pp. 311–36.

Garud, R. and A.H. Van de Ven (1989), ‘Innovation and the emergence of indus-tries’, in A.H. Van de Ven, H. Angle and M.S. Poole (eds), Research on theManagement of Innovations, Cambridge, MA: Ballinger, pp. 489–532.

Gerhards, J. and D. Rucht (1992), ‘Mesomobilization: organizing and framing in twoprotest campaigns in West Germany’, American Journal of Sociology, 98: 555–95.

Giugni, M.G. (1998), ‘Was it worth the effort? The outcomes and consequences ofsocial movements’, Annual Review of Sociology, 98: 371–93.

Giugni, M.G. (1999), ‘How social movements matter: past research, present prob-lems, future developments’, in M.G. Giugni, D. McAdam and C. Tilly (eds), HowSocial Movements Matter, Minneapolis, MN: University of Minnesota Press,pp. xiii–xxxiii.

116 Industrial dynamics

Granovetter, M. and R. Swedberg (1992), The Sociology of Economic Life, Boulder,CO: Westview Press.

Greve, H.R. (1995), ‘Jumping ship: The diffusion of strategy abandonment’,Administrative Science Quarterly, 40: 444–73.

Hargrave, T.J. and A.H. Van de Ven (Forthcoming), ‘A collective action model ofinstitutional innovation’, Academy of Management Review.

Harris, C.D. (1954), ‘The market as a factor in the location of industry in the UnitedStates’, Annals, Association of American Geographers, 44: 315–48.

Hirsh, R. (1999), Power Loss: The Origins of Deregulation and Restructuring in theAmerican Electric Utility System, Cambridge, MA: MIT Press.

Hoffman, A.J. (1999), ‘Institutional evolution and change: environmentalism andthe US chemical industry’, Academy of Management Journal, 42: 351–72.

Ingram, P. and H. Rao (2004), ‘Store wars: the enactment and repeal of anti-chain-store legislation in America’, American Journal of Sociology, 110: 446–87.

Jaffe, A.B., M. Trajtenberg and R. Henderson (1993), ‘Geographic localization ofknowledge spillovers as evidenced by patent citations’, Quarterly Journal ofEconomics, 108: 577–98.

Laird, F. (2001), Solar Energy, Technology Policy, and Institutional Values,Cambridge and New York: Cambridge University Press.

Leblebici, H., G.R. Salancik, A. Copay and T. King (1991), ‘Institutional changeand the transformation of interorganizational fields: an organizational history ofthe U.S. radio broadcasting industry’, Administrative Science Quarterly, 36:333–63.

Lee, B.H. and M.D. Lounsbury (2005), ‘Instituting industry governance: the organ-izational and legal bases of U.S. national organic food standards’, Paper pre-sented at the American Sociological Association annual meeting, Philadelphia,13–16 August.

Lomi, A. (1995), ‘The population ecology of organizational founding: locationdependence and unobserved heterogeneity’, Administrative Science Quarterly, 40:111–44.

Lounsbury, M.D. and M. Ventresca (2002), ‘Social structure and organizationsrevisited’, in M.D. Lounsbury and M.J. Ventresca (eds), Research in the Sociologyof Organizations, Vol. 19, New York: JAI/Elsevier, pp. 3–36.

Lounsbury, M.D., M. Ventresca and P. Hirsch (2003), ‘Social movements, fieldframes and industry emergence: a cultural–political perspective on U.S. recyc-ling’, Socio-Economic Review, 1: 71–104.

Lovins, A.B. (1976), ‘Energy strategy: The road not taken?’, Foreign Affairs, 55: 65–96.Lovins, A.B. and L.H. Lovins (1974), ‘Energy solutions abound: just look around!’,

Business and Society Review, 82: 26–31.Marshall, A. (1922), Principles of Economics, London: Macmillan.Martin, R. (2000), ‘Institutional approaches in economic geography’, in E. Sheppard

and T.J. Barnes (eds), A Companion to Economic Geography, Malden, MA:Blackwell, pp. 77–94.

McAdam, D. (1996), ‘Introduction: opportunities mobilizing structures andframing processes’, in D. McAdam, J.D. McCarthy and M.N. Zald (eds),Comparative Perspectives on Social Movements: Political Opportunities,Mobilizing Structures, and Cultural Framings, Cambridge: Cambridge UniversityPress, pp. 23–40.

McCarthy, J.D. (1996), ‘Constraints and opportunities in adopting, adapting, andinventing’, in D. McAdam, J.D. McCarthy and M.N. Zald (eds), Comparative

Constructing entrepreneurial opportunity 117

Perspectives on Social Movements: Political Opportunities, Mobilizing Structures,and Cultural Framings, Cambridge: Cambridge University Press, pp. 141–51.

McCarthy, J.D. and M.N. Zald (1977), ‘Resource mobilization and social move-ments: a partial theory’, American Journal of Sociology, 82: 1212–41.

McClelland, D.C. (1961), The Achieving Society, Princeton, NJ: Van Nostrand.McCloskey, M. (1992), ‘Twenty years of change in the environmental movement: an

insider’s view’, in R.E. Dunlap and A.G. Mertig (eds), American Environmentalism,Philadelphia, PA: Taylor & Francis, pp. 77–88.

McLaughlin, P. and M. Khawaja (2000), ‘The organizational dynamics of the U.S.environmental movement: legitimation, resource mobilization, and politicalopportunity’, Rural Sociology, 65: 422–39.

Meyer, A.D. (1982), ‘Adapting to environmental jolts’, Administrative ScienceQuarterly, 27: 515–37.

Mezias, S.J. and M. Scarselletta (1994), ‘Resolving financial reporting problems: aninstitutional analysis of the process’, Administrative Science Quarterly, 39: 654–78.

National Audubon Society (1984), The Audubon Energy Plan, New York: NationalAudubon Society.

Nelson, R.R. (1995), ‘Co-evolution of industry structure, technology and support-ing institutions, and the making of comparative advantage’, International Journalof the Economics of Business, 2: 171–84.

Perrow, C. (1986), Complex Organizations: A Critical Essay, New York: RandomHouse.

Perrow, C. (2002), Organizing America: Wealth, Power, and the Origins of CorporateCapitalism, Princeton, NJ: Princeton University Press.

Piore, M.J. and C.F. Sabel (1984), The Second Industrial Divide: Possibilities forProsperity, New York: Basic Books.

Rao, H. (2004), ‘Institutional activism in the early American auto industry’, Journalof Business Venturing, 19: 359–84.

Rao, H., C. Morrill and M.N. Zald (2000), ‘Power plays: social movements, collec-tive action and new organizational forms’, in B.M. Staw and R.I. Sutton (eds),Research in Organizational Behavior, Vol. 22, New York: Elsevier, pp. 239–82.

Romanelli, E. and C. Schoonhoven (2001), ‘The local origins of new firms’, inC. Schoonhoven and E. Romanelli (eds), The Entrepreneurship Dynamic: Originsof Entrepreneurship and the Evolution of Industries, Stanford, CA: StanfordUniversity Press, pp. 40–67.

Romer, P.M. (1986), ‘Increasing returns and long-run growth’, Journal of PoliticalEconomy, 94: 1002–37.

Russo, M.V. (2001), ‘Institutions, exchange relationships, and the emergence of newfields: regulatory policies and independent power production in America,1978–1992’, Administrative Science Quarterly, 46: 57–86.

Sandell, R. (2001), ‘Organizational growth and ecological constraints: the growthof social movements in Sweden, 1881 to 1940’, American Sociological Review, 66:672–93.

Schneiberg, M. (2002), ‘Organizational heterogeneity and the production of newforms: politics, social movements and mutual companies in American fire insur-ance, 1900–1930’, in M.D. Lounsbury and M.J. Ventresca (eds), Research in theSociology of Organizations, Vol.19, Oxford: JAI, pp. 39–89.

Schneiberg, M. and T. Bartley (2001), ‘Regulating American industries: markets,politics, and the institutional determinants of fire insurance regulation’,American Journal of Sociology, 107: 101–46.

118 Industrial dynamics

Schneiberg, M. and S. Soule (2005), ‘Institutionalization as a contested, multi-level process: the case of rate regulation in American fire insurance’, in G. Davis,D. McAdam, W.R. Scott and M. Zald (eds), Social Movements andOrganization Theory, Cambridge and New York: Cambridge University Press,pp. 122–60.

Schumpeter, J.A. (1934), The Theory of Economic Development: An Inquiry intoProfits, Capital, Credit, Interest, and the Business Cycle, Cambridge, MA: HarvardUniversity Press.

Scott, W.R. (2001), Institutions and Organizations, Thousand Oaks, CA: Sage.Sine, W.D. and R. David (2003), ‘Environmental jolts, institutional change, and the

creation of entrepreneurial opportunity in the U.S. electric power industry’,Research Policy, 32: 185–207.

Sine, W.D. and B.H. Lee (2006), ‘The winds of change: antecedents of entrepre-neurial activity in the emerging wind power industry’, Working paper, CornellUniversity.

Snow, D.A. and R.D. Benford (1988), ‘Ideology, frame resonance, and participantmobilization’, in B. Klandermans, H. Kriesi and S. Tarrow (eds), From Structureto Action: Social Movement Participation Across Cultures, Greenwich, CT: JAIPress, pp. 197–217.

Snow, D.A., E.B. Rochford, S.K. Worden and R.D. Benford (1986), ‘Frame align-ment processes, micromobilization, and movement participation’, AmericanSociological Review, 51: 464–81.

Snyder, T.A. (1979), ‘Our energy future – a time to choose’, Sierra, September/October: 4–5.

Sorenson, O. and P.G. Audia (2000), ‘The social structure of entrepreneurial activ-ity: geographic concentration of footwear production in the United States,1940–1989’, American Journal of Sociology, 106: 424–61.

Stern, R.N. and S.R. Barley (1996), ‘Organizations and social systems: organizationtheory’s neglected mandate’, Administrative Science Quarterly, 41: 146–62.

Stinchcombe, A. (1965), ‘Social structure and organizations’, in J. March (ed.),Handbook of Organizations’, Chicago: Rand McNally, pp. 142–93.

Storper, M. and R. Walker (1989), The Capitalist Imperative: Territory, Technology,and Industrial Growth, Oxford: Blackwell.

Strang, D. and E. Bradburn (2001), ‘Theorizing legitimacy or legitimatingtheory? Neoliberal discourse and HMO policy, 1970–89’, in J.L. Campbell andO. Pedersen (eds), The Rise of Neoliberalism and Institutional Analysis, Princeton,NJ: Princeton University Press, pp. 129–58.

Stuart, T. and O. Sorenson (2003), ‘Liquidity events and the geographic distributionof entrepreneurial activity’, Administrative Science Quarterly, 48: 175–201.

Suchman, M.C. (1995), ‘Managing legitimacy: strategic and institutionalapproaches’, Academy of Management Review, 20(3): 571–610.

Suchman, M.C., D.J. Steward and C.A. Westfall (2001), ‘The legal environment ofentrepreneurship: observations on the legitimation of venture finance in SiliconValley’, in C. Schoonhoven and E. Romanelli (eds), The EntrepreneurshipDynamic: Origins of Entrepreneurship and the Evolution of Industries, Stanford,CA: Stanford University Press, pp. 349–82.

Swaminathan, A. and J.B. Wade (2001), ‘Social movement theory and the evolutionof new organizational forms’, in C. Schoonhoven and E. Romanelli (eds), TheEntrepreneurship Dynamic: Origins of Entrepreneurship and the Evolution ofIndustries, Stanford, CA: Stanford University Press, pp. 286–313.

Constructing entrepreneurial opportunity 119

Thornton, P. (1999), ‘The sociology of entrepreneurship’, Annual Review ofSociology, 25: 19–46.

Thrift, N. and K. Olds (1996), ‘Refiguring the economic in economic geography’,Progress in Human Geography, 20: 311–37.

Tilly, C. (1978), From Mobilization to Revolution, Reading, MA: Addison-Wesley.Tushman, M.L. and P. Anderson (1986), ‘Technological discontinuities and organ-

izational environments’, Administrative Science Quarterly, 31: 439–65.Union of Concerned Scientists (UCS) (1980), Energy Strategies Toward a Solar

Future: A Report of the Union of Concerned Scientists, Cambridge, MA: Ballinger.United States Department of Energy (2001), Annual Energy Report, US DOE

website, www.eia.gov, September.Utrup, K.A. (1979), ‘How Sierra Club members see environmental issues’, Sierra,

March/April: 14–18.Wade, J.B., A. Swaminathan and M.S. Saxon (1998), ‘Normative and resource flow

consequences of local regulations in the American brewing industry, 1845–1918’,Administrative Science Quarterly, 43: 905–35.

Weber, A. ([1909] 1928), Theory of the Location of Industries, translated by C.J.Friedrich, Chicago: University of Chicago Press.

Weber, M. (1978), Economy and Society: An Interpretive Sociology, New York:Bedminister Press.

Weick, K.E. (1995), Sensemaking in Organizations, Thousand Oaks, CA: Sage.Zucker, L., M. Darby and M. Brewer (1998), ‘Intellectual human capital and the

birth of U.S. biotechnology enterprises’, American Economic Review, 88:290–306.

120 Industrial dynamics

6. Absorptive capacity and foreignspillovers: a stochastic frontierapproachJojo Jacob and Bart Los*

1. INTRODUCTION

In the literature on economic growth in developing countries, internationaltechnology flows have gained growing attention. International technologycan ‘flow’ from the originating country to the receiving country in severalways. Among them, foreign direct investment (FDI) and trade in inter-mediate inputs have been the subject of a great deal of empirical work.Most studies choose firms or plants as units of analysis and adopt a neo-classical production function framework, in which the average response ofthe endogenous productivity variable to a change in one of the exogenousvariables (such as the intensity of FDI and trade in intermediate inputs) isestimated by means of classical regression analysis. Although econometricscrutiny does not always confirm strong anecdotal evidence, the majorityof studies do find significant positive impacts of international flows.1

Most studies adopt an approach based on production functions, often bymeans of panel data regressions. One of the most prominent disadvantagesis the impossibility of obtaining an understanding of the causes ofobserved heterogeneity. Rather, the focus of these studies is on ‘representa-tive behaviour’ (or, ‘average behaviour’). Deviations from this behaviourare merely seen as realisations of a random noise process. If, for example,productivity performances show an increasing variance over time, the pro-duction function approach does not yield any insights, as the effect is justan increase in the variance of the stochastic random noise process.

In this chapter, we borrow an interesting alternative approach from asubfield in econometrics called ‘stochastic frontier analysis’ (SFA) to studylabour productivity growth in Indonesian plants.2 Typical techniquesbelonging to this subfield do not estimate ‘average relationships’ betweenvariables, but relationships for best-practice plants. Furthermore, theyenable researchers to use plant characteristics as potential explanations of

121

the extent to which other plants’ performances fall below best practice. Inthe context of our chapter, this implies that we shall estimate relationshipsbetween inputs (capital, labour) and output (value added) for best-practicefirms for several years, to get indications of the degree to which internationaltechnology spillovers affected productivity growth. Simultaneously, weshall link the underperformance of other firms to variables that relate to theevolutionary concept of absorptive capacity (Cohen and Levinthal, 1990),such as labour quality, presence and strength of links to foreign markets,ownership, experience and so on. The results are quantifications of thefailure to fully assimilate international technology spillovers, and thereby toraise productivity to its potential level.3

The organisation of the chapter is as follows. In Section 2, we shall brieflyreview some theories of productivity growth that are relevant for ourempirical approach. Section 3 proposes our methodology. It deals with the‘appropriate technology’ accounting framework and discusses the way inwhich frontiers and distances to these frontiers are estimated. Section 4 isdevoted to data issues. Special attention will be paid to the procedures weadopted to clean the dataset. In Section 5, we shall present our results.Section 6 concludes and proposes a few directions for future research.

2. SELECTED THEORIES ABOUT PRODUCTIVITYGROWTH

Convergence (or its absence) of labour productivity levels has attracted a lotof attention, both from economic theorists and from more empirically ori-ented scholars. Although it is hard to classify theories in a field characterisedby synthesis and hybridisation, roughly two categories of theories can bediscerned. We follow Nelson and Pack (1999) in using the labels ‘accumu-lation’ and ‘assimilation’ theories. Accumulation theories basically assumethat raising capital intensities (be it physical capital or human capital) auto-matically leads to labour productivity growth, although increasingly moreinvestment is required for a given productivity gain. In this view, labourproductivity growth is governed by movements along the production func-tion of a given country, sector or firm under consideration. This perspectiveimplies that accumulation of capital is the cause for growth in labour pro-ductivity.

Assimilation theories challenge this view. Here, technology is seen assomething that does not automatically and immediately flow across firms orcountries. Instead, only firms or countries that have invested sufficiently intheir ‘absorptive capacities’ will be able to turn innovations developed else-where into productivity gains for themselves. In the view of assimilationists,

122 Industrial dynamics

policies to stimulate entrepreneurship and eagerness to learn have beenmuch more important. Such a view on macroeceonomic performance can,with relatively minor modifications, be transferred to studies at firm or plantlevel. The resource-based view of the firm (see Teece, 2000, for example)stresses that long-run firm performance is mainly determined by learningcapabilities.

In this chapter, we shall differentiate between two barriers to attainingproductivity levels attained by better-performing plants. The first type ofbarrier relates most strongly to issues mentioned above. Pack (1987) andVan Dijk (2005, Ch. 8) show that plants that are similar in terms of thetypes of machines installed attain widely varying productivity levels.4

Apparently, learning and organisational capabilities are not identicallydistributed across plants, which shows up in different productivity figuresfor plants with more or less identical equipment installed. The second typeof barrier is quite closely associated with what Abramovitz (1989) labelled‘technological incongruence’. A similar idea has recently been proposed inthe form of a formal model by Basu and Weil (1998). They defined tech-nologies as particular combinations of inputs, or, in other words,capital–labour ratios. New knowledge is only ‘appropriate’ for a range ofsuch technologies. Firms or countries will in the short run only be permit-ted to benefit from innovations if these relate to technologies that arecomparable to the ones operated.5 In the longer run, non-appropriateinnovations can become appropriate if the firm or country invests to suchan extent that it shifts its technology to a capital intensity level compar-able to the innovating firm or country.6 The predictions concerning con-vergence and divergence of productivity levels that follow from the Basuand Weil model are based on the assumption that more capital-intensivetechnologies allow producers to attain higher maximum levels of labourproductivity.

Los and Timmer (2005) showed that both types of barriers to catch-upplay an important role in the empirics of macroeconomic growth. We adoptseveral parts of their methodology to investigate how innovations, changesin absorptive capacity and technologies operated contribute to the produc-tivity growth experiences at the plant level in Indonesia’s manufacturingsector.

3. METHODOLOGY

In this section, we shall describe the empirical methodology we use. First,we shall outline how we decompose labour productivity growth (or decline)of a plant into the effects of innovation, assimilation and equipment

Absorptive capacity and foreign spillovers 123

upgrading which creates potential for spillovers. Next, we shall discuss theestimation methods required to arrive at quantifications of these effects. Inthe third subsection, we shall describe the empirical model.

Identifying the Sources of Growth

Los and Timmer (2005) decomposed labour productivity growths rates ofa group of countries, between 1970 and 1990, into the effects of movementstowards the frontier, or changes in technical efficiency (assimilation), move-ments of the frontier (innovation), and capital deepening (creating poten-tial). The decomposition form itself was popularised by Kumar and Russell(2002), but Los and Timmer were the first to link their results to the theo-ries discussed in the previous section. Our approach starts from a similarperspective. It is novel in the sense that it explicitly relates the decompo-sition results to observable characteristics of the plants.

Figure 6.1 shows a plant’s actual labour productivity levels y0 and y1, inan industry with production frontiers f 0 and f1, for periods 0 and 1, respec-tively. The labour productivity change (y1/y0) of this plant can be decom-posed in the following way:

(6.1)

or

(6.2)

In the first term on the right-hand side , a value of larger than0 indicates that the plant under consideration has increased its labour pro-ductivity for the technology operated. In other words, it indicates that theplant has been able to bring about an increased exploitation of technolog-ical potential as compared to the maximum productivity observed for theequipment operated. We call this the ‘assimilation’ effect.7 The secondexplanatory factor indicates the changes in labour productivitydue to increases in capital intensity alone. While a higher capital intensityin itself does not generate higher labour productivity, it can lead to anupward shift in the attainable or the ‘target’ productivity levels, dependingon the slope of the frontier. Therefore, a value greater than 0 for can beinterpreted as ‘creating potential’. The third factor points to theeffect of localised technological change that results in the upward shift ofthe production frontier. Assuming that the plant’s capital intensity remains

(1 � yI)yC

(1 � yC)

yA(1 � yA)

(1 � yT) � (1 � yA) · (1 � yC) · (1 � yI).

y1y0

� �y1yd

· yay0� · �yc

ya ·

ydyb�0.5

· �ybya

· ydyc�0.5

124 Industrial dynamics

constant, a positive value for indicates that it has benefited from anincrease in the maximum attainable labour productivity levels for the giventechnologies. We call this the ‘innovation’ effect.

Los and Timmer (2005) estimated the productivity frontiers for thebeginning and end periods using data envelopment analysis (DEA). Wefollow a similar approach, but with the key difference that we derive thefrontier labour productivity levels by means of stochastic frontier analy-sis (SFA). This change of method has advantages and drawbacks. Themajor drawback is that truly localised innovation cannot be modelled, asthe estimated elasticity of foreign research and development (R&D)spillovers (the source of technological change) is the same across the fullrange of technologies. As a result, the shifts in the frontier labour pro-ductivity levels always amount to an identical proportional growth rate

yI

Absorptive capacity and foreign spillovers 125

Figure 6.1 Labour productivity growth decomposition

Y/L

C/L

F(1)

F(0)

* (0)

* (1)

yd

yb

yc

ya

y1

y0

across the full range of technologies. The distance to the frontier, however,can well change, thereby allowing potentials for spillovers to change.The major advantage is that the location of the frontier is not very sensi-tive to measurement errors for a small number of firms. As is well known(see, for example, Coelli et al., 1998), DEA results can be distorted quitea bit. In view of the sizeable measurement and reporting errors that areoften found in plant-level surveys, especially in developing countries,we feel that the net advantage of SFA as compared to DEA is clearlypositive.

Estimation Method

In recent years, a number of studies have employed stochastic frontier esti-mations for estimating and explaining inefficiencies of firms and plants inindustries. Until recently, the standard approach was a two-stage esti-mation procedure, in which the production frontier is first estimated. In thesecond stage, the resulting inefficiencies (the vertical distances from theobserved productivities to the estimated frontier) are regressed on firm-specific variables (see, for example, Pitt and Lee, 1981).8 Estimation in thesecond stage, however, contradicts the assumption of identically distrib-uted inefficiency effects that underlies the estimation of the stochastic fron-tier in the first stage. To overcome this methodological problem, severalauthors have suggested single-stage procedures for simultaneously estimat-ing both the stochastic frontier and inefficiency functions. The Battese andCoelli (1995) model is one such approach.

Consider the following production function for panel data:

(6.3)

where yit is the dependent variable corresponding to the ith plant and timet, X is a vector of explanatory variables, and �it is the composite error term.It consists of a white noise error vit: and uit. The two setsof disturbances are assumed to be independent. The uits are non-negativerandom variables associated with technical inefficiencies, and are assumedto be independently (but not identically) distributed as truncations (at zero)of the distribution, with:

(6.4)

in which Z is a vector of observable, non-stochastic explanatory variablesassociated with technical inefficiency, and � is a vector of unknowncoefficients.

�it � Zit�,

N(�it, �2u)

vit ~ iid N(0, �2v)

yit � � Xit � �it,

126 Industrial dynamics

In this model, the maximum likelihood method is used for the simul-taneous estimation of the parameters of the frontier and technicalinefficiency models, that is, estimation of the values of the unknown par-ameters s, �s, �u

2 and �v2. We computed the estimates using the FRON-

TIER software package (Coelli, 1996). Battese and Coelli (1995) alsoprovide an expression for the conditional expectation of exp(�uit) given eit.The maximum likelihood estimation of this function is used to estimate thetechnical efficiency index of the ith firm at time t, based on expected valuesconditional on the observed values of the explanatory variables in X andZ. When the productivity frontier is expressed in logarithms, the technicalefficiency index (TEI) can be expressed as follows:

(6.5)

This index has a value between 0 and 1, with 0 (uit��) indicating the leastefficient, and 1 (uit�0) the most efficient plants.

Changes in TEI as defined in equation (6.5) denote a part of the actualshift in labour productivity. When a change in TEI causes an upward shift,as in Figure 6.1, it can be interpreted as associated with the assimilation oftechnology-specific knowledge. It is that part of the assimilation effectwhich can be explained by the changes in the indicators of absorptivecapacity, given their estimated coefficients from the SFA model. Theremainder of the upward shift cannot be explained, and is calculated as thedifference between the actual growth in labour productivity and the pre-dicted growth in labour productivity derived from the SFA model.

The Empirical Model

We begin with a production frontier of the Cobb–Douglas form, aug-mented to account for technological change. In most developing countries,and especially so in Indonesia, foreign technology is the key source of tech-nological progress, because of the virtual absence of own technologicalefforts. We account for technology flows from abroad by constructing ameasure of international R&D stock (IRD). The augmented productionfunction is defined as follows:

(6.6)

where, Yit is value added of plant i at time t, E is total energy use (as a proxyfor capital goods use, as further explained below), L is total number ofworkers, and IRD is the international R&D stock representing the tech-nology flows available to all plants in the industry (see the following section

Yit � AEitL

itIRD�

t,

TEIit � exp(�uit).

Absorptive capacity and foreign spillovers 127

for a fuller description of the variables). The variable IRD is interpreted tobe the driver of shifts in the production frontier in a given industry.Dividing Y and E by L and taking logarithms, equation (6.6) becomes

(6.7)

where the lower-case symbols denote variables in logarithms. In the trans-formation of equation (6.6) to (6.7), we impose the assumption of constantreturns to scale in the rival inputs labour and energy.9 We use equation (6.7)as the frontier function that will be estimated simultaneously with theinefficiency function, based on the procedure described in the previoussubsection.

Given that technology-embodied inputs have often shown to be an impor-tant channel of foreign technology diffusion, a plant’s access to importsmight be a good proxy of its ‘access to foreign technology’. Access to asource of technology does not, however, imply that the acquisition of tech-nology is guaranteed. This is because technology is not entirely ‘codified’,and indeed often takes a highly ‘tacit’ form (Polanyi, 1958). Therefore, theextent to which a plant is able to ‘absorb’ knowledge related to new tech-nologies, also known as a plant’s absorptive capacity (Cohen and Levinthal,1990), can depend on the quality of its labour force. Evenson and Westphal(1995) proxied this quality by the proportion of scientists and engineers ina plant’s workforce.

The ‘ownership structure’ of a plant can also be a significant factorinfluencing the capacity to assimilate knowledge. A plant with foreign man-agement control might be expected to run more productively than, forexample, a non-professional, family-controlled enterprise. The ‘foreignconnection’ may enable the former to adapt itself much more quickly thanthe latter to global changes in technology, production relations and so on.

The performance of enterprises as compared to other enterprises withsimilar technologies may also depend on its ‘size’. As noted by Tybout(2000), in many developing countries, the demand for manufacturedproducts is skewed towards simple items which can be efficiently producedusing cottage techniques. An opposite effect would be the operation ofSchumpeterian dynamics that leads to greater learning efforts by large firms.This may result from scale economies, availability of internal resources inthe presence of imperfect markets and/or uncertainties, synergies betweentechnological, production, marketing and distribution activities, and so on.The empirical evidence, mostly pertaining to advanced economies, shows noconsensus, however (see Marsili, 2001, for an overview).

Another factor that may influence technical efficiency is the ‘age’of a plant.Experienced plants may enjoy the benefits of learning by doing. As Klepper

yit � lit � a � (eit � lit) � IRDt � �it,

128 Industrial dynamics

(2002) argues, with increases in competitive conditions firms with greaterexperience have greater leeway in enhancing their capabilities. Keeping theseconsiderations in mind, we consider the following absorptive capacity vari-ables for the mean inefficiency model represented by equation (6.4):

where, Access represents access to technology spillover, defined as the shareof imported intermediate inputs in total intermediate inputs; LQual standsfor the quality of labour at a plant, defined as the share of non-production(white-collar) workers in total employment; Foreign represents the propor-tion of foreign ownership in a plant; Age is measured as the differencebetween the year of inception and the year of operation; and Size is definedas the logarithm of the total number of workers.

A final aspect to consider is the influence on technical efficiency offactors observable only to the managers of a plant, which are not reflectedin survey-based dataset like ours. As a result, such firm-specific effects (orheterogeneities) may be related to other regressors of the model, which maycause biased results. To overcome this problem, we adopt a specificationthat incorporates plant fixed effects in the inefficiency model.

4. DATA ISSUES

Our analysis focuses on the manufacturing sector of a developing country,Indonesia. Our main data sources are two large plant-level datasets, back-cast and Statistic Industri (SI), constructed by the Indonesian Bureau ofStatistics (Badan Pusat Statistik, BPS) (see Appendix A of Jacob and Los,2005, for a detailed description of the data sources, variables, cleaningprocesses and so on). The datasets cover all large and medium-sized plantsin the manufacturing sector of the country, from 1975 to 2001.10 Afterapplying cleaning procedures to account for duplications, reporting errorsand data entry errors, we focus our analysis on industries defined at a lowlevel of aggregation (5-digit classification). This allows us to investigateproductivity for sets of plants with homogeneous activities. Since the paneldata SFA approach is data intensive, we select 17 industries for which atleast 10 plants are included in the data set (see Appendix Table 6A.1). Theindustries under investigation are quite diverse, which allows us to identifyinter-industry differences in the importance of absorptive capacity for pro-ductivity performance.

Studies aiming at explaining total factor productivity (TFP) growthare often hindered by immeasurable fluctuations in capacity utilisation.

Zit: Accessit; LQualit; Foreignit; Agei; Sizeit,

Absorptive capacity and foreign spillovers 129

Although we do not study TFP growth but labour productivity growth, weencounter similar problems: Our technologies are defined by capital inten-sities, that is, ratios between a (quasi-) fixed input and a much more vari-able input. We circumvent this problem by proxying a plant’s capital use byits energy consumption, about which much information is available in ourdataset.11 This is also convenient from a practical point of view, since ourdata on energy use cover a much longer time span than those on investmenton which we should base our capital stock estimates.

As a determinant of labour quality, we do not have detailed informationabout skills in our database for a sufficiently long period. Therefore, weproxy differences in skill level of labour force across plants by differences inthe share of non-production workers in the total workforce.

Finally, we should describe how we estimate international R&D stocksthat capture technology flows. Since Indonesian firms generally do notundertake any formal R&D activities themselves, it can safely be assumedthat new technology must come from abroad (Hill, 1996). Our admittedlypoor, but widely accepted assumption if suitable output indicators of inno-vation are not available, is that technology production is proportional toR&D expenditures. We have data on R&D expenditures by industry for 10countries that together account for approximately 60 per cent of theimports to Indonesia and about 85 per cent of the total OECD R&Dexpenditure. The selection of this sample is justified because empirical evi-dence suggests that ‘it is not the intensity of import per se that matters, butrather the distribution of the countries of origin. The more you importfrom highly R&D intensive countries, the larger the impact of foreignR&D’ (Lichtenberg and van Pottelsberghe de la Potterie, 1998, p. 1483).Apart from imports, technology purchase, technology collaboration andexports by Indonesian firms as well as foreign investment in the domesticmarket can all act as carriers of technology spillovers. To accommodatethese different channels of technology flow, we do not weight foreign R&Dstock by the volume of imports to Indonesia, which is a standard approachin the spillover literature. Instead, we consider solely the technologicalproximity between the foreign and domestic industries to weight the foreignR&D stock. The specific channels of foreign technology flows are intro-duced in the inefficiency function.

In the first step, foreign R&D stock is weighted by an index of techno-logical distance between the sector of origin and the sector of destination.We use a patent-based measure of technological distance between sectorsderived by Verspagen (1997) from the EPO (European Patent Office) data.In the second step, the resulting R&D stock is weighted by an index of tech-nological congruence between sectors in the advanced economy andIndonesia that are comparable in terms of their classification code. This

130 Industrial dynamics

weight accounts for inter-country differences between sectors. It capturesthe idea that an industry in a follower country benefits more from the globalpool of technology, the greater its technological congruence with industriesin advanced countries. The resulting international R&D stock at the indus-try level can be expressed in terms of the following equation:

(6.8)

where IRDj is the international R&D stock resulting from technology flowsavailable to all plants in the Indonesian industry j; RDck is the R&D stock insector k of partner country c; Pkj is an element of the patent information flowmatrix P, and it captures the flow of sector k’s R&D efforts to sector j; andScj is the technological congruence between sector j of Indonesia and thesame sector of its partner country c. Scj is derived by comparing the inputcoefficient vectors for sector j in the two countries. It takes a value of 1 ifsector j is perfectly similar in the two countries, and zero in the event of com-plete dissimilarity between them (see Appendix B of Jacob and Los, 2005).

Given the fact that the R&D data we use are available only at a level ofaggregation of 2, 3 and, in a few cases, 4-digit (ISIC Rev. 2), IRDj in theabove equation corresponds to these levels. To generate IRD at the 5-digitlevel, we constructed similarity indices between the two sets of classi-fications, using their respective input coefficients vectors.

5. RESULTS

Results for Frontier and Inefficiency Estimation

We begin by documenting some summary statistics of the variables used inthe SFA model in 17 5-digit-industry, plant-level samples. Table 6.1a showsthe means and standard deviations across plants of the levels of the vari-ables. Table 6.1b shows similar statistics of the average annual growth ratesof the variables. The highest average labour productivity level and growthrate is found for ‘drugs’ (35222). The lowest average values for these twovariables are found for ‘tobacco’ (31410) and ‘clay tiles’ (36422), respec-tively. The former industry is also found to have the lowest average energyintensity, while ‘plywood’ (33113) has the highest. Plants in ‘garments’(32210) recorded the fastest average growth in energy intensity, while thosein ‘crumb rubber’ (35523) recorded the slowest growth. Plants in ‘paints’(35210) could benefit from the biggest pool of relevant R&D done inforeign countries. The smallest pool is found for ‘sawmills’ (33111). In other

IRDj(t) � �c, k

(RDckPkjScj) (t),

Absorptive capacity and foreign spillovers 131

132

Tab

le 6

.1a

Sum

mar

y st

atis

tics

of

17 5

-dig

it I

SIC

indu

stri

es:

leve

ls

Indu

stry

(IS

IC)

Pla

nts

Val

/lab

Egy

/lab

IRD

Age

Fore

ign

Acc

ess

Lqu

alS

ize

3117

122

3.78

21.

221

5.13

717

.182

0.00

60.

023

0.08

715

1.99

6(4

.925

)(1

.026

)(0

.000

)(7

.624

)(0

.027

)(0

.044

)(0

.069

)(1

60.8

60)

3117

927

6.50

25.

761

8.53

619

.759

0.03

30.

074

0.17

730

8.57

3(6

.495

)(5

.073

)(0

.000

)(1

5.42

1)(0

.134

)(0

.165

)(0

.160

)(4

07.2

10)

3141

015

3.24

10.

364

7.03

025

.967

0.00

10.

071

182.

561

(3.5

21)

(0.6

61)

(0.0

00)

(10.

542)

(0.0

04)

(0.0

63)

(120

.635

)31

420

348.

472

0.80

17.

532

30.0

590.

048

0.09

973

4.49

3(8

.500

)(0

.983

)(0

.000

)(1

4.03

3)(0

.105

)(0

.065

)(9

84.1

63)

3211

468

7.83

630

.881

36.2

9320

.794

0.03

20.

062

0.11

748

4.55

6(7

.151

)(3

2.01

4)(0

.000

)(9

.159

)(0

.151

)(0

.136

)(0

.070

)(5

50.7

15)

3212

126

5.06

92.

996

29.6

8124

.846

0.07

40.

106

265.

920

(2.8

67)

(2.1

09)

(0.0

00)

(12.

699)

(0.1

44)

(0.0

67)

(260

.919

)32

130

209.

469

8.84

650

.235

16.9

500.

084

0.14

434

0.53

8(5

.964

)(6

.301

)(0

.000

)(9

.202

)(0

.214

)(0

.083

)(3

50.7

49)

3221

074

6.63

21.

144

35.5

0216

.716

0.03

20.

130

0.09

450

1.82

5(6

.830

)(1

.248

)(0

.000

)(1

0.99

6)(0

.132

)(0

.199

)(0

.066

)(6

80.2

47)

3311

113

20.4

8965

0.56

31.

907

14.3

460.

065

0.00

50.

179

452.

494

(12.

767)

(600

.405

)(0

.000

)(5

.757

)(0

.197

)(0

.016

)(0

.071

)(5

77.7

91)

3311

320

20.1

9439

28.3

073.

684

10.4

000.

045

0.02

40.

152

1445

.056

(8.1

83)

(424

1.68

)(0

.000

)(3

.144

)(0

.139

)(0

.035

)(0

.060

)(7

66.1

16)

3321

117

9.09

96.

223

4.65

417

.618

0.05

60.

159

319.

110

(4.1

72)

(4.8

11)

(0.0

00)

(10.

588)

(0.0

81)

(0.0

60)

(396

.681

)34

200

4511

.421

3.80

298

.361

24.1

220.

113

0.26

016

9.70

4(8

.181

)(5

.118

)(0

.000

)(1

6.93

9)(0

.133

)(0

.121

)(1

16.6

85)

133

3521

013

27.4

843.

332

321.

818

25.5

000.

165

0.35

40.

293

162.

660

(30.

403)

(3.0

62)

(0.0

00)

(17.

239)

(0.2

62)

(0.2

33)

(0.1

53)

(104

.254

)35

222

4628

.679

4.85

421

1.06

420

.261

0.24

70.

686

0.39

923

3.13

3(2

2.03

7)(5

.229

)(0

.000

)(1

0.03

1)(0

.349

)(0

.243

)(0

.195

)(1

33.3

55)

3552

320

17.5

9432

9.39

516

.434

16.8

500.

242

0.00

00.

162

239.

948

(16.

738)

(243

.739

)(0

.000

)(3

.843

)(0

.430

)(0

.000

)(0

.067

)(8

5.36

7)35

606

297.

259

7.52

614

.390

14.6

380.

587

0.13

933

2.15

8(5

.550

)(5

.669

)(0

.000

)(4

.998

)(0

.242

)(0

.058

)(4

00.6

33)

3642

218

3.33

52.

255

38.8

1222

.722

0.02

40.

069

135.

398

(1.6

75)

(1.6

73)

(0.0

00)

(15.

973)

(0.0

69)

(0.0

71)

(73.

559)

Not

es:

(i)

Val

/lab

– va

lue

adde

d pe

r em

ploy

ee;E

gy/la

b–

ener

gy u

se p

er e

mpl

oyee

;Siz

e–

num

ber

ofem

ploy

ees;

IRD

– in

tern

atio

nal R

&D

sto

ck in

rea

l19

90 p

urch

asin

g po

wer

par

ity

(PP

P)

US

dolla

rs (

mill

ions

).V

alue

add

ed a

nd e

nerg

y ar

e in

tho

usan

ds o

fre

al 1

990

PP

P U

S do

llars

.(i

i)Se

e m

ain

text

for

the

defi

niti

ons

ofth

e ot

her

vari

able

s.(i

ii)Se

e A

ppen

dix

Tab

le 6

A.1

for

indu

stry

defi

niti

ons.

(iv)

Mea

ns,s

tand

ard

devi

atio

ns in

bra

cket

s.

134

Tab

le 6

.1b

Sum

mar

y st

atis

tics

of

17 5

-dig

it I

SIC

indu

stri

es:

aver

age

annu

al g

row

th r

ates

Indu

stry

(IS

IC)

Pla

nts

Val

/lab

Egy

/lab

IRD

Fore

ign

Acc

ess

Lqu

alS

ize

3117

122

0.19

10.

243

0.05

6�

0.03

8*�

0.27

10.

062

0.04

3(0

.160

)(0

.178

)(0

.000

)(0

.296

)(0

.078

)(0

.063

)31

179

270.

209

0.30

40.

056

0.04

40.

102

0.10

00.

066

(0.2

11)

(0.3

71)

(0.0

00)

(0.0

62)

(0.5

83)

(0.1

56)

(0.0

71)

3141

015

0.40

90.

550

0.05

64.

327*

0.10

80.

067

(0.3

65)

(1.0

34)

(0.0

00)

(0.1

65)

(0.1

03)

3142

034

0.19

50.

978

0.05

6�

0.23

20.

401

0.03

0(0

.228

)(2

.950

)(0

.000

)(0

.488

)(1

.703

)(0

.057

)32

114

680.

262

0.20

8�

0.02

8�

0.02

50.

218

0.06

90.

051

(0.2

08)

(0.2

87)

(0.0

00)

(0.0

44)

(1.5

38)

(0.1

10)

(0.0

69)

3212

126

0.21

90.

270

�0.

028

0.23

60.

114

0.05

1(0

.207

)(0

.225

)(0

.000

)(2

.084

)(0

.142

)(0

.064

)32

130

200.

366

0.19

5�

0.02

8�

0.08

80.

090

0.07

7(0

.513

)(0

.212

)(0

.000

)(0

.327

)(0

.086

)(0

.083

)32

210

740.

286

2.41

7�

0.02

80.

099

0.46

90.

175

0.07

4(0

.260

)(1

7.70

6)(0

.000

)(0

.164

)(2

.822

)(0

.492

)(0

.114

)33

111

130.

314

0.17

90.

032

0.00

3�

0.73

50.

108

0.05

1(0

.210

)(0

.330

)(0

.000

)(0

.005

)(0

.442

)(0

.126

)(0

.051

)33

113

200.

196

0.32

60.

032

0.01

90.

920

0.20

20.

108

(0.1

56)

(0.4

10)

(0.0

00)

(0.0

27)

(3.2

95)

(0.2

98)

(0.0

92)

3321

117

0.26

30.

128

0.03

21.

033

0.03

70.

120

(0.3

77)

(0.1

52)

(0.0

00)

(4.6

09)

(0.0

84)

(0.0

75)

3420

045

0.21

40.

274

�0.

047

�0.

010

0.08

90.

034

(0.1

78)

(0.2

56)

(0.0

00)

(0.7

64)

(0.1

26)

(0.0

65)

135

3521

013

0.23

30.

350

0.03

90.

025

0.16

90.

058

0.04

7(0

.226

)(0

.321

)(0

.000

)(0

.026

)(0

.882

)(0

.091

)(0

.053

)35

222

460.

432

0.24

70.

061

0.00

5�

0.01

40.

039

0.04

2(0

.608

)(0

.216

)(0

.000

)(0

.036

)(0

.304

)(0

.079

)(0

.063

)35

523

200.

314

0.10

00.

030

�0.

022

�0.

656*

0.06

90.

016

(0.1

53)

(0.1

74)

(0.0

00)

(0.0

39)

(0.1

01)

(0.0

33)

3560

629

0.26

80.

310

0.03

0�

0.15

70.

066

0.10

8(0

.202

)(0

.340

)(0

.000

)(0

.336

)(0

.097

)(0

.098

)36

422

180.

174

0.23

30.

019

�0.

212

0.11

70.

038

(0.0

91)

(0.2

46)

(0.0

00)

(0.2

92)

(0.1

92)

(0.0

46)

Not

e:*

The

gro

wth

rat

e of

the

vari

able

und

er c

onsi

dera

tion

is p

osit

ive

in o

nly

one

obse

rvat

ion.

For

oth

er n

otes

,see

Tab

le 6

.1a.

variables too, differences are noticeable. For example, the access-to-spillover variable has a maximum average value of nearly 0.70 (‘drugs’,35222) and a minimum average value of less than 0.01 (‘crumb rubber’,35523). Foreign ownership is absent in eight of the 17 industries. Averageforeign ownership grew fastest in ‘garments’ (32210), while it declined in‘macaroni’ (31171) and ‘crumb rubber’ (35523).12 Another noticeablefeature in both Tables 6.1a and 6.1b is the high standard deviationsreported for many of the variables across plants in the industries consid-ered. Quite often, the standard deviation is larger than the mean value. This

136 Industrial dynamics

Table 6.2 Stochastic frontier estimates for 17 5-digit (ISIC Rev. 2)industriesa

Industry 31171 31179 31410 31420 32114 32121 32130 32210(ISIC Rev. 2)

Production functionConstant �5.132 �11.661 5.484 �3.450 27.665 33.378 54.907 38.426

(1.765)* (19.375) (5.993) (0.998)* (5.495)* (6.428)* (11.056)* (3.900)*Egy/lab 0.121 0.111 0.047 �0.056 0.169 0.139 0.122 0.090

(0.039)* (0.040)* (0.025)* (0.086) (0.029)* (0.037)* (0.053)* (0.022)*IRD 0.885 1.413 0.244 0.852 �1.050 �1.403 �2.543 �1.626

(0.114)* (0.225)* (0.382) (0.089)* (0.314)* (0.376)* (0.533)* (0.229)*

(Mean) Inefficiency functionConstant 1.701 3.924 3.049 �0.220 4.296 1.933 1.649 3.313

(0.406)* (18.938) (1.132)* (0.972) (0.786)* (1.286) (1.373) (0.718)*Age �0.251 0.219 �0.769 �0.028 �0.370 0.105 0.011 �0.037

(0.120)* (0.144) (0.426)* (0.333) (0.180)* (0.232) (0.201) (0.083)Foreign �0.312 �1.409 4.408 �0.630

(0.293) (0.364)* (11.460) (0.255)*Access 0.651 0.049 7.028 0.311 �0.039 0.022 0.164 �0.258

(0.330)* (0.442) (4.324) (0.999) (0.149) (0.179) (0.485) (0.111)*Lqual �0.688 �1.445 �1.967 0.119 �0.903 �0.581 0.184 �0.220

(0.231)* (0.338)* (1.499) (0.997) (0.356)* (0.496) (0.685) (0.376)Size �0.292 0.261 �0.535 �1.097 0.119 �0.021 0.194 0.061

(0.093)* (0.116)* (0.183)* (0.214)* (0.074) (0.100) (0.114)* (0.056) Gamma 0.126 0.601 0.765 0.542 0.492 0.821 0.206 0.342

(0.049)* (1.917) (0.107)* (0.220)* (0.537) (0.082)* (0.811) (0.447) Plants 22 27 15 34 68 26 20 74Observations 264 324 180 408 816 312 240 888

Notes:a Standard errors are in parentheses; * significant at 10%.(i) Egy/lab – e/l.(ii) See main text for the definitions of the other variables.

(iii) See Appendix Table 6A.1 for industry definitions.

phenomenon reflects the highly dual structure of the Indonesian manufac-turing sector, described in detail by, among others, Hill (1996).

Table 6.2 reports the SFA estimation results. For brevity, we do notreport the estimation results for the plant dummies included in theinefficiency function. The results for the frontier production function showthat the coefficients of both energy intensity, Egy/Lab, and the inter-national R&D stock representing knowledge spillovers, IRD, have a posi-tive sign in most industries. The estimated coefficients of energy intensitiesfor the slopes of the productivity frontiers are generally fairly small,however, and even statistically insignificant at the 10 per cent level for fourindustries. This implies that it does not pay off very much for plants just toinvest more, as is suggested by advocates of accumulationist theories.Consequently, accumulation alone cannot be considered as an importantsource of productivity growth in Indonesian manufacturing.13

Absorptive capacity and foreign spillovers 137

33111 33113 33211 34200 35210 35222 35523 35606 36422

Production function�1.746 14.732 �12.599 24.417 �20.476 �14.688 �27.614 �20.899 �40.764(2.131) (8.177)* (1.025)* (1.031)* (4.554)* (1.464)* (4.848)* (1.521)* (15.799)*0.043 0.180 0.062 0.072 0.122 0.006 0.147 0.165 0.144

(0.052) (0.039)* (0.043) (0.035)* (0.049)* (0.031) (0.045)* (0.039)* (0.040)*0.869 �0.466 1.391 �0.818 1.580 1.332 2.126 1.765 2.832

(0.150)* (0.537) (0.071)* (0.056)* (0.237)* (0.077)* (0.288)* (0.090)* (0.850)*

(Mean) Inefficiency function1.627 1.607 �0.298 �1.135 2.671 2.051 0.360 1.520 3.564

(0.691)* (0.582)* (0.818) (0.426)* (1.360)* (0.600)* (0.210)* (0.629)* (4.923)�0.048 �0.221 0.088 0.117 0.200 �0.305 �0.137 �0.304 �0.505(0.171) (0.227) (0.240) (0.123) (0.260) (0.132)* (0.026)* (0.180)* (0.211)*0.860 2.767 �1.103 �0.122 0.111

(0.785) (3.456) (0.422)* (0.391) (0.428)�0.623 �2.181 0.946 �0.142 �0.007 �0.215 0.153 0.119 �0.974(1.037) (2.197) (0.947) (0.257) (0.228) (0.116)* (1.029) (0.136) (0.418)*0.020 �2.717 0.581 �0.670 0.431 0.652 �0.456 �1.713 �0.696

(0.862) (1.055)* (1.202) (0.401)* (0.673) (0.342)* (0.217)* (0.764)* (0.764)�0.410 0.039 0.240 0.088 0.499 0.477 0.010 0.238 �0.319(0.223)* (0.185) (0.278) (0.085) (0.205)* (0.120)* (0.127) (0.143)* (0.144)*1.000 0.442 0.224 0.249 0.354 0.092 0.000 0.096 0.456

(0.000)* (0.179)* (0.128)* (0.044)* (0.095)* (0.021)* (0.000)* (0.027)* (0.875)13 20 17 45 13 46 20 29 18

156 240 204 540 156 552 240 348 216

The sensitivity of the frontier to increases in foreign R&D appears to beprominent. Apparently, the best-practice plants in Indonesia reaped sub-stantial fruits from technological spillovers from abroad. The coefficient forthis variable was statistically significant for 10 of the 17 industries studied,at the 10 per cent level.

Our main interest lies in understanding the factors that cause deviationsfrom the best-practice technology. The estimate for the variance parameter(Gamma in Table 6.2) that corresponds to the estimated share of theinefficiency term in the variance of the composite error term has a positivesign in all industries, and is significant in most industries (12 industries).This should be considered as evidence for the idea that inefficiency effectscontribute substantially to the variety of labour productivity levels indi-cated by positive standard deviations for this variable.

A negative sign for the coefficient of a variable indicates a negativeimpact of that variable on inefficiency. Among all the variables, changes inlabour quality (LQual) variable provide the most promising explanation forchanges in comparison to best-practice performance. Its coefficient has anegative sign in most industries (11 out of 17), and is statistically significantin seven industries.

Foreign ownership (Foreign) has a significantly negative coefficient onlyin three of the nine industries in which the shares held by foreign firms arepositive in one or more establishments. For the remaining industries,changes in the degree of foreign ownership do not appear to matter forassimilation. It might well be that a linear specification of the inefficiencyeffects is not most appropriate here. Explorations to use multiple-regimeeconometrics (to identify critical values of the degree of foreign owner-ship), however, are beyond the scope of this chapter, if only because suchanalyses have hardly been attempted in the SFA branch of econometrics.

We argued before that access to spillover (Access) is likely to exert amajor influence on the technical efficiency of plants. However, this variableyielded a negative coefficient in only eight industries, with statisticalsignificance limited just to three industries. One reason for this result couldbe the narrowness of our measure of access to spillover as it does not con-sider the import of capital goods. Second, the import intensity of inter-mediate inputs use is rather low in most industries as may be required togenerate sufficient within-plant variations. An additional, and probably themost important, cause of very few significant results is the huge measure-ment errors that characterise datasets like ours. Although, we did ‘clean’ thedata extensively, it is unlikely that this has removed all measurement errors.

The Age variable demonstrated a favourable impact on assimilation in amajority of the industries. A negative sign for its coefficient in 11 out of atotal number of 17 industries appear to suggest that a plant’s ability to

138 Industrial dynamics

assimilate knowledge spillovers from similar best-practice plants increaseswith its experience. As argued by Klepper (2002), under competitive pres-sures, firms with greater experience are better positioned to enhance theircapabilities. Our period of analysis covers the export-oriented phase –hence, the more competitive phase – of industrialisation in Indonesia. Wemay therefore conclude that firms which have been in operation for a longerperiod of time have been more successful in enhancing their technologicaland managerial capabilities, and therefore in meeting the challenges ofincreased competition.

The final variable to be discussed is Size, which displays considerableinter-industry variations in its influence. It had an adverse impact on assim-ilation in a majority of the industries (a positive coefficient in 11, with stat-istical significance in five industries). Of the six remaining industries whereits influence has been favourable, in five cases the coefficients displayed stat-istical significance.

The discussion so far has revealed that changes in deviations from bestpractice (due to plant-specific factors) are a significant determinant of aplant’s labour productivity growth. In the following subsection we extendthese results by decomposing labour productivity growth into that result-ing from the shifts in the frontier of an industry’s technology, capital deep-ening and efficiency gains (or losses). We provide theoretically groundedinterpretations for the contributions of these three factors to averagelabour productivity growth in an industry.

Results for the Decomposition Analysis

We decomposed plant-level labour productivity growth during the 1985–96period in each of the 17 industries under consideration. We then calculatedtheir industry average using the geometric average of the initial and finalyear employment shares of plants as the weight.14 Table 6.3 shows thedecompositions of average labour productivity growth rates during the1985–96 period. The first column shows the compound annual averagegrowth in labour productivity and the second column, the period growthrate of productivity. The remaining columns represent the contribution ofthe four components to (period) labour productivity growth.

All industries experienced a positive growth in labour productivity duringthis period, with important inter-industry variations as may be expected ofa heterogeneous group of industries. In a majority of the industries, we seethat productivity growth resulted from a combination of assimilation (unex-plained) and innovation. While the former was the main contributor in mostof the industries, the latter was the leading contributor in industries likepaints & varnishes, plastics and rubber. Explained assimilation was a major

Absorptive capacity and foreign spillovers 139

contributor in about five industries important among which were tobacco,plywood, cigarettes and clay tiles.

The contribution of creating spillover potential was very limited in allindustries. This result is mainly due to the flat shapes of the estimated fron-tiers. Increasing capital intensity hardly contributes to a higher potentiallabour productivity. In other words, learning or assimilation potentialsremained more or less stagnant.

6. CONCLUSIONS AND FUTURE RESEARCH

In this chapter, we proposed an SFA approach to study labour productiv-ity growth in the Indonesian manufacturing sector (in the late 1980s andearly 1990s). We focused specifically on the effects of inflows of foreigntechnology and the role of absorptive capacity changes on plant-leveldifferences in productivity growth.

Our main findings are that foreign R&D played a significant role inmoving the productivity frontier in an upward direction. Hence, the inflow

140 Industrial dynamics

Table 6.3 Decomposition of productivity growth, 1985–1996

Industry Annual Period Contribution to productivity growth (ISIC growth growth

Explained Unexplained Innovation PotentialRev. 2) (%) (%)assimilation assimilation

31171 0.704 8.022 3.776 0.414 3.826 0.00731179 0.567 6.412 �1.783 3.172 4.857 0.16531410 0.421 4.733 6.928 �2.668 1.152 �0.67931420 0.383 4.295 1.818 0.448 2.158 �0.12932114 0.309 3.447 0.717 1.579 0.875 0.27732121 0.518 5.854 �0.253 3.330 2.641 0.13632130 0.756 8.637 �1.103 4.033 6.141 �0.43532210 0.171 1.896 0.092 0.828 1.010 �0.03433111 0.873 10.037 0.991 6.520 3.081 �0.55433113 0.023 0.252 0.997 �0.240 �0.721 0.21633211 0.515 5.810 �2.623 5.081 3.627 �0.27434200 0.480 5.414 �0.077 3.694 1.659 0.13935210 1.138 13.254 �1.098 4.496 8.102 1.75335222 0.660 7.504 0.307 3.247 3.943 0.00635523 0.606 6.871 0.578 1.913 4.816 �0.43635606 0.316 3.529 0.193 1.177 2.586 �0.42636422 0.782 8.944 4.219 �0.268 5.199 �0.206

Note: See Appendix Table 6A.1 for industry definitions.

of technology was an important source of productivity growth for best-practice plants. The creation of spillover potential (by investing to use morecapital-intensive technologies) barely contributed to labour productivitygrowth, since relations between capital intensity and labour productivitywere almost non-existent for best-practice plants. As a consequence, shift-ing to higher capital intensities hardly implied more potential spillovers tobenefit from. This finding is clearly at odds with the major assumptionsunderlying the accumulationist theories of growth.

Assimilation (movements towards the frontier) played an importantrole. We could distinguish between two kinds of assimilation effects. Thefirst type could be explained by our absorptive capacity indicators, suchas labour quality and degree of foreign ownership. For many industries,we found estimation results that underline the importance of buildingabsorptive capacity for assimilating knowledge from best-practice firmsthat operate similar technology. In a quantitative sense, however, theseeffects were often dwarfed by the second kind of assimilation effects.These unexplained assimilation effects were very big and dominated thecomposite effect.

The importance of unexplained assimilation is worrisome on the onehand, in the sense that we cannot explain much. On the other hand, itconfirms our feeling that much heterogeneity of plants is not captured bysurvey-based datasets. Our absorptive capacity indicators are rough ones,and are subject to considerable measurement error. In our view, more in-depth case studies (like Pack, 1987, and Van Dijk, 2005) offer much betteropportunities for assessing the importance of foreign technology anddifferences in absorptive capacity. Studies like ours can play a useful role ininvestigating to what extent case-study results can be generalised. In thatsense, the next steps along the lines of the present chapter could be to probefurther into the inter-industry differences concerning the estimation resultsand to link such differences to differences in the types of technologies used.Industrial taxonomies, for example, as proposed by Pavitt (1984) mightconstitute an interesting and worthwhile starting point in this respect.

NOTES

* A first version of this chapter was presented at the European Meeting on AppliedEvolutionary Economics 2005 (Utrecht, The Netherlands). Useful comments by partici-pants and, in particular, Koen Frenken and Fabio Montobbio are gratefully acknowl-edged.

1. An extensive overview of empirical studies was recently given in Keller (2004).2. A modern textbook is Kumbhakar and Lovell (2000).3. The analysis can be conceived as an empirical approach to part of the technology-

assimilation model of Nelson and Pack (1999).

Absorptive capacity and foreign spillovers 141

4. Pack (1987) studied the performance of textile plants in Kenya, the Philippines and theUK. Van Dijk (2005) focused on the productivity levels of paper-making plants inIndonesia and Finland.

5. Basu and Weil (1998) illustrate this concept by arguing that new knowledge pertainingto the very capital-intensive maglev trains in Japan will not be useful to transporters inBangladesh using very capital-extensive bullock cart technologies.

6. Atkinson and Stiglitz (1969) introduced the concept of ‘localised learning by doing’ bywhich they suggested that firms improve the productivity of a particular mix of capitaland labour over time. Basu and Weil (1998) extended this notion by emphasising theimportance of ‘localized knowledge spillovers’.

7. Below, we shall argue that our estimation framework allows us to decompose assimi-lation effects into ‘explained assimilation’ (explained by means of absorptive capacityindicators) and ‘unexplained assimilation’.

8. For a recent survey, see Wang (2003).9. The scale of operation could be important in learning, for instance because big firms

often have more contacts with suppliers, are often more strongly represented in profes-sional associations and so on. To investigate the learning effect of scale, we include thevariable ‘plant size’ in our inefficiency function.

10. We shall limit our analysis to the 1985–96 period due to the better quality of the datasince 1985 and to the crisis of 1997–99.

11. This approach dates back to the seminal neoclassical macroeconomic growth account-ing study by Jorgenson and Griliches (1967).

12. In Tables 6.1a and 6.1b, the standard deviations for the variable IRD always equal zero.This is due to the fact that we defined the international R&D stocks as industry-levelvariables (see Section 4).

13. It should be noted, however, that this result might be partly due to the fact that newequipment is often more energy saving than more outdated machinery. Consequently,rises in energy consumption might overestimate rises in capital intensity.

14. Prior to deriving the industry average, the multiplicative components of the decomposi-tion equation were transformed, by taking their logarithms, into additive components.

REFERENCES

Abramovitz, M. (1989), Thinking About Growth, Cambridge: CambridgeUniversity Press.

Atkinson, A.B. and Stiglitz, J.E. (1969), ‘A new view of technological change’,Economic Journal, 79, 573–8.

Basu, S. and Weil, D.N. (1998), ‘Appropriate technology and growth’, QuarterlyJournal of Economics, 113, 1025–54.

Battese, G.E. and Coelli, T.J. (1995), ‘A model for technical inefficiency effects in astochastic frontier production function for panel data’, Empirical Economics, 20,325–32.

Coelli, T.J. (1996), ‘A guide to FRONTIER Version 4.1: a computer program forstochastic frontier production and cost function estimation’, CEPA (Centre forEfficiency and Productivity Analysis) Working Paper, No. 7, University of NewEngland, Department of Econometrics, University of New England, Armidale.

Coelli, T.J., Rao, D.S.P. and Battese, G.E. (1998), An Introduction to Efficiency andProductivity Analysis, Boston, MA: Kluwer Academic.

Cohen, W. and Levinthal, D. (1990), ‘Absorptive capacity: a new perspective onlearning and innovation’, Administrative Science Quarterly, 35, 128–52.

Evenson, R. and Westphal, L. (1995), ‘Technological change and technology strat-

142 Industrial dynamics

egy’, in T.N. Srinivasan and J. Behrman (eds), Handbook of DevelopmentEconomics, Vol. 3, Amsterdam: North-Holland, 2209–29.

Hill, H. (1996), The Indonesian Economy since 1966: Southeast Asia’s EmergingGiant, Cambridge: Cambridge University Press.

Jacob, J. and Los, B. (2005), ‘The impact of international technology spillover andabsorptive capacity on productivity growth in Indonesian manufacturing firms’,Paper presented at the European Meeting on Applied Evolutionary Economics(EMAEE), Utrecht, May 19–21.

Jorgenson, D.W. and Griliches, Z. (1967), ‘The explanation of productivity change’,Review of Economic Studies, 34, 249–83.

Keller, W. (2004), ‘International technology diffusion’, Journal of EconomicLiterature, 62, 752–82.

Klepper, S. (2002), ‘The capabilities of new firms and the evolution of the US auto-mobile industry’, Industrial and Corporate Change, 11(4), 645–66.

Kumar, S. and Russell, R.R. (2002), ‘Technological change, technological catch-upand capital deepening: relative contributions to growth and convergence’,American Economic Review, 92, 527–49.

Kumbhakar, S.C. and Lovell, C.A.K (2000), Stochastic Frontier Analysis,Cambridge: Cambridge University Press.

Lichtenberg, F.R. and van Pottelsberghe de la Potterie, B. (1998), ‘InternationalR&D spillovers: a comment’, European Economic Review, 42(8), 1483–91.

Los, B. and Timmer, M. (2005), ‘The “appropriate technology” explanation of pro-ductivity growth differentials: an empirical approach’, Journal of DevelopmentEconomics, 77, 517–31.

Marsili, O. (2001), The Anatomy and Evolution of Industries: Technological Changeand Industrial Dynamics, Cheltenham, UK and Northampton, MA, USA:Edward Elgar.

Nelson, R.R. and Pack, H. (1999), ‘The Asian miracle and modern growth theory’,Economic Journal, 109, 416–36.

Pack, H. (1987), Productivity, Technology, and Industrial Development: A CaseStudy in Textiles, Oxford and New York: Oxford University Press.

Pavitt, K. (1984), ‘Sectoral patterns of technical change: towards a taxonomy anda theory’, Research Policy, 13, 343–73.

Pitt, M.M. and Lee, L.F. (1981), ‘The measurement and sources of technicalefficiency in the Indonesian weaving industry’, Journal of DevelopmentEconomics, 9, 43–64.

Polanyi, M. (1958), Personal Knowledge: Towards a Post-Critical Philosophy,Chicago: University of Chicago Press.

Teece, D.J. (2000), Managing Intellectual Capital, Oxford: Oxford University Press.Tybout, J.R. (2000), ‘Manufacturing firms in developing countries: how well do

they do and why?’, Journal of Economic Literature, 38, 11–44.Van Dijk, M. (2005), ‘Industry evolution and catch up: the case of the Indonesian

pulp and paper industry’, unpublished PhD thesis, University of Eindhoven.Verspagen, B. (1997), ‘Estimating international technology spillovers using tech-

nology flow matrices’, Weltwirtschaftliches Archiv, 133, 226–48.Wang, H.-J. (2003), ‘A stochastic frontier analysis of financing constraints on

investment: the case of financial liberalization in Taiwan’, Journal of Business andEconomic Statistics, 21, 406–19.

Absorptive capacity and foreign spillovers 143

APPENDIX 6A

144 Industrial dynamics

Table 6A.1 Industrial classification

No. Industry ISIC Rev. 2

1 Macaroni, spaghetti, noodles etc. 311712 Bakery products 311793 Dried tobacco and processed tobacco 314104 Clove cigarettes 314205 Weaving mills except gunny and other sacks 321146 Made-up textiles except wearing apparel 321217 Knitting mills 321308 Wearing apparel made of textiles (garments) 322109 Sawmills 33111

10 Plywood 3311311 Furniture and fixtures, mainly wood 3321112 Printing, publishing and allied industries 3420013 Paints, varnishes and lacquers 3521014 Drugs and medicines 3522215 Crumb rubber 3552316 Plastics, bags, containers 3560617 Clay tiles 36422

PART III

Network Analysis

7. Informational complexity and theflow of knowledge across socialboundariesOlav Sorenson, Jan W. Rivkin and Lee Fleming

1. INTRODUCTION

Scholars from a variety of backgrounds – economists, sociologists, strate-gists and students of technology management – have sought a better under-standing of why some knowledge disperses widely while other knowledgedoes not. In this quest, some researchers have focused on the characteris-tics of the knowledge itself (for example, Polanyi, 1966; Reed andDeFillippi, 1990; Zander and Kogut, 1995) while others have emphasizedthe social networks that constrain and enable the flow of knowledge (forexample, Coleman et al., 1957; Davis and Greve, 1997). This chapter exam-ines the interplay between these two factors.

Specifically, we consider how the complexity of knowledge and thedensity of social relations jointly influence the movement of knowledge.Imagine a social network composed of patches of dense connections withsparse interstices between them. The dense patches might reflect firms, forinstance, or geographic regions or technical communities. When doesknowledge diffuse within these dense patches circumscribed by socialboundaries but not beyond them? Synthesizing social network theorywith a view of knowledge transfer as a search process, we argue thatknowledge inequality across social boundaries should reach its peak whenthe underlying knowledge is of moderate complexity.1 To test this hypoth-esis, we analyse patent data and compare citation rates across three typesof social boundaries: within versus outside the firm, geographically nearto versus far from the inventor, and internal versus external to the tech-nological class. In all three cases, the disparity in knowledge diffusionacross these borders is greatest for knowledge of an intermediate level ofcomplexity.

147

2. THE TRANSFER OF COMPLEX KNOWLEDGE

Our hypotheses build on two themes in the literature. First, the transfer ofknowledge from one party to another typically requires effort on the partof the recipient to fill gaps in the transmitted knowledge and to correcttransmission errors. The acquisition of knowledge therefore is best seen notas the receipt of a complete, well-packaged gift, but rather as a searchprocess. Second, social networks, and consequently the social boundariesthat shape them, critically influence that search process.

Knowledge Receipt as Search

Following the lead of evolutionary economists (Nelson and Winter, 1982),we think of a unit of knowledge as analogous to a recipe. The list of ‘in-gredients’ might include both physical components and processes. Therecipe further explains how to combine these components and processes –in what order, in which proportions, under what circumstances – to achievea desired end. Viewing knowledge as a recipe leads naturally to thinking ofinnovation as a search for new recipes. Following a long tradition – begin-ning at least as early as Schumpeter (1939) – we explicitly model innovationas a search process; inventors explore the space of possible combinationsof ingredients (that is, recipes) for new and better alternatives. In discussingthis process, we adopt the idea of a fitness ‘landscape’ as a metaphor for thecharacteristics of the search space. Innovators search these landscapes forpeaks and plateaus, which correspond to good recipes, useful inventionsand profitable strategies.

Once a useful innovation has been discovered, transferring its recipe,even between cooperative actors, can fail for at least two reasons. First, therecipient usually does not fully understand the original recipe, as a result ofimperfections in the transfer process. He/she must therefore begin a searchfor the missing information and to correct the errors in his/her (imperfect)copy of the recipe. Second, the local ingredients and the experience of therecipient rarely match those of the sender; recipients may therefore need toadapt the original recipe to their own context. Knowledge recipients do notact as passive beneficiaries; they actively search, recreate and build upon theoriginal recipes.

In this process, the transfer of certain types of recipes is particularlydifficult. For instance, knowledge characterized by causal ambiguity(Lippman and Rumelt, 1982), a high degree of tacitness (Polanyi, 1966), ordifficult codification (Zander and Kogut, 1995) may resist transfer becauseany communication of such recipes proceeds only with many and largegaps. Our focus, however, is on the informational complexity of transferred

148 Network analysis

knowledge. We consider a piece of knowledge complex if it comprises manyelements that interact richly (Simon, 1962), and we pay special attention tothe intensity of interdependence among the ingredients in a recipe.

To connect the degree of informational complexity to the characteristicsof the space that inventors search, we use Kauffman’s NK model (Kauffman,1993; see also Frenken and Nuvolari, 2004). N denotes the number of(binary) elements in a system while K represents the degree to which thesecomponents interact in determining the fitness of a particular configurationof components. In our context, N is the number of ingredients used in arecipe, and K is the richness of the interactions between those ingredients.

To understand better the way in which the model relates interdependenceto the search process, consider two examples with N�3. Figure 7.1 depicts afitness landscape for a recipe with no interdependence between its compo-nents. Each vertex represents a different potential configuration; the arrowsconnecting them show paths towards higher fitness levels. When K �0,Kauffman randomly assigns a fitness from the uniform unit distribution toeach value (0 or 1) of each element. The overall fitness value for a particularconfiguration is the average of each element’s fitness contribution. As one cansee, any starting point on this landscape leads to the unique optimum (011).

Figure 7.2, on the other hand, illustrates an example with N�2. Thevalue of the fitness contribution for each component then depends not juston its value but also on the values of two other components. Each compo-nent therefore can contribute any of eight (2�2�2) different fitness levels.Kauffman again randomly assigns these values from the uniform unit

Informational complexity and the flow of knowledge across boundaries 149

Note: This relatively correlated landscape has only one minimum and one maximum, 100(0.40) and 011 (0.80), respectively. The component fitness contributions come from auniform [0,1] distribution.

Figure 7.1 Landscape without interdependence (N�3, K�0)

123 w1 w2 w3

000 0.8 0.6 0.2 0.53001 0.8 0.6 0.9 0.77010 0.8 0.7 0.2 0.57011 0.8 0.7 0.9 0.80100 0.4 0.6 0.2 0.40101 0.4 0.6 0.9 0.63110 0.4 0.7 0.2 0.43111 0.4 0.7 0.9 0.67

W ��

N

i�1wi

N

000(0.53)

001(0.77)

010(0.57)

100(0.40)

110(0.43)

111(0.67)

011(0.80)

101(0.63)

distribution. Even in this simple example, one can see that interdependencecomplicates search; depending on where one begins, an agent using a simplehill-climbing algorithm could arrive at either the global maximum (101) ora local one (000).

Complex knowledge resists transfer by making it difficult for a recipientto fill transmission gaps. On the landscapes depicted, a gap is equivalent tonot knowing the correct value for the global optimum of one of the threeelements. Interdependence produces two effects that undermine the recipi-ent’s attempts to regenerate the original recipe (that is, identify theoptimum). First, small errors in transmission cause large problems wheningredients cross-couple in a rich manner. Second, interdependence leadsto a proliferation of ‘local peaks’. These peaks undermine improvementthrough incremental search because changing any single element degradesthe quality of the outcome (Kauffman, 1993). As a result, searchers fre-quently find themselves trapped on local peaks (that is, inferior recipes)when faced with high interdependence.

Complexity and Access to a Template

Success in acquiring and employing complex knowledge depends cruciallyon access to the original success, which serves as a template (Nelson andWinter, 1982: 119–20; Winter, 1995). For reasons considered below, indi-viduals vary in their degree of access to the template. Superior access facili-tates the receipt of knowledge by allowing the recipient to commencesearch with fewer errors and by permitting him/her to solicit advice from

150 Network analysis

Note: This relatively uncorrelated landscape has multiple local minima, 001(0.30) and 100(0.37), and maxima, 000 (0.63) and 101 (0.87).

Figure 7.2 Landscape with maximal interdependence (N�3, K�2)

123 w1 w2 w3

000 0.5 0.8 0.6 0.63001 0.6 0.2 0.1 0.30010 0.2 0.7 0.3 0.40011 0.8 0.6 0.5 0.47100 0.4 0.5 0.2 0.37101 0.9 0.8 0.9 0.87110 0.7 0.4 0.1 0.40111 0.9 0.7 0.3 0.63

W ��

N

i�1wi

N

000(0.63)

001(0.30)

010(0.40)

100(0.37)

110(0.40)

111(0.63)

011(0.47)

101(0.87)

the source during the search process. Consider two actors, both attemptingto assimilate a valuable piece of knowledge but who differ in their access tothe template. The first has superior, though still imperfect, access to andunderstanding of the original, successful recipe. The second has far pooreraccess. How valuable is the first actor’s superior access to the templateduring the search process? We contend that the value depends on the com-plexity of the knowledge being transferred.

Suppose first that the ingredients used in the recipe do not interact (thatis, K�0). In this situation, the first actor’s access to the template does notproduce a persistent advantage. Through routine, incremental search, thesecond actor can reconstruct the recipe. Few local peaks threaten to trapthe poorly informed recipient. As a result, both actors eventually fareequally well; search on the part of a recipient can easily substitute for high-fidelity transmission.

Next consider knowledge with an intermediate degree of interdependence.Local peaks now appear, but they remain relatively few in number. The well-informed actor begins his/her search near, but not precisely at, the originaloptimal set of ingredients. Through incremental search, he/she can find theproper combination of ingredients. The second actor, who begins his/hersearch farther from the target and receives less guidance about the directionin which to explore, more likely becomes ensnared on some local peak, awayfrom and inferior to the original success. Here superior template access givesthe first actor an advantage the second cannot recreate through search.

Finally, imagine a piece of maximally interdependent knowledge (that is,K�N – 1): ingredients depend on one another in an extremely delicate way.Local peaks now pervade the landscape and neither actor’s incrementalsearch will likely build on the original knowledge with any success. The firstactor’s superior access to the template thus has little value beyond thesecond’s inferior access.

Taken together, these arguments imply that the advantage of superiorbut imperfect access to the template reaches its peak at moderate levels ofinterdependence between knowledge components (Rivkin, 2001, developsthis argument further, with the aid of simulations).

Social Boundaries and Template Access

Access to the template depends on the distribution of social relations,which provide the channels through which valuable information flows(Hägerstrand, 1953; Coleman et al., 1957). These social relations do notlink actors at random. Rather, sociologists have consistently noted anddemonstrated that networks concentrate within the boundaries of com-munities and organizations. Our study tests the salience of three types of

Informational complexity and the flow of knowledge across boundaries 151

social boundaries – organizational memberships, geographic regions andtechnological communities – in structuring social networks, and concomi-tantly influencing the flow of knowledge.

Consider organizational boundaries first. A firm attempting to replicateand build on its own prior success has better access to its knowledge thanwould an outside imitator, both because fellow members of an organizationshare codes, specialized languages and beliefs that facilitate high-fidelitytransmission (Arrow, 1974) and because strong interpersonal ties and densesocial networks inside a firm provide access to the template (Granovetter,1985). As argued above, the value of this access peaks for transmittingknowledge of intermediate complexity:

Hypothesis 1 The advantage in receiving and applying knowledge thatmembers of the same firm have over members of different firms reachesits maximum for knowledge of intermediate interdependence.

In other words, the insider’s advantage over the outsider has an inverted U-shaped relation to the interdependence of the knowledge.

Actors belonging to the same geographic unit (for example, city, countryor state) as the innovator also have superior access to the template. The geo-graphic concentration of social relations reflects a variety of factors: thegreater odds that individuals in close proximity encounter one another(Festinger et al., 1950), the high costs of maintaining distant ties (Zipf, 1949;Boalt and Janson, 1957), and the prevalence of local cultures (Benedict,1934). We therefore expect that actors physically close to a source of knowl-edge have better access to it:

Hypothesis 2 A nearby knowledge recipient’s advantage in receivingand applying knowledge over a distant recipient peaks for knowledge ofintermediate interdependence.

To the extent that networks localize geographically, even within a firm,organizations find it difficult to diffuse knowledge beyond its point oforigin. Within a firm, then, we expect simple knowledge to spread broadlyand highly complex knowledge to remain isolated within a single team ordepartment. Knowledge of moderate complexity, however, should spreadwithin a firm to the edges of a facility or locale, but not to geographicallydistant installations:

Hypothesis 3 Within a firm, a nearby knowledge recipient’s advantagein receiving and applying knowledge over a distant recipient reaches itsmaximum for knowledge of intermediate interdependence.

152 Network analysis

An analogous argument applies to technological communities (alsocalled communities of practice, defined in terms of cognitive proximity).Actors who work in the same technological domain as an inventor havesuperior access to the template. Universities, trade associations, profes-sional societies, industry consortia and work experience foster dense socialconnections within such technological communities. These communitiesalso develop common knowledge and communal languages that can facil-itate knowledge transfer. Membership within a common technologicalcommunity thus engenders superior access to the template, which shouldhave its greatest impact when the target knowledge displays moderate inter-dependence:

Hypothesis 4 The advantage in receiving and applying knowledge thata member of a technological community has over a non-member of thecommunity reaches its maximum with knowledge of intermediate inter-dependence.

3. EMPIRICAL CORROBORATION

To test these hypotheses, we analysed prior art citations to all US utilitypatents granted in May and June of 1990 (n�17 264). The data camefrom the Micro Patent database and NBER public access data on patents(Hall et al., 2001). As in many previous studies, we view a prior art citationas evidence of knowledge diffusion. Our statistical approach involves esti-mating the likelihood that a focal patent receives a citation from a futurepatent as a function of several factors: the interdependence of the knowl-edge underlying the focal patent, the status of the citing patent’s inventoras an insider or outsider on some dimension with respect to the focalpatent, the interaction of interdependence and insider/outsider status,and a set of control variables. The results of the estimation allow us toexamine how the likelihood of insider citation compares to the likelihoodof outsider citation and, crucially, whether the gap between the two prob-abilities peaks when the focal patent embodies moderately interdependentknowledge.

Our unit of analysis is a patent dyad, one patent issued in May or Juneof 1990 and one issued later that may or may not cite the first. Hence ourapproach conceptually follows that of other studies of the likelihood of tieformation – in this case, the likelihood that a future patent builds on theknowledge embodied in one of our focal patents. Specifically, our analysisfollows Sorenson and Stuart (2001) in adopting a case-control approach toanalysing the formation of ties. We begin by including all cases of future

Informational complexity and the flow of knowledge across boundaries 153

patents, from July 1990 to June 1996, that cite any of our 17 268 focalpatents: 60 999 in total. Since these citations occur, the dependent variablefor these cases takes a value of ‘1’ to denote a realized citation. In addition,we pair each of the 17 268 focal patents with four future patents that do notcite it (but that could have), with the dependent variable set to zero. Fromthis set of 130 071, we restrict our analysis to the dyads in which both inven-tors reside in the United States, leaving us with a set of 72 801 dyads.

Interdependence

For each dyad, we measure the complexity of the knowledge in the focalpatent, k, by observing the historical difficulty of recombining the elementsthat constitute it (Fleming and Sorenson, 2001). Though the metricinvolves intensive calculation, the intuition behind it is simple: a technol-ogy whose components have, in the past, been mixed and matched readilywith a wide variety of other components has exhibited few sensitive inter-dependencies and receives a low value of k. The measure takes the sub-classes identified in a patent as proxies for the underlying components (seeFleming and Sorenson, 2004, for survey-based validation of the measure).

We compute k in two stages. Equation (7.1) details our calculation of theease of recombination, or inverse of interdependence, for each subclass iused in patent j. We first identify every use of subclass i on patents from 1980to 1990. The denominator is simply the tally of the number of patents witha classification in subclass i. To compute the numerator, we count the numberof different subclasses appearing with subclass i on previous patents. Hence,our measure increases as a particular subclass combines with a wider varietyof technologies, controlling for the total number of applications. This termcaptures the ease of combining a particular technology.

(7.1)

To create our measure of interdependence for an entire patent, we invertthe average of the ease of recombination scores for the subclasses to whichit belongs (equation 7.2).

(7.2)

�Count of subclasses on patent j

�j�i

Ei

.Interdependence of patent j � Kj

�Count of subclasses previously combined with subclass i

Count of previous patents in subclass i .

Ease of recombination of subclass i � Ei

154 Network analysis

Social Boundaries

Three variables capture the insider/outsider status of the potential citinginventor with respect to the holder of a focal patent. The variables reflectmembership in organizational, regional and technical communities. Sameassignee is set to one if two patents in a dyad share a common owner andis zero otherwise. Geographic proximity is equal to the natural log of the dis-tance in miles between the first inventors listed on the two patents in a dyadmultiplied by negative one (so that we expect larger values to increase thelikelihood of citation). Same class is set to one if two patents belong to thesame primary technological class – a proxy for shared membership in acommunity – and is zero otherwise. In all three cases, we test our hypothe-ses by interacting k and its square with the proxy for the density of socialnetworks – whether due to firm boundaries, geographic proximity, or tech-nological similarity. The benefits of social proximity should peak for inven-tions of moderate complexity.

The regressions also include several control variables. Subclass overlapis the number of subclasses that the two patents in the dyad have incommon divided by the number of subclass memberships for the (poten-tially) citing patent. An activity control estimates the typical number ofcitations received by a patent in the same technological areas as the focalpatent (see Fleming and Sorenson, 2001). Recent technology is the averagereference number of the patents listed as prior art, a measure of the close-ness of the patent to the technological frontier. We also include counts ofthe number of backward patent citations and backward non-patent ci-tations, the number of class memberships, and the number of subclassmemberships for the focal patent. We report robust standard errors andcorrect for potential bias in logistic regression of rare events (King andZeng, 2001).

4. RESULTS

The results appear in Table 7.1. Model 1 tests hypothesis 1 by interactingk and k2 with same assignee. As expected, membership within the same firmproduces the greatest diffusion advantage over outsiders for knowledgeof intermediate complexity, as evidenced by the positive coefficient onk � same assignee and the negative coefficient on k2�same assignee. Theinteractions between geographic distance and interdependence in model 2tests hypothesis 2, again showing strong support. Model 3 tests hypothesis3 by re-estimating model 2, but only for dyads where both patents belongto the same firm. In essence it asks: does geography still matter for

Informational complexity and the flow of knowledge across boundaries 155

156 Network analysis

Table 7.1 Rare events logit models of the likelihood of a focal patentreceiving a citation from a future patent

Model 1 Model 2 Model 3 Model 4 Model 5Only same

assignee

k 1.687••• 1.526••• 4.821••• 1.444••• 1.070•• (0.302) (0.307) (0.359) (0.321) (0.362)

k2 �0.892••• �0.793••• �4.208••• �0.704••• �0.359•• (0.082) (0.074) (0.117) (0.098) (0.107)

k�same assignee 2.969••• 6.231••• (0.515) (0.555)

k2�same assignee �3.420••• �9.851••• (0.203) (0.273)

k�–ln (dist) 0.835••• 6.047••• 0.885••• (0.131) (0.213) (0.139)

k2�–ln (dist) �0.794••• �4.566••• �1.025•• (0.044) (0.074) (0.066)

k�same class 3.019•• 5.733••• (1.146) (1.032)

k2 �same class �1.396••• �5.409••• (0.363) (0.330)

Same assignee 0.343 0.389 0.432 0.172 (0.280) (0.276) (0.281) (0.278)

–ln (dist) 0.499••• 0.428••• 0.500••• 0.499••• 0.354••• (0.031) (0.031) (0.066) (0.030) (0.029)

Same class 3.800••• 3.663••• 1.878••• 3.837••• 3.448•••(0.306) (0.307) (0.394) (0.302) (0.299)

Subclass 4.230••• 4.190••• 3.767••• 4.114••• 4.150••• overlap (0.316) (0.317) (0.349) (0.316) (0.314)

Activity control 0.393 0.389 �0. 746•• 0.477 0.466 (0.287) (0.287) (0.248) (0.388) (0.289)

Recent 0.122 0.195 0.010 0.096 �0.024technology (0.171) (0.170) (0.309) (0.165) (0.151)

Backward patent 0.002 0.013 0.025•• �0. 001 �0. 002citations (0.013) (0.013) (0.008) (0.014) (0.014)

Backward non- 0.018•• 0.014• �0.126••• 0.019•• 0.011• patent (0.006) (0.006) (0.037) (0.005) (0.005)citations

Number of 0.070 0.054 0.456 0.041 0.052 classes (0.140) (0.140) (0.249) (0.138) (0.137)

Number of �0. 016 �0. 021 0.170••• 0.001 0.010 subclasses (0.045) (0.045) (0.048) (0.044) (0.044)

Constant �9.224••• �9.953••• �7.148••• �9.206••• �9.162••• (0.714) (0.703) (1.208) (0.684) (0.675)

Log-likelihood �22262.4 �22261.4 �2294.1 �22255.9 �22251.1N 72801 72801 6497 72 801 72801

Note: • p�0.05; •• p�0.01; ••• p�0.001. Model 3 includes only those dyads for whichsame assignee�1.

knowledge diffusion within firms? In support of hypothesis 3, the resultsreveal that even within firm boundaries, social networks influence the flowof knowledge, with the greatest disparity between local diffusion anddistant diffusion arising for knowledge of moderate interdependence.Model 4 tests the salience of technological communities. Once again, theestimates show strong support; the impact of technological communitymembership on citation probability peaks for intermediate k. Model 5includes all three measures of social proximity simultaneously and showsthat each has an independent and significant effect when estimatedtogether, in support of hypotheses 1, 2 and 4.

As expected, the value of superior access to the template reaches amaximum for knowledge of moderate interdependence, regardless ofwhether the superior access comes from organizational membership,geographic proximity, or technological community membership. In ad-dition to being significant, the effects have substantial economic import.For simple or highly complex knowledge, the insider has no greater like-lihood than the outsider of attaining and building on the knowledge in afocal patent. For knowledge of moderate complexity at the gap-maximiz-ing levels of k, a firm insider is 218 per cent more likely than an outsiderto transfer knowledge effectively; an inventor located in the same zip codeis 160 per cent more likely to absorb the knowledge in a region than oneat the average distance (�665 miles); and a technological insider is 238per cent as likely as a technological outsider to build on knowledge inthe class.

5. DISCUSSION

Our analyses considered the impact of superior access to some originalknowledge on the likelihood of diffusion as a function of knowledge com-plexity, using three indicators of social proximity. All knowledge recipients,near and far, compete on equal footing when assimilating simple knowledge.Highly complex knowledge, on the other hand, equally resists diffusion toboth classes of would-be recipients. For knowledge whose ingredientsdisplay a moderate degree of interdependence, however, superior but imper-fect access to the template translates into better knowledge reproduction.Thus in our patent data, the largest gap between the ability of a close recip-ient to build on prior knowledge relative to the ability of a distant recipientarises when the cited patent involves moderate interdependence.

Our findings speak to the question, when does inequality of knowledgearise across social borders? One might initially suspect that highly complexknowledge, the most difficult to reproduce, would create the greatest

Informational complexity and the flow of knowledge across boundaries 157

inequality. But this intuition ignores the fact that inequality in its sharpestform requires some diffusion: to create the most inequity across socialboundaries, knowledge must creep up to the edge of the thick patch of con-nections in which it originated but not beyond. This phenomenon, we haveargued, most likely occurs for moderately complex knowledge. Thus, forexample, one might expect industries based on moderately complex knowl-edge to display especially wide intra-industry dispersion in long-runfinancial returns.

Our argument may also shed light on a conundrum of the literature oneconomic geography. Explanations for agglomeration based on infor-mation spillovers assume that membership in a local community allowsfirms to benefit from the knowledge developed by other firms in the region,but that firms outside the region are excluded from these benefits (forexample, Marshall, 1890; Arrow, 1962). What type of knowledge wouldhave such a characteristic? The literature to date has focused on ‘tacit’knowledge, but typically uses the term simply to refer to uncodified (asopposed to uncodifiable) knowledge (an endogenous outcome of firms’decisions to invest in codification; Brökel, 2005). Our results, building onKauffman’s NK model, point to a different (presumably more exogenous)dimension: informational complexity. Industries that rely on moderatelycomplex knowledge might be especially likely to display geographicagglomeration (for empirical corroboration, see Sorenson, 2004).

Our empirical results come from patent data alone, but the basic logicof our hypotheses applies to knowledge in general, not just the knowl-edge underlying inventions (see Wolter, 2006, for a model based on inter-dependence in production). Hence, future research might usefullyexamine these dynamics across a wide range of applications – includingorganizational learning, the diffusion of management practices, knowl-edge management, and the sustainability of knowledge-based competi-tive advantage.

NOTE

1. This version describes the intuition underlying our theoretical model and reports novelempirical results. For those interested in a more detailed description of the theoreticalmodel, see Rivkin (2001) and Sorenson et al. (2004).

REFERENCES

Arrow, Kenneth J. (1962), ‘The economic implications of learning by doing’, Reviewof Economic Studies, 29: 155–73.

158 Network analysis

Arrow, Kenneth J. (1974), The Limits of Organization, New York: Norton.Benedict, Ruth (1934), Patterns of Culture, New York: Houghton-Mifflin.Boalt, Gunnar and C.G. Janson (1957), ‘Distance and social relations’, Acta

Sociologica, 2: 73–97.Brökel, Tom (2005), ‘The spatial dimension of knowledge transfer and its economic

implications’, Working paper, Max Planck Institute of Economics (Jena).Coleman, James S., Elihu Katz and Herbert Mendel (1957), ‘The diffusion of an

innovation among physicians’, Sociometry, 20: 253–70.Davis, Gerald F. and Henrich R. Greve (1997), ‘Corporate elite networks and gov-

ernance changes in the 1980s’, American Journal of Sociology, 103: 1–37.Festinger, Leon, Stanley Schacter and Kurt W. Back (1950), Social Pressure in

Informal Groups, New York: Harper.Fleming, Lee and Olav Sorenson (2001), ‘Technology as a complex adaptive system:

evidence from patent data’, Research Policy, 30: 1019–39.Fleming, Lee and Olav Sorenson (2004), ‘Science as a map in technological search’,

Strategic Management Journal, 25: 909–28.Frenken, Koen and Alessandro Nuvolari (2004), ‘The early development of the

steam engine: an evolutionary interpretation using complexity theory’, Industrialand Corporate Change, 13: 419–50.

Granovetter, Mark S. (1985), ‘Economic action and social structure: the problem ofembeddedness’, American Journal of Sociology, 91: 481–510.

Hägerstrand, Torsten ([1953] 1967), Innovation Diffusion as a Spatial Process,Chicago: University of Chicago.

Hall, Bronwyn H., Adam B. Jaffe and Manuel Trajtenberg (2001), ‘TheNBER patent citations data file: lessons, insights and methodological tools’,National Bureau of Economic Research Working Paper No. 8498, Cambridge,MA.

Kauffman, Stuart A. (1993), The Origins of Order, Oxford and New York: OxfordUniversity.

King, Gary and Langche Zeng (2001), ‘Logistic regression in rare events data’,Political Analysis, 9: 137–63.

Lippman, Steve and Richard Rumelt (1982), ‘Uncertain imitability: an analysis ofinterfirm differences in efficiency under competition’, Bell Journal of Economics,13: 418–38.

Marshall, Alfred (1890), Principles of Economics, London: Macmillan.Nelson, Richard R. and Sidney G. Winter (1982), An Evolutionary Theory of

Economic Change, Cambridge, MA: Belknap.Polanyi, Michael (1966), The Tacit Dimension, New York: Anchor Day.Reed, Richard and Robert J. DeFillippi (1990), ‘Causal ambiguity, barriers to imi-

tation, and sustainable competitive advantage’, Academy of Management Review,15: 88–102.

Rivkin, Jan W. (2001), ‘Reproducing knowledge: replication without imitation atmoderate complexity’, Organization Science, 12: 274–93.

Schumpeter, Joseph (1939), Business Cycles, New York: McGraw-Hill.Simon, Herbert A. (1962), ‘The architecture of complexity’, Proceedings of the

American Philosophical Association, 106: 467–82.Sorenson, Olav (2004), ‘Social networks, informational complexity and industrial

geography’, in D. Fornahl, C. Zellner and D. Audretsch (eds), The Role of LabourMobility and Informal Networks for Knowledge Transfer, Berlin: Springer-Verlag,pp. 79–96.

Informational complexity and the flow of knowledge across boundaries 159

Sorenson, Olav, Jan W. Rivkin and Lee Fleming (2004), ‘Complexity, networks andknowledge flow’, Harvard Business School Working Paper No. 04-027,Cambridge, MA.

Sorenson, Olav and Toby E. Stuart (2001), ‘Syndication networks and the spatialdiffusion of venture capital investments’, American Journal of Sociology, 106:1546–88.

Winter, Sidney G. (1995), ‘Four Rs of profitability: rents, resources, routines, andreplication’, in C. Montgomery (ed.), Resource-based and Evolutionary Theoriesof the Firm: Towards a Synthesis, Boston, MA: Kluwer, pp. 147–8.

Wolter, Kerstin (2006), ‘Divide and conquer? The role of governance for the adapt-ability of industrial districts’, Advances in Complex Systems, forthcoming.

Zander, Udo and Bruce Kogut (1995), ‘Knowledge and the speed of transfer andimitation of organizational capabilities: an empirical test’, Organization Science,6: 76–92.

Zipf, George K. (1949), Human Behavior and the Principle of Least Effort, Reading,MA: Addison-Wesley.

160 Network analysis

8. Networks and heterogeneousperformance of cluster firmsElisa Giuliani*

1. INTRODUCTION

This chapter explores the relationship existing among the heterogeneousnature of firms in industrial clusters, their structural position in local net-works and their performance. Following the rising interest for spatiallyagglomerated industrial firms (Piore and Sabel, 1984; Pyke et al., 1990;Porter, 1990; Krugman, 1991) and their learning and innovative potential(for example, Maskell, 2001a; Pinch et al., 2003), this chapter shows empiri-cally that the performance of firms in clusters is related to firm-level knowl-edge endowments and their position in the knowledge network. A startingargument of this chapter is that of questioning the widely accepted viewthat knowledge is diffused in clusters in a rather pervasive and unstructuredway, and that this is what affects the enhanced performance of cluster firmsas compared to isolated ones. Most economists and economic geographersshare this view. On the one hand, in fact, economists stress the public natureof knowledge (Arrow, 1962) and argue that geography facilitates inter-firmlearning and innovation because of localized knowledge spillovers (forexample, Jaffe et al., 1993); on the other, recent work done by economicgeographers argues that it is not geography per se that matters for innova-tion, but it is a common institutional endowment and firms’ relationalproximity (later defined), which facilitate the diffusion of knowledge andenhance collective learning in clusters (for example, Maskell andMalmberg, 1999; Capello and Faggian, 2005). A reason for this is the oftenpresumed co-occurrence of firms’ business interactions and knowledgeflows – a view consistent with the Marshallian ‘industrial atmosphere’metaphor (Marshall, 1920).

An increasing number of studies have, however, started to highlight that,in spite of a general homogeneity of conditions in the cluster, firms performdifferently (Lazerson and Lorenzoni, 1999; Rabellotti and Schmitz, 1999;Camison, 2004; Molina-Morales and Martinez-Fernandez, 2004; Zaheerand Bell, 2005). In line with this, some have expressed their conceptual

161

discontent about the pervasive and unstructured view of clusters’ inno-vation (see, for example, Breschi and Lissoni, 2001), and others have exam-ined how key notions of evolutionary economics may be incorporatedinto economic geography (Boschma and Lambooy, 1999; Boschma andFrenken, 2006). In this vein, some have pointed out the need to understandthe heterogeneity of cluster firms’ performance and the characteristics of acluster innovative process, by bringing firm-level learning into the analysis(Bell and Albu, 1999; Maskell, 2001b; Martin and Sunley, 2003).

Using a combination of network analysis (Wasserman and Faust, 1994)and econometrics, this chapter carries out an empirical study of three wineclusters – Colline Pisane and Bolgheri/Val di Cornia in Italy and ColchaguaValley in Chile. It shows that firms perform differently within clusters andthat such differences are due to both firm knowledge bases and to theirdegree of embeddedness in the local knowledge network. In contrast, inter-firm relational proximity is a less powerful factor affecting firm perfor-mance. The chapter concludes by drawing implications for the concept ofclusters in economic geography.

2. GEOGRAPHY, RELATIONAL PROXIMITY ANDTHE DIFFUSION OF KNOWLEDGE:IMPLICATIONS FOR FIRM PERFORMANCE

The process of knowledge diffusion and generation in clusters of firms hastraditionally been based on different reinterpretations of the Marshallian,externality-driven, world of industrial districts. Several empirical studieshave in fact elaborated on the Marshallian notion of knowledge spillovers,referring to Marshall’s description of industrial districts as a place where:

The mysteries of the trade become no mysteries; but are as it were in the air, andchildren learn many of them, unconsciously. . . . Good work is rightly appreci-ated, inventions and improvements in machinery, in processes and the generalorganisation of the business have their merits promptly discussed: if one manstarts a new idea, it is taken up by others and combined with suggestions of theirown; thus it becomes the source of further new ideas. (Marshall, 1920, p. 225;emphasis added)

He envisaged two mechanisms by which knowledge spillovers were gener-ated, the first one was through the embodied capabilities acquired by workersbeing part of the district and the second one was the sharing of ideas amongbusinessmen, a process which Allen (1983) has defined ‘collective invention’(see also Nuvolari, 2004). In Industry and Trade (1919), Marshall defined thisas ‘industrial atmosphere’, highlighting its ‘sticky’ nature:

162 Network analysis

But an industry which does not use massive material, and needs skill that cannotbe quickly acquired, remains as of yore loth to quit a good market for its labour.Sheffield and Solingen have acquired industrial ‘atmospheres’ of their own;which yield gratis to the manufacturers of cutlery great advantages, that are noteasily to be had elsewhere: and the atmosphere cannot be moved. (Marshall, 1919,p. 284; emphasis added)

Thus, the industrial atmosphere was conceived as a highly idiosyncraticmeso characteristic of districts. Several scholars have advanced in this fieldand have elaborated on the original Marshallian ideas. My argument hereis that most of the studies undertaken in this direction have taken a mesoperspective to analyse learning, innovation and performance in clusteredfirms. As such, they have given less emphasis to the micro, and to how themicro can affect the meso. Among these studies, I shall consider here onlythe most influential contributions of both economists and economicgeographers.

As anticipated in the introduction, the economists’ view is that knowl-edge spillovers, which are by definition a public good (Arrow, 1962), tendto be highly localized (Jaffe, 1989; Jaffe et al., 1993), a property that linksconceptually geography and innovation. Within this stream of studies,robust empirical evidence has shown that a relationship exists betweenspatial clustering, knowledge spillovers and firms’ innovative performance(for example, Audretsch and Feldman, 1996; Feldman, 1999; Baptista,2000). This empirical evidence has led scholars and policy makers to believethat geography matters for innovation and for competitiveness (forexample, OECD, 2001). As an example, in his work on industrial clustersand nations’ competitive advantage, Porter (1998) connects the processes oflearning and innovation in clusters to the ‘Marshallian atmosphere’concept, stating that ‘the information flow, visibility, and mutual reinforce-ment within such a locale give meaning to Alfred Marshall’s insightfulobservation that in some places an industry is in the air’ (p. 156). He notesmoreover that ‘more important, however, is the influence of geographicconcentration on improvement and innovation’ (p. 157), since ‘proximityincreases the speed of information flow within the national industry andthe rate at which innovations diffuse’ (p. 157). This view supports the ideathat geography matters for innovation and, implicitly, for economicperformance.

Some economic geographers seem to have moved beyond that. It has beenargued that geographic proximity per se is not sufficient to generate learn-ing, and that other forms of proximity are required for inter-firm learningand innovation to occur (Boschma, 2005). Among these, great emphasis isgiven to the role of social proximity also known as ‘relational proximity’(for example, Maskell and Malmberg, 1999; Amin and Cohendet, 2004).

Networks and heterogeneous performance of cluster firms 163

Industrial clusters, being a spatially localized set of economic activities, arein fact envisaged as ‘embedded’ economies (Granovetter, 1985) where socialrelationships, such as friendship and kinship, are entangled with productiveones. More specifically, relational proximity is believed to favour the form-ation of relational capital, defined as a sort of productive ‘thickening’ basedon market and cooperative inter-firm relationships (Scott, 1998). The re-lational capital, favouring the interaction of productive agents and thediffusion of tacit knowledge (Howells, 2002) is finally said to be the ‘sub-stratum’ of collective learning (Capello, 1999).

Relational proximity and embeddedness are thus considered to operateat the meso level and perform several functions in the context of innovation(Oerlemans and Meeus, 2005), favouring firm performance accordingly. Ina recent study on several district areas in Italy, Capello and Faggian (2005)find support for the importance of relational capital in fostering the inno-vative performance of firms. They therefore argue that

[R]egional economists are . . . correct in underlining that not only are intra-firmcharacteristics crucial for innovation, but also (and maybe most of all) that thelocation of firms in an area where the local labour market and the tight linkswith suppliers foster the exchange of local knowledge are vital for innovation.(p. 82; emphasis added)

In sum, current approaches to the analysis of clusters mainly focus onexplaining why firms that are part of an industrial cluster tend to performbetter than isolated ones. An underlying, implicit assumption, is that firmsthat are part of the same industrial cluster will benefit more or less equallyfrom the presence of external economies at the local level (Marshall, 1920)and, more specifically, from a common geographical, sectoral and re-lational proximity. However, an increasing number of studies have recentlystarted to highlight that, in spite of a general homogeneity of meso con-ditions in clusters, firms perform differently (Lazerson and Lorenzoni,1999; Rabellotti and Schmitz, 1999; Camison, 2004; Molina-Morales andMartinez-Fernandez, 2004; Zaheer and Bell, 2005).

3. WHAT AFFECTS HETEROGENEOUSPERFORMANCE IN CLUSTER FIRMS?

Understanding the factors that lead to heterogeneous firm performance inindustrial clusters requires that the focus of analysis shifts from the mesoto the micro. As suggested by Lazerson and Lorenzoni (1999) and morerecently by Maskell (2001b), individual firms are the key actors in the devel-opment of territorial clusters. In line with this, my argument here is that it

164 Network analysis

is the characteristics of firms, and their inherent heterogeneity, that gener-ate (or inhibit) the conditions at the meso level that ultimately enhancecluster firm performance. Thus, it is a micro to meso perspective that thischapter takes, rather than the meso to micro one commonly found in thecluster literature. Accordingly, the performance of firms should be exploredby considering the interplay between their internal resources and the exter-nal, meso conditions present in the cluster.

The relationship between firms’ internal resources and performance hasalready been investigated by many scholars (Barney, 1991; Grant, 1996).Starting from the evolutionary theory of the firm (Nelson and Winter,1982), I consider firms in the cluster as being characterized by hetero-geneous knowledge bases. By knowledge base I mean here the ‘set of infor-mation inputs, knowledge and capabilities that inventors draw on whenlooking for innovative solutions’ (Dosi, 1988, p. 1126). Knowledge is seenas residing in firms’ skilled knowledge workers, who embody tacit capabil-ities. At the same time, knowledge is not merely the sum of each individ-ual’s knowledge, since it resides in the organizational memory of the firm.As Nelson and Winter (1982, p. 63) put it:

The possession of technical ‘knowledge’ is an attribute of the firm as a whole, asan organized entity, and it is not reducible to what any single individual knows,or even to any simple aggregation of the various competences and capabilitiesof all the various individuals, equipments, and installations of the firm.

The knowledge base is, moreover, considered here as the result of a processof cumulative learning, which is inherently imperfect, complex and pathdependent (Dosi, 1997) and which delivers persistent heterogeneitybetween the firms in the economic system and, understandably, in a cluster.Such heterogeneity, in turn, deepens the uniqueness of resources deployedby firms and explains different growth rates and performance (Penrose,1959). It is reasonable, moreover, that firms that have stronger knowledgebases will perform better than others in the cluster, as they will have easieraccess to external knowledge and rejuvenate their internal capabilitiesaccordingly (Cohen and Levinthal, 1990). This chapter will explore the fol-lowing research question: how does the heterogeneity in firm knowledgebases relate to their performance?

The important issue here is, however, not simply whether micro-level con-ditions affect performance, but what is their interaction with the externalenvironment in the cluster. Innovation rarely occurs in isolation, and, asemphasized by most of the cluster literature, the degree to which firms areembedded in local networks influences their performance (Molina-Moralesand Martinez-Fernandez, 2004; Capello and Faggian, 2005; Zaheer and Bell,2005). Being relationally embedded in a cluster means that firms interact

Networks and heterogeneous performance of cluster firms 165

frequently on business-related matters. For example, if entrepreneurs aremembers of the same local consortium they will meet at local events anddiscuss their productive activities. Similarly, if two firms share machinery,their technical employees will meet and discuss their appropriate use. Allthese interactions generate a trustworthy environment in the cluster, whichmay facilitate the sharing of information and knowledge, thus enhancing theoverall firm capabilities to innovate. In a previous study on wine clusters(Giuliani, 2006), I have shown that business interactions of this type occurin a pervasive way, resembling what Marshall called ‘industrial atmosphere’.This means that firms show a rather homogeneous behaviour in interactingwith the rest of the firms in the cluster. This is a relevant property because,if it is true that firms benefit from being embedded in the local network ofbusiness interactions, their performance will be homogeneously distributed.

The interaction for business-related matters is only one of the severalinformal networks formed by firms in clusters (Boschma, 2005). In fact,different types of networks are likely to carry different informationalcontent and they may affect firm performance differently (Gulati, 1998;Rodan and Galunic, 2004). In Giuliani (2006), I disentangle the knowledgenetwork, based on the transfer of knowledge for the solution of technicalproblems, from the overall network of business interactions (Figure 8.1).The structural properties of the knowledge network suggest that it is builton a more selective basis, if compared to the network of business inter-actions. This means that knowledge diffuses in clusters in a less pervasiveand serendipitous way than is commonly envisaged by the economists andeconomic geographers. This property may have important implications onthe distribution of firm performance in clusters. Given its selectivity, if firmperformance is affected more by the degree of embeddedness in the knowl-edge network than in the network of business interactions, it is reasonableto expect an uneven distribution of firm performances.

4. DATA AND METHOD OF ANALYSIS

Collection of Data

This study is based on micro-level data, collected at the firm level in threewine clusters in Italy and Chile, namely: Colline Pisane (CP), Bolgheri/Valdi Cornia (BVC) and Colchagua Valley (CV). The analysis has requiredcareful data collection through interviews. Interviews were carried out withthe skilled workers (that is, oenologists or agronomists) and the survey wasdirected to producers of fine wines. This analysis includes only horizontalrelationships among firms that operate as wine producers, whereas vertical

166 Network analysis

Networks and heterogeneous performance of cluster firms 167

Note: BI stands for ‘business interaction’; KN for ‘knowledge’; CP for the CollinePisane cluster; BVC for the Bolgheri/Val di Cornia cluster; CV for the ColchaguaValley cluster.

Source: Giuliani (2006) (based on UCINET-Netdraw; Borgatti et al., 2002).

Figure 8.1 Types of networks

(a) BI network in CP (b) KN network in CP

(c) BI network in BVC (d) KN network in BVC

(e) BI network in CV (f) KN network in CV

linkages are not explored here. Data were gathered using the universe offine wine producers populating the three clusters,1 32 in CP, 41 in BVC and32 in CV, making a total of 105 firms. Further information about the popu-lation of firms is reported in Table 8.1.

Apart from general background and contextual information, the inter-views were designed to obtain information that would permit the develop-ment of quantitative indicators in three key areas: (i) the knowledge baseof firms; (ii) the degree of embeddedness of firms in the network of busi-ness interactions; and (iii) the degree of embeddedness of firms in thenetwork of knowledge. These are summarized in Table 8.2.

Econometric Estimation

The analysis is based on an econometric estimation, using a Probit modelwith marginal effects. The estimations are carried out on the aggregate data

168 Network analysis

Table 8.1 Firm characteristics by cluster

Characteristics Clusterof firms by:

CP CV BVC(N�32) (N�32) (N�41)

(a) Size (employees)Small (1–19) 91 28 90Medium (20–99) 9 66 4Large (�100) 0 6 6

(b) Ownership Domestic 100 81 95Foreign 0 19 5

(c) Organization structure1. Part of a group, 3 22 7

vertically integrated firms2. Part of a group, – 13 –

vertically disintegrated firms3. Independent, 88 66 93

vertically integrated4. Other (e.g., cooperatives) 9 – –

(d) Year of localizationUp to 1970s 53 24 241980s 9 16 221990s 31 38 232000s 6 19 15

Note: The numbers refer to percentages within the respective cluster.

Networks and heterogeneous performance of cluster firms 169

Table 8.2 Collection of key variables

1. Knowledge base In the literature, this concept, a key element in the analysis here,is described in terms of the knowledge base of the firm, often associated withtraining, human resources and R&D. Correspondingly, the structured interviewssought detailed information about: (i) the number of technically qualifiedpersonnel in the firm and their level of education and training (human resources),(ii) the experience of professional staff – in terms of months in the industry(months of experience); (iii) the number of other firms in which they had beenemployed (number of firms), and (iv) the intensity and nature of the firms’experimentation activities (experimentation intensity) – an appropriate proxy forknowledge creation efforts, since information about expenditure on formal R&Dwould have been both too narrowly defined and too difficult to obtainsystematically.

2. Network of business interactions In the questionnaire-based interview, relationaldata were collected through a ‘roster recall’ method: each firm was presented witha complete list (roster) of the other firms in the cluster, and was asked the questionreported below:

With which of the cluster firms mentioned in the roster do you interact forbusiness matters?[Please indicate the frequency of interaction according to the following scale:0� none; 1� low; 2� medium; 3� high]

3. Network of knowledge interactions* In the questionnaire-based interview,relational data were collected through a ‘roster recall’ method: each firm waspresented a complete list (roster) of the other firms in the cluster, and was askedthe questions reported below:

i. If you are in a critical situation and need technical advice, to which of the localfirms mentioned in the roster do you turn?[Please rate the importance you attach to the knowledge linkage establishedwith each of the firms according to its persistence and quality, on the basis ofthe following scale: 0�none; 1� low; 2�medium; 3�high].

ii. Which of the following firms do you think have benefited from technicalsupport from this firm?[Please indicate the importance you attach to the knowledge linkageestablished with each of the firms according to its persistence and quality, onthe basis of the following scale: 0�none; 1�low; 2�medium; 3�high].

Note: * Respondents were asked to rate each of the mentioned relationships on a scale of0 to 3. A value of 0 is given when no linkage is formed. A value of 1 corresponds to anoccasional knowledge linkage with limited content in terms of quality of the knowledgeflow, whereas a value of 3 corresponds to a persistent knowledge linkage that carries fine-grained knowledge.

It is worth remembering here that this study was designed to collect cross-sectional dataonly. Therefore, the relational data gathered through the above-mentioned roster studiesrefer to the very recent past in which the interviews have been carried out (maximum of twoyears).

obtained by pooling together the three clusters’ variables. Since data arecoming from three different geographical clusters of firms, I applied theMoulton method (Moulton, 1990) to control for the possibility that therandom disturbances in the regression are correlated within each cluster.

Dependent Variable

Performance at the firm level is measured here by an indicator of the qualityachieved by each firm’s wines. The quality of worldwide wines is annuallyassessed and rated by international panels of experts and published inseveral specialized wine journals (for example, Wine Spectator, Decanter,Wine Enthusiast and Robert Parker’s Guide). Having a wine rated by any ofthese international specialized journals is, first, an acknowledgement of thequalitative properties of the wines, and second, a very powerful marketingdevice for a firm, since experts’ ratings strongly influence market prices(Nerlove, 1995; Combris et al., 1997, 2000; Landon and Smith, 1997). Theperformance of a firm is therefore seen here as its capacity to develop newwines, which are valued as ‘quality wines’ by international experts.

The indicator adopted in this chapter is drawn from one of the aboveinternational wine journals: Wine Spectator.2 This journal wine rating isbased on the quality assessment of an international panel of expert oenol-ogists, who review more than 12 000 wines each year in blind tastings. Aftertasting, oenologists assign a score to each wine brand according to a 100-point scale, ranging from 100, when the wine is of outstanding quality, to50 when it is of poor quality.3 Certain information is listed with each ratedwine: the wine vintage, the wine area and the market price. The indicatorused here (RATING) is valued 1 when any of the firm’s wines has beenassigned at least 70 points in years 2002–04, the minimum threshold for awine to be considered of drinkable quality and to be recommended by thejournal. It is valued 0 otherwise. A lag of two years is allowed between theyear in which the interviews were carried out and the vintage of the mostrecent rated wines.

It is worth noting that the majority of the wines tasted are submitted toWine Spectator by the wineries or their US importers. Additionally, thejournal spends substantial effort in buying and reviewing wines that are notsubmitted, at all price levels. Accordingly, a firm’s wines may not be ratedfor three main reasons. First, due to a selection bias, firms with poor-quality achievements will have little incentive to send their wines to thejournal for assessment. Second, US importers will not recommend andsignal wines to Wine Spectator when they consider them of poor quality.Third, Wine Spectator itself selects out all the wineries producing verypoor-quality wines. These considerations suggest therefore that firms

170 Network analysis

whose wines are not rated tend to be poor performers. The same applies forfirms whose wines are rated but are assigned less than 70 points.

Using Wine Spectator as a unique source of information, however, maypose some robustness problems. Even if it is highly unlikely that a success-ful winery will not be spotted by the journal, it is still possible that somewineries are overlooked. In order to control for that, I correlated RATINGand two other indicators drawn from Wine Spectator’s ratings – the sum ofscores per planted vineyard hectares (SSH) and the average price of ratedwines normalized by the average price of rated wines in the cluster (PRICE)– with an indicator of the relative performance of firms in the cluster as per-ceived by its members.4 I find strong correlations between the perceived per-formance and the three Wine Spectator indicators: RATING (0.65**), SSH(0.65**) and PRICE (0.62**). These results suggest that the performanceindicator RATING is robust enough to measure the quality of winesachieved by cluster firms.

Independent Variables

Firm knowledge base (KB)The knowledge base of the firm is measured by extracting a factor from theprincipal component analysis of the four variables listed in Table 8.2 (Point1). The factor explains more than 75 per cent of variance and it has beencalculated considering the pooled sample of firms.

Embeddedness in the network of business interactions (BI_DC)This variable measures the extent to which a firm has established linkagesfor business matters with other firms in the cluster. The existence of a busi-ness interaction is mapped by the question in Table 8.2 (Point 2). The degreecentrality of the network of business interactions (BI_DC(j)) is consideredhere a proxy of firms’ the relational embeddedness in the cluster. It is meas-ured by the extent to which an actor j is central in a network on the basisof the ties that it has directly established with other i actors of the network(�(x(ji))). This measure uses undirected dichotomous data. The value hasbeen normalized by its theoretical maximum (g – 1), where g is the numberof firms in each cluster.

Embeddedness in the knowledge network (KN_DC)This variable measures the degree to which a firm is central in the knowl-edge network, mapped using questions (i) and (ii) in Table 8.2 (Point 3).Also in this case the embeddedness of firms is measured by actor-level

BI_DC( j) � �(x(ji)) �g � 1.

Networks and heterogeneous performance of cluster firms 171

degree centrality (KN_DC). This measure uses undirected dichotomousdata. The value has been normalized by its theoretical maximum (g – 1),where g is the number of firms in each cluster:

Control Variables

I control here for the following firm-level variables that are commonly as-sociated with performance: the size of firms, measured by the log ofemployees (LEMP), the age of the firm (AGE), measured as the number ofyears since the start of operations until 2002. The ownership (OWN) whichis a dummy variable indicating whether the firm is foreign (1) or domestic(0) owned. I also control for the type of organization. As shown by Table8.1, firms in the clusters have four different types of organizational struc-tures: ORG1 corresponds to firms that are part of a national group andperform all phases of the production process within the cluster; ORG2refers to firms that are also part of a national group but perform only partof the production process, usually grape-growing, within the cluster; ORG3refers to firms that are independently owned and that perform all produc-tion phases in the cluster, where the headquarter is also located; finally,ORG4 represents a residual category including cooperatives.

5. EMPIRICAL RESULTS

Table 8.3 reports the descriptive statistics and the correlation matrix andTable 8.4 the results of the Probit estimation. Model 1 in Table 8.4 includesonly the control variables, showing that only size is positively related to thelikelihood of a firm being rated by Wine Spectator and therefore with itsperformance.

Model 2 shows that the value of firm knowledge base is strongly and pos-itively related to performance, a result that is in line with several otherrecent contributions (for example, Camison, 2004; Zaheer and Bell, 2005).This is explained by the fact that firms that have better-educated or more-experienced knowledge workers (Drucker, 1993), and that carry higherinternal experimentation intensity, are more likely to exploit knowledge forthe generation of successful innovations (March, 1991). This in turn drivesa firm to achieve higher performances (Wernerfelt, 1984).

Model 3 provides support for the view that firms that interact for busi-ness matters are more likely to be good performers. This result is consistentwith the fact that being co-located in the same industrial cluster, and being

KN_DC(j) � �(x( ji)) �g � 1.

172 Network analysis

173

Tab

le 8

.3D

escr

ipti

ve s

tati

stic

s an

d co

rrel

atio

n m

atri

x

Mea

nSt

d de

v.R

AT

ING

LE

MP

AG

EO

WN

ER

OR

G1

OR

G2

OR

G3

OR

G4

KB

BI_

DC

KN

_DC

RA

TIN

G0.

300.

461.

00L

EM

P1.

991.

390.

47**

*1.

00A

GE

32.6

166

.95

0.14

0.01

1.00

OW

NE

R0.

070.

250.

080.

18*

�0.

091.

00O

RG

10.

100.

310.

19*

0.34

***

�0.

050.

41**

*1.

00O

RG

20.

040.

190.

31**

*0.

23**

*�

0.04

�0.

05�

0.07

1.00

OR

G3

0.83

0.38

�0.

26**

*�

0.42

***

0.06

�0.

28**

*�

0.75

***

�0.

44**

*1.

00O

RG

40.

030.

17�

0.11

0.07

�0.

01�

0.05

�0.

06�

0.03

�0.

38**

*1.

00K

B0.

001.

000.

50**

*0.

66**

*�

0.09

0.21

***

0.51

***

0.09

�0.

43**

*�

0.08

1.00

BI_

DC

26.6

917

.05

0.37

***

0.22

***

0.14

0.04

0.03

0.12

�0.

090.

000.

22**

*1.

00K

N_D

C8.

118.

300.

64**

*0.

43**

*0.

050.

22**

*0.

17*

0.35

***

�0.

26�

0.12

0.52

***

0.55

***

1.00

Not

e:*

Sign

ifica

nt a

t 10

%;*

* Si

gnifi

cant

at

5%;*

** S

igni

fica

nt a

t 1%

.Bas

ed o

n P

ears

on c

oeffi

cien

ts.

embedded in the local network, may facilitate access to relevant infor-mation or may ease inter-firm transactions. In this case, firms may benefitfrom several types of externalities and enhance their performance accord-ingly. This evidence seems to support the view that both intra-firmresources and relational proximity matter for firm performance (Capelloand Faggian, 2005). However, no evidence is found that the latter mattersmore than the former.

Model 4 finds a strong and positive relationship between the degree offirms’ embeddedness in the knowledge network and their performance.This is in line with most of organizational sociology’s literature (forexample, Powell et al., 1996; Smith-Doerr and Powell, 2003), since theaccess to external sources of knowledge for the solution of internal prob-lems favours innovation and enhances firm performance. The interestingresult here is that the coefficient of KN_DC is higher in Model 4 than the

174 Network analysis

Table 8.4 Probit estimations with marginal effects

Dependent Model 1 Model 2 Model 3 Model 4 Model 5variables Control Knowledge Knowledge Knowledge Knowledge

variables base only base with base with base withonly BI_DC KN_DC BI_DC and

KN_DCdF/dx (s.e.) dF/dx (s.e.) dF/dx (s.e.) dF/dx (s.e.) dF/dx (s.e.)

LEMP 0.140 0.063 0.058 0.064 0.063 (0.052)** (0.070) (0.052) (0.071) (0.770)

AGE 0.000 0.000 0.001 0.000 0.000 (0.000)* (0.000)*** (0.000)** (0.000)*** (0.000)***

OWNER 0.019 �0.146 �0.015 �0.146 �0.146(0.099) (0.082) (0.138) (0.084) (0.082)

ORG3a �0.067 �0.024 0.067 �0.022 �0.024(0.225) (0.335) (0.231) (0.334) (0.335)

KB 0.085 0.181 0.083 0.085 (0.018)*** (9.296)*** (0.017)*** (0.018)***

BI_DC 0.006 0.000 (0.003)** (0.001)

KN_DC 0.027 0.026 (0.005)*** (0.006)***

N 96 96 96 96 96Log pseudo- �46.003222 �41.330488 �38.942215 �33.722975 �33.698371

likelihoodPseudo R2 0.1934 0.2745 0.3170 0.4087 0.4092

Note: * Significant at 10%; ** Significant at 5%; *** Significant at 1%. aORG1, ORG2 andORG4 have been dropped due to collinearity.

coefficient of BI_DC in Model 3. Furthermore, when both BI_DC andKN_DC are considered in the estimation (Model 5), BI_DC ceases to besignificant, while both KB and KN_DC persist in being positive andstrongly significant.5

The results of this study indicate that both firm internal capabilities, andthe degree to which firms are embedded in local networks, matter for theirperformance. However, it is relevant to note that it is a specific type ofnetwork that affects performance most: the knowledge network. It is reason-able to argue that this is connected to the structural properties of thisnetwork. As illustrated by Giuliani (2006), the knowledge network is formedon a selective rather than pervasive, collective basis. This is due to twofactors: first, firms with stronger knowledge bases have more to transfer andare understandably more likely to be targeted by other firms for technicaladvice. Second, firms with stronger knowledge bases will seek technicaladvice from equally advanced firms, thus targeting firms with strong knowl-edge bases in their search for external knowledge (Giuliani and Bell, 2005).On the basis of this, communities of knowledge are formed in clusters by arestricted group of equally advanced firms. An implication of selectivity isthat the knowledge that is circulated within the community is likely to be ofvaluable content, which in turn enriches the knowledge base of the memberfirms and, consequently, their performance. It is therefore reasonable toargue that this evidence casts doubts on the importance of meso-level con-ditions, such as geographical, sectoral and relational proximity, for perfor-mance. More convincingly, it seems that knowledge endowments affectperformance both directly and indirectly through the generation of a localknowledge network, which serves to enhance individual firms’ capabilities.

6. CONCLUSION

This chapter has attempted to understand the factors that influence firmperformance in industrial clusters. Taking an evolutionary approach to thisdomain of studies, it has shown that the heterogeneous distribution of firmknowledge bases is related to their performance, both directly and indi-rectly through the participation in the local knowledge community.Following up on a previous study (Giuliani, 2006), this novel empirical evi-dence shows that, in spite of pervasive business interactions, the perfor-mance of firms is unevenly distributed in the clusters. The econometricestimations provide support for this and show that firm performance inclusters depends on their internal capabilities (that is, knowledge bases) andtheir capability of being connected to the local knowledge network. Moreimportantly, this empirical evidence seems consistent with the fact that

Networks and heterogeneous performance of cluster firms 175

similar meso characteristics – that is, the geographic and relational prox-imity of firms – constitute the substratum neither for collectively sharedknowledge flows (Giuliani, 2006).

On this basis, two considerations can be raised. First, one should beextremely careful in associating the concept of industrial clusters withenhanced performance and competitiveness, even when firms are geo-graphically and relationally proximate. Instead, more rigorous studiesshould be carried out in the future that analyse the interplay between firms,the cluster knowledge network, and performance. Second, as recently sug-gested by Markusen (2003), more rigorous analysis in regional studies willprovide better indications for policy makers. Indeed, this study supportsthe view that cluster performance is more likely to be enhanced bystrengthening firms’ knowledge bases rather than by pooling firmstogether in the same geographical area (as is the case of ‘technopoles’(OECD, 2000)) or by promoting inter-firm networking per se (UNCTAD,2001; UNIDO, 2001).

NOTES

* The author would like to thank Gustavo Crespi and Koen Frenken for comments on aprevious version of this chapter. Thanks go also to Marcelo Lorca Navarro, Cristian DiazBravo, Erica Nardi and Elena Bartoli for their support during fieldwork. Financialsupport by the UK Economic and Social Research Council (PTA-026-27-0644) is alsogratefully acknowledged.

1. The lists of firms are drawn from official sources: the S.A.G. (Servicio Agricola yGanadero) for Chile and the provinces of Pisa and Livorno for Italy.

2. The choice of the journal was done on the basis of two criteria: free availability on theweb and coverage (countries, vintages, wine areas).

3. The rating is based on the following criteria: 95–100 Classic: a great wine; 90–94Outstanding: superior character and style; 85–89 Very good: wine with special qualities;80–84 Good: a solid, well-made wine; 70–79 Average; drinkable wine that may have minorflaws; 60–69 Below average; drinkable but not recommended; 50–59 Poor; undrinkable,not recommended.

4. The questionnaire included a question asking the respondents to name the firms in theircluster that they perceived as having achieved high performance in terms of quality ofwines and commercial success.

5. Given the existence of a positive relationship between two of the independent variables,KB and KN_nDC (Giuliani, 2005), simultaneous equation modelling would have givenmore robust or insightful econometric estimations. This model is applied in other forth-coming works by the author.

REFERENCES

Allen, R.C. (1983), ‘Collective invention’, Journal of Economic Behaviour andOrganization, 4: 1–24.

176 Network analysis

Amin, A. and Cohendet, P. (2004), Architectures of Knowledge: Firms, Capabilitiesand Communities, Oxford: Oxford University Press.

Arrow, K.J. (1962), ‘Economic welfare and the allocation of resources for inven-tion’, in R.R. Nelson (ed.), The Rate and Direction of Inventive Activity,Princeton, NJ: Princeton University, pp. 609–26.

Audretsch, D. and Feldman, M.P. (1996), ‘R&D spillovers and the geography ofinnovation and production’, American Economic Review, 86: 630–40.

Baptista, R. (2000), ‘Do innovations diffuse faster within geographical clusters?’,International Journal of Industrial Organization, 18: 515–35.

Barney, J.B. (1991), ‘Firm resources and sustained competitive advantage’, Journalof Management, 17: 99–120.

Bell, M. and Albu, M. (1999), ‘Knowledge systems and technological dynamism inindustrial clusters in developing countries’, World Development, 27: 1715–34.

Borgatti, S.P., Everett, M.G. and Freeman, L.C. (2002), Ucinet 6 for Windows,Harvard, MA: Analytic Technologies.

Boschma, R.A. (2005), ‘Proximity and innovation: a critical assessment’, RegionalStudies, 39(1): 61–74.

Boschma, R.A. and Frenken, K. (2006), ‘Why is economic geography not an evo-lutionary science? Towards an evolutionary economic geography’, Journal ofEconomic Geography, 6(3): 273–302.

Boschma, R.A. and Lambooy, J.G. (1999), ‘Evolutionary economics and economicgeography’, Journal of Evolutionary Economics, 9: 411–29.

Breschi, S. and Lissoni, F. (2001), ‘Knowledge spillovers and Local InnovationSystems: a critical survey’, Industrial and Corporate Change, 10: 975–1005.

Camison, C. (2004), ‘Shared, competitive and comparative advantages: a competence-based view of industrial-district competitiveness’, Environment and Planning, A36:2227–56.

Capello, R. (1999), ‘Spatial transfer of knowledge in high technology milieux:learning versus collective learning processes’, Regional Studies, 33: 353–65.

Capello, R. and Faggian, A. (2005), ‘Collective learning and relational capital inlocal innovation processes’, Regional Studies, 39(1): 75–87.

Cohen, W.M. and Levinthal, D.A. (1990), ‘Absorptive capacity – a new perspectiveon learning and innovation’, Administrative Science Quarterly, 35(1): 128–52.

Combris, P., Lecocq, S. and Visser, M. (1997), ‘Estimation of hedonic price equationfor Bordeaux wine: does quality matter?’, Economic Journal, 107: 390–402.

Combris, P., Lecocq, S. and Visser, M. (2000), ‘Estimation of hedonic priceequation for Burgundy wine’, Applied Economics, 32: 961–67.

Dosi, G. (1988), ‘Sources, procedures and microeconomic effects of innovation’,Journal of Economic Literature, 26: 1120–71.

Dosi, G. (1997), ‘Opportunities, incentives and the collective patterns of techno-logical change’, Economic Journal, 107: 1530–47.

Drucker, P. (1993), Post Capitalist Society, New York: Harper Business.Feldman, M.P. (1999), ‘The new economics of innovation, spillovers and agglom-

eration: a review of empirical studies’, The Economics of Innovation and NewTechnology, 8: 5–25.

Giuliani, E. (2006), ‘The selective nature of knowledge networks in clusters: evi-dence from the wine industry, Journal of Economic Geography, in press.

Giuliani, E. and Bell, M. (2005), ‘The micro-determinants of meso-level learningand innovation: evidence from a Chilean wine cluster’, Research Policy, 34(1):47–68.

Networks and heterogeneous performance of cluster firms 177

Granovetter, M. (1985), ‘Economic action and social structure: the problem ofembeddedness’, American Journal of Sociology, 91: 481–510.

Grant, R.M. (1996), ‘Toward a knowledge-based theory of the firm’, StrategicManagement Journal, 17: 109–22.

Gulati, R. (1998), ‘Alliances and networks’, Strategic Management Journal, 19:293–317.

Howells, J.R.L. (2002), ‘Tacit knowledge, innovation and economic geography’,Urban Studies, 39(5–6): 871–84.

Jaffe, A.B. (1989), ‘Real effects of academic research’, American Economic Review,79: 984–99.

Jaffe, A.B., Trajtenberg, M. and Henderson, R. (1993), ‘Geographic localization ofknowledge spillovers as evidence from patent citations’, Quarterly Journal ofEconomics, 188: 577–98.

Krugman, P. (1991), Geography and Trade, Cambridge, MA: MIT Press.Landon, S. and Smith, C.E. (1997), ‘The use of quality and reputation indicators by

consumers: the case of Bordeaux wine’, Journal of Consumer Policy, 20: 289–323.Lazerson, M.H. and Lorenzoni, G. (1999), ‘The firms that feed industrial districts:

a return to the Italian source’, Industrial and Corporate Change, 8: 235–66.March, J.G. (1991), ‘Exploration and exploitation in organizational learning’,

Organization Science, 2(1): 71–87.Markusen, A. (2003), ‘Fuzzy concepts, scanty evidence, policy distance: the case for

rigour and policy relevance in critical regional studies’, Regional Studies, 37:701–17.

Marshall, A. (1919), Industry and Trade, London: Macmillan.Marshall, A. (1920), Principles of Economics, London: Macmillan.Martin, R. and Sunley, P. (2003), ‘Deconstructing clusters: chaotic concept or

policy panacea?’, Journal of Economic Geography, 3(1): 5–35.Maskell, P. (2001a), ‘Towards a knowledge-based theory of the geographical

cluster’, Industrial and Corporate Change, 10: 921–43.Maskell, P. (2001b), ‘The firm in economic geography’, Economic Geography, 77(4):

329–44.Maskell, P. and Malmberg, A. (1999), ‘Localised learning and industrial competi-

tiveness’, Cambridge Journal of Economics, 23(2), 167–86.Molina-Morales, F.X. and Martinez-Fernandez, M.T. (2004), ‘How much differ-

ence is there between industrial district firms? A net value creation approach’,Research Policy, 33: 473–86.

Moulton, B.R. (1990), ‘An illustration of a pitfall in estimating the effects of aggre-gate variables on micro units’, Review of Economics and Statistics, 72(2): 334–8.

Nelson, R.R. and Winter, S.G. (1982), An Evolutionary Theory of Economic Change,Cambridge, MA: Belknap Press of Harvard University Press.

Nerlove, M. (1995), ‘Hedonic price function and the measurement of preferences:the case of Swedish wine consumers’, European Economic Review, 39: 1697–716.

Nuvolari, A. (2004), ‘Collective invention during the British Industrial Revolution:the case of the Cornish pumping engine’, Cambridge Journal of Economics, 28(2):347–63.

Oerlemans, L.G. and Meeus, M.T.H. (2005), ‘Do organizational and spatial prox-imity impact on firm performance?’, Regional Studies, 39(1): 89–104.

Organization for Economic Cooperation and Development (OECD) (2000), ‘Localpartnership, clusters and SME globalisation’, Conference for Ministers respon-sible for SMEs and Industry Ministers, Bologna, Italy, 14–15 June.

178 Network analysis

Organization for Economic Cooperation and Development (OECD) (2001),Innovative Clusters: Drivers of National Innovation Systems, Paris: OECD.

Penrose, E. (1959), The Theory of the Growth of the Firm, Oxford: Blackwell.Pinch, S., Henry, N., Jenkins, M. and Tallman, S. (2003), ‘From “industrial dis-

tricts” to “knowledge clusters”: a model of knowledge dissemination and com-petitive advantage in industrial agglomerations’, Journal of Economic Geography,3: 373–88.

Piore, M.J. and Sabel, C.F. (1984), The Second Industrial Divide: Possibility forProsperity, New York: Basic Books.

Porter, M. (1998), On Competition, Cambridge, MA: Harvard Business ReviewSeries.

Porter, M. (1990), The Competitive Advantage of Nations, London: Macmillan.Powell, W.W., Koput, K.W. and Smith-Doerr, L. (1996), ‘Interorganizational col-

laboration and the locus of innovation: networks of learning in biotechnology’,Administrative Science Quarterly, 41: 106–45.

Pyke, F., Becattini, G. and Sengenberger, W. (1990), Industrial Districts and Inter-firm Co-operation in Italy, Geneva: International Institute for Labour Studies.

Rabellotti, R. and Schmitz, H. (1999), ‘The internal heterogeneity of industrial dis-tricts in Italy, Brazil and Mexico’, Regional Studies, 33(2): 97–108.

Rodan, S. and Galunic, C. (2004), ‘More than network structure: how knowledgeheterogeneity influences managerial performance and innovativeness’, StrategicManagement Journal, 25: 541–62.

Scott, A.J. (1998), Regions and the World Economy: The Coming Shape of GlobalProduction, Competition and Political Order, Oxford and New York: OxfordUniversity Press.

Smith-Doerr, L. and Powell, W.W. (2003), ‘Networks and economic life’, inN. Smelser and Swedberg, R. (eds), The Handbook of Economic Sociology,London and Princeton, NJ: Russell Sage Foundation and Princeton UniversityPress.

UNCTAD (2001), World Development Report: Promoting Linkages, Geneva, NewYork: United Nations Conference on Trade and Development.

UNIDO (2001), Development of Clusters and Networks of SMEs, Vienna: UnitedNations Industrial Development Organization.

Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods andApplications, Cambridge: Cambridge University Press.

Wernerfelt, B. (1984), ‘A resource-based view of the firm’, Strategic ManagementJournal, 5: 171–80.

Zaheer, A. and Bell, G.G. (2005), ‘Benefiting from network position: firm capabili-ties, structural holes and performance’, Strategic Management Journal, 26:809–25.

Networks and heterogeneous performance of cluster firms 179

9. Social networks and the economicsof networksDaniel Birke*

1. INTRODUCTION

The work by Arthur (1989) and the other literature on path dependence areprominent in many evolutionary arguments. Path dependence arises inincreasing return technologies and describes the property that for manytechnologies, historically small events can have a profound and lastingimpact on the development path of technologies. It is well understood thattechnologies with increasing returns can exhibit multiple equilibria andthat these equilibria can often be local rather than global optima. Causesfor increasing returns can be learning and network effects.

The models of path dependence, however, have been based on rathersimple assumptions about the choice behaviour of consumers in markets withnetwork effects. The empirical question remains: how do consumers choosebetween rival products in a market with network effects? Consumers interactwith other consumers in a variety of ways. Information about products andservices is often spread by word of mouth and individuals are more likely tochoose a product about which they have heard from a friend or which theyhave tried out with the help of a friend. In many cases, consumers also try toassociate themselves with their peer group by consuming similar goods, tryto imitate consumption behaviour of groups which they regard as havinghigher ‘social status’and try to distinguish themselves from groups with lower‘social status’ (see Cowan et al., 1997). Network effects have been treated inthe literature mainly as an only positive peer effect, where every individual isa peer of everyone else, that is, where this peer effect is anonymous. Here, weanalyse the impact of peer groups on choice behaviour applied to the case ofmobile telephony.

After the seminal article of Rohlfs (1974), and the influential papers ofFarrell and Saloner (1985) and Katz and Shapiro (1985), there has been aplethora of theoretical studies into the nature of network effects and bynow network effects theory has reached a rather mature state. The litera-ture on network effects typically distinguishes between two types of such

180

effects: direct and indirect. Direct network effects refer to the case whereusers directly benefit from other users of the same network. In mobiletelecommunications, a direct network effect arises when a user can call alarger set of other users. Indirect network effects, on the other hand, arisebecause bigger networks support a larger range of complementary prod-ucts and services. In second generation mobile networks, indirect networkeffects are only of second-order significance, but they will play an increas-ing role after the introduction of third generation networks, where usagewill be heavily influenced by the availability of data services.

Mobile networks are highly compatible with each other from a technicalpoint of view. Users typically do not experience any quality differencebetween calls made to the same network and calls made to other networks.Network effects are mainly induced by network operators through higherprices for calls to other networks (off-net calls) than for calls to the samenetwork (on-net calls). This pricing strategy is pursued by operators inmost European countries. In a previous paper (Birke and Swann, 2006), wehave shown that choice of mobile phone operators is strongly coordinatedwithin households and that this effect is far more important for operatorchoice than the effect of overall network size. This is interesting in thecontext of the network effects literature, as an often assumed equivalencebetween direct and indirect network effects hinges on the assumption thatonly overall network size matters and not who is on the network. Whereasthis assumption seems tenable for markets with indirect network effects, itis doubtful for markets with direct network effects. In these markets, con-sumers are interested primarily in which of their interaction partners usesthe same technology and rather less in the overall number of users.

This chapter looks directly at operator choice in a social network. For thispurpose, we conducted a survey of a class of undergraduate students atNottingham University Business School, asking them to identify their socialnetwork and filling in a questionnaire about their mobile phone usage. Wethus were able to obtain a relatively well-bounded network. Obviously, stu-dents do have a social network apart from their university class, but theresults show that interaction between students was strong within class andthat students coordinated operator choice within this social network and inparticular among students sharing geographical origin.

Network data analysis exhibits two distinctive features that have to beaddressed by the researcher. First, observations are not independent of eachother. We shall try to overcome this problem of structural autocorrelation byusing a technique called ‘quadratic assignment procedure’ for permutation-based estimation of standard errors (see Krackhardt, 1988). Second, obser-vations often are not drawn from a random sample. In our case, the studentsample is a convenience sample and we certainly cannot claim to be close to

Social networks and the economics of networks 181

a random sample. This limits the generalisability of the results, but, as weshall discuss later, it also yields some interesting results that we would nothave obtained with a random sampling approach.

This chapter is organised as follows. Section 2 discusses social networkanalysis and its usefulness for researchers interested in evolutionary eco-nomics. Section 3 gives a brief introduction to the UK mobile telecommuni-cations market. Section 4 outlines the approach taken with the survey. Itpresents descriptive analysis of students’ attitudes towards mobile telecom-munications and a description of the social network within the class. This isfollowed by a graphical and statistical analysis. Section 5 discusses the results.

2. SOCIAL NETWORK ANALYSIS ANDEVOLUTIONARY ECONOMICS

Research on social networks (and indeed networks in general) hasincreased rapidly in the last two decades and is undertaken in a variety ofdifferent disciplines (see Borgatti and Foster, 2003). Social network analy-sis (SNA) has its root in sociology and in its early years has been widelyused to analyse interaction between individuals. These interactions cantake on a variety of forms, from friendship over business to sexual re-lationships. However, SNA is not limited to social networks, but can beapplied to the analysis of networks in general. Economics has been slowto make use of advances in social network analysis and only recently havesuch studies been conducted in higher numbers and especially in fieldsassociated with evolutionary economics (see, for example, Breschi andLissoni, 2004; Giuliani, this volume; Maggioni and Uberti, this volume).

Why is social network analysis particularly helpful for researchers in thetradition of evolutionary economics? SNA is inherently relational, contex-tual and systemic, which is similar to how many researchers associated withevolutionary economics see the world and their discipline. Evolutionaryeconomics emphasises the role that diversity of people or organisationsplays for the understanding of economic processes (see Metcalfe, 1998).Furthermore, constraints in learning and in the selection process makeevolutionary economics much more inclined to structural theories ofeconomic interaction than mainstream economics, where most of the inter-actions occur in an (anonymous) marketplace. Both, learning and selectionenvironment are argued to be highly localised – both socially and geo-graphically (see Antonelli, 1995) and one main reason for this localisationis the underlying structural patterns of interaction.

A central theme of evolutionary economics is the importance of inno-vation and a rich variety of academic papers focus on the innovation

182 Network analysis

process. Closely intertwined with this topic is the role of knowledge andhow knowledge is acquired by firms. It is generally recognised that a highpercentage of knowledge is not readily codifiable and is transmittedthrough direct interaction between individuals. The paper by Cowan andJonard (2004), for example, is work along these lines; it looks at therelationship between network structure and diffusion performance in aknowledge barter process and is based on the idea of small-world networks.In general, there is high interest among researchers in the evolutionary tra-dition in using agent-based simulations to capture economic phenomenaand here again in particular in innovation research (see, for example, thediscussion in Frenken (2006) and the references therein). A common themeto a lot of this work is a rejection of the standard assumption of uniformagents interacting with other agents through an anonymous marketplace.Network structure plays an important role as an underlying assumption ofhow agents interact or as a key parameter that is varied and analysed by theresearcher. However, for simplicity the literature often assumes completelyregular network structures or easy network generation mechanisms and itis therefore important to confront these models with empirical findings.

In the recent past, there has also been an increasing interest in empiricalwork examining and using network structure. Cantner and Graf (2005)employ SNA to describe and analyse the evolution of the innovator networkin the town of Jena, Germany and Giuliani’s chapter in this volume looksat knowledge networks in the Italian and Chilean wine industry. Otherworks, like Murmann’s (2003) book on the economic history of the dyeindustry, have used social network analysis in a descriptive way to portraythe academic industrial dye knowledge network in the nineteenth century.

Likewise, there is a growing literature in sociology about how economictransactions are embedded in social relationships (see Uzzi, 1996). Friend-ship ties with competitors, for example, have been found to improve the per-formance of hotels in Sydney due to enhanced collaboration, mitigatedcompetition and better information exchange (see Ingram and Roberts,2000). These networks can further be used for the effective social enforce-ment of rules and norms of doing business.

Contrary to these contributions, this chapter will focus exclusively on thedemand side of the market – an area that remains scarcely studied.

3. THE MOBILE TELECOMMUNICATIONSINDUSTRY IN THE UK

In the United Kingdom, there are four main Global System for MobileCommunications (GSM) operators: Vodafone, O2, T-Mobile and Orange

Social networks and the economics of networks 183

and a purely third generation operator: ‘3’. O2 (Cellnet) and Vodafonestarted operation in 1985 with analogue mobile networks.1 There had beenrelatively slow growth until the entry of T-Mobile (One-to-One) and Orangeafter 1993 introduced stronger competition to the market. However, themarket really took off with the widespread use of prepaid cards, which mademobile telephony attractive for the mass market and especially for low-usageconsumers. Although a first prepaid tariff was launched by Vodafone inSeptember 1996, prepaid usage became popular only after mid-1998. AsFigure 9.1 shows, this led to a period of rapid expansion in the number ofsubscribers, which lasted roughly until early/mid-2001.

After the burst of the stock market bubble in mid-2001, operatorscleaned up their customer base of inactive consumers (note the shortdecline in Figure 9.1) and have since continued to grow, but this time moregradually and with a stronger focus on increasing the average revenue peruser (ARPU) and on upgrading prepaid customers to post-paid customers.In 2004, operators alone generate around £11 billion of revenues per year.However, as the market is by now reaching saturation with a penetrationrate of over 85 per cent, future revenue growth has to come from anincreased ARPU rather than from a bigger customer base.

In May 2000, the four GSM operators and ‘3’ were awarded licences forthird generation Universal Mobile Telephone System (UMTS) networksfor about £4 billion each. Many expect that 3G networks will further boostrevenues in the mobile telecommunications market. In 2003, ‘3’ introducedthe first third generation network in the UK. After a slow start, thecompany now has over 3 million users. The other companies recently fol-lowed with their own third generation networks.

184 Network analysis

Figure 9.1 Number of subscribers in the UK (in millions)

10

20

30

40

50

60

Q1 19

94

Q1 19

95

Q1 19

96

Q1 19

97

Q1 19

98

Q1 19

99

Q1 20

00

Q1 20

01

Q1 20

02

Q1 20

03

Q1 20

04

No. of subscribers

Especially interesting for our analysis is the development of marketshares (see Figure 9.2). At the end of 1998, the market was dominated bythe incumbent operators O2 and Vodafone, which together accounted foralmost 70 per cent of the market. However, by the beginning of 2001 thislead had dissipated and subscriber market shares were levelled. Today, themarket is about equally split among the four GSM operators.2

The ability of T-Mobile and Orange to catch up with the incumbentoperators is somewhat unique to the UK market and is different from, forexample, the German market in which the two biggest operators (T-Mobileand Vodafone) still control about 80 per cent of the market and havereported stable market shares in recent years. With strong network effectspresent in the market, this ‘catch-up’ by T-Mobile and Orange is surpris-ing, as network effects result in a strong tendency towards higher marketconcentration. It could be argued that the development in the UK marketis due to the high compatibility between networks. However, in Birke andSwann (2006) we showed that network effects do play an important role inthe adoption of mobile telephones and in operator choice.

4. COORDINATION OF CHOICE OF MOBILEPHONE OPERATORS

The Survey

To analyse how consumers coordinate their choice of mobile phone oper-ator with their social network, we looked at a network from a class ofundergraduate students. We expected mobile phones to be a rather impor-tant tool of interaction for this group and the network to be rather well

Social networks and the economics of networks 185

Figure 9.2 Development of subscriber market shares

10%

15%

20%

25%

30%

35%

40%

Q4 19

98

Q2 19

99

Q4 19

99

Q2 20

00

Q4 20

00

Q2 20

01

Q4 20

01

Q2 20

02

Q4 20

02

Q2 20

03

Q4 20

03

Q2 20

04

Vodafone

O2

T-Mobile

Orange

bounded with a high percentage of intra-network interaction. The sampleconsists of students from a second-year undergraduate course called‘Economics of Organisation B’, which was held at Nottingham UniversityBusiness School in spring 2005. Most students from this course study for athree-year degree and have been studying together in a variety of othercourses for about 18 months. Although most students can be expected touse their mobile phone to interact with many people outside the class, areasonably regular interaction between the students can be assumed.

The questionnaire on which the survey is based consists of two parts. Inthe first part, we asked students for some demographic details and abouttheir attitudes and behaviour with regard to mobile phones. Information onoperator choice and demographic variables like gender and nationality areused as input for the regression analysis. The other answers are mainly usedto gain insights into how the respondents use their mobile phones and fora first descriptive analysis. In the second part of the questionnaire, studentswere handed out a list of course participants and were asked to identifythemselves and the people they communicate with. The exact wording ofthe question was ‘Please tick the people that you call’.3 Thus, we are able tobuild up a network of the relevant communication relations within theclass. Both parts took about 10 minutes to fill in and were distributed andcollected during one session of ‘Economics of Organisation B’.

A total of 236 students registered for this course. From these students,171 filled in the first part of the questionnaire (the ‘questions’ part) ofwhom 158 were identified as students from the course list. Of the remain-ing 13 students, three respondents indicated that their name was not on thelist and another 10 did not identify themselves. To every student for whomwe did not receive an identified questionnaire (which mostly included stu-dents who missed class), we sent out a reminder email and consequentlyreceived another four responses.

In total, this resulted in 175 completed ‘question’ parts and 159 completed‘roster’ parts, which is a response rate of 74 and 67 per cent, respectively. Formost of the descriptive statistics, all 175 responses are used and for all analy-sis relating to social networks only the subsample of 159 students is used.

All of the students are undergraduate students and are almost of thesame age. There is about an even number of male (48 per cent) and female(52 per cent) students among the respondents. Importantly for our analy-sis, the share of foreign students is rather high, with only slightly over halfof the students originating from the UK and with another 8 per centcoming from other European countries (see Table 9.1). Chinese studentsare the second biggest group in the course (22 per cent).

The high number of foreign students is interesting for us, as we expectthat students who come to the UK primarily to study would have a social

186 Network analysis

network that revolves more around their university class than that ofBritish students. As can also be seen from Table 9.1 the majority of Chineseand other Asian students are female, while the majority of English andother European students are male.

Criteria for Choice of Mobile Phone Operator

A first option to analyse why the respondents chose their operators is to askthem directly about their choice criteria. Table 9.2 gives the responsefrequencies for a number of criteria. Quality, special offers, cost of calls andoperator choice of friends and family all seem to be important.

The obvious drawbacks of directly asking respondents about theirchoice criteria are the difficulties of comparing the relative importance ofthe different factors. Furthermore, it is not always clear whether the givenanswers are the actual reasons for choosing an operator or whether it isan ex post rationalisation of the choice process. Quality of the network, forexample, is named as an important criterion by most respondents. How-ever, the quality of the four GSM networks in the UK is roughly equiva-lent in terms of most quality characteristics such as network coverage,international roaming and customer service.4 We might rather measure thegeneral importance that the respondents attribute to quality when choos-ing a product than the particular importance for mobile phone networks.

Of course, a precondition for consumers coordinating their operatorchoice is some knowledge about what operators their peers use. This is farfrom given, as mobile networks are hard to identify from telephonenumbers. In the UK, there are several hundred prefixes associated with thedifferent networks.5 This is contrary to, for example, Germany where thereare only 23 prefixes, which makes it far easier to identify people who areusing the same operator – especially in the earlier years of mobile phoneadoption when not all of these prefixes were used. In general, operatoridentification from telephone numbers is easier in smaller countries wherefewer different prefixes are needed to cover all subscribers.

Social networks and the economics of networks 187

Table 9.1 Gender and nationality of respondents

British Other Chinese Other Africans The TotalEuropean Asian Americas

Male 60 9 6 4 5 0 84(65.2%) (64.3%) (15.4%) (23.5%) (50%) (48%)

Female 32 5 33 13 5 3 91(34.8%) (35.7%) (84.6%) (76.5%) (50%) (100%) (52%)

Information on who else is using the same network therefore has to beobtained by other channels in the UK. It seems likely that the availabilityof this information is directly linked to the closeness of two individuals inthe social network. This information could then be obtained either throughdirect conversation or through identifying the operator from a mobilephone. The latter requires that the name or logo of the operator is con-spicuously placed on the mobile phone, which is only the case for co-branded mobiles. In recent years, operators try to raise the awareness oftheir brand and increasingly place their logos on mobile phones next to thelogos of the mobile phone manufacturer, or mobile phones exclusivelycarry the brand of the network operator.

According to Table 9.3, respondents claim to know the mobile networkoperator for a high percentage of their peers. Especially the operators

188 Network analysis

Table 9.2 Frequencies for choice criteria

Strongly Agree Neither Dis- Strongly Don’tagree nor agree disagree know

(1) (2) (3) (4) (5)

Quality of the network 27 80 28 11 4 6(e.g. network coverage,roaming possibilities etc.)

Special offer 52 59 31 10 6 2Cost of calls, text messages 48 62 25 16 4 2

in generalIt is cheaper, because my 49 43 38 19 9 1

friends/family use the same network

Cost of handset 28 55 41 17 9 4Handsets available from 21 43 40 31 14 5

this operatorMore services available 3 21 54 46 26 4

(games etc.)Good customer service 15 49 55 16 13 7

Table 9.3 Do you know which operator your friends/family/partner use?

Know it Know it for some Don’t know it

My friends 78 (45.4%) 80 (46.5%) 14 (8.1%)My family members 123 (76.4%) 22 (13.7%) 16 (9.9%)My partner 62 (77.5%) 18 (22.5%)

used by family members and partners are known by the large majority.The lower figure of knowledge for operators used by friends is an indica-tor that operator coordination between friends might be lower thanwithin households.

Network Statistics

Besides an analysis of individual respondents, we can also analyse thecharacteristics of the social network as a whole. A first measure character-ising the overall network is its density. For directed graphs, it is calculatedas ��L/N(N–1) with L being the number of lines present and with N (N–1)giving the maximum number of lines potentially present in a directednetwork of N people. In our case, N is the number of students that filled inthe roster of the questionnaire, which is 159. In total, 815 different relationswere identified. However, 195 relations are to non-respondents in theclass and only 620 are to respondents, resulting in a density of �g�620/(159*158)�0.0247. At the individual level this network density stemsfrom an average of 3.90 nominations.

It is hard to judge whether this is a rather dense or a rather loose network.Overall network density obviously depends on network size, as people haveonly a limited capacity to communicate with other people. Furthermore,our measure of ‘Who do you call?’ can be assumed to refer to rather closecontacts compared, for example, with the question ‘Who do you know?’.

Network density can also be measured at an individual level, that is, onecan calculate the network density of each individual’s network. If indi-vidual x has n friends, then this local network density measure gives the per-centage of ties present between the n friends. Individual 1, for example, hassix friends. Those six friends have 28 ties between each other, out of a pos-sible t�n(n – 1)�6*5�30 ties, resulting in a local network density of �1�0.93. Table 9.4 gives the frequencies of local network density for differentnumbers of friends. The local networks observed in the course are rathertight cliques with an average local density of 0.52. Rather dense cliquesshould favour a coordination of operator choice. Table 9.4 also shows thatlocal network density decreases with the number of friends. Clearly, largernetworks are less closely interconnected.

If every node in a graph is connected to all other nodes via a path, thenthe graph is said to be connected and consists of one single component. Ifthere are disconnected subsets, several components exist. The number ofcomponents can be calculated in two different ways. First, one can calcu-late the number of weak components, that is, two persons are in the samecomponent, if one of them can reach the other. Second, strong compo-nents require that both persons can reach each other. In undirected or

Social networks and the economics of networks 189

symmetrical networks, these measures coincide, but in our case, we have adirected network and consequently we get two measures.

Calculating weak components, we end up with three components in thenetwork (plus two students who only nominated students who did notrespond). The largest component consists of the large majority of students– altogether 146 students are part of this component (92 per cent). The nextlargest component consists of nine students and then there is a third groupof two students who nominated each other, but did not nominate anyoneelse (and were not nominated by other students).

The result is an indication that the class network can really be seen as onenetwork. The average distance between two reachable nodes in the networkis 5.7 (s.d. 2.3) with a maximum of 12 steps needed to reach the final node.Out of a potential N(N – 1)�159*158�25122, 18151 pairs can reach eachone another (72 per cent). The distribution of distances between these pairsis displayed in Figure 9.3. The distances are roughly normally distributedand the distribution has a mode of 6.

Although the large majority of nodes belong to the same component, theaverage path length is rather long and has a considerable variance, which

190 Network analysis

Table 9.4 Local network density

No. of friends

2 3 4 5 6 7 8 9 10 11 Mean

No. of 28 28 22 26 15 7 7 3 0 2 138observations

Local network 0.71 0.57 0.47 0.48 0.41 0.41 0.33 0.35 – 0.1 0.52density

Figure 9.3 Distribution of distances between nodes

0%2%4%6%8%

10%12%14%16%18%

1 2 3 4 5 6 7 8 9 10 11 12

already hints at some clustering in the network, which can be furtheranalysed using graphical techniques.

Graphical Analysis of Social Network

Social networks can very usefully be analysed by graphical representationsof these networks, in particular in the case of medium-sized networks witha couple of hundred nodes.6 Figure 9.4 depicts the social network withinthe class of students, based on their communication pattern.

It is a directed graph and arrows depict the direction of the nominationsfrom the roster. The graph was created using a spring embedding algorithmfrom UCI-NET, which is based on the idea of simulating the social networkgraph as a system of mass particles. Nodes are the mass particles and theedges are springs between the particles, while the algorithm tries to minimisethe energy of this system. Some form of clustering immediately becomesobvious. First, shapes of the objects, depicting nationalities, are highly clus-tered. Chinese students for example (up triangles) almost exclusively com-municate with other Chinese students. At the bottom right of the graph,there is a group of Asian students who even form a distinct component andhave only communication links within the group. This is the second largestcomponent from the previous section. Two Spanish students form the thirdlargest component, which can also be found at the bottom right of thegraph. Finally, there are two isolates at the upper left.

Second, the graph shows a clustering of shades, which are depicting mainoperator chosen.7 This clustering of shades clearly occurs along national-ity lines. Chinese students are in the majority using Vodafone and similartendencies can be observed for other nationalities. However, there alsoseems to be a coordination of operators within nationalities, that is, alsowithin nationalities students that call each other tend to use the samemobile phone operator.

The strong correlation of operator choice within nationalities alsobecomes clear by a cross-tabulation of nationality and operator used (seeTable 9.5). Almost 80 per cent of Chinese students choose Vodafone.Similarly, almost half of the British students opted for an O2 mobile,whereas six out of 10 Africans use T-Mobile.

Regression Results

The original data of communications patterns are organised in a squarematrix of 159 rows and columns with 1’s indicating a communication re-lationship and 0’s indicating the absence of a communication relationship.As is common for network data, diagonal values are not allowed. For a

Social networks and the economics of networks 191

192

Fig

ure

9.4

Inte

ract

ion

netw

ork

ofst

uden

ts

Vod

afon

e

T-M

obile

Ora

nge

Virg

in

‘3’

Oth

erop

erat

ors

O2

Key

Sha

pes

Brit

ish

Oth

er E

urop

ean

Chi

nese

Oth

er A

sian

s

Afr

ican

s

The

Am

eric

as

regression analysis, this matrix is transformed into dyadic relationships(relationships between two nodes). We therefore get N(N – 1)�25122different dyads.

When using network data, the assumption of ordinary least squares (OLS)and logit models that observations are independent fails. Observations areclearly not independent as there are at least N – 1 dyads involving every indi-vidual. This correlation between observations involving the same nodesstems, for example, from the fact that it is far more likely to have the sameoperator as your friends if you use an operator with a high market share inthe network. This would result in a positive correlation between observationsfrom the same row or column and consequently, while parameter estimatesare unbiased, estimated p-values overstate the significance level.

One possibility to adjust for incorrect standard errors is the quadraticassignment procedure (QAP) as proposed by Krackhardt (1987, and 1988)for social network data. The idea of QAP is to permute rows and columns ofthe original data matrix for the dependent variable and then to re-estimatethe original regression model. This process is reiterated to obtain an empiri-cal sampling distribution from which QAP p-values are calculated. Detailsof the estimation procedure used are described in Birke and Swann (2005).

We are estimating a logit model with same_operator as the dependentvariable. This variable takes on the value ‘1’ if two students use the sameoperator and ‘0’ otherwise. There are four independent variables that areconstructed in a similar way: same_nation (respondents of the dyad havethe same nationality/come from the same group of nations as defined

Social networks and the economics of networks 193

Table 9.5 Mobile phone operators and nationality

British Other Chinese Other Africans The TotalEuropean Asian Americas

3 10 3 5 3 2 0 23(11.1%) (23.1%) (13.5%) (17.7%) (20%) (13.5%)

O2 42 5 1 5 0 1 54(46. 7%) (38.5%) (2.7%) (29.4%) (33.3%) (31.8%)

Orange 17 2 2 1 2 0 24 (18.9%) (15.4%) (5.4%) (5.9%) (20%) (14.1%)

T-Mobile 6 0 0 0 6 0 12 (6.7%) (60%) (7.1%)

Virgin 2 0 0 0 0 1 3(2.2%) (33.3%) (1.8%)

Vodafone 13 3 29 8 0 1 54(14.4%) (23.1%) (78.4%) (47.1%) (33.3%) (31.8%)

Total 90 13 37 17 10 3

above), friend8 (respondents call each other on their mobile phone),same_sex (nodes have the same gender) and same_payment (respondentsuse the same type of payment: contract versus pre-paid). Table 9.6 showsthe results from a logit estimation of the model with QAP p-values.

Same_nation, friend and same_sex are highly significant and show theexpected sign, confirming the graphical analysis from Figure 9.4. Tworespondents of the same nationality, who are friends and of the same sexare significantly more likely to use the same operator. Same_nation andfriend have a particularly high significance level and in fact no permutationresulted in a parameter estimate higher than the observed values from theoriginal regression. Same_sex is still significant at the 5 per cent level, butthe coefficient is far lower than the other two.

To get a better intuition of the importance of the different variables,Table 9.7 lists the predicted probabilities of having the same operator for anumber of constellations. For two respondents who are friends, of the samenationality and of the same gender, there is an almost 50 per cent prob-ability that they also use the same operator. On the other hand, for tworespondents who don’t call each other, are not of the same nationality andnot of the same gender, this probability is only 18 per cent. Most of thisvariation is due to the friend and same_nation parameters.

The model from Table 9.6 excludes the operator ‘3’, because of itsdifferent pricing structure. Analysing the same model for different oper-ators yields a positive coefficient for the ‘friend’ parameter for all oper-ators, but ‘3’.9 This is further support for our hypothesis that network

194 Network analysis

Table 9.6 Determinants of choosing the same operator

Parameter estimates QAP p-values

same_nation 0.814 p�0.000friend 0.676 p�0.000same_sex 0.130 p�0.014same_payment �0.041 p�0.426Constant �1.434 p�0.000

Table 9.7 Predicted probabilities of using the same operator

Not same nationality Same nationality

Not friends Not same sex 0.18 0.32Same sex 0.20 0.34

Friends Not same sex 0.28 0.45Same sex 0.30 0.47

effects are the reason for consumers coordinating their operator choice.Operator ‘3’ is the only UK operator that typically does not inducenetwork effects, but rather offers packages of calling time regardless ofthe network to which calls are made. The incentive for ‘3’ users to coordi-nate with their peers is therefore lower. This can also clearly be seen inFigure 9.4, where ‘3’ users are evenly distributed over the graph. Theresults are also contrary to the argument that learning effects or word ofmouth might be the prime cause of this coordination. The third gener-ation network and handsets of ‘3’ are arguably more difficult to masterthan other mobile phones and we would expect a coordination of oper-ator choice for ‘3’ if these effects were strong.

The correlation of operator choice within nationalities is particularlyinteresting and this might have several causes. All UK operators alsooperate networks in a number of other countries; sometimes under thesame brand, sometimes under different brands. Non-UK students mighthave simply continued to use the same operator they already used in theirhome country. However, concentration of operators worldwide is far lowerthan in the market for mobile phone handsets. Furthermore, most studentscome from countries where these operators do not have a network, as mostoperators have a rather European focus.

This coordination of operators within nationalities might be due tocommon unobserved characteristics and attitudes of respondents with thesame background or it could be a coordination mechanism. We thereforeregress friend on same_nation and same_sex. Both having the same nation-ality and being of the same sex are important predictors of friendshipbetween two respondents (see Table 9.8). Students from the same nationand from the same sex interact far more frequently.

Both parameters are highly significant, but the coefficient of same_nationis about three times bigger than for same_sex. Like for operator choice, wecan again calculate the predicted probabilities for different constellations(see Table 9.9). The predicted probability of an interaction between tworespondents is generally rather low, but for two respondents from the samenationality and the same sex this probability is 10 times higher than for tworespondents of different nationalities and different gender.

Social networks and the economics of networks 195

Table 9.8 Friendship determinants

Parameter estimates QAP p-values

same_nation 1.823 0.000same_sex 0.701 0.000Constant �5.119 0.000

The group of students for which coordination is strongest are Chinesestudents who in the large majority used Vodafone. To the best of ourknowledge, at the time of the survey10 there was no special tariff offered byVodafone targeting Chinese students (like, for example, cheap calls toChina) and Vodafone does not have an own network in the PRC, which stu-dents might have used prior to their study in England.11 Asking fellow stu-dents why Chinese students choose Vodafone as their operator, they repliedthat other Chinese students told them on arrival that all Chinese studentsuse Vodafone and that they should also use it, if they want other people tocall them. This has afterwards also been confirmed by other Chinese stu-dents and by students from other nationalities.12 If nationality is a strongdeterminant of friendship, it is a good choice to use the same network asother people from the same nation in order to reduce the number of off-netcalls. Furthermore, even when accounting for this effect, friends are stillmore likely to choose the same operator.

5. DISCUSSION

We have shown that consumers coordinate their choice of mobile phoneoperators not only within households, but also in their wider socialnetwork. We further found that this depends on the price differencebetween on- and off-net calls induced by most operators. Like the resultsfrom Birke and Swann (2006), this is further evidence that in markets withdirect network effects it matters to the consumer who is on the samenetwork. In this respect, direct network effects are different from indirectnetwork effects, where only the total number of users matters.

As discussed earlier, the sample on which this chapter is based is far fromrandom and it is therefore difficult to generalise the findings to, say, theBritish population. The high percentage of foreign students might havefavoured the results to a certain extent. However, we can also observe astrong coordination of operators among British students. Furthermore, itcan be assumed that a significant part of a student’s communication takesplace outside the ‘Economics of Organisation B’ class. Results from thesurvey (and common sense) suggest that, for example, calls to the partner

196 Network analysis

Table 9.9 Predicted probabilities of calling each other

Not same nationality Same nationality

Not same sex 0.006 0.036Same sex 0.012 0.069

are a significant share of all calls. Consequently our results might ratherunderstate the extent of coordination between close friends. In furtherresearch using calling records from land lines, we have found that the largemajority of calls are made to three to five parties, but that these interac-tion partners vary over time. If this also holds for mobile phones then, atany one point in time, it should be relatively straightforward to coordinateoperator choice with these peers. If geographical origin of students is animportant predictor of friendship, then these geographical patterns mightbe even more stable than the friendships existing at the time of choosingthe operator (for international students this typically is the time of arrivalto the UK).

The results pertaining to differences in interaction within and betweenstudents from different nationalities are interesting in their own right.First, they show that not only is geographic distance an important factorin many diffusion and coordination processes but also geographic origin.Second, the analysis gives a powerful illustration of how individual choicecan be constrained by network structure. Even if coordination of con-sumption choices in markets with global network effects is important,different operators (or technologies) can easily coexist with one anotherdue to local network effects, as long as global network effects are not toostrong. Finally, we can speculate that Chinese students overwhelminglyusing Vodafone might be a case of path dependence where a relativelysmall past event made some Chinese students choose Vodafone as theiroperator and where it is now beneficial for new arrivals to choose Vodafonesimply because the large majority of other Chinese students use the sameoperator.

The empirical analysis in this chapter focuses on one specific point intime and therefore cannot capture dynamics in the system. A longitudinalanalysis based on social network data is even harder to achieve than for tra-ditional datasets, due to the more severe consequences of missing valuesand the related problems with sampling new participants. However, thisanalysis could in theory be done relatively easily with access to electronicdata on mobile calling patterns between individuals. This would alsodirectly relate the frequency and cost of interactions with network choice.By using electronic calling records, it would be possible to get a far biggersample size, which in theory could include all subscribers to a particularnetwork. The obvious difficulty here is the confidentiality of this data andthe resulting reluctance of companies to grant access to it.

This chapter has demonstrated the fruitfulness of the use of socialnetwork analysis techniques for the analysis of economic problems. Onedrawback of SNA is the need for network data which is often not readilyavailable. With the digitalisation of many aspects of society, much more

Social networks and the economics of networks 197

network data are easily accessible in electronic forms to researchers now-adays. However, there remains a considerable number of research areas thatrequire the collection of primary data by the researcher, which is anapproach traditionally seldom pursued by economists. Social networkanalysis enables a more realistic and rich modelling of many economicdecision-making processes and therefore in many cases justifies the add-itional efforts in data collection. Although this should be beneficial to eco-nomics as a whole, as we have argued, this could be particularly interestingfor work in the tradition of evolutionary economics.

NOTES

* I am grateful to Peter Swann, David Paton, Robin Cowan, seminar partici-pants at Chimera, University of Essex, Nottingham University Business School andparticipants at the EMAEE conference in Utrecht, The Netherlands for helpful com-ments. I would also like to acknowledge financial support from the University ofNottingham Business School and the ESRC. The analysis has been conducted using theeconometrics package STATA, SAS and the social network software UCI-NET andPAJEK.

1. See Valletti and Cave (1998) for an analysis of the UK market from 1985 to 1998.2. Note that this holds for subscriber market shares. Although there has been a similar

trend in revenue market shares, Vodafone still boasts the highest revenue, as its customersgenerate a higher ARPU.

3. For simplicity, we shall call respondents who communicate with each other ‘friends’ inthe rest of the study.

4. See Birke and Swann (2006) for details.5. Number portability also makes it more difficult to identify mobile networks from

telephone numbers. However, in the UK, less than 7 per cent of all mobile numbers areported (Q3 2004).

6. See Freeman (2005) for an overview of graphical representations for social networkanalysis.

7. Some of the respondents had up to three different mobile phone operators. Most of themindicated a main operator used; for three respondents the main operator was decidedrandomly as no information about which operator was mainly used was obtainable.

8. The friend matrix has been symmetrised for the regression analysis, that is, we assumethat if a tie is not reciprocated, this is because the respondent forgot to nominate theinteraction partner and not because it is a true one-way relationship. This is a standardassumption of social network analysis. The results are virtually the same when using thenon-symmetrised matrix.

9. The coefficient for Orange is not significant, but has the right sign.10. More recently, special international tariffs have been offered by some operators. O2 has

taken the lead here and there is anecdotal evidence that some Chinese students areswitching to O2 to profit from these discounts.

11. Vodafone has a minor stake in China Mobile, but it is a rather small stake (approximately3.27 per cent) and is most likely not known to the average consumer.

12. One of the comments I received from international seminar and conference partici-pants was that they encountered similar coordination mechanisms when they movedabroad.

198 Network analysis

REFERENCES

Antonelli, C. (1995), The Economics of Localized Technological Change andIndustrial Dynamics, Dordrecht etc.: Kluwer Academic Publishers.

Arthur, W.B. (1989), ‘Competing technologies, increasing returns, and lock-in byhistorical events’, Economic Journal, 99(394), 116–31.

Birke, D. and Swann, G.M.P. (2005), ‘Social networks and choice of mobile phoneoperator’, Nottingham University Business School: Industrial EconomicsDivision Occasional Paper Series, No. 2005-14.

Birke, D. and Swann, G.M.P. (2006), ‘Network effects in mobile telecommuni-cations – an empirical analysis’, Journal of Evolutionary Economics, 16(1):65–84.

Borgatti, S.P. and Foster, P.C. (2003), ‘The network paradigm in organizationalresearch: a review and topology’, Journal of Management, 29(6), 991–1013.

Breschi, S. and Lissoni, F. (2004), ‘Knowledge networks from patent data: method-ological issues and research targets’, in Glänzel, W., Moed, H. and Schmoch, U.(eds), Handbook of Quantitative Science and Technology Research, Dordrecht:Kluwer, pp. 613–43.

Cantner, U. and Graf, H. (2005), ‘The network of innovators in Jena: an applicationof social network analysis’, Paper presented at the 4th European Meeting onApplied Evolutionary Economics (EMAEE), 19–21 May, Utrecht, TheNetherlands.

Cowan, R., Cowan, W. and Swann, G.M.P. (1997), ‘A model of demand with inter-actions among consumers’, International Journal of Industrial Organization,15(6): 711–32.

Cowan, R. and Jonard, N. (2004), ‘Network structure and the diffusion of knowl-edge’, Journal of Economic Dynamics & Control, 28(8): 1557–75.

Farrell, J. and Saloner, G. (1985), ‘Standardization, compatibility, and innovation’,RAND Journal of Economics, 16(1): 70–83.

Freeman, L.C. (2005), ‘Graphic techniques for exploring social network data’,in Carrington, P.J., Scott, J. and Wasserman, S. (eds), Models and Methodsin Social Network Analysis, Cambridge: Cambridge University Press,pp. 248–69.

Frenken, K. (2006), ‘Technological innovation and complexity theory’, Economicsof Innovation and New Technology, 15(2): 137–55.

Ingram, P. and Roberts, P.W. (2000), ‘Friendships among competitors in the Sydneyhotel industry’, American Journal of Sociology, 106(2): 387–423.

Katz, M.L. and Shapiro, C. (1985), ‘Network externalities, competition, and com-patibility’, American Economic Review, 75(3): 424–40.

Krackhardt, D. (1987), ‘OAP partialling as a test of spuriousness’, Social Networks,9(2): 171–86.

Krackhardt, D. (1988), ‘Predicting with networks: nonparametric multiple regres-sion analysis of dyadic data’, Social Networks, 10(4): 359–81.

Metcalfe, J.S. (1998), Evolutionary Economics and Creative Destruction, The GrazSchumpeter lectures, London: Routledge.

Murmann, J.P. (2003), Knowledge and Competitive Advantage. The Co-evolution ofFirms, Technology, and National Institutions, Cambridge: Cambridge UniversityPress.

Rohlfs, J. (1974), ‘A theory of interdependent demand for a communicationsservice’, Bell Journal of Economics and Management Science, 5(1): 16–37.

Social networks and the economics of networks 199

Uzzi, B. (1996), ‘The sources and consequences of embeddedness for the economicperformance of organizations: the network effect’, American Sociological Review,61(4): 674–98.

Valletti, T.M. and M. Cave (1998), ‘Competition in UK mobile communications’,Telecommunications Policy, 22(2): 109–31.

200 Network analysis

PART IV

Spatial systems

10. Diversity, stability and regionalgrowth in the United States,1975–2002Jürgen Essletzbichler*

1. INTRODUCTION

Recent years have witnessed important changes in economic governancesystems represented as a scalar shift of economic and political power fromnational states to supra-national entities and subnational entities such ascities and regions (Jessop, 1990, 1994; Brenner, 1998, 2004; Scott, 1998).Cities, regions and city-regions are increasingly forced into direct compe-tition with each other that prompts regional policy makers to activelydesign and shape regional economic development (Harvey, 1989; Leitnerand Sheppard, 1998). Baden-Württemberg, the Third Italy and SiliconValley exemplify the paradigmatic model of economic development thatother regions attempt to emulate (Bartik, 1996). Regional policies aredesigned to attract clusters of functionally related industries with highgrowth potential although the value of cluster-based policies is not uncon-tested (Hudson, 1999; Lovering, 1999; Begg, 2002; Martin and Sunley,2003; Boschma, 2004; Kitson et al., 2004). Duranton and Puga (2000: 533)caution that many of these policies seem to ‘lack a clear rationale or evento be based on common misconceptions’. The value of industrial special-ization for regional economic development is uncertain as theoretical andempirical work on specialization and diversity of cities suggests (Baldwinet al., 2003; Black and Henderson, 2003; Duranton and Puga, 2000, 2001;Feldman and Audretsch, 1999; Henderson, 1997; Henderson et al., 1995).In particular, there appears to be a trade-off between growth and stabilityof regional economies that is largely ignored by policy makers (Baldwinand Brown, 2004).

This chapter examines the relationship among diversity, growth and sta-bility of regional production systems. Empirical work by regional scientistsand new geographical economists yields ambiguous results. While Kort(1981) and Baldwin and Brown (2004) find strong evidence for a positive

203

relationship between stability and diversity and a negative relationshipbetween employment growth and diversity, Attaran (1986) and Smith (1990)contest these findings. Overall, the review by Dissart (2003) suggests thatmore diversity leads to more stability and less growth in unemployment. Theregional science literature is strongly focused on the identification of empir-ical relationships, while the theoretical links are not fully developed (Conroy,1974, 1975; Siegel et al., 1995; Chandra, 2003). New geographical econo-mists examine the theoretical and empirical relationship between diversityand economic growth (Krugman, 1991; Glaeser et al., 1992; Henderson,1997; Brakman, et al., 2001). Some of these researchers are influenced by theideas of Alfred Marshall (1920) and Jane Jacobs (1969) linking variety (tech-nological and/or industrial) to external economies, efficiency of regional pro-duction systems and economic growth (Henderson, 2003). Conclusions fromboth areas of research have potentially important policy implications.Should regional policy makers stimulate or curtail the production of diver-sity to foster regional economic growth? Should policy makers focus on gen-erating conditions for high rates of economic growth or should they focus onminimizing growth rate fluctuations?

The existing work has developed important theoretical arguments tounderstand the relationship between diversity and growth (Henderson,1988; Glaeser et al., 1992, Quigley, 1998), but the relationship betweendiversity and stability has been undertheorized. In part, this might beexplained by the influence of neoclassical economics on regional scienceand new geographical economics focusing on market competition as theonly allocation mechanism. Although formulated at the level of the firm,these concepts have been scaled up to the regional and national levels(Porter,1990, 1998). What is often overlooked is the fact that firm compe-tition is based on very different principles from regional competition.While firms have to maximize profits to stay in business, regions cannot gobankrupt (Krugman, 1994). Furthermore, regional policy makers areresponsible to different interests in the region including the provision oftechnical and social infrastructures and social services. Only if it isassumed that increased ‘regional efficiency’ translates into welfare gainsfor everybody can the exclusive focus on economic growth be justified.Instead of using economic growth as a vehicle to achieve other goals suchas equity or sustainability, higher rates of economic growth become thepolicy target. This exclusive focus on growth might be problematic if itleads to a reduction in technological, industrial, social and institutionaldiversity in the region. Because growth rates are maximized if less-efficientroutines, technologies, skills and industries are eliminated, this is likely tobe the case. A lack of diversity might reduce the adaptive potential of theregion to future change.

204 Spatial systems

This chapter addresses explicitly the trade-off between regionalemployment growth and regional economic stability drawing on insightsfrom evolutionary theory and ecological economics. Evolutionary theo-ries highlight the importance of diversity as fuel for the selection process(Nelson, 1995). Selection winnows on existing variation and, given astable selection environment and no introduction of new diversity, willensure that only the most efficient entities survive. In reality, new diversityis added through innovation and firm entry and coupled with a continu-ally changing environment, efficiency and optimality criteria are perpetu-ally redefined. Perfect adaptation towards a global optimum is thereforeimpossible (Hodgson, 1993, 1997). Applications in evolutionary econom-ics focus primarily on the impact of firm diversity in populations of com-peting firms on population (for example, industry) averages. In this work,Fisher’s principle is employed, stating that the rate of change is propor-tional to the variance in efficiency characteristics (for example, profitrates, unit costs or productivity levels) (Metcalfe, 1994, 1998). Recentwork suggests that intra-population dynamics has to be linked to inter-population dynamics and include selection processes at various analyticalscales such as the firm, industry, region and nation (Gowdy, 1992;Andersen, 2004).

Moving the focal level to the regional scale complicates the analysis con-siderably. Ecological economists and evolutionary biologists have longargued that a trade-off between adaptive efficiency and the adaptability (theability to adapt to environmental changes) of ecosystems exists (Gould andLewontin, 1979; Levins and Lewontin, 1985; Vrba and Gould, 1986). Likeecosystems, regions might be confronted with an explicit trade-off betweenadaptation and flexible adaptivity/resilience. Adaptation refers to the optimaladjustment to current environmental circumstances. Adaptation is achievedthrough enhanced efficiency of individual agents (for example, through inno-vation and imitation) and the elimination of redundant features such as unde-sired skills, inefficient technologies, industries, organizations and institutions.While boosting current efficiency levels (and rates of economic growth), lowerlevels of diversity decrease the likelihood of pre-adaptive features and thepotential to react to changing environmental conditions ushered in by tech-nological paradigm shifts, exogenous shocks or changes in the institutionalenvironment (Holling, 1973, 2001). However, there are limits on the extent ofdiversity. Without commonalities between different entities, no synergiesarise, and certain efficiency thresholds necessary for the economic survival ofregions might never be reached.

In this chapter, the theoretical arguments from evolutionary theory andecological economics are summarized to put the trade-off between regionaleconomic diversity and regional economic growth on stronger theoretical

Diversity, stability and regional growth in the United States 205

foundations. For this purpose, Section 2 reviews work by evolutionarytheorists and ecological economists and their emphasis on the relationshipamong diversity, growth and stability. Section 3 presents a simple empiricalmodel that links regional economic diversity to stability and growth andSection 4 concludes this chapter.

2. DIVERSITY, STABILITY AND GROWTH

The case for diverse regional production systems hinges on the premise thatdiversity reduces volatility1 (or enhances stability). Stability is seen as a pos-itive property of regional production systems for two reasons. First, highlevels of volatility are often coupled with higher rates of unemployment,because contracting economies destroy jobs and release workers andbecause it takes time to match workers to new jobs. Second, high volatilitycomplicates planning decisions to provide adequate investment in technicaland social infrastructures (Schoening and Sweeney, 1992; Baldwin andBrown, 2004). The maintenance of diversity is therefore useful from apolicy point of view. However, diversity affects not only stability but alsoregional efficiency by stimulating or constraining innovation, technologyspillovers and supplier–customer interaction (Jacobs, 1969). Whether ornot diversity will generate external economies through spillovers is likely todepend on the exact mix of industries, firms, workers, organizations andinstitutional practices in a region. Too much diversity might stifle theformation of spillovers through a lack of synergies. Too little results inincreasing specialization and makes the region vulnerable to changes intechnological paradigms, demand and supply shocks.

Current market-driven policies largely based on David Ricardo’s theoryof comparative advantages drive the formation of trade areas and theglobalization process. The erosion of national boundaries is likely to resultin increasing regional specialization. This might increase overall efficiency(at the supra-national or global level) but at the expense of increased vul-nerability at the subnational or regional level. Unfortunately a theoreticaldiscussion of these intertemporal and interspatial trade-offs is largelyabsent from economic geography. A substantial body of literature theoriz-ing these trade-offs, however, is emerging in evolutionary theory, ecologicaleconomics and complex systems analysis (Giampietro and Mayumi, 1997;Holling, 2001, 2004; Rammel and van den Bergh, 2004; Ulanowicz, 1997).In the following, the arguments emerging from this literature are summar-ized. Although there are limitations to the transferability of knowledge andconcepts from the physical and biological sciences to the social realm, someof the conclusions from this literature pertain to all complex, adaptive

206 Spatial systems

systems (Giampietro and Mayumi, 1997). The review of this literature isnot expected to generate a series of testable hypotheses but a series ofgeneral principles on the relationship among diversity, stability and growthof regional production systems that can be explored through empiricalwork.

Evolutionary Theory, Diversity and Stability

Diversity and selectionEvolution is driven by the creation and destruction of diversity. Diversityis expressed as variation at the genetic level, as biodiversity at the level ofecosystems, as technological diversity at the level of industries, as industrialand institutional diversity at the level of regions and countries. In biologi-cal systems, diversity is created by random mutation. In socio-economicsystems, diversity is generated primarily by the processes of innovation andplant entry (Nelson and Winter, 1982; Saviotti and Metcalfe, 1991;Hodgson, 1993; Dosi and Nelson, 1994; Nelson, 1995; Saviotti, 1991, 1996;Rigby and Essletzbichler, 1997, 2006; Essletzbichler and Rigby, 2005a, b).Reduction of diversity is driven by imitation and selection. Selectionrewards those species, firms, regions or countries that are best adapted tonarrow conditions at the moment. In this sense, selection operates as ashort-term adaptive force (Rammel and Staudinger, 2002). Adaptation isinterpreted as ‘temporary feature providing a benefit over its alternativesunder specific environmental conditions’ (Rammel and van den Bergh,2003: 123, emphasis added). This view on adaptation has a number ofimportant implications relating to questions of optimality, efficiency, equi-librium and causal relationships.

Without going into details about debates on adaptationism in evo-lutionary biology (see, for instance, Gould and Lewontin, 1979; Depew andWeber, 1995), a few clarifications need to be made in the context of this con-tribution. Spencer interpreted selection as a process that guaranteed the‘survival of the fittest’. In this view, selection is regarded as a (global) opti-mization process. This reading of selection entails a closed universe, onethat can be described by a unique and optimal equilibrium configurationtowards which the system gravitates. Natural selection is the mechanismthat ensures that this state will be reached eventually. In equilibrium, onlythose traits, species and populations survive that are perfectly adapted toenvironmental conditions describing this equilibrium. In equilibrium aglobal optimum is reached. This view of evolution was adopted by neo-classical economists who interpreted competitive markets as selectionenvironments that ensured that only firms with optimal technologies andorganizational routines survived. Because inefficiency was considered to be

Diversity, stability and regional growth in the United States 207

only a transitory phenomena, diversity in firm behavior could be ignoredand firms be treated as if they were profit maximizers (Friedman, 1953).During the movement towards global optimum, diversity becomes elimi-nated (see Vromen, 1995).

In the absence of changing environmental conditions and creation ofnew diversity, selection would indeed reduce variation until only the profitmaximizers would survive (Alchian, 1950; Jovanovic, 1982; Iwai 1984a, b;Metcalfe and Gibbons, 1986; Metcalfe, 1994, 1998). In reality, firms areconfronted with moving targets in the form of shifting fitness landscapes,continuous introduction of new diversity in the form of innovation andtechnological change, random shocks and non-linear feedback mech-anisms and complex patterns of interactions whose outcomes cannot bepredicted ex ante. In this environment of uncertainty and unpredictabil-ity, optimization must be understood as local and myopic (Nelson, 1995).In that sense it might be better to talk about ‘survival of the fitter orsufficiently fit’ (Rammel and van den Bergh, 2003) or ‘survival of thefitting’ (Boulding, 1981). According to this view, selection does notentirely eliminate diversity. Although the persistence of diversity might beundesirable from a neoclassical point of view, the rejection of the exis-tence of a global optimum makes diversity in the form of redundant, sub-optimal and inefficient technologies, skills, firms and industries not onlyacceptable but a necessary condition for long-term survival of firms andregions.

Diversity, optimality and stabilityAs in regional science, the exact relationship between diversity and stabilityis still debated in ecological theory (Holling, 2001; Rammel and Staudinger,2002). In ecology, diversity is negatively related to stability ‘if species diver-sity reflects a diversity in functional entities in an ecosystem with minimumredundancy’ (Rammel and Staudinger, 2002: 305). Translated to economicgeography, this case would describe a region with a diversity of sectorswhose technological inputs and demand are highly correlated, that is, whoseinput–output structures are almost identical. In this case, the existingindustrial diversity would not protect the region from demand shocksand/or shifts in technological paradigms (Wagner and Deller, 1998; Frenkenet al., 2005).

Independent of the specific expression of diversity, the rejection of theassumption of a global optimum complicates definitions of efficiency2

suggesting that current regional policies based on some notion of economicand social efficiency are driven by the ‘ideology of efficiency’ (Bromley,1990) and not derived from solid theoretical foundations. In the presence ofshifting adaptive landscapes and moving equilibria, the focus on efficiency

208 Spatial systems

(in the ‘maximum power’ sense) entails a prioritization of short-term adap-tation/optimization that comes, potentially, at the expense of long-termstability. ‘If optimality exists it will be temporary, because through evolu-tion, selection, and innovation it is easily transformed into maladaptivetraits. Under such conditions, diversity is a key element of long term stabil-ity and even survival’ (Rammel and van den Bergh, 2003: 127).

Diversity, enhanced adaptive flexibility and ‘evolutionary potential’One of the main arguments to maintain diversity is its role as ‘repertoire ofalternative options’, which increases the probability that pre-adaptations toaltered conditions exist. This is referred to as ‘evolutionary potential’(ibid.). While selection operates as a short-term adaptive force that reducesdiversity to narrow and temporally adapted features, selection does notguarantee survival in the long run (Matutinovic, 2001). Diversity persistsbecause of imperfect adaptation and the counter acting influence of othersorting mechanisms. Selection rewards those individuals or firms that arerelatively more efficient (generally characterized by lower input–outputratios) but a firm’s competitive position is also improved by exaptation andexogenous shocks (Gowdy, 1992). For instance, exaptation could refer toan increase in efficiency of suppliers of a firm A that translates intolower costs of inputs and in turn a lower input–output ratio of firm A.This improvement of efficiency is achieved without any actual technologi-cal or organizational changes by firm A. Exogenous shocks, such as a risein energy prices, can influence firm A’s efficiency through a shift in rela-tive prices of input factors. Firms that use relatively small amounts ofenergy will improve their efficiency relative to firms that use largeramounts of energy (for an empirical example, see Berman and Bui, 2001).Selection is thus only one of many sorting mechanisms that drive evolution.Exaptation and exogenous shocks might stimulate diversity, becausethese sorting mechanisms might reward relatively ‘inefficient’ firms.Diversity might therefore be as much an evolutionary outcome as special-ization. Within bounds, regions should therefore embrace rather than elim-inate redundancy.

In ecology, redundancy of agents (and pathways) stands for ecosystemoverhead. Ulanowitz in Matutinovic (1992) argues that ‘ecosystem over-head evolves: (1) as a response to the opportunity for the complete use ofavailable resources (efficiency in the “second law” sense); (2) to preventsystem brittleness; (3) to preserve its adaptive response and creativity; and(4) to preserve its reliability’ (Matutinovic, 2002: 434). Similarly, there aregood reasons for economic systems to embrace overhead or redundancy.From an evolutionary point of view, firms, institutions, regions and coun-tries are likely to be forced into a trade-off between realizing short-term

Diversity, stability and regional growth in the United States 209

profits (adaptation to current conditions to achieve a local optima) andlong-term flexibility to enhance the adaptive potential and the ability toreact to technological paradigm shifts, exogenous shocks and industrialshifts (Schütz, 1999). Mayumi and Giampietro (2001: 13, emphasis added)suggest that long-term (regional) competitiveness is achieved through‘increases in efficiency . . . by amplifying the most performing activities,without eliminating completely the obsolete ones’. From a regional point ofview this entails a strengthening of existing well-performing sectors (prob-ably clusters) but without completely eliminating those firms and sectorsthat appear less efficient and redundant at present. This view also resonateswith arguments made by innovation system researchers (Lundvall andJohnson, 1994; Edquist, 1997).

Contrary to biological systems, socio-economic systems actively producediversity. This means that the diversity-selection feedback works muchfaster in social systems and hence, any reduction in diversity might be trans-lated more rapidly into adaptability problems. Furthermore, economicsystems are often characterized by increasing returns based on internal andexternal economies, cumulative technological change, learning andnetwork externalities and complementary production factors that canresult in path-dependent evolution and lock-in. Diversity helps to breaklock-in and path dependence (Grabher, 1993; Arthur, 1994; Grabher andStark, 1997). Diversity at the level of the region refers to diversity inlabor (skills), firms, industrial sectors, organizations and institutional en-vironments but also the network connections between local and non-localagents (Granovetter, 1973; Grabher and Stark, 1997; Matutinovic, 2002).Diversity can thus be seen as a risk-minimizing strategy similar to port-folios in business economics (Chandra, 2003).

The theoretical arguments on the relationships among diversity, stabilityand resilience are rather general and biology, ecology and complex adapt-ive systems theory have yet to solve the exact linkages between them.Despite these shortcomings, the theoretical arguments put forward demon-strate that a narrow policy focus on regional efficiency is problematic.

Strategies to maximize efficiency in the short term might pose problemsfor economic prosperity over longer time horizons and hence, prioritizeimplicitly the needs of current generations at the expense of future gener-ations. To complicate matters further, the impact of economic policies willlead to conflicting outcomes not only at various temporal but also atvarious spatial scales. Competition between regions might yield positiveeconomic returns for some regions, but also result in the unnecessary dupli-cation of infrastructure, services and organizations that appear wastefulfrom the perspective of the national state (Harvey, 1989; Hubbard andHall, 1998). On the other hand, regional specialization might result in

210 Spatial systems

positive region-specific externalities that maximize wealth at the nationallevel at the expense of intra-regional diversity (Krugman, 1991; Fujitaet al., 1999, Neary, 2001). Nations might manage risks by maintaining aportfolio of specialized regions similar to assets of companies. Decline insome regions will be compensated by growth in other regions. In this case,the national scale receives priority over the regional scale where the averagewell-being of national citizens will increase at the expense of decliningwelfare in declining regions. If these intertemporal and interspatial trade-offs do indeed exist, the trade-offs at various temporal and spatial scaleshave to be made explicit in regional policy templates rather than hiddenbehind the assumption that free markets will lead to a (global) welfareoptimum.

While evolutionary theory provides us with interesting insights into thetrade-off between diversity, stability and growth, the exact relationshipbetween technological and industrial diversity and economic growth andstability have been insufficiently developed so far. The relationship betweendiversity and economic growth has been addressed extensively by new geo-graphical economists (this literature cannot be discussed in this chapter butfor overviews on the new geographical economics, see Martin (1999),Sheppard (2000a, b), Neary (2001), Duranton and Puga (2004), Frenken etal. (2005) and Robert-Nicoud (2005) and for an empirical attempt to dis-entangle the effects of urbanization and localization economies on metro-politan labor productivity, see Rigby and Essletzbichler (2002)) whileregional scientists applied portfolio theory to examine the empirical re-lationship between diversity and stability.

Industrial Diversity and Portfolio Theory

In business economics and industrial organization, the concept of portfoliorefers to the valuation of the collection of a company’s assets to examinethe impact of product diversity on corporate profitability growth. Thebasic underlying principle is that diversity of assets reduces risk. Ideally acompany diversifies into technologically related industries/products inorder to maximize economies of scope, but also industries that are charac-terized by unrelated demand in order to protect overall sales from demandshocks in individual product markets. This reasoning has a striking simi-larity to the arguments by Giampietro and Mayumi (1997) on the behaviorof complex adaptive systems. Although regions cannot go bankrupt in thesame way as corporations do, regions expand and contract over the busi-ness cycle. Regional contraction manifests itself through plant closures, lowentry rates and a shrinking employment base. Once a negative cumulativecycle is set in motion it is often hard to switch to a new path of regional

Diversity, stability and regional growth in the United States 211

economic growth, to attract businesses and jobs. In severe cases of eco-nomic decline, regions are confronted with very fast rates of employmentdecline. This is particularly the case if the economic base is dependent ona few companies and/or industrial sectors. If individual plants are closeddown because of structural problems occurring in this sector, related indus-tries follow rapidly and whole areas can be transformed into ghost townsin a short period of time. Detroit in the United States, Liverpool in the UK,Ivanovo in Russia and Halle in Germany are examples of these unfoldingprocesses (Oswalt, 2004). Although the region does not go bankrupt in thesame sense as firms do, capital has to be scrapped, workers laid off and, inthe case of prolonged crisis, have to move to other regions.

In most circumstances, not all sectors of an economy decline at the sametime or at equal rates. Borrowing from portfolio theory, it is therefore pos-sible to think of regional diversification as a strategy to reduce the risk ofeconomic decline. Developing a portfolio of industries whose demand islargely uncorrelated might be a useful strategy of regions to avoid bigfluctuations in rates of economic growth and to shield them in part fromeconomic decline during recessions (Baldwin and Brown, 2004). Clearlythese arguments are rather abstract and require refinement. The same levelsof regional diversity might result from very different industry mixes, andsome of them might be more favorable than others. Even if the levels ofregional diversity remain constant, the underlying industry mix of regionsmight change over time. And finally, what are the appropriate temporal andspatial scales to examine the evolution of regions? Diversity might be usefulfor the long run at the expense of short-run economic growth. Regionaleconomic specialization might yield high levels of efficiency that benefitactors at the regional and national scales and that maintain industrialdiversity at the national scale.

Frenken et al. (2005) describe new geographical economics and portfolioapproaches as static because variety at a single point in time relates toregional growth. Boschma and Lambooy (1999) argue that urbanizationeconomies and Jacobs externalities are more important during the emer-gence of new industries and technological paradigms when industries havenot yet generated their specific skills, or supplier and institutional require-ments, but that localization economies might become more important oncethese factors are created (see also Boschma and Frenken, 2003). Thisliterature is important as it brings a dynamic perspective to the literatureand demonstrates how industrial and technological diversity influenceregional growth at various stages of industry life cycles, but regions aresomehow considered as containers in which industrial evolution unfolds.Instead of following a single industry (or cluster) through time and space,it is possible to start with regions and examine the relationship between the

212 Spatial systems

distribution of characteristics within regions and the changes in regionalaggregates such as growth, productivity, profitability and stability. Thisavenue is pursued in the following empirical analysis: regions are concep-tualized as an assembly of industries and the distribution of employmentamong these industries (indicating regional industrial diversity/concen-tration) is expected to exert an influence on changes in regional economicgrowth and stability.

3. EMPIRICAL ANALYSIS

The empirical part of this chapter is based on employment data from theUS county business patterns (1975–2002). The goal of the analysis is theestablishment of a negative statistical relationship between economicgrowth and stability and a positive relationship between industrial diversityand economic stability at the level of the economic areas of the Bureau ofEconomic Analysis (BEA) through the application of spatial econometrictechniques. In order to measure the impact of industrial diversity onregional volatility, it is necessary to control for the influence of additionalexplanatory variables. The choice of variables is informed by the theoreticaldiscussion and empirical work in regional science (for example, Baldwin andBrown, 2004). Regional stability/volatility is measured as the variance ofannual regional employment growth rates. According to Baldwin andBrown, the variance will be influenced by the diversity of a region’s indus-trial structure, the variance of its industries’ growth rates and the covariancebetween those growth rates. The correlates of volatility are chosen from aset of structural characteristics of regions. The names, definitions andexpected signs of these variables are summarized in Table 10.1.

The simplest and most widely used measure of diversity is probably theHerfindahl index (Duranton and Puga, 2000; Chandra, 2003; Baldwin and

Diversity, stability and regional growth in the United States 213

Table 10.1 Correlates of volatility

Variable Variable description Hypothesized name sign

HERF75 Herfindahl measure of diversity/specialization (1975) �GROWTH Average annual compound rate of growth �SIZE75 Total employment in 1975 � /�PLSIZE75 Average plant size (total employment per plant 1975) �R75 Percent of employment in resource based industries �

(SIC 12, 13, 14, 21, 24, 29)

Brown, 2004). Experimentation with entropy measures of diversity did notchange the conclusions of the chapter. The Herfindahl index H of a BEAregion r is measured as

(10.1)

where and refers to employment in sector i in BEA regionr. The index varies between 1 (all employment is concentrated in one sector)and 1/n (employment is distributed equally among all sectors). A highervalue indicates greater concentration of employment in fewer sectors(lower diversity), while a lower value indicates a more even distribution ofemployment across sectors (higher diversity). The index has been con-structed for the base year (1975) using the 1972 SIC3 3-digit system. Basedon the theoretical discussion on diversity and volatility, a positive relation-ship between the level of concentration (a high Herfindahl index) andvolatility (high variance of annual growth rates) is expected.

Employment levels and growth rates are also expected to relate tovolatility. Although it is expected that total regional employment is cor-related with diversity and hence, the effect of employment size subsumedby the effect of diversity on volatility, Malizia and Ke (1993) argue thatlarger regions are more stable than smaller regions and that size (meas-ured as total employment) might have a positive effect on regional stab-ility independent of the effect of diversity. The impact of size on volatilitywill also depend on the geographic concentration of markets for products.If firms in larger regions sell a larger share of their product in localmarkets, then the growth rates of a region’s industries are more likely tobe correlated because they will be dependent on the same market andsubject to the same economic shocks. In this case, larger regions will bemore volatile than smaller regions even if they are characterized bysimilar levels of industrial diversity. The relationship between size andvolatility is therefore ambiguous.

Malizia and Ke found a U-shaped relationship between growth andvolatility, suggesting that regions that have concentrations in fast-growingindustries have higher growth rates, while regions with concentrations infast-declining industries have lower growth rates. On the other hand, morediverse regions are characterized by more stability and average growthrates. The U-shaped relationship between growth and volatility has beenconfirmed by Baldwin and Brown (2004) for Canadian census regions andmanufacturing industries. Contrary to the U-shaped relationship detectedby Malizia and Ke and Baldwin and Brown, a linear (positive) relationshipbetween growth and volatility for BEA regions was discovered (see Figure

Eirsir � Eir �� iEir

Hr � �i

s2ir

214 Spatial systems

10.2, below). The squared growth rate was thus omitted from the set ofindependent variables.

The average plant size in a region is expected to exert a positive influenceon regional economic stability because smaller plants are generally newerand more likely to exit the industry than larger plants (Davis et al., 1996;Baldwin et al., 1998). Furthermore, large firms tend to produce a variety ofcommodities and are thus less vulnerable to market fluctuations affecting aparticular product. Product variety tends to be lower in smaller firms whichwill decrease their ability to adapt to market fluctuations affecting a specificproduct. Hence, we would expect a negative relationship between averageplant size and volatility. The negative correlation coefficient between averageplant size and stability confirms this relationship (see Table 10.4, below).

Baldwin and Brown (2004) add export shares and the share of employ-ment in different types of industries (for example, resource based) asexplanatory variables. Unfortunately, export data are not available for BEAregions but the share of regional employment in resourced-based industrieswas included. It can be expected that regions with a high share of resource-based industries such as mining, agriculture and forestry, logging, lumberand petroleum, are characterized by higher volatility in growth rates,because resource-based economies are often influenced by global (that is,exogenous) price fluctuations. A positive relationship between the share ofresource-based industries and volatility is expected.

The data used in this analysis are based on county business patterns from1975 to 2002. The data have been aggregated to the level of BEA regionsbecause they are probably closest to functional economic regions in the US(similar to labor market regions or travel-to-work areas in Europe)(Johnson and Kort, 2004). County business patterns provide employment,establishment and wage data for SIC 4-digit industries between 1975 and1986. For many counties, actual figures have been suppressed and replacedby employment size classes. Because of the large amount of undisclosedinformation (in particular for smaller counties) and in order to reducemeasurement error, county employment at the SIC 4-digit level has beenaggregated to SIC 3-digit employment and the diversity measures have beencalculated at the SIC 3-digit level. If information for individual countieswas not reported at the SIC 3-digit level, the average value of the employ-ment size class (for example, 10 for employment size class 0–19) was usedto impute the missing information. Because counties were then aggregatedto the new BEA regions, measurement error will be relatively small and isunlikely to influence the results. Changes in county and BEA definitionshave been considered in order to keep the geography constant over thewhole period. Overall, the dataset spans 27 years and includes the 177 BEAareas of the continental United States.

Diversity, stability and regional growth in the United States 215

Figure 10.1 maps the key variables volatility, diversity and growth for the177 BEA areas. Growth and volatility are measured over the wholeperiod from 1975 to 2002 while diversity is measured for the base year, 1975.Table 10.2 lists the top 10 and bottom 10 cities with respect to stability, diver-sity and growth. Figure 10.1 and Table 10.2 reveal the following geographicpattern. The regions characterized by the highest levels of stability tend to

216 Spatial systems

Source: County Business Patterns, 1975–2002.

Figure 10.1 Volatility (a), diversity (b), growth (c)

Key

12345

(1 = lowest; 5 = highest)

(a)

(b)

(c)

217

Tab

le 1

0.2

BE

A a

rea

rank

ings

by

stab

ility

,div

ersi

ty a

nd g

row

th

Ran

kB

EA

are

aSt

abili

tyB

EA

are

aD

iver

sity

BE

A a

rea

Gro

wth

1P

hila

delp

hia,

PA-N

J-0.

0003

942

Lit

tle

Roc

k,A

R0.

0103

007

Las

Veg

as,N

V6.

87D

E-M

D2

Lin

coln

,NE

0.00

0452

3H

arri

sbur

g,PA

0.01

0435

7A

usti

n,T

X5.

983

New

Yor

k,N

Y-N

J-C

T-

0.00

0458

9Ja

ckso

n,M

S0.

0105

806

Sara

sota

,FL

5.42

PA4

Alb

any,

NY

0.00

0502

6N

ew Y

ork,

NY

-NJ-

CT

-0.

0107

762

Pho

enix

,AZ

5.16

PA5

Dov

er,D

E0.

0005

044

Mem

phis

,TN

-MS-

AR

0.01

0803

3F

ayet

tevi

lle,A

R-M

O5.

056

Ric

hmon

d,V

A0.

0005

127

Atl

anta

,GA

-AL

0.01

0846

7B

end,

OR

5.01

7H

arri

sbur

g,PA

0.00

0566

Phi

lade

lphi

a,PA

-NJ-

0.01

1209

9F

lags

taff

,AZ

4.94

DE

-MD

8Sc

rant

on,P

A0.

0005

689

Dal

las-

For

t W

orth

,TX

0.01

1443

7Sa

cram

ento

,CA

-NV

4.77

9O

mah

a,N

E-I

A0.

0005

821

Chi

cago

,IL

-IN

-WI

0.01

1445

9C

olor

ado

Spri

ngs,

CO

4.72

10R

oche

ster

,NY

0.00

0606

4C

olum

bia,

SC0.

0115

646

Orl

ando

,FL

4.70

168

San

Ang

elo,

TX

0.00

2358

3M

orga

ntow

n,W

V0.

0344

388

Stat

e C

olle

ge,P

A1.

0616

9P

endl

eton

,OR

0.00

2435

2B

end,

OR

0.03

4525

8W

ater

loo,

IA1.

0217

0E

ugen

e,O

R0.

0024

807

Port

Art

hur,

TX

0.03

5662

Spri

ngfi

eld,

IL0.

9417

1Sa

raso

ta,F

L0.

0025

731

Pue

blo,

CO

0.03

7247

6E

rie,

PA0.

9017

2F

arm

ingt

on,N

M0.

0028

575

Ren

o,N

V0.

0376

502

Buff

alo,

NY

0.89

173

Ren

o,N

V0.

0031

622

Ode

ssa,

TX

0.04

1524

1O

dess

a,T

X0.

8917

4O

dess

a,T

X0.

0034

569

Fla

gsta

ff,A

Z0.

0416

221

Dav

enpo

rt,I

A-I

L0.

8617

5L

afay

ette

,LA

0.00

3963

6W

ater

loo,

IA0.

0693

846

Cle

vela

nd,O

H0.

8017

6C

aspe

r,W

Y0.

0039

645

Gul

fpor

t,M

S0.

0725

675

Pit

tsbu

rgh,

PA0.

8017

7B

end,

OR

0.00

4271

7L

as V

egas

,NV

0.07

4008

7Po

rt A

rthu

r,T

X0.

76

Sou

rce:

Cou

nty

Bus

ines

s P

atte

rns,

1975

–200

2.

be concentrated in the Mid-Atlantic region including areas such asPhiladelphia, New York and Rochester, while among the most volatile aremany of the resource-based economies of Oregon, Texas and Wyoming. Themost diverse regions are those surrounding large urban areas such as NewYork, Philadelphia, Dallas, Chicago, Atlanta or Memphis, while the morespecialized areas tend to be either resource-based economies or those focus-ing on tourism such as Las Vegas or Reno. The fastest-growing regions arelocated in Florida and the southwest of the country and include retirementareas and high-tech centers. The slowest-growing regions are found in theold manufacturing heartland and include regions such as Buffalo, Clevelandand Pittsburgh. Although some regions score high/low on several variables,others do not follow a clear pattern. Figure 10.2 plots the relationshipbetween volatility on the vertical axis and growth/diversity on the horizon-tal axes. Both relationships are positive and significant at the 0.0001 level.

Descriptive statistics of all dependent and independent variables as wellas the logarithmic values are presented in Table 10.3, while the raw corre-lation coefficients are presented in Table 10.4. Table 10.3 highlights consid-erable variation in diversity, growth and stability, although the variation isconsiderably smaller than for Canadian Census regions (Baldwin andBrown, 2004). The correlation coefficients reveal that most of the indepen-dent variables are correlated with stability. Table 10.4 indicates thatspecialized regions and those characterized by higher rates of economicgrowth, smaller employment size and smaller average plant size tend to bemore stable (that is, have lower variances of growth rates). Also of interestis the positive relationship between growth and diversity: specializedregions appear to grow more rapidly but are also characterized by highervolatility. Table 10.3 suggests the presence of extreme outliers with respect

218 Spatial systems

Source: County Business Patterns, 1975–2002.

Figure 10.2 Relationship between volatility and growth/diversity

0

0.0005

0.001

0.0015

0.002

0.0025

0.003

0.0035

0.004

0.0045

0 0.02 0.04 0.06 0.08

Growth

Volatility

Corr = 0.31 (p = 0.0001)

0

0.0005

0.001

0.0015

0.002

0.0025

0.003

0.0035

0.004

0.0045

0 0.02 0.04 0.06 0.08

Diversity

Volatility

Corr = 0.44 (p = 0.0001)

to the dependent and independent variables that might drive overall results.Hence, the natural logarithms of the variables are taken to remove the effectof those outliers. Table 10.4 suggests that the linear relationships betweenthe logged variables tend to become stronger.

The basic regression model estimated may be written as:

(10.2)

where y is an N�1 vector of observations on the dependent variable, X is anN�K matrix of observations on K independent variables, is a K�1 vectorof regression coefficients, and � is an N�1 vector of errors assumed to benormally and independently distributed. As discussed above, the dependentvariable is the variance of regional growth rates (VARGROWTH) and theindependent variables are HERF75, GROWTH, EMP75, PLSIZE75 andR75. Figure 10.1 suggests considerable spatial autocorrelation in the depen-dent variable. Employing spatial contiguity weights, a Moran’s I value of0.3348 (significant at the 0.001 level) suggests the presence of strong spatialautocorrelation of the independent variable. In the presence of spatial auto-correlation, ordinary least squares (OLS) estimates may be inconsistent(Anselin, 1988; Anselin and Rey, 1991).

Spatial dependence is of two basic forms, error dependence and lagdependence. In the spatial error model the errors can no longer be assumedindependent and identically distributed and the regression model takes thefollowing form:

(10.3)y � X � �W� � �,

y � X � �,

Diversity, stability and regional growth in the United States 219

Table 10.3 Basic statistics of dependent and independent variables

Variables n Mean Std dev. Minimum Maximum

VARGROWTH 177 12.56 6.38 3.94 42.72HERF75 177 0.02 0.01 0.01 0.07MGROWTH 177 2.53 1.09 0.76 6.87EMP75 177 336866.00 680227.00 11288.00 6561322.00PLSIZE75 177 12.71 2.89 7.56 19.80R75 177 0.05 0.05 0.01 0.29

LOGVARGROWTH 177 2.43 0.43 1.37 3.75LOGHERF75 177 �4.01 0.35 �4.58 �2.60LOGMGROWTH 177 0.84 0.44 �0.27 1.93LOGEMP75 177 11.91 1.19 9.33 15.70LOGPLSIZE75 177 2.52 0.23 2.02 2.99LOGR75 177 �3.36 0.88 �5.16 �1.25

Source: County Business Patterns, 1975–2002.

where is the spatial autoregression coefficient, W is an N�N matrix ofspatial weights representing the geography of the observational units(BEA’s), and � is an N�1 vector of errors assumed to possess the usualproperties. In this form, spatial dependence influences the error term onlyand it has been shown to influence the power of tests for heteroscedasticityand the structural stability of regression coefficients. In the spatial lagmodel, the standard regression equation may be rewritten as:

(10.4)

where � is the spatial autoregression coefficient. In this form, the value ofthe dependent variable at a particular location is jointly determined by its

y � �Wy � X � �,

�

220 Spatial systems

Table 10.4 Correlation coefficients between dependent and independentvariables (p-values in parentheses)

VARGROWTH HERF75 MGROWTH EMP75 PLSIZE75 R75

VARGROWTH 1 0.44 0.31 �0.21 �0.39 0.60(0.0001) (0.0001) (0.0042) (0.0001) (0.0001)

HERF75 0.44 1 0.09 �0.22 �0.13 0.30(0.0001 (0.246) (0.003) (0.0796) (0.0001)

MGROWTH 0.31 0.09 1 �0.15 �0.30 �0.03(0.0001) (0.246) (0.0514) (0.0001) (0.7212)

EMP75 �0.21 �0.22 �0.15 1 0.44 �0.23(0.0042) (0.003) (0.0514) (0.0001) (0.0018)

PLSIZE75 �0.39 �0.13 �0.30 0.44 1 �0.37(0.0001) (0.0796) (0.0001) (0.0001) (0.0001)

R75 0.60 0.30 �0.03 �0.23 �0.37 1(0.0001) (0.0001) (0.7212) (0.0018) (0.0001)

LOG LOG LOG LOG LOG LOGVARGROWTH HERF75 MGROWTH EMP75 PLSIZE75 R75

LOG 1 0.50 0.29 �0.42 �0.42 0.48VARGROWTH (0.0001) (0.0001) (0.0001) (0.0001) (0.0001)LOGHERF75 0.50 1 �0.04 �0.49 �0.25 0.29

(0.0001) (0.6416) (0.0001) 0.0007 (0.0001)LOG 0.29 �0.04 1 �0.16 �0.30 0.07MGROWTH (0.0001) (0.6416) (0.0343) (0.0001) (0.3718)LOGEMP75 �0.42 �0.49 �0.16 1 0.78 �0.48

(0.0001) (0.0001) (0.0343) (0.0001) (0.0001)LOGPLSIZE75 �0.42 �0.25 �0.30 0.78 1 �0.45

(0.0001) (0.0007) (0.0001) (0.0001) (0.0001)LOGR75 0.48 0.29 0.07 �0.48 �0.45 1

(0.0001) (0.0001) (0.3718) (0.0001) (0.0001)

values at other locations and OLS estimation is no longer consistent(Anselin and Rey, 1991).

The results of the linear model are presented under Model 1 in Table 10.5.Specialized and faster-growing regions and those with higher shares ofresource-based industries are characterized by more volatility in growthrates. All relationships are significant at the 0.01 level. The average plant size

Diversity, stability and regional growth in the United States 221

Table 10.5 Determinants of volatility

Variables Model 1 Model 2 Model 3 Model 4 OLS OLS ML spatial ML spatial

lag errorForm Linear Log-linear Log-linear Log-linearDependent LN LN LNvariable � VARGROWTH VARGROWTH VARGROWTH VARGROWTH

Constant 4.87* 5.27** 3.87** 5.14**(2.05) (14.03) (8.41) (12.76)

(LN)HERF75 182.09** 0.53** 0.44** 0.48**(4.70) (6.39) (8.41) (5.85)

(LN)MGROWTH 1.60** 0.24** 0.16** 0.18**(4.90) (4.15) (2.89) (2.93)

(LN)EMP75 4.54�10e7 0.05 0.02 0.04(0.81) (1.25) (0.44) (1.01)

(LN)PLSIZE75 �0.25� �0.38* �0.21 �0.38�

(�1.72) (�2.06) (�1.19) (�1.89)(LN)R75 58.94** 0.15** 0.11** 0.12**

(8.30) (4.75) (3.52) (3.57)W_lnvargrowth 0.37**

(4.67)(Lambda) 0.43**

(4.83)

R-square (adj.) 0.515 0.462 0.539p 0.537p

Log-likelihood �512.161 �47.8736 �36.8477 �36.8319AIC 1036.32 107.747 87.6953 85.6638SC 1055.38 126.804 109.928 104.720

Diagnostics for heteroscedasticityBreusch–Pagan 51.22** 2.58425 8.57 8.97Koenker–Basset 14.79* 2.35506 – –

Diagnostics for spatial dependenceLM-ERROR 5.85* 21.76** – –LM-LAG 5.51* 27.38** – –

Notes: OLS�ordinary least squares; ML�maximum likelihood; **, *, �: significant atthe 0.01, 0.05, 0.1 levels; spatial weights are based on queen spatial contiguity of BEA areas;p indicates a pseudo R-square measure, because the standard R-square is invalid in MLestimation. AIC�Akaike information criterion; SC�Schwartz criterion.

is negatively related to volatility, supporting the theoretical arguments dis-cussed above. The relationship is significant at the 0.1 level only. The size ofthe region is positively related to volatility, lending some support to Fujitaet al.’s (1999) argument that the demand for products in large regions ismore likely to be correlated, exacerbating demand shocks. However, Table10.4 revealed high correlations between regional size and most other inde-pendent variables, suggesting that the size effects might have been picked upby other variables (for example, HERF75). Furthermore, Table 10.5 showsthat the positive relationship between EMP75 and VARGROWTH is notstatistically significant. In addition to the parameter estimates and t-values,a set of diagnostic statistics have been added. An adjusted R-square valueof 0.515 indicates a relatively good fit of the original model. However, testson heteroscedasticity and spatial dependence reveal that both are present inthe model. Heteroscedasticity could have been the result of correlatedspatial errors. However, even after correcting for spatial lags/errors, hetero-scedasticity posed a problem. The scatterplots depicted in Figure 10.2seemed to suggest that the variance of volatility increases with higher ratesof both growth and diversity, and that no single variable could be easilyidentified to cause heteroscedasticity. Gujarati (2003) suggests that the logtransformation of variables often helps to eliminate heteroscedasticity.

Model versions 2–4 present the results for the log-linear models. The factthat parameter estimates can be interpreted as elasticities is an addedadvantage of the log-linear model. Model 2 presents the results for the OLSestimates without correction for spatial dependence. The signs of the par-ameter estimates for the log-linear version do not change, although theparameter estimate for average plant size is now significant at the 0.05 leveland the adjusted R-square value indicates a moderately worse model fit. Onthe other hand, the Breusch–Pagan and Koenker–Bassett tests reveal noheteroscedasticity, while spatial dependence is still present in the models.Lagrange multiplier tests suggest the presence of both spatial lag andspatial error. Model 3 provides the results for the spatial lag model. Themodel results are based on maximum likelihood estimation and the valuesin parentheses are z- rather than t-values. The signs of the parameter esti-mates are consistent with those of Models 1 and 2. The negative estimatefor the size of plants is no longer significant, but all the goodness-of-fitmeasures indicate a clear improvement from model version 2. Furthermore,the parameter estimate for the spatially lagged dependent variable is posi-tive and significant at the 0.01 level. The results suggest that volatility ofgrowth in a region is also influenced by volatility of growth in the neigh-boring regions. The results for the spatial error model (Model 4) are similarto the results of the spatial lag model. The estimate for average plant size issignificant again at the 0.1 level and the goodness-of-fit statistics suggest an

222 Spatial systems

almost identical performance when compared to the spatial lag model. Thespatial lag parameter, �, is also positive and significant at the 0.01 level.

Anselin (1988, 1992) advises using performance indicators such as thelog-likelihood, Akaike information or Schwartz criteria to inform modelselection. Because the performance indicators for both models are almostidentical, no obvious choice for the ‘best’ model emerges. Because theparameter estimates are very similar and since the purpose of the model isnot predication but the establishment of statistical relationships betweenvariables, this does not pose a major predicament. Independent of the exactmodel specification, a comparison of the elasticities suggests that a changein diversity has the highest impact on change in regional economic stabil-ity, followed by a change in average growth rate and the change in share ofresource-based industries (keeping in mind that the impact of average plantsize is barely significant). The results confirm the work by other researchersand highlight the importance of industrial diversity for regional economicstability.

4. CONCLUSION

This chapter employed evolutionary theory to develop arguments on thetrade-off between short-term adaptation and long-term adaptability. Atpresent, little thought is given to the potentially negative impacts of cluster-based regional policies predicated on the spatial concentration of func-tionally integrated sectors. The concentration of economic activity in a feweconomic areas is likely to boost short-term productivity growth and profitrates through the exploitation of externalities based on the local skill base,knowledge spillovers, and traded and untraded interdependencies. Thenegative side of specialization is a decline in adaptive flexibility, the abilityto react to continually changing economic environments.

The chapter examined the relationship among stability, growth and diver-sity for 177 BEA areas over the 1975–2002, period using employment datafrom county business patterns. The analysis revealed a strong positive re-lationship between diversity and stability on the one hand, and growth andinstability, on the other. While these results confirm work in other countriesand provide some credibility to evolutionary theories of regional economicchange, it is important not to overstate the results. Although regional eco-nomic stability is desirable because it facilitates planning for technical andsocial infrastructure and avoids the pitfalls of fast growth (congestion, risinghouse prices, environmental degradation, overinvestment in infrastructure),stability (small variances in growth rates) coupled with economic decline isproblematic if the decline is rapid and investments have to be written off

Diversity, stability and regional growth in the United States 223

rapidly. In other words, from a policy point of view, it still has to be workedout what kind of regional stability is desirable, bearing in mind the trade-offbetween growth and stability. Furthermore, the Herfindahl index provides ageneral measure of industrial diversity, but does not capture the degree offunctional relation between sectors. Regions can contain a large number ofdifferent but functionally integrated economic sectors that react in a similarfashion to demand shocks. The impact of diversity on stability will beinfluenced strongly by the degree of functional integration of sectors andhence, future work will have to pay more attention to this aspect of diversity(Frenken et al., 2005). However, based on insights from evolutionary theoryand the empirical results, industrial, institutional, skill, technological andsocial diversity should be elevated to a general principle of regional eco-nomic development even at the cost of short-term welfare losses. This isimperative if we drop the assumptions of global optimality and equilibrium.

NOTES

* This research was partially funded by an Annual Grant of the University of Southampton(A2001/19). I would also like to thank Koen Frenken for valuable comments on an earlierdraft of this chapter. The usual disclaimer applies.

1. Volatility is interpreted as the opposite of stability and will be measured as the variancein annual rates of employment change.

2. Ecology offers three different notions of efficiency: ‘(1) “first law” efficiency, or simply thefraction of energy input that appears as output; (2) efficiency in the “second law” sensewhere resources are being used more thoroughly by a diverse set of agents, having differentsingle-use efficiencies (the most efficient agent is the one that effects the most complete useof the available resource, regardless of the rate of use); (3) efficiency in the “maximumpower” sense, where an agent uses a resource to provide either the quickest return or thegreatest rate of output’ (Matutinovic, 2002: 433). Economics prioritizes definition (3).From a ‘maximum power’ efficiency perspective, less-efficient firms are considered redun-dant. ‘Second law’ efficiency is generally absent from economic policy discourse althoughit might be desirable from an equity point of view.

3. Standard Industrial Classification.

REFERENCES

Alchian, A. (1950), ‘Uncertainty, evolution and economic theory’, Journal ofPolitical Economy, 58: 211–22.

Andersen, E.S. (2004), ‘Population thinking and evolutionary economic analysis:exploring Marshall’s Fable of Trees’, Danish Research Unit for IndustrialDynamics (DRUID) Working Paper, 04-05.

Anselin, L. (1988), Spatial Econometrics, Dordrecht: Kluwer Academic.Anselin, L. (1992), SpaceStat tutorial: a workbook for using SpaceStat in the analy-

sis of spatial data, Santa Barbara: NCGIA National Center for GeographicInformation and Analysis.

224 Spatial systems

Anselin, L. and S. Rey (1991), ‘The performance of tests for spatial dependence inlinear regression models’, Geographical Analysis, 23: 110–31.

Arthur, B. (1994), Increasing Returns and Path Dependence in the Economy, AnnArbor, MI: University of Michigan Press.

Attaran, M. (1986), ‘Industry diversity and economic performance in U.S. areas’,Annals of Regional Science, 20: 44–54.

Baldwin, J., T. Dunne and J. Haltiwanger (1998), ‘A comparison of job creation andjob destruction in Canada and the United States’, The Review of Economics andStatistics, 80: 347–56.

Baldwin, J.R. and W.M. Brown (2004), ‘Regional manufacturing employmentvolatility in Canada: the effects of specialization and trade’, Papers in RegionalScience, 83: 519–41.

Baldwin, R.E., R. Forslid, P. Martin, G.I.P. Ottaviano and F. Robert-Nicoud(2003), Economic Geography and Public Policy, Princeton, NJ: PrincetonUniversity Press.

Bartik, T. (1996), ‘Strategies for economic development’, in R. Aronson and E.Schwartz (eds), Management Policies in Local Government Finance, Washington,DC: International City/County Management Association for the ICMAUniversity, pp. 287–311.

Begg, I. (ed.) (2002), Urban Competitiveness, Bristol: Policy Press.Berman, E. and L. Bui (2001), ‘Environmental regulation and productivity: evi-

dence from oil refineries’, Review of Economics and Statistics, 83: 498–510.Black, D. and V. Henderson (2003), ‘Urban evolution in the US’, Journal of

Economic Geography, 3: 343–72.Boschma, R.A. (2004), ‘Competitiveness of regions from an evolutionary perspec-

tive’, Regional Studies, 38: 1001–14.Boschma, R.A. and K. Frenken (2003), ‘Evolutionary economics and industry

location’, Review of Regional Research, 23: 183–200.Boschma, R.A. and J.G. Lambooy (1999), ‘Evolutionary economics and economic

geography’, Journal of Evolutionary Economics, 9: 411–29.Boulding, K. (1981), Evolutionary Economics, Beverly Hills, CA: Sage.Brakman, S., H. Garretsen and C. van Marrewijk (2001), An Introduction to

Geographical Economics, Cambridge: Cambridge University Press.Brenner, N. (1998), ‘Global cities, glocal states: global city formation and state ter-

ritorial restructuring in contemporary Europe’, Review of International PoliticalEconomy, 5: 1–37.

Brenner, N. (2004), New State Spaces, Oxford: Oxford University Press.Bromley, D. (1990), ‘The ideology of efficiency: searching for a theory of policy

analyses’, Journal of Environmental Economics and Management, 19: 86–107.Chandra, S. (2003), ‘Regional economy size and the growth-instability frontier: evi-

dence from Europe’, Journal of Regional Science, 43: 95–122.Conroy, M.E. (1974), ‘Alternative strategies of regional industrial diversification’,

Journal of Regional Science, 14: 31–46.Conroy, M.D. (1975), ‘The concept and measurement of regional industrial

diversification’, Southern Economic Journal, 41: 492–505.Davis, S.J., J.C. Haltiwanger and S. Schuh (1996), Job Creation and Destruction,

Cambridge, MA and London: MIT Press.Depew, D. and B. Weber (1995), Darwinism Evolving, Cambridge, MA: MIT Press.Dissart, J.C. (2003), ‘Regional economic diversity and regional economic stability:

research results and agenda’, International Regional Science Review, 26: 423–46.

Diversity, stability and regional growth in the United States 225

Dosi, G. and R. Nelson (1994), ‘An introduction to evolutionary theories in eco-nomics’, Journal of Evolutionary Economics, 4: 153–72.

Duranton, G. and D. Puga (2000), ‘Diversity and specialization in cities: why, whereand when does it matter?’, Urban Studies, 37: 533–55.

Duranton, G. and D. Puga (2001), ‘Nursery cities: urban diversity, process inno-vation and the life-cycle of products’, American Economic Review, 91: 1454–77.

Duranton, G. and D. Puga (2004), ‘Micro-foundations of urban agglomerationeconomies’, in V. Henderson and J.-F. Thisse (eds), Handbook of Regional andUrban Economics, vol. 4. Amsterdam: North-Holland, pp. 2063–117.

Edquist, C. (1997), Systems of Innovation: Technologies, Institutions andOrganizations, London and Washington, DC: Pinter.

Essletzbichler, J. and D.L. Rigby (2005a), ‘Technological evolution as creativedestruction of process heterogeneity: evidence from US plant level data’,Economic Systems Research, 17: 25–45.

Essletzbichler, J. and D.L. Rigby (2005b), ‘Competition, variety and the geographyof technology evolution’, Tijdschrift voor Economische en Sociale Geografie, 95:48–62.

Feldman, M. and D. Audretsch (1999), ‘Innovation in cities: science-based diver-sity, specialization and localized competition’, European Economic Review, 43:409–29.

Frenken, K., F.G. van Oort, T. Verburg and R. Boschma (2005), ‘Variety andregional economic growth in the Netherlands’, Papers in Evolutionary EconomicGeography no. 05.02. Utrecht University, Urban and Regional research centerUtrecht.

Friedman, M. (1953), Essays in Positive Economics, Chicago: University of ChicagoPress.

Fujita, M., P. Krugman and A.J. Venables (1999), The Spatial Economy: Cities,Regions and International Trade, Cambridge, MA: MIT Press.

Giampietro, M. and K. Mayumi (1997), ‘A dynamic model of socioeconomicssystems based on hierarchy theory and its application to sustainability’,Structural Change and Economic Dynamics, 8: 453–69.

Glaeser, E., H. Kallal, J. Schenkman and A. Shleifer (1992), ‘Growth in cities’,Journal of Political Economy, 100: 1126–52.

Gould, S.J. and R.C. Lewontin (1979), ‘The spandrels of San Marco and thePanglossian paradigm: a critique of the adaptionist programme’, Proceedings ofthe Royal Society of London series, 205(1161): 581–98.

Gowdy, J.M. (1992), ‘Higher selection processes in evolutionary economic change’,Journal of Evolutionary Economics, 2: 1–16.

Grabher, G. (1993), ‘The weakness of strong ties: the lock-in of regional develop-ment in the Ruhr area’, In G. Grabher (ed.), The Embedded Firm, New York andLondon: Routledge, pp. 255–77.

Grabher, G. and D. Stark (1997), ‘Organizing diversity: evolutionary theory,network analysis and postsocialism’, Regional Studies, 31: 533–44.

Granovetter, M. (1973), ‘The strengths of weak ties’, American Journal ofSociology, 78: 1360–80.

Gujarati, D.N. (2003), Basic Econometrics, 4th ed. New York: McGraw-Hill.Harvey, D. (1989), ‘From managerialism to entrepreneurialism: the transformation

in urban governance in late capitalism’, Geografiska Annaler, 71B(1): 2–17.Henderson, J.V. (1988), Urban Development: Theory, Fact and Illusion, Oxford:

Oxford University Press.

226 Spatial systems

Henderson, J.V. (1997), ‘Externalities and industrial development’, Journal ofUrban Economics, 42: 449–70.

Henderson, J.V. (2003), ‘Marshall’s scale economies’, Journal of Urban Economics,53: 1–28.

Henderson, J.V., A. Kuncoro and M. Turner (1995), ‘Industrial development incities’, Journal of Political Economy, 103: 1067–85.

Hodgson, G. (1993), Economics and Evolution, Cambridge: Polity Press.Hodgson, G. (1997), ‘Notes on reductionism and emergence’, Available on request

from the author.Holling, C.S. (1973), ‘Resilience and stability of ecological systems’, Annual Review

of Ecology and Systematics, 4: 1–23.Holling, C.S. (2001), ‘Understanding the complexity of economic, ecological, and

social systems’, Ecosystems, 4: 390–405.Holling, C.S. (2004), ‘From complex regions to complex worlds’, Ecology and

Society, 9(11), available at www.ecologyandsociety.org/vol9/iss 1/art11/Hubbard, P. and T. Hall (1998), ‘The entrepreneurial city and the “new urban

politics” ’, in T. Hall and P. Hubbard (eds), The Entrepreneurial City:Geographies of Politics, Regime and Representation, New York: John Wiley,pp. 1–23.

Hudson, R. (1999), ‘The learning economy, the learning firm and the learningregion: a sympathetic critique to learning’, European Urban and Regional Studies,6(1): 59–72.

Iwai, K. (1984a), ‘Schumpeterian dynamics II: technological progress, firm growthand “economic selection”’ Journal of Economic Behavior and Organization, 5:321–51.

Iwai, K. (1984b), Schumpeterian dynamics: an evolutionary model of innovationand imitation’, Journal of Economic Behavior and Organization, 5: 577–98.

Jacobs, J. (1969), The Economy of Cities, New York: Vintage.Jessop, B. (1990), State Theory, University Park, PA: Pennsylvania State University

Press.Jessop, B. (1994), ‘Post-Fordism and the state’, in A. Amin (ed.), Post-Fordism: A

Reader, Oxford: Blackwell, pp. 251–79.Johnson, K.P. and J.R. Kort (2004), ‘2003 redefinition of the BEA economic areas’,

Survey of Current Business, November: 68–75.Jovanovic, B. (1982), ‘Selection and the evolution of industry’, Econometrica, 50(3):

649–70.Kitson, M., R. Martin, and P. Tyler (2004), ‘Regional competitiveness: an elusive

yet key concept?’, Regional Studies, 38: 991–9.Kort, J.R. (1981), ‘Regional economic instability and industrial diversification in

the U.S.’, Land Economics, 57: 596–608.Krantz, R. (2002), ‘Employment in business services: a year of unprecedented

decline’, Monthly Labor Review, April: 17–24.Krugman, P. (1991), ‘Increasing returns and economic geography’, Journal of

Political Economy, 99: 483–99.Krugman, P. (1994), ‘Competitiveness: a dangerous obsession’, Foreign Affairs,

March/April: 1–17.Leitner, H. and E. Sheppard (1998), ‘Economic uncertainty, inter-urban compe-

tition and the efficacy of entrepreneurialism’, in T. Hall and P. Hubbard (eds),The Entrepreneurial City: Geographies of Politics, Regime and Representation,New York: John Wiley, 285–307.

Diversity, stability and regional growth in the United States 227

Levins, R. and R. Lewontin (1985), The Dialectical Biologist, Cambridge, MA:Harvard University Press.

Lovering, J. (1999), ‘Theory led by policy: the inadequacies of the New Regionalism(illustrated from the case of Wales)’, International Journal of Urban and RegionalResearch, 23: 379–95.

Lundvall, B.-A. and B. Johnson (1994), ‘The learning economy’, Journal of IndustryStudies, 1: 23–42.

Malizia, E.E. and S. Ke (1993), ‘The influence of economic diversity on unemploy-ment and stability’, Journal of Regional Science, 33: 221–35.

Marshall, A. (1920), The Principles of Economics, London: Macmillan.Martin, R. (1999), ‘The new “geographical turn” in economics: some critical

reflections’, Cambridge Journal of Economics, 23, 63–91.Martin, R. and P. Sunley (2003), ‘Deconstructing clusters: chaotic concept or policy

panacea?’, Journal of Economic Geography, 3: 5–35.Matutinovic, I. (2001), ‘The aspects and the role of diversity in socioeconomic

systems: an evolutionary perspective’, Ecological Economics, 39: 239–56.Matutinovic, I. (2002), ‘Organizational patterns of economies: an ecological per-

spective’, Ecological Economics, 40: 421–40.Mayumi, K. and M. Giampietro (2001), ‘The epistemological challenge for modelling

sustainability: exploring the difference in meaning of risk, uncertainty and ignor-ance’,Paperpreparedfor theECHighLevelScientificConference–ESEE‘Frontiers1: Fundamental Issues of Ecological Economics’, Cambridge, UK, 4–7 July.

Metcalfe, S. (1994), ‘Competition, Fisher’s principle and increasing returns in theselection process’, Journal of Evolutionary Economics, 4: 327–46.

Metcalfe, S. (1998), Evolutionary Economics and Creative Destruction, London andNew York: Routledge.

Metcalfe, S. and M. Gibbons (1986), ‘Technological variety and the process of com-petition’, Economie Appliquée, 39: 493–520.

Neary, J.P. (2001), ‘On hypes and hyperbolas: introducing the new economic geog-raphy’, Journal of Economic Literature, 39: 536–61.

Nelson, R.R. (1995), ‘Recent evolutionary theorizing about economic change’,Journal of Economic Literature, 33: 48–90.

Nelson, R. and S. Winter (1982), An Evolutionary Theory of Economic Growth,Cambridge, MA: Harvard University Press.

Oswalt, P. (ed.) (2004), Schrumpfende Städte. Städtischer Wandel im Zeichen vonPostfordismus und Globalisierung, Hatje Canz Verlag.

Porter, M. (1990), The Competitive Advantage of Nations, New York: Free Press.Porter, M. (1998), ‘Clusters and the new economics of competition’, Harvard

Business Review, November–December: 77–98.Quigley, J.M. (1998), ‘Urban diversity and economic growth’, Journal of Economic

Perspectives, 12: 127–38.Rammel, C. and M. Staudinger (2002), ‘Evolution, variability and sustainable

development’, International Journal of Sustainable Development and GlobalEcology, 9: 305.

Rammel, C. and J.C.J.M. van den Bergh (2003), ‘Evolutionary policies for sustain-able development: adaptive flexibility and risk minimizing’, EcologicalEconomics, 47: 121–33.

Rigby, D.L. and J. Essletzbichler (1997), ‘Evolution, process variety and regionaltrajectories of technological change in US manufacturing’, Economic Geography,73: 269–84.

228 Spatial systems

Rigby, D.L. and J. Essletzbichler (2002), ‘Agglomeration economies and productiv-ity differences in US cities’, Journal of Economic Geography, 2: 407–32.

Rigby, D.L. and J. Essletzbichler (2006), ‘Technological variety, technologicalchange and the geography of production techniques’, Journal of EconomicGeography, 6: Forthcoming.

Robert-Nicoud, F. (2005), ‘The structure of simple “New economic Geography”models (or, on identical twins)’, Journal of Economic Geography, 5: 201–34.

Saviotti, P. (1991), ‘The role of variety in economic and technological development’,in P. Saviotti and S. Metcalfe (eds), Evolutionary Theories of Economic andTechnological Change, Chur: Harwood Academic Publishers, pp. 172–208.

Saviotti, P.P. (1996), Technological Evolution, Variety and the Economy,Cheltenham, UK and Brookfield, USA: Edward Elgar.

Saviotti, P.P. and J.S. Metcalfe (1991), ‘Present development and trends in evo-lutionary economics’, in P.P. Saviotti and J.S. Metcalfe (eds), EvolutionaryTheories of Economic and Technological Change, Chur: Harwood AcademicPublishers, pp. 1–30.

Schoening, N.C. and L.E. Sweeney (1992), ‘Proactive industrial developmentstrategies and portfolio analysis’, Review of Regional Studies, 22: 227–38.

Schütz, J. (1999), ‘The value of systemic reasoning’, Ecological Economics, 31: 23–9.Scott, A.J. (1998), Regions and the World Economy, Oxford and New York: Oxford

University Press.Sheppard, E. (2000a), ‘Geography or economics? Conceptions of space, time, inter-

dependence and agency’, in G.L. Clark, M.P. Feldman and M.S. Gertler (eds),The Oxford Handbook of Economic Geography, Oxford: Oxford University Press,pp. 99–119.

Sheppard, E. (2000b), ‘Competition in space and between places’, in E. Sheppardand T. Barnes (eds), A Companion to Economic Geography, Oxford: Blackwell,pp. 169–86.

Siegel, P.B., T.G. Johnson and J. Allwang (1995), ‘A structural decomposition ofregional economic instability: a conceptual framework’, Journal of RegionalScience, 35: 457–70.

Smith, E.D. (1990), ‘Economic stability and economic growth in rural communi-ties: dimensions relevant to local employment creation strategy’, Growth andChange, 21: 3–18.

Ulanowicz, R.E. (1997), Ecology, the Ascendent Perspective, New York: ColumbiaUniversity Press.

Van den Bergh, J.C.J.M. (2004), ‘Evolutionary thinking in environmental econom-ics; retrospect and prospect’, in J. Foster and W. Hölzl (eds), Applied EvolutionaryEconomics and Complex Systems, Cheltenham, UK and Northampton, MA,USA: Edward Elgar, pp. 239–75.

Vrba, E.S. and S.J. Gould (1986), ‘The hierarchical expansion of sorting and selec-tion: sorting and selection cannot be equated’, Paleobiology, 10: 217–28.

Vromen, J.J. (1995), Economic Evolution: An Enquiry into the Foundations of NewInstitutional Economics, London and New York: Routledge.

Wagner, J.E. and S.C. Deller (1998), ‘Measuring the effects of economic diversityon growth and stability’, Land Economics, 74: 541–56.

Diversity, stability and regional growth in the United States 229

11. Inter-regional knowledge flows inEurope: an econometric analysisMario A. Maggioni and T. Erika Uberti*

1. INTRODUCTION

The aim of this chapter is to analyse the impact of knowledge on regionaleconomic development and, consequently, on regional disparities acrossfive major European countries: France, Germany, Italy, Spain and theUnited Kingdom. In particular our focus is on the nature of knowledge,not only as a fixed cost in the production process (leading to scaleeconomies), an investment good (influenced by accumulation and depreci-ation dynamics) and an experience good (whose quality attributes can bedetected only upon using, or consuming, the good), but also as a ‘re-lational’ good displaying network externalities.

In this chapter we analyse the manifold nature of knowledge through theanalysis of four distinct but complementary phenomena (Internet hyper-links, European research networks, European Patent Office (EPO) co-patentapplications, Erasmus student mobility) which characterise knowledge as anintrinsic relational structure (directly) connecting people, institutions and(indirectly) regions across five European countries. Two main research ques-tions are addressed: the first deals with the notion of regional disparities; thesecond refers to the different concepts of distance, namely geographical,functional and sectoral.

Regional disparities can no longer be defined only in terms of statisti-cal differences in the values of standard macroeconomic indicators.Knowledge matters more and more in defining both the level and thegrowth rate of a given region GDP (Sapir et al., 2004). For this reason, newrelational indicators have to be built and compared in order to develop anew kind of (relational) analysis able to complement the usual ‘attribu-tional’ one.

Traditionally, regional economic disparities have been ascribed toperipherality – measured by the distance from the main centres of popu-lation and economic activity – and/or to a high level of dependence ondeclining sectors (mainly ‘mature industries’). The scale of regional and

230

other disparities, as well as the political approach and the specific policyinstruments used at the European level to deal with this problem, havechanged very much over the years. Europe is lagging behind the USA interms of growth and investments in knowledge infrastructures, but thisgeneral statement, while true, hides a huge variance across Europeanregions and nations (DG Enterprise, 2003).

In the last 15 years, income differences among European member stateshave been strongly narrowing while the process has been matched with awidening of the inter-regional variance within single countries (Martin,1998). All this casts a shadow on the whole range of European regionalpolicies, explicitly designed to reduce geographical imbalances andstrengthen regional cohesion, and raises questions about the consequencesof the future Europe enlargement, as the gap is expected to widen. A veryodd and worrying aspect of the European context is that the productivecapacity agglomeration process – as a consequence of market forces – maybecome too strong and risky to be socially unacceptable.

In addition, at the Lisbon 2000 European Council, the European Union(EU) set itself the ambitious goal of becoming ‘the most competitive anddynamic knowledge-based economy in the world, capable of sustainable eco-nomic growth with more and better jobs and greater social cohesion’(European Commission, 2000) and the Council requested the Commissionto report annually on the structural indicators of progress in member statestowards the EU’s strategic goal. These calls for robust evidence and rigorousmonitoring of outcomes led to the development of a set of comprehensivestructural indicators to underpin further analyses. In particular, it is inter-esting to study the different effects of geographical, functional and sectoraldistance on the relational activity of different territories. This is the object ofthe present analysis which, within a ‘gravitational’ framework, looks at fourdifferent relational variables (Internet hyperlinks, European research net-works, EPO co-patenting applications and Erasmus student flows) between110 European NUTS2 (Nomenclature des Unités Territoriales Statistiques)regions located in five European countries: Germany (40 regions), Spain(16), France (22), Italy (20) and United Kingdom (12).

Gravitational models usually include geographical distance (based ongeodesic path or road distance) between two areas to capture a series ofdistance-related phenomena which are difficult to measure (such as: trans-port costs, time elapsed during shipment, synchronisation costs, comm-unication costs, transaction costs and cultural distance). Here, we use‘geographical’ distance, calculated as the shortest road distance existingbetween two NUTS2 ‘capitals’, but we add two concepts of distance, the‘functional’ distance, calculated as the difference (in absolute value)between the level of innovative performance of different regions (based on

Inter-regional knowledge flows in Europe 231

the Regional Summary Innovation Index (RSII) contained in the EuropeanInnovation Scoreboard (EIS) and the ‘sectoral’ distance (based on the sec-toral distribution of the patenting activity).

The chapter is organised as follows: Sections 2 and 3 describe the vari-ables used in the different analyses; Section 4 presents different types ofcorrelations (Pearson, Spearman and quadratic assignment procedure);Section 5 illustrates the use of social network analysis to detect structuralproperties of different knowledge exchange flows; Section 6 is devoted tothe econometric analysis of two ‘gravitational’ models; and Section 7concludes.

2. FOUR TYPES OF KNOWLEDGE FLOWS

Krugman, in his Geography and Trade, stated that ‘knowledge flows . . . areinvisible; they leave no paper trail by which they may be measured andtracked’ (Krugman, 1991, p. 53).

Jaffe et al. (1993, p. 578) reacted to the previous statement by suggestingthat ‘knowledge flows do sometimes leave a paper trail, in the form of ci-tations in patents. Because patents contain detailed geographical informa-tion about their inventors, we can examine where these trails actually lead’.

We attempt to move the approach a little further by focusing on fourknowledge-based relational phenomena: digital information exchange(transmitted through Internet hyperlinks), participation in the sameresearch networks (funded by the EU Fifth Framework Programme), EPOco-patent applications and Erasmus student exchange flows. Through thesevariables we attempt to measure the intrinsic relational structure of knowl-edge flows which directly connects people and institutions and, indirectly,regions, across five European countries.

These four variables capture different types of knowledge (spanningfrom ‘pure tacit’ to ‘pure codified’ knowledge) and different stages of theknowledge creation process. Although information and communicationtechnologies (ICTs) reduce communication and transmission costs, thenature of knowledge and its creation process are very complex and requiresocial processes involving different modalities of interactions. Even in theInternet era, face-to-face relations remain crucial (Feldman, 2002).

It is worth noting that the relational variables considered in the analysisspan the entire spectrum of ‘relational’ aspects of knowledge creation, sug-gesting alternative ways to detect the knowledge trail: from a new andnon-material way of information exchange (that is, Internet hyperlinks), tophysical and virtual institution-based interactions developed to improveknowledge creation (that is, research networks) by exchanging mostly

232 Spatial systems

codified knowledge, to physical and virtual individual-based relationshipsaimed at developing marketable innovations (that is, co-patent appli-cations) by exchanging mostly tacit knowledge and know-how, to the phys-ical movement of people leaving their own region in order to acquire a partof their university education in a foreign institution (Erasmus studentexchange).

Digital Information Exchange (through Internet Hyperlinks)

The recent diffusion of ICTs, and in particular of the Internet, stimulatedseveral analyses to measure the diffusion of such a phenomenon acrosscountries, regions, cities, ethnic groups and social classes in order to mapthe current state and to detect the presence of a ‘digital divide’. Differentindicators may be used to detect the diffusion of ICTs. The simplest way isto measure the ‘endowment’ of the ICT equipment (that is, the number ofInternet hosts, personal computers and broadband connections) and, moregenerally, all telecom infrastructures that allow efficient connections. Asecond way concerns the measurement of the ‘access’ conditions to ICTservices, in terms of the market structures (and prevailing pricing strate-gies) of the relevant markets (telecoms, Internet service providers (ISPs)and so on). Another way is related to the ‘use’ of ICT, which may bedetected by measuring the number of people on-line, to time spent on-line,to size of different on-line activities (e-commerce, e-government and so on).

A further way concerns the relational nature of the physical infrastruc-ture of the Internet (comprising cables, routers, satellite and radio con-nections), and of the www (world wide web), the Internet virtual interfaceand service platform that allows us to visualise and exchange the infor-mation. Since the www is a network of web pages linked through Internethyperlinks, it can be used to map the inner structure of communicationchannels and to detect the producers and consumers of digital infor-mation. When an Internet hyperlink button is clicked, the content of thetarget web page is transferred to the clicking computer. One may thusthink that the web page containing the hyperlink button acts as animporter of digital information and the ‘target’ web page represents theinformation ‘exporter’ or, more precisely, think of an Internet hyperlink asan index of revealed comparative advantages in the production of specifictypes of digital information.

Internet hyperlinks therefore may be used as an indicator of ‘potentialuse’,1 since the existence of a hyperlink from one web page to anothersignals the willingness of the ‘owner’ of one page to import digital infor-mation from another and increases the probability that the targeted webpage is actually accessed.2 One may argue that the number of Internet

Inter-regional knowledge flows in Europe 233

hyperlinks in a web page is uninformative since the inclusion of a newhyperlink button is not constrained by a monetary budget as it does notcost anything. However, the presence of buttons within a page is subject toa harder ‘graphical’ budget constraint. Web design handbooks show thatwhile the number of hyperlinks is a key element in determining the attrac-tiveness of a web page, such attractiveness is a non-monotonic function ofthe number of hyperlinks: it is good to have a few buttons but not too many.

The www has been thoroughly analysed by mathematicians, physicists,information scientists and engineers,3 in order to detect its structure anddevelopment laws. Albert and Barabasi (2002) argue that the www has ascale-free topology in which a small number of ‘central’ web pages are verypopular (are targeted by a huge number of hyperlinks), while the rest of theweb is composed of peripheral pages which are almost unconnected andvirtually unknown. Other studies show that different typologies of webpages have different organisational structures. Some types of web pages(that is, those of universities and newspapers) display a random networkstructure in which there are a few extremely central and extremely periph-eral nodes, while most of the nodes are targeted by a number of hyperlinksaround the average (Maggioni and Uberti, 2005).

Uberti and Maggioni (2004) analysed the connectivity of web pages ofdifferent institutions (universities, local authorities and chambers of com-merce) at the regional level and showed that universities are the most active‘traders’ of digital information. In this chapter we therefore include thenumber of Internet hyperlinks between 308 university web pages located inGerman, Spanish, French, Italian and UK regions at the NUTS2 level.4

We chose these 308 universities, members of the European UniversityAssociation (EUA), located in Germany, Spain, France, Italy and the UnitedKingdom. The selected sample accounted for 51 universities in France, 61 inGermany, 53 in Italy, 45 in Spain and 95 in the UK. However, its represen-tation of the total population of European universities largely differs fromcountry to country because of the exclusion of Hochschulen (in Germany)and Écoles Supérieures (in France) which are not members of the EUA.

Since our analysis is devoted to the analysis of information and knowl-edge flows, we transposed the matrix of the Internet hyperlinks (that is, thepresence of an Internet hyperlink from region j to country i, is analysed asthe presence of an information channel flowing in the opposite direction,from region i to region j).

Research Networks

Since the early 1980s, the EU has promoted the creation of research con-sortia (among firms, universities, research centres and public agencies) in

234 Spatial systems

order to increase the competitiveness of the European industry and tofoster intra-European cohesion through the exchange and diffusion ofscientific and technological knowledge. In both the Single European Actand the Maastricht Treaty (Article 130G), the European institutions havebeen given competence in the area of science and technology and havedeveloped several actions in order to promote research and development(R&D), create European networks, coordinate R&D and stimulate theEuropean mobility of researchers (Breschi and Cusmano, 2004).

Framework programmes always constituted the main planning instru-ment and funding source for R&D policies in the EU, but as time passed,priorities changed: ‘the latest programmes have shifted the emphasis fromsupply-side factors, central in the design of the first policies, to diffusion-oriented projects and the increase of central skills and knowledge amongEuropeans’ (ibid., p. 752). In 1998, the European Council and the Euro-pean Parliament approved the Fifth Framework Programme (5FP), a pro-gramme with a different structure from the previous ones, valid for fiveyears (from 1998 to 2002), and financed with about €15 million. This 5FPis divided into 10 thematic and horizontal programmes, and provides for 12different types of contracts.5

In our analysis we focus on two contracts – explicitly dedicated to theestablishment and use of scientific networks, namely thematic network con-tracts and research network contracts – whose coordinator is a universitylocated in one the 110 regions of the sample, and we included all partici-pants located in these regions, irrespective of their typology (that is, uni-versities, research centres or business organisations).

Co-patents

Patents (and patent applications) are one of the most established outputindicators of innovative activities. Since the seminal contribution ofScherer (1965), patents have been used in the economic literature (Griliches,1981, 1990), in order to measure knowledge spillovers and other spatialexternality effects which, in contrast to what was argued by Krugman(1991), do leave a paper trail (Jaffe et al., 1993).

The constitution of the EPO in Munich in 1973 allowed researchers touse a common dataset to analyse the innovative performance of differentEuropean countries and regions. In particular, Paci and Usai (2000) andBreschi and Lissoni (2004) have developed systematic analyses of patent-ing activity throughout Europe at different NUTS levels, showing the exist-ence of significant clustering phenomena (whose agglomeration indices areeven higher than those registered by high-tech manufacturing) within acore–periphery geographical pattern.

Inter-regional knowledge flows in Europe 235

Later studies analyse patent data as relational variables. Unlike Breschiand Lissoni’s (2004) analysis on patent citations, Maggioni and Usai (2005)look at patents as a relation between inventors and applicants at NUTS2level and study the distributions of these relationships within differentEuropean countries. They seek industry-specific patterns and test thehypotheses of a diffused ‘brain-drain’ dynamics by which peripheralregions host inventors, but do not exploit the economic outcomes of theirscientific and technological creativity since applicants (mostly firms) arelocated in the core regions.

In this chapter we consider another relational aspect of patents: the co-invention process. Co-invention (and thus co-patenting) is a process involv-ing both tacit and codified knowledge exchanges. For this reason it impliesa series of both ‘face-to-face’ and ‘over-a-distance’ relationships betweeninventors. This is why it is interesting to analyse the relative importance of‘geographic’ versus ‘functional’ distance as forces shaping the inter-regional (international) structure of this knowledge flow network. Out ofa total of more than 170 900 patent applications belonging to everyInternational Patent Classification (IPC) section (coming from inventorslocated in the above-mentioned five countries in the 1998–2002 period) –extracted by the CRENoS files which were based on the original EPO data-base – we selected only those patents whose applications were recorded bymore than one inventor. Next we split each patent into equal shares attrib-uted to each inventor. We then added these data for each NUTS2 region inorder to built a matrix in which a generic cell ij represents the share ofpatents6 recorded jointly by inventors located in regions i and j (whereregions i and j could belong to different nations). A total of nearly 30 000co-patents was detected.

Erasmus Student Exchange

The Erasmus student exchange represents another relevant part of the spec-trum of relational activities involving knowledge flows among Europeanregions: the mobility of tertiary education students, who represent the basicchannel for international training and education.7 The Erasmus pro-gramme, introduced in 1987 – and, since 1995, part of the Socrates pro-gramme8 – is a European programme devoted to fostering higher educationand to creating a ‘European dimension’ of education. Its popularity, interms of student participation, is constantly increasing, and has undoubt-edly widened after the Lisbon Council emphasised the enforcement of edu-cation and training, and student mobility as important goals to be achieved.The Erasmus student exchange reflects several important features, equallycontributing to ‘[strengthening] the whole fabric of relations existing

236 Spatial systems

between the peoples of Europe’ (European Commission, 2005): the‘institutional’ integration among European countries; the ‘openness’ ofnational tertiary systems; and the ‘relative attractiveness’ of a country,either in terms of its culture or in terms of the reputation of its tertiaryeducation system.

As in the case of digital information flowing through hyperlinks, we areinterested in the flow of knowledge; hence we consider the region in whichthe hosting university is localised as the ‘emitting’ region (region i) of theknowledge flows embedded in the ‘learned’ students returning to their‘receiving’ region (region j) after their period of study abroad.9

3. EXPLANATORY (ATTRIBUTIONAL ANDRELATIONAL) VARIABLES

In this chapter we use some attributional variables to detect differences inthe knowledge-based characteristics of 110 European regions. The attri-butional variables include: GDP, R&D intensity (the ratio between totalR&D expenditure and GDP), and three measures of distance (or dissimi-larity): the geographical distance (based on road distance between ‘capi-tals’), the functional distance (based on the RSII contained in the EIS), andthe sectoral distance (based on the 2-digit sectoral composition of regionalpatent application10).

What follows is a brief description of variables, their transformationsand data sources. Note that throughout the chapter, subscript i refers to the‘emitting’ region and j to the ‘receiving’ region, while I and J refer to coun-tries.

GDPi and GDPj: Gross domestic product of regions i and j expressed inpurchasing power standards (pps). GDP data, expressed in millions ofeuros refer to year 2000.

RDi and RDj: Research and development intensity of regions i and j. It iscalculated as the ratio between the regional levels of gross expenditure onresearch and development (GERD) and GDP and refers to various years.

GDistij: Geographical distances among 110 European regions are calcu-lated according to the shortest road distance (in kilometres) betweenregional ‘capitals’. The notion of ‘regional capital’ implied the use of acertain degree of arbitrariness since NUTS2 levels are administrativemeaningful entities in Italy, Germany, Spain and France, but not in theUK. In this last case we used population as the selecting criteria to

Inter-regional knowledge flows in Europe 237

identify the most relevant city (which we called ‘capital’), irrespective ofthe presence of an administrative capital.

FDistij: Functional distance is measured as the difference (taken inabsolute value) of the values registered by the two regions on the RSIIcontained in the EIS. The RSII measures the ‘European technologicalleadership’ and ranks the absolute innovative performance ofEuropean regions. It is calculated by re-scaling the regional values ofthe 13 available indicators11 according to the following formula, andthen taking the unweighted average of the re-scaled values per eachregion:

(11.1)

where XfjJ is the value of an indicator f for region j in country J, and mis the number of available indicators for the j region (DG Enterprise,2003).

This composite index is based in data recorded on different years butofficially refers to 2003. (DG Enterprise, 2003).

SDISTij: Sectoral distance is measured as the inverse of the technologicalnearness index (Moreno et al., 2005) calculated as a correlationcoefficient between the sectoral composition of patent applicationregistered by region i and by region j at the EPO in the 1997–2000period.

CONTIGij: Contiguity, or adjacency, is a dummy variable which takesthe value 1 for contiguous regions (that is, which share a border), and 0elsewhere.12

COUNTRYIJ: A dummy variable which is used to control for fixednational effects both on the ‘emitting’ and the ‘receiving’ regions.

4. FOUR TYPES OF KNOWLEDGE FLOWS: ACORRELATION ANALYSIS

This section shows some results on the correlation existing between the fourdependent variables we used to describe knowledge flows (Internet hyper-links, research networks, co-patents and Erasmus student flows) used in theeconometric exercise. In particular we shall focus our attention on simple

RSIIjJ � �m

f�1� XfjJ � min(XfjJ)max(XfjJ) � min(XfjJ)�,

238 Spatial systems

correlation (Pearson), rank correlation (Spearman) and quadratic assign-ment procedure (QAP) correlation coefficients.

Table 11.1 presents both Pearson’s simple correlation coefficients andSpearman’s rank correlation coefficients for our relational dependentvariables. All coefficients are positive and significant, showing that thefour variables selected for the analysis measure different elements of thesame phenomenon: information and knowledge flows. Digital infor-mation and research networks, on one side, and digital information andErasmus exchange, on the other side, show the highest correlationcoefficients.13 This may be interpreted as a sign of complementaritybetween virtual and physical interactions among European universities(and regions). One may also note that the high correlation coefficientbetween research networks and the Erasmus exchange programme showsthe existence of hysteresis in the university inter-regional (and inter-national) relationships. Once a relationship is established, both professorsand students continue to exploit it.

The EU attempts to build research networks aimed at producing not only‘pure research’, but also applied research and marketable innovationswhich seem to be partially successful: in fact Pearson’s correlation betweenresearch networks and co-patenting is quite high (0.264).14

We further analysed the relationships among these knowledge flows byusing the QAP correlation, a bootstrap method that computes correlationindices between entries of two square matrices and assesses the frequencyof random measures as large as actually observed. The QAP algorithm pro-ceeds in two steps. In the first step it computes Pearson’s correlationcoefficients between corresponding cells of the two data matrices. In thesecond step, it randomly (synchronously) permutes rows and columns ofone matrix and re-computes the correlation to the other matrix. The secondstep is carried out hundreds of times (in our case: 5000 times) in order tocompute the proportion of times that a random measure is larger than orequal to the observed measure calculated in step 1. A low proportion

Inter-regional knowledge flows in Europe 239

Table 11.1 Pearson and Spearman correlations between knowledge flowvariables

Pearson’s correlation Spearman’s rank correlation

Diginfoij RNij Patij Erasij Diginfoij RNij Patij Erasij

Diginfoij 1.000 Diginfoij 1.000RNij 0.458 1.000 RNij 0.313 1.000Patij 0.167 0.264 1.000 Patij 0.276 0.058 1.000Erasij 0.331 0.339 0.212 1.000 Erasij 0.322 0.241 0.196 1.000

(smaller than 0.05) suggests a strong relationship between the two matricesthat is unlikely to have occurred by chance (Borgatti et al., 2002).

Table 11.2 shows the results of such a procedure.15 The highest corre-lation is registered for research networks and Internet hyperlinks (0.302),followed by Erasmus student flows and Internet hyperlinks (0.270),confirming the simple correlation results and showing the high comple-mentarities between these flows of knowledge.

QAP procedure shows that the correlation between co-patenting andresearch networks and co-patenting and Erasmus exchange flows (whichregistered low Spearman and Pearson correlation coefficients) is notsignificant, indicating the persistence of frictions between different worlds(that is, the business and academic environments). Although the sample ofresearch networks included in the analysis has been selected,16 hence suffersfrom some biases, these results may also show that EU programmes seemto fail in connecting different actors, hence these actions need to beredefined to be really effective across different institutions (and in particu-lar between profit and non-profit organisations).

5. NETWORK ANALYSIS OF KNOWLEDGE FLOWS

Network analysis (NA) uses quantitative techniques, derived from graphtheory, to study and describe the structure of interactions (edges) betweengiven entities (nodes) (Wasserman and Faust, 1994). Initially used by soci-ologists and ethnologists to study complex personal interactions, NA hasrecently been used in economic analyses (Snyder and Kick, 1979;Maggioni, 1993 and 2000; Leoncini et al., 1996; Uberti, 2002; Breschi and

240 Spatial systems

Table 11.2 QAP correlation between knowledge flow variables

Pearson’s correlation

Diginfoij RNij Patij Erasij

Diginfoij 1.000RNij 0.302** 1.000

(0.031)Patij 0.220** 0.062 1.000

(0.043) (0.040)Erasij 0.270** 0.245** 0.103 1.000

(0.038) (0.036) (0.048)

Note: ** Significant at 5%; standard error in parenthesis.

Cusmano, 2004; Breschi and Lissoni, 2004) to analyse institutional, tech-nological and commercial relationships among agents, industries, regionsand countries. Therefore, in this chapter, 110 NUTS2 European regions aretreated as nodes, while their different knowledge flows are treated as edges.

Orthodox approaches describe the innovation process through an atom-istic principle that assumes the existence of individual utility maximisationprocedures and does not take into account the wider social, economic andinstitutional framework. By contrast, NA highlights some relevant struc-tural features.

The ‘behaviour’ of a node (in terms of strategy and performance) has to beinterpreted in terms of both structural limits and internal features. Internodalrelationships must be examined from two complementary perspectives: thesingle node’s and the whole system’s perspective. Neither a single node nor apair of nodes can be meaningfully analysed, when isolated from the systemframework (holistic principle). Systems display a surprisingly intrinsic fractalnature: both the macro level (whole system) and the micro level (nodes) arecomposed by a plurality of structurally interrelated elements. The interdepen-dence of observations does not hinder NA techniques, allowing a wider use ofthis methodology even when more traditional statistical and econometric tech-niques based on pure attributional variables suffer. (Bramanti and Maggioni,1997, p. 327)

In the following analysis we shall use the density and clusteringcoefficients as the index of systemic connection. Density is defined as theratio between the actual number of edges e and the maximum number ofdirected edges in a network composed of n nodes17 or:

(11.2)

The clustering coefficient of node i characterises the extent to whichnodes adjacent to it are adjacent to each other (Watts, 1999):

(11.3)

where vi is the number of nodes connected to i and � is the total number ofpossible edges in i’s neighbourhood. The clustering coefficient for the wholenetwork is obtained by averaging the clustering coefficient of all nodes inthe network.

The networks analysed in the chapter describe different knowledge flows:co-patenting and research networks are symmetric, and Internet hyperlinksand Erasmus exchange flows are a-symmetric. In the latter case, for each

Ci ��i�

,

D � en(n � 1).

Inter-regional knowledge flows in Europe 241

node an outdegree (number of outward connections) and an indegree(number of inward connections) have been calculated.

Furthermore, NA indices have been calculated from a dichotomizedversion18 of the original innovation flow matrices. The customary pro-cedure implies the choice of an ‘appropriate’ (often ad hoc) threshold;however, it must be considered that the choice of a given threshold is stra-tegic because different values produce different dichotomised networks. Inthis analysis we choose the network average as the threshold value. In orderto detect the most central actor within the system and the definition of ascale of hierarchy (inequality), centrality and centralisation indices havebeen designed.19 Formally, the degree centralisation of a network (system)of n nodes (regions) can be defined as follows:

(11.4)

where C*b is the centrality value of the most central region in the system andthe denominator reflects the maximum level of centrality obtainable in asystem of n regions. The centralisation indices (which lies between 0 and 1)measure the difference in centrality levels between the most central regionand the others. A high centralisation index identifies a very hierarchicsystem where differences in positions are maximised, and a pivotal nodeexists. A low centralisation index identifies a structure where most of thepositions are similar and interchangeable.

Table 11.3 shows the network indices for different knowledge flows.Density indices of dichotomised networks show that the digital infor-

mation and the co-patenting networks are the least dense, while theresearch network is the most dense. The data show that knowledge flows donot spread evenly between European regions, which suggests that tra-ditional face-to-face interactions remain one of the most active phenom-ena of knowledge creation, although virtual ones are cheaper.

C�

ib˛

��

i(C*b � Ci

b)

(n � 1)(n � 2),

242 Spatial systems

Table 11.3 Network analysis indices of knowledge flow structures

Density Clustering Isolated nodes Centralisation

Outdegree Indegree

Diginfoij 0.111 0.741 33 0.425 0.379RNij 0.240 0.542 15 0.361 0.361Patij 0.140 0.727 22 0.281 0.281Erasij 0.199 0.464 3 0.586 0.410

The ranking based on clustering coefficient is almost the opposite of thedensity-based one. This can be explained by referring to the number of iso-lated nodes. The most clustered networks (digital information and co-patenting) have many isolated nodes (private club structure): meaning thatif one region is connected to another one, then it is very likely that the sameregion is also connected to the original node neighbours.

Centralisation indices show that, in general, the Erasmus studentnetwork is the most centralised, while co-patenting is the least centralisednetwork. However – since Internet hyperlinks and Erasmus are asymmet-ric, while research networks and co-patenting are symmetric – it is moreuseful to consider each couplet in isolation.

As far as symmetric relationships are concerned, research networksexhibit a more hierarchical regional structure than co-patenting ones, sug-gesting that educational institutions are tied for better or worse to theregion’s performance, while individual inventors are more evenly diffusedand their interactions follow a more uniform pattern. The co-patentingnetwork has a rather non-hierarchic structure due to the presence of somevery central regions (Oberbayern, Darmstadt, Düsseldorf and Île deFrance) and to a series of other regions that are connected not exclusivelyto the most central ones but also with their national neighbourhood. Infact, by removing the most central nodes from the network, highly con-nected national ‘islands’ emerge (see Figures 11.1 and 11.2).

As far as a-symmetric relationships are concerned, Erasmus studentflows display a more hierarchical structure (some European regions arehighly engaged in the Erasmus programme either as a source or a desti-nation of student flows, while others are almost not involved) than thedigital information one (differences in the number of hyperlinks are notso relevant). The difference in centralisation values (referred to as out-degree and indegree) of the Erasmus programme may be interpreted as theexistence of a larger difference in participation in the Erasmus programmeof different European regions as recipients of students than as senders (agreater number of regions send their students abroad, but their destinationis concentrated in a smaller number of regions).

A similar (although smaller) difference is shown by the centralisationindices of the digital information networks. European regions show agreater difference in their information exports than in their informationimports. In other words while universities (and regions) are more similarin the number of hyperlink buttons inserted in their web pages, few uni-versities (and regions) record a larger share of total Internet hyperlinkdestinations.

Inter-regional knowledge flows in Europe 243

244

Fig

ure

11.1

Co-

pate

nt n

etw

ork,

1998

–200

2,in

clud

ing

Obe

rbay

ern,

Dar

mst

adt,

Düs

seld

orf

and

Île

de F

ranc

e

York

shire

and

the

Hum

ber

245

Fig

ure

11.2

Co-

pate

nt n

etw

ork,

1998

–200

2,ex

clud

ing

Obe

rbay

ern,

Dar

mst

adt,

Düs

seld

orf

and

Île

de F

ranc

e

York

shire

and

the

Hum

ber

6. GRAVITY EQUATIONS

Looking for the source of regional disparities we use gravity equations inorder to assess whether ‘geographic distance’ was responsible for such aphenomenon (that is, peripherality exogenously causes poor perfor-mances of regions, therefore determining the polarisation of a rich coreand a depressed periphery) or whether ‘functional distance’ (that is,difference in the scientific and technological levels) endogenously plays amajor role in determining the existence of a much more dense core (thenetwork of more advanced regions) and a residual sparse set of relationswithin the periphery. Finally, we tested the influence of the similarity/dis-similarity of the productive structure of different regions by detecting theeffects of ‘sectoral distance’ (measured through patent activity) on knowl-edge flows.

The gravity equation model is a powerful tool of empirical analysis toexplain social interactions (for example, international trade, foreign directinvestment, migration and tourism) according to the existence of ‘attract-ing’ and ‘impeding’ forces. This range of models is derived from the ‘law ofuniversal gravitation’ proposed by Isaac Newton in 1687, which stated that‘gravitational force between masses decreases with the distance betweenthem, according to an inverse-square law. . . . [T]he theory notes that thegreater an object’s mass, the greater its gravitational force on another mass’(Wikipedia, 2005).

In the economic literature these models are commonly used to explaininternational trade: bilateral trade between two countries is proportional totheir economic mass (that is, GDP or population) and inversely related totheir geographical distance. These models have been a successful tool forempirical analysis since the 1960s: the signs of parameters of importingand exporting countries’ GDPs are positive, roughly equal to unity andsignificant, and the sign of geographical distance is negative and significant(Tinbergen, 1962; Poyhonen, 1963). Recently this empirical success hasbeen theoretically demonstrated (Anderson, 1979; Bergstrand, 1985;Helpman, 1988; Deardorff, 1998; Feenstra, 2002; Dalgin et al., 2004).

We specify a gravitational model which explains the level of a particulartype of knowledge flows between two generic regions i and j as a functionof a series of relational and attributional variables. All variables are takenin logs in order to interpret the estimated coefficients as elasticities.

The generic dependent variable, KFij, stands for four different typolo-gies of knowledge flows: digital information (KFij�Diginfoij), researchnetworks (KFij�RNij), co-patenting (KFij�Patij), or Erasmus studentexchange (KFij�Erasij); the independent variables are as defined inSection 3. Table 11.4 shows the results of eight ordinary least squares (OLS)

246 Spatial systems

247

Tab

le 1

1.4

Gra

vity

equ

atio

n fo

r kn

owle

dge

flow

s

Dig

ital

info

rmat

ion

flow

s R

esea

rch

netw

orks

RN

ijC

o-pa

tent

s P

atij

Era

smus

stu

dent

s E

ras ij

Dig

info

ij

(Ia)

(Ib)

(I

Ia)

(IIb

) (I

IIa)

(III

b)(I

Va)

(IV

b)F

unct

iona

lSe

ctor

alF

unct

iona

lSe

ctor

alF

unct

iona

lSe

ctor

alF

unct

iona

lSe

ctor

aldi

stan

cedi

stan

cedi

stan

cedi

stan

cedi

stan

cedi

stan

cedi

stan

cedi

stan

ce

Inde

pend

ent

vari

able

GD

Pi

0.76

1***

0.75

8***

0.07

6***

0.07

5***

0.74

8***

0.73

3***

0.62

1***

0.60

4***

(0.0

21)

(0.0

21)

(0.0

04)

(0.0

04)

(0.0

25)

(0.0

25)

(0.0

18)

(0.0

18)

GD

Pj

0.60

0***

0.59

4***

0.07

6***

0.07

5***

0.74

7***

0.73

3***

0.46

1***

0.44

1***

(0.0

22)

(0.0

21)

(0.0

04)

(0.0

04)

(0.0

26)

(0.0

25)

(0.0

18)

(0.0

18)

RD

i0.

240*

**0.

236*

**0.

045*

**0.

041*

**0.

391*

**0.

394*

**0.

065*

0.05

6**

(0.0

26)

(0.0

26)

(0.0

05)

(0.0

05)

(0.0

30)

(0.0

30)

(0.0

25)

(0.0

24)

RD

j0.

035

0.03

30.

045*

**0.

041*

**0.

392*

**0.

394*

**0.

140*

**0.

137*

**(0

.027

)(0

.026

)(0

.005

)(0

.005

)(0

.030

)(0

.030

)(0

.024

)(0

.024

)G

Dis

t ij�

0.44

9***

�0.

445*

**�

0.02

0**

�0.

024*

**�

0.52

5***

�0.

580*

**�

0.00

8�

0.03

2(0

.038

)(0

.038

)(0

.006

)(0

.007

)(0

.043

)(0

.039

)(0

.038

)(0

.037

)C

onti

g ij�

0.16

8**

�0.

139

�0.

011

�0.

015

1.09

8***

0.97

5***

1.22

0***

1.16

9***

(0.1

00)

(0.0

98)

(0.0

17)

(0.0

17)

(0.0

74)

(0.0

72)

(0.1

58)

(0.1

54)

FD

ist ij

�0.

051*

*�

0.02

0***

0.00

4�

0.03

3**

(0.0

16)

(0.0

03)

(0.0

18)

(0.0

14)

SD

ist ij

�0.

403*

*�

0.05

2**

�0.

520*

**�

0.67

7***

(0.1

18)

(0.0

17)

(0.1

35)

(0.1

08)

Con

stan

t�

7.67

8***

�7.

400*

**�

1.45

4***

�1.

346*

**�

10.2

15**

*�

9.53

9***

�8.

960*

**�

9.52

5***

(0.4

11)

(0.3

97)

(0.0

67)

(0.0

66)

(0.4

77)

(0.4

65)

(0.4

46)

(0.4

41)

Num

ber

of65

1367

0911

643

1177

246

2347

5251

0051

94ob

serv

atio

nsR

-squ

ared

0.53

90.

541

0.21

80.

217

0.64

20.

643

0.35

90.

365

F-t

est

608.

9263

6.37

109.

7911

0.37

679.

3869

5.77

264.

5427

6.19

Not

e:**

*si

gnifi

cant

at

1%;*

*si

gnifi

cant

at

5%;*

sign

ifica

nt a

t 10

%;r

obus

t st

anda

rd e

rror

in p

aren

thes

is;c

ount

ry d

umm

ies

are

incl

uded

in a

llre

gres

sion

s bu

t no

t re

port

ed.

regressions,20 where we considered alternatively functional and sectoral dis-tance in the regressors.

Regression (a):

Regression (b):

where �I indicates country dummy variables and, �ij and �ij are standarderror terms.

Table 11.4 presents the results of the econometric analysis.

● Regression I describes the structure of information flows runningthrough Internet hyperlinks established between European universi-ties. These flows are positively influenced by both regions’ GDP,confirming the existence of a positive relation between the ‘economicsize’ of a region and its involvement in ICT (in terms of endowments,access and use) which may well lead to a ‘digital divide’ phenomenon.Note also the coefficient of the emitting region is slightly larger thanthat of the receiving one. A positive and significant coefficient isregistered by the R&D intensity of the emitting region, while thecoefficient of the ‘receiving’ region is not significant. This suggeststhat the level of intensity of innovation inputs of a region determinesthe ‘visibility’ of the local university’s website (perhaps via a relation-ship between public funding of R&D and university relevance21),while the positioning of hyperlink buttons follows a different logic.Geographical distance is negative and significant suggesting that, atleast for our sample of university websites, the advent of the Internetdid not cause the ‘death of distance’.22 Digital relationships areconsidered in academia as complementary to physical ones and face-to-face contacts are still crucial. It is, however, worth noting that thecoefficient of the contiguity variable is also negative and significant.Such a result may be explained in terms of a limited use ofInternet-based information flows between neighbouring universitiesand some hidden forms of spatial competition on the local pool ofprospective students. Functional distance bears a significant and neg-

5ln(GDistij) � 6Contigij � 7ln(SDistij) � �I � �ij

ln(KFij) � 0 � 1ln(GDPi) � 2ln(GPDj) � 3ln(RDi) � 4(RDj) �

5ln(GDistij) � 6Contigij � 7ln(FDistij) � �I � �ij

ln(KFij) � 0 � 1ln(GDPi) � 2ln(GPDj) � 3ln(RDi) � 4ln(RDj) �

248 Spatial systems

ative coefficient, thus signalling that university networks of relationsas measured by Internet hyperlinks tend to develop between similarregions. A similar result is shown by the measure of sectoral distance,perhaps suggesting the existence of a deep relation between a region’sindustrial structure and the characteristics of its universities.

● Regression II analyses the joint participation of research institutionsbelonging to different regions in two different types of researchnetwork under the fifth EU research framework programme. Theseflows are positively influenced by both regions’ GDP, confirming theexistence of a positive relation between the ‘economic size’ of aregion, its research potential and its scientific networking activity. TheR&D intensity coefficients of the emitting and receiving regions arepositive and very similar, suggesting that the propensity to be involvedin the network is positively correlated with the ‘scientific and techno-logical level’ of both regions. The coefficient of geographical distanceis negative and significant; however, its size is very small, suggesting alimited influence of spatial effects in this activity whose aim is explic-itly to link research units from different places all over Europe. This isconfirmed by the insignificance of the coefficient on contiguity. Thecoefficients of both functional and sectoral distance are negative andsignificant: meaning that both the scientific and technological levelsand the sectoral specialisation of a region play a positive role in deter-mining the probability of joining the same research network. In otherwords, research networks have a ‘club’ structure in which agentsmatch with others that are similar. If this is the case, then research net-works cannot be used as policy tools to support cohesion and inclus-iveness since their structure is a ‘segmented’ one in which strongerregions cooperate with other stronger ones, and weaker with weaker.It is also worth noting that the coefficient on the sectoral distance hasa higher value than the geographic distance.

● Regression III describes the structure of scientific relationships, whichderives from the exchange of knowledge and know-how betweenEuropean inventors. Co-patenting relationships are strongly and pos-itively influenced by both regions’ GDP and R&D intensity; thecoefficients of both variables are very similar. This confirms that bothsize (that is, larger and richer regions have a greater number ofpatentable inventions) and technological level play an importantrole in determining the amount of knowledge exchange neededto develop a patentable innovation. Geographic distance has asignificant and negative coefficient. This could be explained in termsof the need for face-to-face contacts in the R&D activity (based ontacit knowledge exchange) leading to a patent application. Since the

Inter-regional knowledge flows in Europe 249

coefficient of functional distance is not significant, we focus the atten-tion on model b in order to test whether the sectoral distance plays amore relevant role. This is exactly the case: the negative and significantcoefficient shows that a common sectoral specialisation of the techno-logical activity of the two regions is important to determine the levelof scientific collaboration between inventors. The positive andsignificant coefficient on contiguity registered in both specifications (aand b) confirms Jaffe et al.’s (1993) results and shows that the innova-tion process is deeply rooted in a given territory and that knowledgespillovers easily overcome regional borders. The coefficient of the geo-graphic distance is larger than not only the coefficient of the functionaldistance (not significantly different from zero) but also the coefficientof the sectoral distance: in the innovation process space does matter.

● Regression IV looks at student flows within the Erasmus programme.As already explained, since in this chapter we are focusing on knowl-edge flows, we consider the region in which the ‘hosting’ university islocalised as the ‘emitting’ region of the knowledge flows embeddedin the ‘learned’ student returning to his/her original and ‘receiving’region after their studies. Regional GDP and R&D intensitycoefficients are significant and positive. Larger, richer and techno-logically advanced regions are more aware of the advantages of aninternational education process and more involved in this pro-gramme. The coefficients on both functional and sectoral distanceare negative and significant, while the coefficient of geographical dis-tance is insignificant. Taken together, this may be interpreted asshowing that the Erasmus programme does foster the geographicalmobility of European students but not as much as cohesion and con-vergence of the scientific level of European regions. Geographicaldistance does not influence the flow of Erasmus students; however,students from top regions (in terms of their respective RSII) tend tostudy abroad in ‘better’ foreign regions than their counterpartscoming from the bottom regions.

In every model shown in Table 11.4 country dummies – included in theestimation to take into account the institutional factors of the emitting andreceiving regions which may be determined by national characteristics –record significant coefficients. The regression constant – in the gravitationalmodels literature – refers to a regional fixed effect which is sometimes inter-preted as an indirect measure of remoteness (that is, the distance of oneregion to every other region). If one accepts this reasoning, our resultssupport the common wisdom that peripheral (in a geographical sense)regions are also peripheral in a functional sense and that knowledge and

250 Spatial systems

information flows have a hierarchically segmented structure with limitedevidence of a ‘filtering-down process’.

7. CONCLUSION

The reduction of regional disparities has been one of the main targets of EUpolicies since its very beginning. However, the digital revolution has givennew meanings to this concept. Per capita GDP and unemployment rates arestill relevant economic indicators, but so are knowledge and ICT indicators(in terms of endowments, access and use). This chapter – which focuses onthe structure of knowledge flows as measured by four distinct but comple-mentary variables (Internet hyperlinks, research networks, EPO co-patentapplications and Erasmus student mobility) – has underlined the intrinsicrelational nature of knowledge. We showed that there is a positive corre-lation between knowledge exchange flows and that these flows are influencedby different types of distance: geographical, functional and sectoral.

NA techniques showed that Erasmus student flows and Internet hyper-links have a more hierarchical structure in their outdegree than in their in-degree. These results confirm the existence of a polarised centre–peripheryhierarchy of European regions which is reflected in the structure of knowl-edge flows.23 The NA perspective showed that although the co-patentnetwork displays some international relations connecting European regions,co-patenting still remains a mainly intra-national activity, mostly connect-ing regions in the same nation.

By using a ‘gravitational’ model we demonstrated that, far from the claimof the ‘death of distance’, geographical distance is still relevant for deter-mining the structure of inter-regional knowledge flows.

Functional and sectoral distance also play a crucial role, suggesting thatknowledge flows easily between similar (according to their scientific, tech-nological and sectoral characteristics) regions. Convergence between less-developed regions towards income levels of richer regions in the EU is thushampered by the observed network dynamics.

If the EU intends to build a ‘truly European’ research area in which thenetworking of ‘centres of excellence’ acts as a ‘catalyst for backward areas’,this target may still be far away.

NOTES

* The authors would like to thank S. Beretta, B. Dettori, M. Nosvelli, M. Riggi, G. Turatiand S. Usai for fruitful discussions, and M. Gioè for research assistance. The following

Inter-regional knowledge flows in Europe 251

people and institutions have been helpful in the process of unpublished data gathering:N. Poupard and P. Vareschi (UK Socrates–Erasmus Council, London), M. Preda (Istitutodi studi su popolazione e territorio, Università Cattolica del Sacro Cuore, Milan); S. Usaiand B. Dettori (Centre for North South Economic Research (CRENoS), Università degliStudi di Cagliari) and M. Thelwall (School of Computing and Information Technologies,University of Wolverhampton). We acknowledge the financial support of the Ministry ofEducation, University and Research (MIUR) under the Research Project ‘Dinamicastrutturale: tecnologie, reti, istituzioni’ (2003131274_001).

1. While it is possible for a single website to count and map all access, it is extremely difficultto collect this information on a wider scale since it would involve the cooperation of theweb masters of all websites.

2. A further confirmation of the informational content of Internet hyperlinks relates to thefact that several search engines (and in particular the popular Google) use hyperlinkcounting as ranking criteria of web pages since they consider it a good proxy of thequality and relevance of the web page.

3. A new discipline (webmetrics) devoted to the analysis of the www using bibliometric pro-cedures has been created, as well as a scientific association (the International Society forScientometrics and Informetrics, ISSI) (Maggioni and Uberti, 2005).

4. The retrieval of Internet hyperlinks – following Thelwall and Smith (2002) – was run in2003 using Altavista, a public search engine.

5. Cooperative research contracts, coordination of research actions, cost-sharing contracts,demonstration contracts, explanatory awards, explanatory awards (demonstration),explanatory awards (thematic network), preparatory, accompanying and support meas-ures, research grants (individual fellowship), research network contracts, study con-tracts, assessment contracts and thematic network contracts. Some of these contracts areassigned to single applicants (that is, research grants), while others require the creationof research networks among the participants (that is, research and thematic networkcontracts).

6. A patent registered by three inventors located in three distinct regions i, j and z, wouldbe split into n*(n – 1) cells and, respectively, into i with j and z, j with i and z, and z withi and j. Hence an invention co-patented by three individuals in three different regions isregistered with a value of 0.1666 in the cells corresponding to six different couplets.

7. The EU has devoted a new programme to post-graduate student exchange(Erasmus–Mundus); it started in 2002 and has not yet been adequately monitored, hencein this analysis we focus only on the Erasmus student flows.

8. The Socrates programme is the European programme for education, and includes eightactions; it was developed ‘to promote the European dimension and to improve thequality of learning by encouraging cooperation between the participants’ countries’(European Commission, 2005).

9. The data are courtesy of the UK Socrates–Erasmus Council, the UK National Agencyresponsible for the administration of the Erasmus programme in the UK.

10. Data kindly provided by CRENoS.11. The RSII indicators are: population with tertiary education; participation in lifelong

learning; employment in medium–high and high-tech employment; employment in high-tech services; public R&D; business R&D; EPO high-tech patent applications; EPOpatent applications; share of innovative enterprises in the manufacturing and servicesectors; innovation expenditures in manufacturing and in services; and sales of ‘new tothe firm but not new to the market’ products.

12. Sometimes a regional border in our sample may also be a national border, and bordersare a significant variable in many empirical papers based on the gravitational model.However, we did not distinguish between these two cases since, with the joint use of thecontiguity and country dummies, we are able to identify these cases.

13. With the exception of a Spearman correlation between co-patents and digital information.14. The lower value for the Spearman correlation could be partially explained by the typol-

ogy of the research networks considered in this study, which excludes those coordinatorsthat are not universities.

252 Spatial systems

15. These correlations were calculated using binary matrices dichotomised according to theaverage of raw matrices: the cell ij value above the mean would be registered 1 and 0otherwise.

16. See ‘Research networks’ in Section 2.17. For symmetric networks with undirected edges, the density is calculated as follows:

18. A value equal to 1 is substituted for the actual value of the edge when it is greaterthan or equal to the cut-off; and 0 when the actual value is smaller than the cut-off.The use of valued versus unvalued networks is widely discussed in the literature(Wasserman and Faust, 1994). In the econometric analysis performed in Section 6, net-works have been used in their valued (that is, containing all different numerical values)version, while in this section networks are dichotomised according to their average.

19. If both degree centrality (for the single node) and centralisation (for the whole system)indices are used on a directed network, then it must be stressed that inward and outwardmeasures (relative to the inward and outward links of a node) are, in general, not equal.In the chapter, therefore, centrality and centralisation indices – without any furtherspecification – identify the outward measure of the indices.

20. Since we are mainly interested in the significance and signs of the coefficient, simple OLSestimation provides valid results. Alternative estimation procedure (either count datamodels or OLS with Box–Cox transformation) would allow detailed analysis of thecoefficient values.

21. In terms of international ranking of its research output.22. This is reinforced by the fact that the coefficient of the geographical distance is larger

than the coefficient of the functional distance and (slightly) the coefficient of the sectoraldistance.

23. A larger number of regions send their students abroad, but their destination is concen-trated in a small number of regions. Also, it is easier to be the origin of a hyperlink thanto be the target.

REFERENCES

Albert, R. and Barabasi, A.L. (2002), ‘Statistical mechanics of complex networks’,Reviews of Modern Physics, 74, 47–99.

Anderson, J.E. (1979), ‘A theoretical foundation for the gravity equations’,American Economic Review, 69, 106–16.

Bergstrand, J.H. (1985), ‘The gravity equation in international trade: some micro-economic foundations and empirical evidence’, Review of Economics andStatistics, 71, 474–81.

Borgatti, S.P., Everett, M.G. and Freeman, L.C. (2002), UCINET 6 for Windows:Software for Social Network Analyis, Cambridge, MA: Harvard, AnalyticTechnologies.

Bramanti, A. and Maggioni, M.A. (1997), ‘The dynamics of milieux: the networkanalysis approach’, in Ratti, R., Bramanti, A. and Gordon, R. (eds), TheDynamics of Innovative Regions, Aldershot: Ashgate pp. 321–41.

Breschi, S. and Cusmano, L. (2004), ‘Unveiling the texture of a European researcharea: emergence of oligarchic networks under EU framework programmes’,International Journal of Technology Management, 27, 747–72.

Breschi, S. and Lissoni, F. (2004), ‘Knowledge networks from patent data: method-ological issues and research targets’, CESPRI Working Papers 150, Centre ofResearch on Innovation and Internationalization (CESPRI), Milan.

D � 2e� [n(n � 1)].

Inter-regional knowledge flows in Europe 253

Dalgin, M., Mitra, D. and Trindade, V. (2004), ‘Inequality, nonhomothetic prefer-ences, and trade: a gravity approach’, NBER Working Paper Series 10800,National Bureau of Economic Research, Cambridge, MA.

Deardorff, A. (1998), ‘Determinants of bilateral trade: does the gravity work in aneoclassical world?’, in Frankel, J.A. (ed.), The Regionalization of the WorldEconomy, Chicago: University of Chicago Press, pp. 7–32.

DG Enterprise (2003), ‘2003 European Innovation Scoreboard’, Technical PaperNo. 3, Regional innovation performances, http:// trendchart.cordis.lu/, accessedJanuary 2005.

European Commission (2000), ‘The Lisbon strategy for economic, social and envi-ronmental renewal’, available at http://europa.eu.int/comm/lisbon_strategy/index_en.html, accessed February 2005.

European Commission (2005), http://europa.eu.int/comm/education/index_en.html,accessed: February 2005.

Feenstra, R.C. (2002), Border effects and the gravity equation: consistent methodsfor estimation’, Scottish Journal of Political Economy, 49, 491–506.

Feldman, M.P. (2002), ‘The Internet revolution and the geography of innovation’,International Social Sciences Journal, 54, 47–56.

Griliches, Z. (1981), ‘Market value, R&D and patents’, Economic Letters, 7, 183–7.Griliches, Z. (1990), ‘Patent statistics as economic indicators: a survey’, Journal of

Economic Literature, 28, 1661–707.Helpman, E. (1988), ‘Imperfect competition and international trade: evidence from

fourteen industrial countries’, in Spence, A.M. and Hazard, H.A. (eds),International Competitiveness, Cambridge, MA: Ballinger.

Jaffe, A.B., Henderson, R. and Trajtenberg, M. (1993), ‘Geographic localization ofknowledge spillovers as evidenced by patent citations’, Quarterly Journal ofEconomics, 108, 557–98.

Krugman, P. (1991), Geography and Trade, Cambridge, MA: MIT Press.Leoncini, R., Maggioni, M.A. and Montresor, S. (1996), ‘Intersectoral innovation

flows and national technological systems: network analysis for comparing Italyand Germany’, Research Policy, 25, 415–30.

Maggioni, M.A. (1993), ‘Network analysis of regional industrial dynamics andlocal economic policies’, paper presented at the 5th Society for theAdvancement of Socio-Economics (SASE) conference, New York, 25–28March.

Maggioni, M.A. (2000), ‘Intersectoral innovation flows within and between nationsand regions: network analysis and systems of innovation’, in Punzo, L., Farina,F. and Fabel, O. (eds), European Economies in Transition, London: Macmillan,pp. 148–73.

Maggioni, M.A. and Uberti, T.E. (2005), ‘Webmetrics’, in Pagani, M. (ed.),Encyclopedia of Multimedia Technology and Networking, London: Idea GroupInc., pp. 1091–95.

Maggioni, M.A. and Usai, S. (2005), ‘Patents as relations: the organisation and theevolution of the innovative activity in two European countries’, paper presentedat the Open Conference on Knowledge and Regional Economic Development,Barcelona, 9–11 June.

Martin, P. (1998), ‘Can regional policies affect growth and geography in Europe?’,World Economy, 21, 757–74.

Moreno, R., Paci, R. and Usai, S. (2005), ‘Spatial spillovers and innovation activ-ity in European regions’, Environment and Planning A, 37: 1793–1812.

254 Spatial systems

Paci, R. and Usai, S. (2000), ‘Technological enclaves and industrial districts. Ananalysis of the regional distribution of innovative activity in Europe’, RegionalStudies, 34, 97–104.

Poyhonen, P. (1963), ‘A tentative model for the volume of trade between countries’,Weltwirtschaftliches Archiv, 90, 93–9.

Sapir, A., Aghion, P., Bertola, G., Hellwig, M., Pisani-Ferry, J., Rosati, D., Viñals,J. and Wallace, H. (2004), An Agenda for a Growing Europe: The Sapir Report,Oxford: Oxford University Press.

Scherer, F.M. (1965), ‘Firm size, market structure, opportunity and the output ofpatenting inventions’, American Economic Review, 55, 1097–125.

Snyder, D. and Kick, E.L. (1979), ‘Structural position in the world system and eco-nomic growth, 1955–1970: a multiple network analysis of transnational inter-actions’, American Journal of Sociology, 84, 1096–126.

Thelwall, M. and Smith, A. (2002), ‘Interlinking between Asia-Pacific universityweb sites’, Scientometrics, 55, 335–48.

Tinbergen, J. (1962), Shaping the World Economy: Suggestions for an InternationalEconomic Policy, New York: 20th Century Fund.

Uberti, T.E. (2002), ‘Flussi commerciali, tecnologici ed informativi: elementi perun’analisi strutturale del processo di globalizzazione’, doctoral thesis, Milano:Università Cattolica del Sacro Cuore.

Uberti, T.E. and Maggioni, M.A. (2004), ‘Infrastrutture ICT e relazionalità poten-ziale. Un esercizio di “hyperlinks counting” a livello sub-nazionale’, QuadernoDiseis 0402, Milano: Università Cattolica del Sacro Cuore.

Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods andApplications, Cambridge: Cambridge University Press.

Watts, D. (1999), Small Worlds: The Dynamics of Networks between Order andRandomness, Princeton, NJ: Princeton University Press.

Wikipedia (2005), ‘Low of universal gravitation’, http://en.wikipedia.org/wiki/Law_of_universal_gravitation, accessed April 2005.

Inter-regional knowledge flows in Europe 255

12. Explaining the territorial adoptionof new technologies: a spatialeconometric approachAndrea Bonaccorsi, Lucia Piscitello andCristina Rossi*

1. INTRODUCTION

The idea that information and communication technology (ICT) will reducethe economic importance of geographic distance has been expounded ener-getically in post-Internet literature (Cairncross, 2001). According to thisview, the New Economy works in a space rather than a place, transport costswill be drastically reduced, distance will be less important, and peripheralregions will benefit from opportunities that are not available in an economybased on manufacturing industry (Negroponte, 1995; Kelly, 1998;Compaine, 2001). Since it is mostly based on non-material and humancapital investment, ICT offers new areas of potential growth to regions orareas that have historically suffered from isolation, high transport costs ora lack of private and public physical infrastructure. Consequently, accord-ing to this view, the concentration of income opportunities and wealthshould decrease over time. Although other predictions have also been madein the debate over the impact of the digital economy (for example, UNDP,2001; Norris, 2002), this view is still prevalent.

However, the reality is not so rosy. Not only are there huge disparities inthe intensity with which ICT is adopted across countries, but big differencesstill persist within industrialised countries. Indeed, differences in economicdevelopment still shape the rate of the adoption of these technologies, atfirm, regional and national levels. The reasons for these stylised facts havebeen investigated at length in recent years.

This chapter contributes to current literature in several ways. First, itfocuses on intra-national or regional differences, which is a much lessexplored dimension of the ‘digital divide’. Second, it uses a new metric forICT adoption, namely the number of second-level Internet domain namesregistered under the country code Top Level Domain (ccTLD) ‘.it’. Finally,

256

it explicitly combines the analysis of determinants with a spatial econo-metric approach. The chapter is organised as follows. Section 2 surveys theliterature on the digital divide and the relation between local developmentand adoption of ICT. Section 3 describes data and methodology. Section 4contains a description of the model and the empirical results. Section 5summarises the main conclusions of the chapter.

2. THE LOCAL DIGITAL DIVIDE: THE RELATIONBETWEEN DEVELOPMENT AND ICT ADOPTION

The conceptual link between economic development and ICT adoption isa widely researched issue in economics literature. It can be claimed that thenature of ICT makes it possible to overcome territorial peripherality.Unlike traditional heavy and light manufacturing investment, ICT mayincrease regional attractiveness as a strategic location factor, thus enhanc-ing territorial competitiveness (Gillespie et al., 1989; Kraemer andDedrick, 1996; Steinmuller, 2001; Camagni and Capello, 2005). The suc-cessful experiences of Ireland and India as emerging regions in the pro-vision of software services, due to the availability of efficientcommunication infrastructures, are often quoted. Contrary to most expec-tations, however, the overall empirical reality is one of big geographicdifferences in the rate of ICT adoption, which seems to reinforce ratherthan reduce disparities and inequalities.1

Most studies have revealed astonishing differences in Internet and com-puter penetration between North America and Europe on the one hand andAfrican and Asian countries on the other (see Chinn and Fairlie, 2004, fora comprehensive survey of this literature). These disparities have mainlybeen explained in relation to differences in income, but also in humancapital, telecommunication infrastructures (Dasgupta et al., 2001;Oyelaran-Oyeyinka and Lal, 2003; Pohjola, 2003; Wallsten, 2003), demo-graphic variables and regulatory regimes (Wallsten, 2003).2 Although theseexplanations are fairly convincing, it is puzzling as to why there is still scantevidence of a process of convergence on the part of less-developed countriesin the adoption of these technologies.

Less investigation has been devoted to the local dimension of the phe-nomenon; indeed digital inequalities do not divide only developed fromdeveloping countries but also regions within the same country (‘local digitaldivide’; see, for instance, Gareis and Osimo, 2004; Ramsay, 2004). Bothdeveloped and developing countries suffer from severe regional disparitiesin ICT adoption. Evidence has been gathered with reference to the UnitedStates (NTIA, 2002; Mills and Whitacre, 2003), Canada (Dryburgh, 2001),

Explaining the territorial adoption of new technologies 257

Portugal (Nunes, 2004), Spain (Billon Curras and Lera Lopez, 2004), Italy(Bonaccorsi et al., 2002; Assinform, 2004) and China (Qingxuan andMingzhi, 2002; Wensheng, 2002).

One clear-cut stylised fact that emerges from this literature is thatregional disparities are at least as important as cross-country differences,at least within industrialised nations. In Italy, for instance, Bonaccorsi etal. (2002) found that the geographic concentration of Internet adoptionis much greater than that of population or income. This would seem tosuggest that, far from reducing regional disparities, ICT actually re-inforces them.

Empirical studies have shown that the determinants of local inequalitiesare associated with economic, social and demographic disparities. In par-ticular, differences in the spatial diffusion of ICT have been explained interms of differences in technological levels, infrastructural endowments(Marrocu et al., 2000; Iammarino et al., 2004) and local spillover effects(Jaffe et al., 1993; Audretsch and Feldman, 1996; Galliano and Roux,2004). However, local inequalities might also be influenced by spatialfactors. Investigating the geography of second-level domain names inPortugal (.pt), Nunes (2004) recently proposed that the Internet might con-tribute to reinforcing the tendency to territorial disintegration, promotinggeographic disparities in a more pronounced way than is the case in the realeconomy space. He found that the role of ICT in overcoming spatialinequalities in Portugal is less important than expected, since these tech-nologies, far from changing the existing spatial structure, are deeplyinfluenced by it. The importance of a spatial perspective for the analysis ofInternet adoption has been addressed by Aztema and Weltevreden (2004)and by Weltevreden et al. (2005), who combine a conventional innovation-adoption approach with a detailed geographical analysis to study the adop-tion of business-to-consumer e-commerce by Dutch retailers.

In line with the most recent studies, mainly framed within models oftechnology diffusion (Geroski, 2000), several groups of factors thatpotentially influence the territorial adoption of ICT were distinguished(for an excellent recent survey, see OECD, 2004). One category of factorsthat are positively related to ICT adoption concerns the local technolog-ical endowment and the relevant absorptive capacity. The latter refersboth to the ability of firms to assess technological opportunities (whichdepends on their endowment of human and knowledge capital, Cohenand Levinthal, 1989), and to learning effects. The former may arisefrom earlier use of ICT or a predecessor of a specific ICT element whichalready embodies constituent elements of later applied, more advancedvintages (McWilliams and Zilberman, 1996). In addition, according toHollenstein (2004: 41).

258 Spatial systems

[T]hese aspects of absorptive capacity refer to the standard epidemic model oftechnology diffusion and to the relevant information spillovers from users to nonusers of the technology. This model basically states that a firm’s propensity toadopt a technology at a certain point in time is positively influenced by thepresent (or lagged) degree of its diffusion in the economy as a whole or in theindustry to which the firm is affiliated.

A second category of variables concerns market characteristics. It hasbeen amply demonstrated that the sectoral specialisation of a regionimpacts significantly upon the adoption of ICT (Pohjola, 2003).

Likewise, the characteristics of firms have traditionally been employed asexplanatory variables in most studies on adoption. In particular, firm sizecaptures the Schumpeterian hypothesis about the positive relation betweeninnovativeness and dimensional scale. The same holds for firm age, althoughthe theoretical arguments are not conclusive (positive experience effectsversus negative adjustment cost effects in the case of older firms, see Lal,2001; Hollenstein, 2004). The adoption of ICT may also be affected by themarket conditions in which firms are operating, particularly the competitivepressure to which they are exposed. In markets where competition isstronger, firms are expected to be more inclined to innovative activities orrapid technology adoption (Porter, 1990; Majumdar and Venkataraman,1993; Feldman and Audretsch, 1999; Hollenstein, 2004).

Finally, we explicitly take into account the role that spatial externalitiesplay in current thinking about innovative activity (see Audretsch, 2003).

3. METHODOLOGY AND DATA

Domain Names as a Proxy for ICT Adoption

The term ICT encompasses a wide range of technologies. According tothe Canadian Statistics Bureau, it ‘includes desktop and laptop comput-ers, software, peripherals and connections to the Internet that areintended to fulfil information processing and communications func-tions’.3 Such a variety poses severe methodological problems when itcomes to measuring the level of territorial adoption of these assets.According to Pohjola (2003), two kinds of metrics reveal disparities inICT adoption across countries: data on ICT equipment and its use, andindicators of ICT spending.

Most of the studies that have analysed geographical inequalities at theinternational level have identified ICT with the Internet, focusing on thenumber of Internet hosts (OECD, 2001; Kiiski and Pohjola, 2002)and users (Norris, 2002; NTIA, 2002),4 although restricting the issue of

Explaining the territorial adoption of new technologies 259

differences in ICT adoption simply to Internet access is misleading (Odenand Rock, 2004). Indeed, while data regarding Internet hosts are readilyavailable and highly reliable (Press, 1997; Wolcott et al., 2001),5 this metricsuffers from two major shortcomings: the data are gathered only at anational level and they do not provide any information about the adopters.

Regional-level analysis benefits from the availability of larger sets ofindicators, ranging from the share of sales of electronic goods to mobilephones; survey data are also available.6

Recently, the use of domain names as a proxy of Internet diffusion hasbeen proposed (Zook, 2000; Zook et al., 2004). Domains may be a validproxy for ICT adoption, mainly because they operationalise the intentionto actively supply contents through the Net. Those who register a domainname use the Internet in a more conscious manner, with a view not only todemanding but also to adding content to it.7 In general, the registration ofa domain name by a firm is the first step towards setting up a websitethrough which to present their products or even to undertake electroniccommerce activities. Domains therefore provide an underestimation ofICT adoption8 as: (i) ICT adoption does not necessarily involve registeringa domain; and (ii) Internet service providers (ISPs) often offer their usersroom (on their servers) for adding new content. Thus, domains constitutea lower bound as any registrant is unquestionably an ICT adopter.Additionally, every domain name is uniquely associated with a registrantwhose geographical location and nature are unambiguously recorded in thedatabases of the organisations that manage the ccTLD (Mueller, 1998;Grubesic, 2002). The availability of information at the subnational levelmakes domains a valid metric to explore the territorial dimension of ICTadoption, while data on the nature of the registrants make it possible totake into account different adoption determinants for different populationsof potential adopters.

In the present study, domain name registrations by Italian firms are usedas a proxy for ICT adoption at the NUTS3 (Nomenclature des UnitésTerritoriales Statistiques) level (103 provinces). In the 2002–03 period, theInstitute of Informatics and Telematics (IIT) of the National ResearchCouncil (CNR), the Sant’Anna School of Advanced Studies and theUniversity of Pisa built up a database of domain name registrations, organ-ised on a subregional basis into different categories of actors (individuals,business firms, universities and research centres, third sector associationsand local government bodies). Data were extracted from the databases ofthe registrations under the ccTLD ‘.it’ that are managed by the ItalianRegistration Authority (RA) hosted by IIT. A total of 500 000 domainnames were inspected for classification; multiple names registered by thesame registrant were carefully checked and eliminated.

260 Spatial systems

The Empirical Evidence on ICT Adoption in Italy: The North–South Divide

In order to use domain name registrations as a proxy for the level of ICTadoption, the penetration rate in each province was calculated as the per-centage of firms in the province that had at least one domain name regis-tered in the RA databases as of July 2001. Table 12.1 summarises thedescriptive statistics of the variable.

As expected, the data show the existence of a North–South divide. Theregions of Italy are highly differentiated from social, cultural, demographicand economic points of view, and ICT adoption is no exception. Thefigures in Table 12.2 reveal that the level of ICT adoption in Italy is quitelow, with an average penetration rate of less than 4 per cent, and thatdifferences between the macro areas are highly significant (theKruskall–Wallis test is significant at p � 0.01).

No southern province ranks in the top 50. The best-performing provincein the South ranks 55th, with just eight northern provinces below that po-sition. Conversely, the 20 worst-performing provinces are all located in theSouth (see Table 12.3 for the top 10 and bottom 10 positions). The geo-graphical disparities in ICT adoption mirror the inequalities in economicdevelopment (measured by value added per employee) both among andwithin geographical macro areas (Figure 12.1).

As the analysis of the map in Figure 12.1 suggests a possible spatial auto-correlation with regard to the provinces located in the North–Centre, theoverall pattern in the local spatial pattern was disaggregated by means ofthe local indicator of spatial association (LISA). In fact, Figure 12.2 showsthe emergence of 3 H-H clusters, all located in the North–Centre, and 2 L-L

Explaining the territorial adoption of new technologies 261

Table 12.1 ICT adoption: descriptive statistics

ICT No. Min Max Mean Std Skewness Kurtosisadoption dev.

103 1.20 9.11 3.76 1.65 0.42 2.72

Table 12.2 ICT adoption in macro areas

Area No. Mean Std dev. Kruskal–Wallis test – p-value

North 46 4.76 1.31Centre 21 4.40 1.29South 36 2.11 0.66 0.000

Total 103 3.76 1.65

clusters in the South. This is in line with Fabiani et al. (2003), who foundvery big differences between firms in the South of Italy and in the Northand Centre in the rate of adoption of almost all ICTs. Iammarino et al.(2004) highlight the same divide as far as the production of ICT is con-cerned. In addition, Figure 12.2 also evidences that ICT adoption is morelocally concentrated than added value per employee.9

Following the literature on the spatial distribution of innovation(Audretsch and Feldman, 1996; Audretsch, 2003), spatial dependence wasexpected to exist between the observations. As Le Sage notes, ‘spatialdependence in a collection of sample data observations refers to the factthat one observation associated with a location which we might label idepends on other observations at locations j � i’ (Le Sage, 1998: 3).

Table 12.4 reports the results of tests normally used for detecting spatialdependence.10 ICT adoption is the percentage of firms in each province that

262 Spatial systems

Table 12.3 Ranking of Italian provinces by ICT adoption and per capitaincome (PCI) in 2001 top 10 and bottom 10 provinces

Ranking Ranking Province Region Macro region ICT ICT PCI (NUTS3) (NUTS2) (NUTS1) adoptionadoption

1 1 Milano Lombardia North 9.12 4 Firenze Toscana Centre 7.13 53 Bolzano Trentino Alto North 6.7

Adige4 26 Lecco Lombardia North 6.55 2 Bologna Emilia Romagna North 6.56 17 Torino Piemonte North 6.47 21 Varese Lombardia North 6.38 25 Udine Friuli Venezia North 6.2

Giulia9 28 Como Lombardia North 6.2

10 12 Roma Lazio Centre 6.2

94 82 Benevento Campania South 1.595 95 Potenza Basilicata South 1.596 101 Agrigento Sicilia South 1.597 84 Brindisi Puglia South 1.498 91 Matera Basilicata South 1.499 103 Caltanissetta Sicilia South 1.3

100 100 Vibo Valentia Calabria South 1.3101 102 Crotone Calabria South 1.3102 98 Enna Sicilia South 1.2103 75 Campobasso Molise South 1.2

had at least one domain name registered as of 2001. All three tests confirmthe existence of a spatial dependence, and we can conclude that the adoptionof ICT by each province i is related to the adoption of other provinces j � i.This shows that geography matters.

Explaining the territorial adoption of new technologies 263

Figure 12.1 Distribution of ICT adoption and added value per inhabitantacross Italian provinces: the ‘Three Italies’

North North

Centre Centre

South

Std Deviation: ICT_END Std Deviation: VAA

South

1.20–2.11 (19)

2.11–3.77 (37)

3.77–5.42 (26)

5.42–7.08 (19)> 7.08 (2)

9 704.00–12 978.30 (25)

12 978.30–17 276.21 (19)

17 276.21–21 574.13 (44)

21 574.13–25 872.04 (13)> 25 872.04 (2)

4. ECONOMETRIC MODELS OF TERRITORIAL ICTADOPTION

First of all a standard cross-section specification was estimated. Using thevariables mentioned in Section 2, ICT adoption was modelled as a func-

264 Spatial systems

Figure 12.2 ICT adoption and added value per employee, univariate LISA

LISA Chister Map

ICT adoption Added value per employee,LISA

Not Significant

High–Low

High–High

Low–Low

Low–High

Table 12.4 Spatial dependence tests for the dependent variable (ICT adoption)

Moran’s I E(I) Sd(I) z�

ICT adoption 0.589 �0.010 0.064 9.385 ***Geary’s c c E(c) Sd(c) z�ICT adoption 0.480 1.000 0.080 �6.494 ***Getis & Ord’s G G E(G) Sd(G) z�ICT adoption 0.053 0.044 0.002 6.001 ***

Note: � two-tail test; *** significant at p�0.01.

tion of a province’s absorptive capacity, the characteristics of firms, thesectoral composition of the province and technological endowment.Checks were also made for urbanisation economies. A dummy wasinserted to address the North–South divide. The proxies employed, andtheir statistical descriptives and correlations, are reported in Tables 12.5and 12.6.

Explaining the territorial adoption of new technologies 265

Table 12.5 Specification of dependent and independent variables

Variables Description Source

Dependent variableICT ADOPTION Percentage of firms that have registered Registration

at least one domain name authority for the ccTLD ‘.it’ – elaboration

Explanatory variablesAbsorptive capacity

PATENTS Ratio of the number of patents granted USPTO – by the United States Patent and elaborationTrademark Office (USPTO) in each province in the 1991–99 period and the number of firms in that province

PUBLICATIONS Ratio between the number of scientific ISI Citation Indexpublications by university researchers databases – in each province and the number of elaborationfirms in that province

CompetitionCOMPETITION Percentage of districtual local units Infocamere –

elaboration

Characteristics of firmsSIZE Ratio of the number of employees and ISTAT

the number of firms in manufacturing

Sectoral compositionSTRUCTURE Percentage of firms in the primary sector. Infocamere –

This is a dummy variable that assumes elaborationvalue 0 if the province is below the national average, 1 otherwise

Technological endowmentINFRASTRUCTURE Facilities and networks for telephony and Istituto Tagliacarne

telematics (Index of endowment, Italy � 100)

ControlsDUMMY METRO- Dummy variable, value 1 for Our elaborationPOLITAN metropolitan, 0 otherwise

DUMMY NORTH Dummy variable, value 1 for northern Our elaborationprovinces, 0 otherwise

266

Tab

le 1

2.6

Cor

rela

tion

mat

rix

Var

iabl

eS

IZE

PAT

EN

TS

INF

RA

ST

RU

CT

UR

EP

UB

LIC

AT

ION

SC

OM

PE

TIT

ION

SIZ

E1.

000

PAT

EN

TS

0.52

01.

000

INF

RA

ST

RU

CT

UR

E0.

256

0.42

81.

000

PU

BL

ICA

TIO

NS

0.18

80.

305

0.37

81.

000

CO

MP

ET

ITIO

N0.

378

0.25

60.

227

0.00

11.

000

Table 12.7 reports the results from two ordinary least squares (OLS)specifications; as they are highly correlated, the variables PATENTS andINFRASTRUCTURE are included alternately. As expected, ICT adoptionis positively influenced by:

● Local absorptive capacity: the coefficient of the variable PATENTSis positive and significantly different from zero at p � 0.10, asis the one for PUBLICATIONS, which is significant at least atp � 0.05.

Explaining the territorial adoption of new technologies 267

Table 12.7 Standard OLS models

Variable Model

OLS 1 OLS 2Coefficient Coefficient

Constant 1.174* �0.207(1.797) (�0.350)

PATENTS 0.464*(1.594)

PUBLICATIONS 0.352*** 0.256**(2.899) (2.326)

COMPETITION 0.011*** 0.008***(3.781) (3.144)

SIZE 0.276 0.479***(1.481) (3.056)

STRUCTURE 0.199 �0.116(0.993) (0.614)

INFRASTRUCTURE 0.011***(5.223)

DUMMY NORTH 1.602*** 1.408***(7.120) 6.907

DUMMY METROPOLITAN 1.975*** 1.260***(5.349) (3.511)

Adj. R-squared: 0.749 0.800F-statistic 44.477*** 59.196***Log likelihood �122.698 �111.059Akaike information criterion 261.396 238.118Schwarz criterion 282.474 259.196Moran’s I (error) 1.817* 2.224**Robust LM (lag) 0.999 0.001Robust LM (error) 0.000 1.324

Note: Standard errors in brackets; *** p-value�0.01; ** p-value�0.05; Obs.�103.

● Characteristics of firms: SIZE is positive, although it is significant atp � 0.01 only in the second specification.

● Sectoral composition of the province: this seems to play no role inexplaining the dependent variable, although it is worth noting thatit is strongly and positively correlated with the variable SIZE(Pearson correlation index 0.517, p-value � 0.000); we might there-fore hypothesise a positive contribution of STRUCTURE to ICTadoption.

● Market conditions: the variable COMPETITION is positive andhighly significant (p � 0.01), revealing that competitive pressure iscrucial in stimulating ICT adoption at the local level. However, theproxy utilised (percentage of districtual local units per province)might also be considered a proxy for imitation processes among firmsin their use of new technologies.

● Technological endowment: as expected, technological infrastructuresdedicated to telephony and telematics (INFRASTRUCTURE) posi-tively and significantly affect (at p � 0.01) ICT adoption in theprovince.

Given that in Italy there is a significant North–South divide in social, cul-tural, demographic and economic terms, the dummy variable NORTHhighlights spatial heterogeneity in ICT adoption. Likewise, the dummy formetropolitan areas reveals that ICT adoption is affected by urbanisationeconomies, which is in line with the literature on the role of cities in thediffusion of the Internet (see, for instance, Zook, 1999).11

It is worth observing that even when controlling for spatial heterogen-eity (through the inclusion of the dummy variable), the diagnostics forspatial dependence signal the possible existence of spatial autocorre-lation. In both specifications, Moran’s I tests proved significant at least atp � 0.10, although none of the robust LM tests was significant.12

Therefore, following Anselin (1988, 2004) and Anselin and Moreno(2003), spatial lag and spatial error models were run in order to dis-entangle the spatial effect.

The results of the spatial models, reported in Table 12.8, show that thecoefficient ‘lambda’ is highly significant at least at p � 0.05, suggesting thatspatial dependence might work through omitted variables with a spatialdimension (that is, that the errors from different provinces are spatially cor-related, see Abreu et al., 2004). The spatial lag of the dependent variable(W_ICT) also turned out to be significant (at p � 0.05), although only inthe first specification, revealing that the adoption of new technology in agiven province depends not only on the values of the explanatory variablesin the province, but also on ICT adoption in other provinces.

268 Spatial systems

5. CONCLUSIONS

This chapter contributes to the literature on the determinants of the adop-tion of new technologies at the local level in several ways. First, it corrob-orates some robust findings in current literature. It was found that variablesdescribing the vitality of general economic activity are relevant. Economicenvironments with traditional economic activities are less vibrant in ICTadoption: the larger the proportion of firms in the primary sector, the lowerthe intensity of advanced-level Internet use. This general effect is reinforced

Explaining the territorial adoption of new technologies 269

Table 12.8 Spatial lag and spatial error models

Variable Model

Spatial lag 1 Spatial error 1 Spatial lag 2 Spatial error 2Coefficient Coefficient Coefficient Coefficient

Constant 1.017 1.979*** �0.230 0.599(1.559) (3.007) (�0.405) (0.888)

W_ICT 0.213** 0.153(2.117) (1.542)

PATENTS 0.385 0.534*(1.387) (1.867)

PUBLICATIONS 0.355*** 0.331*** 0.260** 0.249***(3.106) (2.993) (2.491) (2.582)

COMPETITION 0.010*** 0.010** 0.007*** 0.008***(3.406) (3.692) (2.861) (3.272)

SIZE 0.185 0.059 0.397*** 0.274*(1.056) (0.321) (2.6069) (1.675)

STRUCTURE 0.226 0.331* �0.081 0.0251.192 (1.691) (�0.451) (0.136)

INFRASTRUCTURE 0.0100*** 0.011***(5.141) (5.793)

DUMMY NORTH 1.190*** 1.652*** 1.117*** 1.407***(4.022) (6.312) (4.125) (5.618)

DUMMY 2.031*** 1.977*** 1.323*** 1.313***METROPOLITAN (5.848) (5.789) (3.872) (4.089)

Lambda 0.290** 0.369***(2.428) (3.286)

R-squared: 0.776 0.778 0.818 0.829Log likelihood �121.045 �121.213 �110.092 �108.568Akaike info criterion 260.09 258.427 238.184 233.137Schwarz criterion 283.803 279.505 261.896 245.215LR test 3.306* 2.969* 1.935 4.982**

Note: Standard errors in brackets; *** p-value�0.01; ** p-value�0.05; Obs.�103.

by a specific technological effect related to ICT: the higher the index oftechnological endowment, measured in terms of the local-level telecom-munications network, the greater the probability of advanced-level use ofICT. These findings corroborate the notion that very traditional, highly‘material’ investments play a major role in explaining the local digitaldivide. As anticipated in the literature on telecommunications investment(Biehl, 1982; Gillespie et al., 1989; Kraemer and Dedrick, 1996), regionaldevelopment may be adversely affected by disparity in the level of infra-structure. Contrary to expectations, the spatial diffusion seems to follow theexisting geography of development rather than dramatically changing it.Our results are also consistent with existing evidence on the geographicconcentration of ICT production and differences in the adoption of ICTby firms in Italy (Pagnini, 2002; Fabiani et al., 2003; Iuzzolino, 2003;Iammarino et al., 2004). It must be admitted that our data do not capturethe structure of ICT supply, but rather the structure of demand or utilis-ation. Firms are only part of the adoption process as described by our dataon domain names. At the same time, it is clear that general economicfactors and the localisation and activity of firms in these industries stronglyinfluence ICT utilisation in the business sector, in households and in societyat large.

Second, the adoption of ICT is strongly influenced by the level of knowl-edge available at the province level, as measured by the flow of patent regis-trations and scientific publications. This effect was related to the notionof absorptive capacity, and a clear analogy was drawn with the idea thatonly firms that invest in in-house R&D are able to capture externallycreated knowledge. According to our results, areas that are poor in generaltechnological activity and in research are less likely to make active use ofICT, thus suggesting the benefits from local effects of human capital ac-cumulation. While this effect may be intuitive for production activities, dueto input pooling and knowledge spillovers (Ellison and Glaeser, 1997;Pagnini, 2002), it is interesting to observe how important it is for the adop-tion of new technologies as well. In addition, the larger the proportion offirms in a province that is part of an industrial district, the more intense theadoption of ICT, thus confirming the positive impact of competitive pres-sure. This contributes to the debate about the ability of industrial districts(mainly based on small and medium-sized firms in traditional industries)to absorb ICT, but also casts some light on the role of the imitationprocesses that motivate firms to choose new technologies.

Third, a spatial econometric approach was specifically introduced for theanalysis of the relationship between the digital divide and the adoption ofnew technologies. Spatial proximity is very important as spillovers flowacross provinces. However, as the empirical literature has shown (Jaffe

270 Spatial systems

et al., 1993; Keller, 2002) that benefits from spillovers actually decline withdistance, peripherality is still expected to be an obstacle to ICT adoption.As a matter of fact, our empirical evidence regarding Italy shows that areasfar from centres suffer from severe difficulties in adjusting to new technol-ogy. Consequently, there is a need for models that include contiguitymatrices at further levels of spatial lags.

Finally, the crucial role of complementarities is well reflected in our data.The literature on the impact of ICT on productivity and economic growthhas strongly emphasised the crucial importance of the coexistence and co-evolution of investment in physical infrastructure and equipment, invest-ment in human capital and profound changes in organisational structuresand procedures in both the private and public sectors (Brynjolfsson, 1993;Brynjolfsson and Hitt, 1996; Bresnahan et al., 1999; Black and Lynch,2001; OECD, 2004).

NOTES

* We gratefully acknowledge Koen Frenken, an anonymous referee and the participantsat the 4th EMAEE Conference, Utrecht, May, 19–21 2005 for their insightful commentsand suggestions.

1. According to the OECD (2001), the ‘digital divide’ refers to ‘the gap between individu-als, households, businesses and geographic areas at different socio-economic levels withregard both to their opportunities to access information and communication technolo-gies (ICTs) and to their use of the Internet for a wide variety of activities’.

2. The cost of monthly connection to broadband services is estimated to be 0.9 per cent ofaverage income in Japan, 1.2 per cent in Belarus and 9.1 per cent in Cameroon(Fundación AUNA, 2004).

3. See www.statcan.ca/english/freepub/81-004-XIE/def/ictdef.htm.4. An analysis of cross-country diffusion of personal computers can be found in Caselli

and Coleman (2001).5. For instance, every six months Network Wizard publishes the results regarding all the

TLD on its website, whereas the RIPE (http://www.ripe.net) publishes the data aboutthe ccTLD in its area (Europe, North Africa, Middle East) on a monthly basis. Hostsbelong to the so-called ‘endogenous metrics’, which are obtained in an automatic orsemiautomatic way from the Internet itself (Diaz-Picazo, 1999). The organisationsthat manage the different ccTLD and gTLD perform the hostcount under theirTLD on a regular basis and provide these data on the Web or by File Transfer Protocol(FTP).

6. The bi-annual survey ‘A nation on line’, conducted on more than 3,000 US citizens(NTIA, 2002), collects data on the number of PCs purchased by families and the activ-ities they carry out on the Internet.

7. Domain grabbing must also be taken into account. However, this phenomenon does notaffect our data, as the unit of analysis is the registrant rather than the domain: multipleregistrations were discarded from the database.

8. It is worth observing that hosts suffer from the same drawback. Indeed, hostcount pro-grams do not reach private networks (Intranets) and machines protected by firewalls.The use of dynamic Internet Protocol (IP) addresses by ISPs should also be taken intoaccount. In addition, they are prone to overestimation due to a number of factors, forinstance the association of multiple IP addresses with the same computer.

Explaining the territorial adoption of new technologies 271

9. There is a considerable amount of literature about the spatial concentration of economicactivity (for a review of this literature, see Arbia, 2001).

10. The most general specification for the matrix was used, that of geographical distancebetween the centroids of provinces i and j (km). However, as the specification of thespatial weights is contentious in the literature, alternative contiguity matrices (Rook-based and Queen-based contiguity) were also tried, but the results did not change sub-stantially. The results of these alternative models have not been reported here, but areavailable on request.

11. A common proxy for urbanisation economies is population density, but this is not a validoption with regard to Italy. The whole country is highly populated (average populationdensity: 191.7 inhabitants per square kilometre), and when the provinces were ranked bypopulation density, it turned out that several of the top 10 ranking provinces do notcontain big cities.

12. Lagrange multiplier tests were used to assess the extent to which remaining unspecifiedspatial spillovers may be present in the estimation of expression (see Anselin, 1988;Florax and Nijkamp, 2004; Arbia, 2006).

REFERENCES

Abreu, M., de Groot, H.L.F. and Florax, R.J.G.M. (2004), ‘Spatial patterns of tech-nology diffusion: an empirical analysis using TFP’, Tinbergen InstituteDiscussion Paper 2004-079/3.

Anselin, L. (1988), Spatial Econometrics: Methods and Models, Dordrecht: KluwerAcademic Publishers.

Anselin, L. (2004), ‘Exploring spatial data with GeoDa: a workbook’, www.geoda.uiuc.edu/pdf/geodaworkbook.pdf.

Anselin, L. and Moreno, R. (2003), ‘Properties of tests for spatial error com-ponents’, Regional Science and Urban Economics, 33, 595–618.

Arbia G. (2001), ‘Modelling the geography of economic activities on a continuousspace’, Papers in Regional Science, 80, 411–24.

Arbia G. (2006), Spatial Econometrics: Statistical Foundations and Applications toRegional Convergence, Springer: Berlin.

Assinform (2004), ‘Rapporto sull’informatica e le telecomunicazioni in Italia’,Assinform, Milan.

Audretsch, D. (2003), ‘Innovation and spatial externalities’, International RegionalScience Review, 26, 167–74.

Audretsch, D.B. and Feldman, M.P. (1996), ‘Knowledge spillovers and the geog-raphy of innovation and production’, American Economic Review, 86, 630–40.

Aztema, O.L.A.C. and Weltevreden, J.W.J. (2004), ‘Adoption of B2C commerce byretail outlets at city centres: the role of place, products and organisation’, Paperpresented at the international specialist meeting on ICT, everyday life and urbanchange, Doorn, The Netherlands, 4–7 November.

Biehl, D. (1982), ‘The contribution of infrastructure to regional development’,Commission of the European Communities, Infrastructure Study Group, FinalReport. Office for Official Publications of the European Communities,Luxembourg.

Billon Curras, M. and Lera Lopez, F. (2004), ‘The North–South digital divide ininformation and communication technologies development: the case for Spain’sregions’, Paper presented at the 44th European Congress of the Regional ScienceAssociation, 25–29 August.

272 Spatial systems

Black, S. and Lynch, L. (2001), ‘How to compete: the impact of workplace prac-tices and information technology on productivity’, Review of Economics andStatistics, 83(3), 434–45.

Bonaccorsi, A., Martinelli, M., Rossi, C. and Serrecchia, I. (2002), ‘Measuring andmodelling Internet diffusion’, Paper presented at the Danish Research Unit forIndustrial Dynamics (DRUID) Summer Conference, Copenhagen, 6–8 June.

Bresnahan, T.F., Brynjolfsson, E. and Hitt, L.M. (1999), ‘Information technology,workplace organization, and the demand for skilled labor: firm-level evidence’,NBER Working Paper 7136, National Bureau of Economic Research,Cambridge, MA.

Brynjolfsson, E. (1993), ‘The productivity paradox of information technology’,Communication of the ACM, 36(12), 67–77.

Brynjolfsson, E. and Hitt, L. (1996), ‘Paradox lost? Firm-level evidence on thereturns to information systems’, Management Science, 42(4), 541–58.

Cairncross, F. (2001), The Death of Distance: How the Communications RevolutionWill Change Our Lives, Cambridge, MA: Harvard Business School Press.

Camagni, R. and Capello, R. (2005), ‘ICT and regional competitiveness in the eraof Internet’, The Annals of Regional Science, 39(3), 421–38.

Caselli, F. and Coleman, W.J. (2001), ‘Cross-country technology diffusion: the caseof computers’, NBER Working Paper 8130, National Bureau of EconomicResearch, Cambridge, MA.

Chinn, M.D. and Fairlie, R.W. (2004), ‘The determinants of the global digital divide:a cross country analysis of computer and Internet penetration’, Discussion Paper881, Economic and Growth Center, Yale University, New Haven, CT.

Cohen, W.M. and Levinthal, D.A. (1989), ‘Innovation and learning: the two facesof R&D’, Economic Journal, 99, 569–96.

Compaine, B.M. (2001), The Digital Divide: Facing a Crisis or Creating a Myth?,Cambridge, MA: MIT Press.

Dasgupta, S., Somik L. and Wheeler, D. (2001), ‘Policy reform, economic growthand the digital divide: an econometric analysis’, World Bank Working Paper2567, Washington, DC.

Diaz-Picazo, G.F. (1999), ‘An analysis of international Internet diffusion’, PhDThesis, MIT, Cambridge, MA.

Dryburgh, H. (2001), ‘Changing our ways: why and how Canadians use theInternet’, Statistics Canada, Housing; Family and Social Statistics Division,Catalogue no. 56F004MIE, No. 3.

Ellison, G. and Glaeser, E.L. (1997), ‘Geographic concentration in U.S. manufac-turing industries. A dartboard approach’, Journal of Political Economy, 105(5),889–927.

Fabiani, S., Schivardi, F. and Trento, S. (2003), ‘Quale impresa italiana investe intecnologie digitali?’, in S. Rossi (ed.), La Nuova Economia. I fatti dietro il mito,Bologna: Il Mulino, pp. 125–50.

Feldman, M. and Audretsch, D. (1999), ‘Innovation in cities: science-based diversity,specialization and localized competition’, European Economic Review, 43, 409–29.

Florax, R. and P. Nijkamp (2004), ‘Misspecification in linear spatial regressionmodels’, in K. Kempf-Leonard (ed.), Encyclopaedia of Social Measurement, SanDiego: San Diego Academic Press.

Fundación AUNA (2004), ‘eEspaña 2004, IV Informe anual sobre el desarrollo dela Sociedad de la Información en España’, www.fundacionauna.com/areas/25_publicaciones/publi_251_4.asp, accessed on 28th September 2006.

Explaining the territorial adoption of new technologies 273

Galliano, D. and Roux, P. (2004), ‘Spatial inequalities in the adoption of infor-mation and communication technologies. The case of French industrial firms’,Paper presented at the 4th Congress on Proximity Economics. Proximity,Networks and Coordination, Marseille, 17–18 June.

Gareis, K. and Osimo, D. (2004), ‘Benchmarking regional performance in the infor-mation society: turning it into practice’, Paper presented at the Workshop‘Measuring the information society: what, how, for whom and what?’, 5th AnnualConference of the Association of Internet Researchers, Brighton, UK, 19–22September.

Geroski, P. (2000), ‘Models of technology diffusion’, Research Policy, 29, 603–25.Gillespie, A., Goddard, J., Hepworth, M. and Williams, H. (1989), ‘Information

and communications technology and regional development: an informationeconomy perspective’, Science, Technology and Industry Review, 5, 86–111.

Grubesic, T.H. (2002), ‘Spatial dimensions of Internet activity’, Telecom-munications Policy, 26(7–8), 363–87.

Hollenstein, H. (2004), ‘The decision to adopt information and communication tech-nologies (ICT): firm-level evidence from Switzerland’, in OECD (2004), Ch. 3.

Iammarino, S., Jona-Lasinio, C. and Mantegazza, S. (2004), ‘Labour productivity,ICT and regions. The resurgence of the Italian “dualism”?’, LLEE WorkingDocument 12, Luiss Lab on European Economics.

Iuzzolino, G. (2003), ‘Si producono in Italia, e dove, le tecnologie digitali?’, in S. Rossi(ed.), La Nuova Economia. I fatti dietro il mito, Bologna: Il Mulino, pp. 95–124.

Jaffe, A.B., Trajtenberg, M. and Henderson, R. (1993), ‘Geographical localisationof knowledge spillovers, as evidenced by patent citations’, Quarterly Journal ofEconomics, 58, 577–98.

Keller, W. (2002), ‘Geographical localisation of international technology diffusion’,American Economic Review, 92, 120–42.

Kelly, K. (1998), New Rules for the New Economy. Ten Ways the Network Economyis Changing Everything, London: Fourth Estate.

Kiiski, S. and Pohjola, M. (2002), ‘Cross-country diffusion of the Internet’,Information Economics and Policy, 14, 297–310.

Kraemer, K.L. and Dedrick, J. (1996), ‘IT and economic development: inter-national competitiveness’, in V.H. Dutton (ed.), Information and CommunicationTechnologies. Visions and Realities, Oxford: Oxford University Press, pp. 61–111.

Lal, K. (2001), ‘The determinants of the adoption of information technology: acase study of the Indian garments industry’, in M. Pohjola (ed.), InformationTechnology, Productivity and Economic Growth, Oxford, UK: Oxford UniversityPress.

Le Sage, J.P. (1998), ‘Econometrics toolbox’, www.spatial-econometrics.com, 17November 2004.

Majumdar, S.K. and Venkataraman, S. (1993), ‘New technology adoption in UStelecommunications: the role of competitive pressures and firm-level induce-ments’, Research Policy, 22, 521–36.

Marrocu, E., Paci, R. and Pala, R. (2000), ‘Estimation of total factor productivityfor regions and sectors in Italy. A panel cointegration approach’, Centre for NorthSouth Economic Research (CRENOS) Working Paper, 00/16.

McWilliams, B. and Zilberman, D. (1996), ‘Time of technology absorption andlearning by using’, Economics of Innovation and New Technology, 4, 139–54.

Mills, B.F. and Whitacre, B.E. (2003), ‘Understanding the non-metropolitan–metropolitan digital divide’, Growth and Change, 34(2), 219–43.

274 Spatial systems

Mueller, M. (1998), ‘The battle over Internet domain names: global or nationalTLDs?’, Telecommunications Policy, 22(2), 89–107.

Negroponte, N. (1995), Being Digital, New York: Knopf.Norris, P. (2002), Digital Divide: Civic Engagement, Information Poverty, and the

Internet Worldwide, Cambridge: Cambridge University Press.NTIA (National Telecommunications and Information Administration), Economics

and Statistics Administration (2002), ‘A nation online: how Americans are expand-ing their use of the Internet’, www.ntia.doc.gov/ntiahome/dn/nationonline_020502.htm, 3 October 2004.

Nunes, F. (2004), ‘The geography of .pt top level domain’, Paper presented at the44th European Congress of the European Regional Science Association, Porto,25–29 August.

Oden, M. and Rock, K. (2004), ‘Beyond the digital access divide: towardmeaningful measures of information’, Paper presented at the 5th AnnualConference of the Association of Internet Researchers, Brighton, UK, 19–22September.

Organization for Economic Cooperation and Development (OECD) (2001),‘Understanding the digital divide’, www.oecd.org/dataoecd/38/57/1888451.pdf, 1October 2004.

Organization for Economic Cooperation and Development (OECD) (2004), ‘Theeconomic impact of ICT. Measurement, evidence and implications’, www1.oecd.org/publications/e-book/9204051E.PDF, 1 October 2004.

Oyeralan-Oyeyinka, B. and Lal, K. (2003), ‘The Internet diffusion in Sub-SaharanAfrica: a cross-country analysis’, Discussion Paper of INTECH, Institute forNew Technology, United Nations University.

Pagnini, M. (2002), ‘Misure e determinanti della agglomerazione spaziale nei com-parti industriali italiani’, Banca d’Italia, Discussion Paper 452, October.

Pohjola, M. (2003), ‘The adoption and diffusion of ICT across countries: patternsand determinants’, in D.C. Jones (ed.), New Economy Handbook, San Diego, CA:Elsevier–Academic Press, pp. 77–100.

Porter, M. (1990), The Comparative Advantage of Nations, New York: Free Press.Press, L. (1997), ‘Tracking the global diffusion of the Internet’, Communications of

the ACM, 40(11), 11–17.Qingxuan, M. and Mingzhi, L. (2002), ‘New economy and ICT development in

China’, Information Economics and Policy, 14(2), 275–95.Ramsay, R. (2004), ‘The Internet and its geographic dilemma: constructing the

digital divide’, Paper presented at the 5th Annual Conference of the Associationof Internet Researchers, Brighton, UK, 19–22 September.

Steinmuller, W.E. (2001), ‘ICTs and the possibilities of leapfrogging by developingcountries’, International Labour Review, 140(2), 193–210.

United Nations Development Programme (UNDP) (2001), Human DevelopmentReport 2001. Making New Technologies Work for Human Development, Oxford,UK and New York: Oxford University Press.

Wallsten, S. (2003), ‘Regulation and Internet use in developing countries’, AEIBrookings Joint Center for Regulatory Studies Related Publications 03-8,Washington, DC.

Weltevreden, J.W.J., de Kruif, K., Atzema, O.A.L.C., Frenken, K. and vanOort, F.G. (2005), ‘The b2c e-commerce landscape of the Dutch retail sector’,Paper presented at the 45th Congress of European Regional Science Association(ERSA), Free University, Amsterdam, The Netherlands, 23–27 August.

Explaining the territorial adoption of new technologies 275

Wensheng, W. (2002), ‘Bridging the digital divide inside China’, Paper presented atthe 3rd Annual Conference of the Association of Internet Researchers,Maastricht, NL, 13–16 October.

Wolcott, P., Press, L., McHenry, W., Goodman, S. and Foster, W. (2001), ‘A frame-work for assessing the global diffusion of the Internet’, Journal of the AIS, (2)6.

Zook, M.A. (1999), ‘Internet cities of the United States and the world: under-standing new geographies’, Paper presented at the Cities in the GlobalInformation Society Conference, Newcastle upon Tyne, UK, 22–24 November.

Zook, M.A. (2000), ‘Internet metrics: using hosts and domain counts to map theInternet globally’, Telecommunications Policy, 24(6/7), 613–20.

Zook, M.A., Dodge, M., Aoyama, Y. and Townsend, A. (2004), ‘New digital geog-raphies: information, communication, and place’, in S.D. Brunn, S.C. Cutter, andJ.W. Harrington (eds), Geography and Technology, Dordrecht: Kluwer AcademicPublishers, pp. 155–76.

276 Spatial systems

PART V

Planning

13. Evolutionary urban transportationplanning? An explorationLuca Bertolini

1. INTRODUCTION

For urban transportation planners these are challenging times. On the onehand, and in spite of all the hype about dematerialization of society, phys-ical mobility systems appear ever more crucial in granting individualsand organizations the access to the spatially and temporally disjointedresources they need to thrive or even just to survive. On the other, becauseof a heterogeneous mix of mounting financial and fiscal constraints toinfrastructure expansion, and growing awareness of and social resistanceto the negative impacts of mobility, the traditional ‘predict and provide’approach to planning is no longer an option.

Practical concerns are echoed by more fundamental critiques (see,for instance, Dimitriou, 1992; Gifford, 2003). Central to this more fun-damental criticism is the contention that conventional planning methodsdo not adequately account for the irreducible uncertainty of developmentsaffecting transport and its relationship with the broader context.Uncertainty is, of course, inherent to any future-oriented activity. Thereare, however, different forms of uncertainty. As discussed by Van derHeijden (1996), a first form of uncertainty is risk, where there is enoughhistorical precedent in terms of similar events to allow the estimation ofprobabilities for various outcomes (this is the core realm of forecasting);structural uncertainties are a second form, where the event, while stillconceivable in terms of chains of cause and effect, is unique enough not toprovide any indication of likelihood (think of the complex interplay ofrising wealth, social emancipation, mass motorization and urban decen-tralization, as it unfolded in industrialized nations in the post-war period);a third form, unknowables, is where the event cannot even be imagined(think of the 1973 oil crisis). While all three forms of uncertainty mayapply at any time, the likelihood of uncertainties of types two and threeincreases as the time horizon gets longer and the system gets more complex(that is, with more components and more relationships), up to a point

279

where prediction is no longer possible. As a discipline that also deals withthe long term and with highly complex systems, urban transportationplanning should also be able to come to terms with fundamentally unpre-dictable events, that is, irreducible forms of uncertainty (the second andthe third forms). Yet, a convincing response is still lacking. In the words ofMeyer and Miller (2001, p. 519): ‘No aspect . . . is as pervasive to the[transport planning] process, and yet as often ignored, as uncertainty’. Inresponse, Meyer and Miller stress the need to improve present land use andtransportation models. More adequate forecasting models would need toexplore the full range of system responses (short and long run) to a broadvariety of policy combinations (transportation, land use and other), anddo this at the level of individual responses (by means of disaggregatedbehavioural models). However, and crucially, they also recognize that‘Even with “ideal” . . . models, uncertainty will still exist with respect tothe exact nature of future activity systems’ (p. 340). Banister (2002, p. 141)strikes a similar note:

Some of the limitations of the TPM [transport planning model] have been metby the ILUTM [integrated land use transport model]. But the complexity of theland development process, travel decisions and the rapidly changing forms ofindustry, of population structure, of lifestyles, and of the use of time all contriveto make progress difficult, if not impossible.

This chapter attempts to take this more fundamental level of criticismto its logical conclusion. In particular, its aim is to explore if and how anevolutionary approach to urban transportation planning can help to over-come some of the limits mentioned above, and develop planning methodsthat can usefully complement forecasting-based ones. The inspiration isdrawn from much more advanced conceptualizations in other disciplines,and most notably evolutionary economics and the application of com-plexity theory to the understanding of cities. Evolutionary and complex-ity approaches seem especially appropriate because they both recognizethe high level of interdependency between the different components of thesystem and the limits to dealing with such interdependency by means ofprediction, because of irreducible uncertainty. Building upon this line ofreasoning, two core hypotheses are formulated with respect to the objectand the scope of urban transportation planning. The first hypothesis isthat the urban transportation system indeed behaves in an evolutionary,complex fashion. The second, related hypothesis is that because of this,urban transportation policies also need to focus on enhancing theresilience and the adaptability of the system. Changes in transport andland use development patterns and policies and in the broader context in

280 Planning

the post-war period in the Amsterdam region are analysed in order toillustrate the two core hypotheses. In the conclusions more general impli-cations are drawn.

2. AN EVOLUTIONARY APPROACH

Evolutionary thinking originated in the natural sciences but is increas-ingly being applied in the social sciences and most explicitly in economics(Nelson and Winter, 1982; Dosi and Nelson, 1994; van den Bergh andFetchenhauer, 2001; Boschma et al., 2002). Intriguing parallels can alsobe found in works adopting theories and methods of the emerging scienceof complexity – and particularly the concept of self-organization – mostnotably including applications to the analysis of cities (Allen, 1997;Portugali, 1999; Batten, 2001; but see also the earlier work by Jacobs,1961). Characteristically of all these streams of work, the assumption of(a single) equilibrium as the ‘natural’ state of the system is questioned,and attention is rather directed to far-from-equilibrium processes ofchange. It is acknowledged that different social actors can react differentlyto similar system-wide perturbations, depending on both the specificitiesof the local context and their personal features. Individual decisionsand actions eventually cumulate into development processes that areboth path dependent – as earlier experiences largely determine theresponse to new stimuli – and unpredictable – as even small, localdifferences can have (due to self-reinforcing mechanisms) big, global con-sequences. Underlying this thinking is the recognition that social systemsare complex systems, that is, that they are systems characterized by a highdegree of interdependency between a wide range of components andprocesses. Such complexity fundamentally bounds the rationality ofsocial actors.

A focus on evolutionary economics can help further develop the argu-ment. While there are different interpretations within the field, somebasic principles are aptly captured by the notion of microevolution intro-duced by Nelson and Winter (1982). According to Nelson and Winter,because of irreducible uncertainty, the existence of transaction costs andthe difficulty of change in the short term, firms tend to follow ‘organiz-ational routines’, or proven ways of conducting business, rather thanconsider each time all possible alternative courses of action. On the otherhand, the evaluation of current routines can lead firms to make adjust-ments and even substitution. The results of such a ‘search process’ are,however, also uncertain. Furthermore, because past experiences influenceboth existing routines and the search for new ones, different firms will

Evolutionary urban transportation planning? 281

typically have different routines and try different alternatives, resulting ina variety of economic behaviour. Eventually, the actual performance ofa firm will constitute the major incentive to maintaining or changing aroutine. Such performance is largely determined by the characteristics ofthe ‘selection environment’, that is, the interplay of demand and supplyin the marketplace. The selection environment is not a static entity either,as it will also change as a result of the accumulation of firm-specificprocesses. In this sense, there is ‘co-evolution’ between the market andindividual firms.

The resulting economic reality is one characterized by continuous suc-cessions of disturbances and adaptations, which preclude the attainment ofa stable equilibrium. Continuous change means that initially successfulroutines can become less efficient or effective, or even have unexpected con-sequences. There is no once-and-for-all optimal routine. Furthermore, thenature of the process underlies the incremental nature of change, and thedifficulty of more than marginally altering an existing routine. The risk thatfirms be ‘locked in’ in a non-optimal routine is therefore always present. Theimplication is that beyond a certain threshold, marginal change will notsuffice and coordinated change will be required. However, because it isuncertain which routine will be able to break the impasse, diversity of andcompetition among alternatives should be stimulated. It is precisely suchredundancy of routines that makes the economic system resilient, that is,capable of continuous performance in the face of changing, uncertaincircumstances.

The above conceptualization of economic reality can also be usefullyapplied to the object of this chapter. Existing transport and land use poli-cies can be seen as organizational routines. The broader, changing urbansocio-demographic and economic context can be seen as the selectionenvironment in which existing policies must continuously prove theirworth and the search process for better policies takes place. As also poli-cies, in their turn, affect the selection environment, there is co-evolutionbetween environment and policies. The analogy further suggests thatthere is no universally valid, optimal policy. Accordingly, while learningfrom practical experience elsewhere and from theoretical models isimportant, the value of a solution can only be appreciated in a specific,continuously evolving situation. Understanding the unique set of oppor-tunities and constraints determined by a given historical developmentpath and local configuration of factors is thus crucial. However, becauseof limits to predictability, only real-life ‘policy experiments’ (SzejnwaldBrown et al., 2004), that is, actual engagement with the selection en-vironment, can give full understanding of these. At the same time, recog-nition of the unpredictability of the outcome – particularly when the long

282 Planning

term is concerned – should also result in recognition of the need to lookfor ways of improving the ability of the system to react and perform inthe face of unforeseen (and unforeseeable) change. More specifically, anurban transport and land use system capable of performing in the face ofunpredictable change would be, in the first place, one capable of contin-uing to function in the face of change, that is, it must be a resilient system.This seems particularly important for system components that cannotchange rapidly, or easily (such as a transportation network morphology).Second, it would be a system capable of changing itself in response tochange in the socio-economic environment, that is, it must also be anadaptable system. This would especially apply to system components that,given their nature, can change relatively swiftly (think of a road-pricingregime). As the requirements of resilience and adaptability might be con-tradictory, finding a workable balance between them lies at the heart ofthe task.

Building upon this line of reasoning, two core hypotheses can be formu-lated with respect to the object and goals of urban transportation planning.The first hypothesis is that the urban transport system indeed behaves in anevolutionary, complex fashion. The second, related hypothesis is thatbecause of this, urban transportation policies also need to focus on enhanc-ing the resilience and the adaptability of the system. Changes in the trans-port and land use development patterns and policies and in the economicand socio-demographic context in the Amsterdam region in the post-warperiod are analysed in the following sections to illustrate the two corehypotheses. The goal of this exercise is not so much that of providing aninterpretation, let alone a conclusive one, of these events, but rather that ofexploring to which degree they can be characterized as evolutionary andcomplex. For this purpose, the two core hypotheses are further articulatedin the following sub-hypotheses:

1. The behaviour of the urban transport system can be characterized asevolutionary and complex because:

● the system alternates periods of incremental, quantitative changeand periods of radical, qualitative change, or system transitionphases;

● in all periods, change in the system is path dependent, that is,existing patterns of transportation networks, land uses andtransport and land use policies limit the scope for change;

● however, during transition phases both the scope for policies toinfluence the outcome and the unpredictability of such anoutcome are greatest.

2. Because of path dependency, policies need to:

Evolutionary urban transportation planning? 283

● build upon the unique set of opportunities and constraints forchange determined by a specific historical development path andlocal combination of factors.

3. Because of unpredictability, policies need to:● increase the resilience of the system, that is, its ability to keep

functioning in the face of unexpected change. This seems es-pecially important for the shape of transportation networks, asthis is relatively difficult/slow to change;

● increase the adaptability of the system, that is, its ability to reactto unexpected change. This seems especially important for landuse regulations and mobility management measures, as these arerelatively easy/fast to change.

Developments in the Amsterdam region are summarized in Table 13.1,Boxes 13.1 and 13.2 and Figure 13.1. In the following section, they will beused to illustrate the two sets of hypotheses.

3. ILLUSTRATING THE HYPOTHESES

The Nature of Change

Behaviour of the urban transport system in the Amsterdam case can becharacterized as evolutionary and complex because:

The system alternates periods of incremental, quantitative change andperiods of radical, qualitative change, or system transition phases.

The alternating of periods of (more predictable) quantitative change andperiods of (less predictable) qualitative change, or transition phases can beobserved in all the domains of change described in Table 13.1. This patternof development is best shown by the existence of trend-breaking points.Trend-breaking points are defined here as changes in the nature of devel-opment rather than in the amount of development (for instance a shiftfrom an industry- to a service-based economy as opposed to a change inthe rate of growth of an economy). For clarity, in the following transitionsthe different streams of change will be discussed sequentially. This shouldnot conceal the fact that they are in reality strongly interrelated. Some feelfor these interdependencies can be obtained from Boxes 13.1 and 13.2,where two major transport and land use policy transitions are discussed insome detail.

284 Planning

285

Tab

le 1

3.1

Ove

rvie

w o

fdi

ffer

ent

dom

ains

of

chan

ge in

the

Am

ster

dam

urb

an r

egio

n,19

46–1

9991

Per

iod

Soci

o-de

mog

raph

icE

cono

mic

Lan

d us

e po

licy

Lan

d us

eT

rans

port

pol

icy

Tra

nspo

rt

Pre

-R

elat

ive

econ

omic

1935

:Am

ster

dam

Lim

ited

cen

tral

1901

:firs

t pl

an fo

r a

1876

:Am

ster

dam

-19

46st

agna

tion

‘gen

eral

exp

ansi

onbu

sine

ss d

istr

ict

frei

ght

railw

ay r

ing

Nor

th S

ea C

anal

pl

an’(

AU

P);

(CB

D)

form

ing

in t

hear

ound

Am

ster

dam

open

s (s

hift

ing

the

conc

entr

ated

,sta

r-ci

ty c

entr

e;ex

tens

ive

1935

:(A

UP

)fo

cus

ofha

rbou

r lik

e ur

ban

expa

nsio

npu

blic

hou

sing

‘cyc

labl

e di

stan

ce’

acti

vity

fro

m e

ast

to

as a

lter

nati

ve t

ode

velo

pmen

ts in

the

from

the

cit

y ce

ntre

wes

t)su

burb

aniz

atio

nre

st o

fth

e ci

ty;

defi

nes

the

oute

r18

89:C

entr

al

trai

n/tr

am s

uppo

rted

limit

of

urba

nSt

atio

n op

ens

subu

rban

izat

ion

expa

nsio

n,pr

ovin

cial

(sep

arat

ing

harb

our

else

whe

re in

the

road

s an

d tr

am li

nes

from

cit

y)re

gion

conn

ect

city

and

1917

:Sch

ipho

l ope

nssu

rrou

ndin

gs,

as m

ilita

ry a

irpo

rtac

cess

ibili

ty o

fth

eha

rbou

r is

a c

entr

alco

ncer

n,la

ndre

serv

atio

ns fo

rfu

ture

tan

gent

ial

road

s,ca

nals

and

frei

ght

railw

ays

are

cons

truc

ted

1946

– A

mst

erda

m f

rom

Nat

iona

lIn

the

cit

y ce

ntre

:19

55:(

firs

t ci

tyIn

fras

truc

ture

5877

0,00

0 to

870

,000

1958

:nat

iona

l rep

ort

‘CB

D fo

rmin

g’an

dce

ntre

rep

ort)

the

1958

:Sch

ipho

lin

habi

tant

s (p

ost-

intr

oduc

es R

ands

tad

‘slu

m c

lear

ance

’st

udy

ofan

beco

mes

the

nat

iona

lw

ar m

axim

um)

(‘ri

m c

ity’

) an

dun

derg

roun

d ra

ilway

airp

ort

Gro

ene

Har

t (‘

gree

nIn

the

per

iphe

ry:

new

syst

em is

pro

pose

d,he

art’

) co

ncep

ts,

resi

dent

ial

disc

ussi

ons

and

286

Tab

le 1

3.1

(con

tinu

ed)

Per

iod

Soci

o-de

mog

raph

icE

cono

mic

Lan

d us

e po

licy

Lan

d us

eT

rans

port

pol

icy

Tra

nspo

rt

Dem

ogra

phic

whi

ch a

re t

o gu

ide

neig

hbou

rhoo

dsal

tern

ativ

e pl

ans

Mob

ility

tren

ds:

land

use

pla

nnin

g in

(wes

t);e

xpan

sion

of

follo

wG

row

th o

ffi

rst

posi

tive

,lat

erth

e w

est

ofth

ein

dust

rial

and

com

mut

ing

tow

ards

nega

tive

mig

rati

onN

ethe

rlan

ds fo

r th

eha

rbou

r si

tes

Am

ster

dam

and

bala

nce

(tow

ards

rest

of

the

cent

ury

othe

r re

gion

alsu

burb

s an

dce

ntre

sab

road

)L

ocal

1955

:firs

t ci

ty c

entr

eM

odal

spl

it r

egio

n re

port

,env

isag

ing

(hom

e to

wor

k)it

s ra

dica

l19

47:

tran

sfor

mat

ion

(will

Pub

lic t

rans

port

no

t be

ado

pted

)75

%

1958

–65:

plan

s fo

rC

ar 5

%pe

riph

eral

res

iden

tial

Bik

e 20

%

expa

nsio

ns19

60:

Pub

lic t

rans

port

67%

Car

16%

Bik

e 18

%19

59–

Am

ster

dam

fro

m

Gen

eral

tre

nds:

Nat

iona

lIn

the

cit

y ce

ntre

:N

atio

nal

Infr

astr

uctu

re84

870,

000

to 6

75,0

00

sust

aine

d gr

owth

In

the

sec

ond

half

offi

rst

urba

n re

new

al

1966

:nat

iona

lE

xpan

sion

of

the

inha

bita

nts

(pos

t-of

the

econ

omy

the

1960

s an

d up

to

and

late

r m

otor

way

pla

n (i

nm

otor

way

net

wor

k,w

ar m

inim

um)

unti

l the

ear

ly

the

end

ofth

e 19

70s

neig

hbou

rhoo

dth

e re

gion

oft

enlit

tle

inve

stm

ent

in19

70s;

long

per

iod

‘con

cent

rate

dre

gene

rati

onfo

llow

ing

exis

ting

the

railw

ays,

plan

sD

emog

raph

icof

decl

ine

follo

win

g de

cent

raliz

atio

n’hi

stor

ical

rou

tes,

for

new

can

als

are

tren

ds:

the

firs

t oi

l cri

sis

polic

y�

grow

thIn

the

per

iphe

ry:

new

such

as

the

A2

aban

done

dne

gati

ve n

atio

nal

(197

3) a

nd

cent

res

at a

roun

d 30

resi

dent

ial

tow

ards

Utr

echt

,but

1966

–91:

real

izat

ion

287

mig

rati

on b

alan

ce,

cont

inui

ng in

to t

hekm

fro

m t

he m

ajor

neig

hbou

rhoo

ds;

also

incl

udin

g ne

wof

the

mot

orw

ay

posi

tive

1980

sci

ties

furt

her

expa

nsio

n of

tang

ents

,suc

h as

the

ri

ng A

10,l

arge

lyin

tern

atio

nal

Fro

m t

he b

egin

ning

indu

stri

al a

ndA

10 m

otor

way

rin

g)us

ing

exis

ting

rig

hts

mig

rati

on b

alan

ceS

pati

al t

rend

s:of

the

1980

s ha

rbou

r si

tes;

ofw

ayN

atio

nal m

igra

tion

:fi

rms

mov

e to

the

‘c

ompa

ct c

ity’

polic

yin

crea

sing

ly a

lso

1960

s:ru

nway

tr

adit

iona

lur

ban

peri

pher

y �

den

sifi

cati

on o

foffi

ce d

evel

opm

ents

Loc

alex

pans

ion

ofho

useh

olds

and

the

subu

rban

exis

ting

cen

tres

and

U

nder

grou

nd r

ailw

aySc

hiph

ol a

irpo

rt(f

amili

es)

to t

he

cent

res,

in s

earc

h su

burb

an g

row

th

In t

he s

ubur

bs:

deba

tes

1972

:Am

ster

dam

-su

burb

s,sm

alle

r of

chea

p sp

ace

clos

e (a

t 15

–30

km)

deve

lopm

ent

ofSe

e B

ox 1

3.1:

‘The

Nor

th S

ea c

anal

ho

useh

olds

to th

e ci

tyan

d/or

labo

ur;t

heto

the

maj

or c

itie

sgr

owth

cen

tres

late

196

0s a

nd e

arly

broa

dene

d an

d In

tern

atio

nal

harb

our

decl

ines

(Pur

mer

end,

Alm

ere,

1970

s:a

tran

spor

tde

epen

edm

igra

tion

:Tur

ksan

d th

e ai

rpor

tH

oofd

dorp

)an

d la

nd u

se p

olic

y19

78–9

7:Sc

hiph

olan

d M

oroc

can

grow

s st

rong

lyL

ocal

tran

siti

on d

isse

cted

’ra

ilway

con

nect

ions

gues

t w

orke

rs in

Urb

an r

enew

al

1975

/197

6/19

78:fi

rst

open

the

1960

s,br

ingi

ngde

bate

s‘t

raffi

c ci

rcul

atio

n19

77–8

2:ea

st li

neov

er t

heir

fam

ilies

See

Box

13.

1:‘T

he la

tepl

an’;

a ba

lanc

eun

derg

roun

d ra

ilway

in t

he 1

970s

;19

60s

and

earl

ybe

twee

n ac

cess

ibili

tyop

ens

grow

ing

shar

e fr

om19

70s:

a tr

ansp

ort

and

livea

bilit

y is

Dut

ch A

ntill

es a

ndan

d la

nd u

se p

olic

yne

eded

,the

mea

nsM

obili

tySu

rina

me

(fol

low

ing

tran

siti

on d

isse

cted

’ar

e pu

blic

tra

nspo

rtL

onge

r an

d m

ore

inde

pend

ence

in19

74–8

1:st

ruct

ure

(mos

t no

tabl

y th

eca

r-ba

sed

trip

s;19

75)

plan

par

ts A

,B a

ndtr

am),

a lim

ited

,co

nges

tion

C;s

hift

of

focu

s fr

omco

arse

pri

mar

y ro

ades

peci

ally

urba

n de

sign

to

netw

ork,

a re

stri

ctiv

ein

the

cen

tre

Cul

tura

l tre

nds:

plan

ning

pro

cess

,pa

rkin

g po

licy

in t

heG

row

th o

fem

erge

nce

ofm

ass

pres

erva

tion

of

the

city

cen

tre,

and

new

com

mut

ing

tow

ards

cons

umpt

ion,

new

resi

dent

ial f

unct

ion

cycl

e ro

utes

Am

ster

dam

and

,lif

esty

les,

in o

ld n

eigh

bour

-19

82:d

raft

tra

nspo

rtm

ore

nota

bly,

othe

rem

anci

pati

on o

fho

ods,

iden

tifi

cati

onst

ruct

ure

plan

;the

regi

onal

cen

tres

wom

en,y

outh

ofsu

b-ce

ntre

s on

the

focu

s is

on

impr

ovin

gcu

ltur

eur

ban

frin

ge,s

tudi

esth

e ex

isti

ng s

itua

tion

288

Tab

le 1

3.1

(con

tinu

ed)

Per

iod

Soci

o-de

mog

raph

icE

cono

mic

Lan

d us

e po

licy

Lan

d us

eT

rans

port

pol

icy

Tra

nspo

rt

for

a re

side

ntia

lan

d on

an

Mod

al s

plit

reg

ion

conv

ersi

on o

fth

ein

crem

enta

l app

roac

h(h

ome

to w

ork)

east

ern

harb

our

1960

:P

ublic

tra

nspo

rt

67%

Car

16%

Bik

e 18

%19

71:

Pub

lic t

rans

port

41

%C

ar 4

6%B

ike

13%

Bik

e sh

are

in t

he

city

(al

l tri

ps)

1950

:ca.

80%

1975

:ca.

25%

1985

–A

mst

erda

m f

rom

Gen

eral

tre

nds:

Nat

iona

lIn

the

cit

y ce

ntre

:N

atio

nal

Infr

astr

uctu

re99

675,

000

to 7

25,0

00st

ruct

ural

cha

nge

Sinc

e 19

88:‘

urba

nne

w h

ousi

ng a

nd19

88:‘

Seco

nd19

78–9

7:Sc

hiph

olin

habi

tant

san

d st

rong

rec

over

yno

des’

polic

y;cu

ltur

al a

nd le

isur

etr

ansp

ort

stru

ctur

era

ilway

con

nect

ions

ofth

e ec

onom

y de

velo

pmen

t to

be

faci

litie

spl

an’;

‘dou

ble

goal

’op

enD

emog

raph

icsi

nce

the

seco

nd

conc

entr

ated

in t

heof

impr

ovin

g bo

th

1987

–88:

Alm

ere

tren

ds:

half

ofth

e 19

80s

maj

or c

itie

sac

cess

ibili

ty a

nd

and

Fle

vola

ndle

ss s

ubur

bani

z-an

d in

the

199

0s;

Sinc

e 19

91:‘

VIN

EX

In t

he p

erip

hery

:ne

wliv

eabi

lity,

to b

e ra

ilway

con

nect

ions

atio

n,m

ore

birt

hs,

grow

th

neig

hbou

rhoo

ds’

resi

dent

ial

achi

eved

by

redu

cing

open

mor

e im

mig

rant

s co

ncen

trat

ed in

po

licy;

resi

dent

ial

neig

hbou

rhoo

ds;

mob

ility

by

car,

1990

:Am

stel

veen

th

an in

the

busi

ness

ser

vice

s,ex

pans

ion

to b

ere

side

ntia

lth

roug

h a

mix

of

surf

ace

light

rai

l lin

e

289

prec

edin

g pe

riod

logi

stic

s,IC

T a

nd

real

ized

wit

hin

orco

nver

sion

of

olde

rpr

icin

g an

d lo

cati

on

open

sne

w m

edia

,lei

sure

ad

jace

nt t

o th

eha

rbou

r si

tes

(eas

t);

polic

ies,

and

sele

c-19

91:

Cul

tura

l tre

nds:

and

tour

ism

exis

ting

cit

yfu

rthe

r ex

pans

ion

ofti

ve in

fras

truc

ture

Z

eebu

rger

tunn

elgr

owin

gyo

unge

r in

dust

rial

impr

ovem

ents

open

s (c

ompl

etio

n ap

prec

iati

on o

fS

pati

al t

rend

s:L

ocal

and

harb

our

site

sof

the

mot

orw

ay

urba

n liv

ing;

cont

inui

ng d

e-19

80s:

whi

le e

xist

ing

(wes

t,so

uthe

ast)

Loc

alri

ng A

10)

olde

r ur

ban

conc

entr

atio

n bu

tpl

ans

(gro

wth

1985

:(st

ruct

ure

1997

:sur

face

met

ro

neig

hbou

rhoo

dsal

so g

eogr

aphi

cal

cent

res,

new

pl

an)

the

met

ro h

asri

ng li

ne o

pens

incr

easi

ngly

spec

ializ

atio

n of

mot

orw

ays

and

beco

me

a ‘n

o-no

’,po

pula

r;co

ntin

uous

econ

omic

act

ivit

yra

ilway

s) a

reIn

the

sub

urbs

:fo

cus

on e

xpan

sion

19

97:P

iet

grow

th o

fet

hnic

(new

med

ia,l

eisu

reim

plem

ente

d in

the

hous

ing

deve

lopm

ents

ofth

e tr

am n

etw

ork,

Hei

ntun

nel o

pens

m

inor

itie

s (u

p to

and

tour

ism

in t

here

st o

fth

e re

gion

,di

ffus

ed (

but

stro

ngly

enfo

rcem

ent,

smal

l(r

oad

link

36%

in 1

999)

;ci

ty c

entr

e,bu

sine

ssA

mst

erda

m a

dopt

sco

nstr

aine

d by

noi

seim

prov

emen

ts,

conn

ecti

ngem

erge

nce

ofse

rvic

es a

nd I

CT

ina

com

pact

cit

y po

licy

boun

dari

es a

roun

dco

ordi

nati

on a

ndre

conv

erte

d ea

ster

net

hnic

the

peri

pher

al s

ub-

1985

:str

uctu

re p

lan

Schi

phol

air

port

and

m

aint

enan

ce (

also

dock

land

s to

ne

ighb

ourh

oods

ince

ntre

s lo

gist

ics

‘The

cit

y ce

ntra

l’;na

ture

pre

serv

atio

nbe

caus

e of

fina

ncia

lm

otor

way

rin

g)th

e pe

riph

ery

arou

nd t

he a

irpo

rt,

com

pact

cit

y,hi

ghar

eas)

;offi

ce a

ndpr

oble

ms)

harb

our

and

dens

itie

s an

din

dust

ry

1986

:fol

low

ing

Mod

al s

plit

reg

ion

mot

orw

ays)

func

tion

al m

ix,

deve

lopm

ents

recu

rrin

g co

nges

tion

(hom

e to

wor

k)ho

usin

g-le

d,(c

once

ntra

ted

ofth

e C

oent

unne

l19

71:

cons

olid

atio

n of

the

arou

nd a

irpo

rt a

ndan

d st

rong

gro

wth

of

Pub

lic t

rans

port

hi

stor

ical

cen

tre

mot

orw

ay c

orri

dors

)th

e ai

rpor

t,41

%

1991

:str

uctu

re p

lan

cons

truc

tion

of

a ne

wC

ar 4

6%‘A

mst

erda

m’;

Cit

y ce

ntre

tre

nds,

wes

tern

mot

orw

ayB

ike

13%

dow

nsiz

ing

ofth

e19

75–9

9ta

ngen

t (t

he A

5) is

1991

:pe

riph

eral

Inha

bita

nts

�11

%de

cide

dP

ublic

tra

nspo

rt

expa

nsio

ns;

Dw

ellin

gs �

39%

Lat

e 19

80s:

a st

udy

22%

ambi

tiou

sJo

bs –

24%

ofth

e C

ham

ber

ofC

ar 6

0%re

deve

lopm

ent

plan

Offi

ces

–10%

Com

mer

ce s

how

s th

eB

ike

18%

Res

taur

ants

/caf

esne

w p

ossi

bilit

ies

of

290

Tab

le 1

3.1

(con

tinu

ed)

Per

iod

Soci

o-de

mog

raph

icE

cono

mic

Lan

d us

e po

licy

Lan

d us

eT

rans

port

pol

icy

Tra

nspo

rt

for

the

IJ b

anks

�53

%tu

nnel

ling

wit

hout

Bik

e sh

are

in t

he c

ity

Shop

s �

40%

disr

upti

on o

fth

e(a

ll tr

ips)

IJ-b

anks

deb

ates

exis

ting

urb

an f

abri

c19

75:c

a.25

%Se

e B

ox 1

3.2:

‘The

1991

:(st

ruct

ure

1996

:ca.

35%

late

198

0s a

nd e

arly

plan

) se

vera

l new

1990

s:a

land

use

infr

astr

uctu

re

polic

y tr

ansi

tion

deve

lopm

ents

are

di

ssec

ted’

envi

sage

d,in

clud

ing

1996

:str

uctu

re p

lan

the

new

wes

tern

‘O

pen

city

’;m

otor

way

tan

gent

defi

niti

vean

d ra

ilway

fre

ight

aban

donm

ent

oflin

e,a

new

nor

th–

plan

s fo

r w

este

rnso

uth

met

ro li

ne,a

expa

nsio

ns,f

ocus

on

light

rai

l rin

g lin

e,th

e gr

een

stru

ctur

e,an

d a

light

rai

l lin

eth

e So

uth

Axi

s w

illto

con

nect

the

beco

me

the

mai

nre

conv

erte

d ea

ster

nbu

sine

ss c

entr

e,th

edo

ckla

nd a

rea

IJ b

anks

will

dev

elop

1993

:reg

iona

l a

live–

wor

k–le

isur

etr

ansp

ort

plan

m

ix,a

new

sub

-(R

VV

P);

the

goal

isce

ntre

in A

mst

erda

mto

impr

ove

Nor

th is

indi

cate

dac

cess

ibili

ty w

hile

pr

eser

ving

live

abili

ty,

the

mos

t im

port

ant

mea

ns a

re r

educ

tion

291

ofth

e gr

owth

in

car-

kilo

met

res

and

am

odal

shi

ft t

o pu

blic

tran

spor

t an

d bi

ke

Not

e:1.

The

tw

o de

mog

raph

ic t

rend

-bre

akin

g po

ints

of

1959

(fr

om p

opul

atio

n gr

owth

to

popu

lati

on d

eclin

e in

Am

ster

dam

) an

d 19

85 (

from

decl

ine

to g

row

th)

are

used

to

dist

ingu

ish

thre

e m

ain

phas

es.T

his

is d

one

for

conv

enie

nce

and

does

not

impl

y th

at t

hese

are

als

o re

leva

nt d

ates

for

othe

r st

ream

s of

chan

ge,i

nclu

ding

mor

e qu

alit

ativ

e as

pect

s of

dem

ogra

phic

dev

elop

men

t.

Sou

rces

:H

onig

(19

96);

Rom

mer

ts (

1997

);B

ruhè

ze a

nd V

eraa

rt (

1999

);D

ijkst

ra e

t al

.(19

99);

Reg

iona

al O

rgaa

n A

mst

erda

m (

2000

,200

4);

Win

ters

hove

n (2

000)

;Bra

nd (

2002

);le

Cle

rcq

(200

2);D

iens

t R

uim

telij

ke O

rden

ing

Am

ster

dam

(20

03a,

2003

b);P

oels

tra

(200

3);T

erho

rst

and

van

deV

en (

2003

);C

entr

aal B

urea

u vo

or d

e St

atis

tiek

(w

ww

.cbs

.nl)

.

BOX 13.1 THE LATE 1960s AND EARLY 1970s: ATRANSPORT AND LAND USE POLICYTRANSITION DISSECTED

The 1960s are the theatre of an extensive production of far-reachingurban renewal and infrastructure plans for Amsterdam’s historic city,following a first city centre report in 1955.The underlying philosophyseems straightforward: population growth is to be accommodated innew expansions on the urban periphery and in growth centres in theregion (in line with national policy), service growth is to be concen-trated in an enlarged and restructured city centre, and a new under-ground urban railway network is to be developed to link the newconcentrations of population, jobs and services. The new trans-portation system is seen as a tool to reinforce the position of the citycentre in the region, and as a way to fight mounting congestion there(by both providing an alternative to the car and giving the car morespace above ground). From 1963 to 1966 an urban railway office isinstalled to work out the plans. Conclusion of the study is to start withthe construction of an eastern line – connecting the central railwaystation to urban renewal areas in the centre and the newly plannedsoutheastern urban expansion – and to follow later with anorth–south line. However, and signalling expert disagreement, analternative, incremental plan is also developed, envisaging a firstphase with a north–south line only and expansion of the bus andtram network as a substitute – at least for the time being – for anextensive underground railway network. Other plans follow, includingin 1967 one by the American professor D.A. Jokinen who proposesa system of radial urban motorways to connect a drastically restruc-tured city centre.This plan in particular has a shock effect on a publicopinion increasingly concerned with the fate of the historic city. In1968, however, conflicting plans and ongoing discussions notwith-standing, the city council decides ‘in principle’ to build the under-ground railway.The decade of urban renewal debates also seems toreach its resolution point with the publication in 1969 of rigorousplans envisaging the demolition of as many as 75000 dwellings inthe historic city.

While there is still enough consensus on the policy course withinthe city council, the railway and urban renewal proposals meetunexpected, strenuous resistance from the public. Leading thecontestation is an unorthodox coalition of local inhabitants fearingdisplacement and emerging urban youth movements wanting to

292 Planning

affirm their alternative visions of urban life. The planning machin-ery seems, however, unstoppable. In 1970 an agreement isreached with the national state on financing the eastern line and inthe same year the city council decides to start its construction aswell as preparations for a north–south line. Shortly thereafter dis-approval starts. The contestation, however, explodes and peopletake to the streets, seamlessly merging with protest against urbanrenewal. Popular pressure mounts to the point where the citycouncil has to reverse its decisions. The first change is regardingland use. In 1972 the council decides to build houses instead of athroughway on top of the inner-city section of the undergroundrailway tunnel. Amendments to transport follow: in 1974 the councildecides to complete the eastern line but to halt indefinitely furtherimplementation of the rest of the plan. The decision does not im-mediately calm the situation, and in 1975 there are violent riotsagainst the underground railway.

The first stretch of the eastern line opens in 1977, but a year laterthe policy change of course is made official. With regard to landuse, a local government report sanctions the shift by trading ‘urbanrenewal’ with ‘building for the neighbourhood’, that is, incremental,housing-led adaptation of the historic city, without displacement ofthe existing inhabitants. A ‘traffic circulation plan’ performs thesame function for transport, by stressing the need to strike abalance between accessibility and liveability, and to do this bymeans of improvement of the existing tram system, developmentof a coarse primary road network, a restrictive parking policy in thecity centre, and new cycle routes.

The contrast between the vision of the city and its transportsystem before and after these turbulent years can still be appreci-ated at a glance in the Mr. Visser plein, where the undergroundrailway eastern line enters the medieval city centre. Looking towardsthe periphery of the city one sees a large traffic thoroughfare flankedby modern, tall office buildings. Looking towards the centre onesees a much smaller street with plenty of space for bicycles andpedestrians and a mix of preserved and new residential buildings,with mostly small-scale retail on the ground floor. In between the twois the mouth of a never completed road tunnel, since converted intoan indoor playground. In the Nieuwmarkt underground railwaystation, pictures of the 1975 riots remind us how this all came about.

Sources: Honig, 1996; Dijkstra et al., 1999; le Clercq, 2002; Dienst RuimtelijkeOrdening Amsterdam, 2003a; Poelstra, 2003.

Evolutionary urban transportation planning? 293

BOX 13.2 THE LATE 1980s AND EARLY 1990s:A LAND USE POLICY TRANSITIONDISSECTED

In the late 1980s, while the municipality is struggling to attractoffice developments to the IJ banks area next to AmsterdamCentral Station, an intense, spontaneous market dynamics istaking place at peripheral locations along the southern andwestern motorway rings (see Figure 13.1, below). The traditionalorientation of the city on its port on the north side, which the IJbanks project tries to continue, is thus being subverted by an ori-entation of new developments towards the south side, better con-nected with the airport and the rest of the Randstad. In 1988 anexhibition and publication (ARCAM, 1988) first gives a synthetic,and to many shocking, impression of the new spatial reality takingshape. Putting together information on plans and projects untilthen available to the general public only in piecemeal form, theindependent ARCAM foundation shows how peripheral develop-ments are turning Amsterdam ‘inside out’.

The answer of the municipality to this evidence remains ambigu-ous. The official policy is that IJ banks is the most important lo-cation to develop, in order to reinforce the economic base of thecity centre, and that developments along the motorway ring are notto be allowed. But the possibility that firms leave or bypass the city– which desperately needs both the jobs and the land rents theycarry with them – altogether is too great a risk to adopt a hardstance.Thus, one after the other, exceptions to the policy are madeto allow companies to remain or locate at least within the cityboundaries. Nothing more, though: in 1993, an officer of the muni-cipality still declares: ‘An integral vision [for peripheral develop-ments] has no priority.Sometimes you just have to allow somethingin order to avoid firms escaping to competing locations’ (VanNierop, 1993, p. 95).

However, this approach is meeting with growing criticism. Marketactors lament that an urban design framework and coordination ofdevelopment would boost property demand and values in thesouthern urban fringe. Also, concerned local district authorities(these are the lower-level municipal governments installed in 1990)protest that transformations are occurring without reference to oneanother or to the context, making local impacts difficult to identifyand to manage. Quite strikingly, the big transport infrastructure

294 Planning

providers and operators – the department of public works, thenational railway company (NS) and the local transport company(GVB) – are not participating in the debate at this point. However,ambitious interventions and plans that will boost the accessibility ofthe southern urban fringe are in train, quite independently from theurban development debate. These include a light rail ring line, anorth–south underground railway line, capacity and connectionimprovements in the national and regional railway networks, andhigh speed train links to France and Germany.

In the early 1990s, unfolding events are making the position ofthe municipality increasingly difficult. While little happens at IJbanks, exceptions continue to accumulate along the ring: later onthe municipality itself will estimate the amount of office space thusdeveloped at around 300000 m2! Also, the large bank concernABN-AMRO demands the authorization to build its headquartersnext to Zuid station, on the southern urban fringe, exacerbatingtensions that are also maturing inside the municipality. Thencomes the proverbial last straw: in February 1994, the privatepartner of the IJ banks initiative withdraws, because of a lack offaith in the financial feasibility of the operation. A policy U-turnappears inevitable.

In the spring of 1994, following the local elections, a programmeagreement is voted by the new council, in which a new policy isindeed agreed. With regard to IJ banks, rather than an office con-centration, a mixed live–work area will be aimed at, anchored toactivities in the cultural and tourist spheres, thus reinforcing theemerging character of the historic city centre. Along the Zuidas(south axis), an office district of international standing will be pro-moted, bringing together in an integrated plan developments thatare at the moment occurring piecemeal. Contradicting what wasaffirmed only a year earlier, the city council states that: ‘For theZuidas, the area for large-scale offices, an integral plan will be pre-pared in order to avoid the situation whereby the development con-tinues incidentally’ (reported in Gemeente Amsterdam, 1996).

Sources: ARCAM (1988); Van Nierop (1993); Gemeente Amsterdam (1996);Bertolini and Spit (1998).

● Socio-demographic transitions There are two trend-breaking pointsin the population development of Amsterdam: in 1959, when thepopulation in the central city starts to decline in absolute terms, andin 1985, when it starts to grow again. Contributing to this are other

Evolutionary urban transportation planning? 295

demographic trend-breaking points (Wintershoven, 2000): 1973,when the national migration balance first starts improving; 1975,when the birth rate in the city first starts to rise again; the 1970s, whenthe international migration balance becomes positive (albeit it hasbeen strongly fluctuating since then). Crucially, change is not just ofa quantitative nature. In particular, the 1960s and 1970s were also aperiod of radical, and fairly abrupt cultural change, including far-reaching phenomena such as the emergence of mass consumption,female emancipation, a youth culture and so on. While many of thenew lifestyles thrived in the city, the more traditional, middle-classfamily households started a massive migration to the suburbs. Lateron, and further enriching the picture, the 1980s and 1990s saw theemergence of a new, extensive multicultural dimension in the cityfuelled by national and international migration trends, but also – andperhaps more surprisingly – a return to urban living by choice.

● Economic transitions As far as economic trends are concerned thereis a first major trend-breaking point in the 1970s, when, followingglobal developments epitomized by the first oil crisis of 1973, andafter a long period of sustained growth, the urban, regional andnational economy all fall into a decline that will last until the secondhalf of the 1980s. Older centres in the region, such as Amsterdam andHaarlem, are the worst hit. The 1990s were, on the contrary, growthyears, with the Dutch economy consistently performing above theEuropean average and the Amsterdam economy performing abovethe national average. The economic upheaval of the 1990s is of a de-cisively qualitative nature: growth has taken place within the contextof a radical shift from an industry-based, nationally coordinatedeconomy to a service-based, rapidly globalizing economy, and hasbeen spurred by the emergence of new leading sectors and locationsas business and financial services in the southern urban fringe,tourism, leisure and new media in the historic city centre, logistics inthe airport area. The emergence of Schiphol airport as one of a fewmajor European passenger and freight hubs is tightly linked to allthis.

● Land use policy transitions The 1970s are also a major transitionphase in land use policies. These are years of extreme policy turbu-lence, resulting in a radical change of course. This is particularly thecase in the dominating attitude towards the historic city centre, wherea shift from transformation to conservation occurs (see Box 13.1).This shift will also have a long-standing impact on land use policyelsewhere (think of the emergence and consolidation of the notion ofcomplementary subcentres in the urban periphery, where office

296 Planning

growth banned from the historic city centre has been increasinglyaccommodated). A second period of land use policy instability canbe observed at the beginning of the 1990s, with the failed attempt ata large-scale, CBD-like redevelopment of the IJ banks, adjoining thehistoric city centre (see Box 13.2). In contrast to these recurring‘city centre turbulences’, policies concerned with expansions inAmsterdam’s urban periphery and in the suburbs show much lessdebate and a remarkable stability (essentially the acceptance ofdecentralization, albeit in concentrated form).

● Land use transitions Changes in land use policies appear to havebeen both a consequence and a factor of actual land use changes (seeland use column in Table 13.1). In both Amsterdam and theNetherlands, land use policies in the post-war period can be seen asa reaction to wider urbanization trends first and suburbanizationtrends later. On the other hand, the conservative policies for the citycentre adopted at the end of the 1970s have been a factor in deter-mining the unique land use mix that has developed there in the sub-sequent, partial reurbanization phase (see ‘City centre trends,1975–99’ in Table 13.1). More indirectly, they have also had a role inthe emergence of alternative centres in the periphery and in thesuburbs, which has assumed a systematic rather than incidental char-acter following the disappearance of expansion opportunities in thehistoric city centre. There are sharp discontinuities in both broadertrends and local policies: the shift from urban expansion to subur-banization as the dominating force in the early post-war years, thepartial counterweight provided by reurbanization trends sincerestructuring of the economy in the late 1980s, and the transitions inland use policy discussed in Boxes 13.1 and 13.2.

● Transport policy transitions The land use policy debate in the 1960sand 1970s has been mirrored by intense transport policy debate in thesame period (see Box 13.1). Also in this case the focus was the his-toric city centre, here in the form of contestation of urban motorwayand, particularly, railway plans. The resulting shift in transportpolicy was perhaps even more pronounced than that in land usepolicy, with an effective halt to both urban motorway and urbanrailway expansion and a shift to an approach dominated by mobilitymanagement and marginal infrastructure interventions (see thetraffic plans of the 1970s in Table 13.1). Only during the 1990s, andon the condition of there being no harm to the existing urban fabricwill new urban infrastructure proposals be allowed to re-enter thepolitical arena (see the local transport policies in the 1985–99 periodin Table 13.1). Even then, the no-harm condition appears to have

Evolutionary urban transportation planning? 297

been a main rationale determining which links were going to be givenpriority, at least in terms of implementation (compare the relativelyshort time span from plan to realization of the elevated light rail ringline that opened in 1997 with the much longer and contested plan-ning and development process of the underground north–south line,which is set to open in 2011).

● Transport transitions Finally, continuities and discontinuities alsocharacterize actual infrastructure and mobility developments. Duringthe whole period, and particularly since the second half of the 1960s,the motorway system has been dramatically expanded. The same canbe said of the inter-regional railway network. The change was far fromquantitative alone, as the superimposition of motorway and railwaytangents on the existing radial systems profoundly affected the relativeaccessibility of locations in the region. These developments, illustratedin Figure 13.1, have had profound, largely unanticipated impacts onmobility and activity patterns in the region. The implications of thenew structure (the emergence of a new, polycentric and unstable set ofactivity centres and mobility flows) seem to have been fully appreciatedonly in the 1990s, when perception of the need to address issues at theurban regional scale has become ubiquitous.

As far as mobility is concerned the main radical change in the periodunder examination is the advent of mass motorization. Even in this case,policy makers were at first caught off guard, if not by the phenomenon initself, certainly by its pervasiveness. However, the reaction as maturedthrough the transport policy transition phase mentioned above and dis-cussed in Box 13.1 produced, at least within the historic city centre, a set ofmeasures that have long been able to manage mobility (see the local trans-port plans of the 1970s and 1980s in Table 13.1). The impact of this policyshift on actual behaviour is perhaps best epitomized by the reversal in thedecline in bike use since the 1970s (see trends in ‘Bike share in the city’ inTable 13.1), which has fundamentally contributed to give Amsterdam anexceptional share of non-motorized traffic (at 51 per cent of all trips in 1995it is by far the highest in the industrialized world in a large sample of citiesanalysed by Kenworthy and Laube, 2005).

Change in the system is path dependent, that is, existing patterns oftransportation networks, land uses and transport and land use policieslimit the scope for change.

The issue of path dependency is a wide-ranging one, cutting across multipleaspects and different layers of economic, social and cultural trends. There is

298 Planning

Evolutionary urban transportation planning? 299

Source: Adapted from Jansen (2003).

Figure 13.1 Changes in the built-up area and infrastructure in theAmsterdam region, 1967–2001

Built-up area

Railways

Motorways

Centre

Built-up area, 1967

Built-up area, 1967–2001

Railways, 1967

Railways, 1967–2001

Motorways, 1967

Motorways, 1967–2001

Centre, 1967

Centre, 1967–2001

Zaandam

Zaandam

Purmerend

Purmerend

1967

Amsterdam

Amsterdam

Almere

Haarlem

Amstelveen

Amstelveen

Haarlem

2001

N

N

5 km

Haarlemmermeer

Haarlemmermeer

5 km

SchipholAirport

SchipholAirport

path dependency in facts, but also in ideas, and the two are intertwined.Adequate treatment would require much more than a couple of paragraphs.Its consideration in this chapter will have to be limited and will only refer toland use and transport morphological aspects. The existence of opportuni-ties for and constraints to policy change determined by the existing urbanand network morphology is seen here as evidence of path dependency.

The pre-existing shape of land uses and transport infrastructure has con-ditioned subsequent developments in the Amsterdam region in many ways.The pre-war modernization of the Amsterdam city centre was compara-tively late and limited in nature. This can be related to the comparativelylate and limited industrialization of the Netherlands, but also to the sheersize of the city centre itself, in its turn a legacy of the global role ofAmsterdam in the seventeenth century (Terhorst and van de Ven, 2003).The existence of such a large, preserved city centre was a factor that firsthampered large-scale urban renewal and later favoured the policy shiftfrom transformation to conservation as described in Box 13.1. While therehave since been attempts to somewhat turn away from this decision (see Box13.2), the halt to large-scale, downtown-like restructuring of the city centrehas up to now proved irreversible, even if reasons have shifted (first largelybecause of social resistance, and later largely because of market prefer-ences: compare Boxes 13.1 and 13.2).

With regard to infrastructure, there is also a long line connecting theradial road and railway structure in place before the war, infrastructureplans and land reservations dating from as early as the first half of thetwentieth century, and actual transport and land use developments inthe period under discussion. Just focusing on the road and rail tangentsthat have proved so crucial for successive developments, the mostimportant decisions and actions contributing to the final result include:land reservations for a railway freight line around the city made at thebeginning of the twentieth century; land reservations for local roadsmade in the 1932 Amsterdam ‘general expansion plan’; the opening –starting in the 1970s and profiting from those rights of way – of railwaylinks to connect the airport of Schiphol to the rest of the country (anational government-led process); the realization since the 1970s of amotorway ring as part of the national motorway plan of 1966 – partlyusing the reservations for the freight line and partly those for the localroads; and the realization in the 1990s of a light rail line following theroute of the airport railway links.

The transport systems that were eventually put in place are of a totallydifferent nature from anything even thinkable prior to the war (not roadsbut motorways, not freight but passenger rail). Analogously, the nationalmotorway and railway planners of the 1960s and 1970s were not anticipat-

300 Planning

ing (if at all concerned with) significant impacts on the urban form.However, previous decisions largely determined where new infrastructuredevelopments, and thus indirectly also land use developments, were tooccur. The unique network morphology that eventually emerged, with acombination of radial and tangential connections, both road and rail,intersecting well inside the existing urban fabric, and since providing theinfrastructure backbone to urban and regional development (see Figure13.1), would never have emerged in this form without such early land reser-vations and national infrastructure plans.

During transition phases both the scope for policies to influence theoutcome and the unpredictability of such an outcome are greatest.

Proof for the first part of this hypothesis (scope) is the occurrence of qual-itative (rather than quantitative) change consistent with policy goals.Evidence for the second part (unpredictability) is the concomitant occur-rence of qualitative change that was not aimed at. Let us focus on the mainpolicy transition phase, the late 1960s and early 1970s (see Box 13.1). Thiswas a unique period because instability in different domains connected(think of the central role of the emerging youth culture in the contestationof urban renewal plans), resulting in radical policy change. It can beargued (even if it should be tested in more detail) that it is precisely becauseof this diffused turbulence that such radical, deliberate change was pos-sible. In other periods policy seems rather a reaction to broader, stable (andthus difficult to reverse) trends, having at best the effect of marginally con-ditioning the outcome (as with the relative concentration of suburbaniz-ation). On the contrary, qualitative policy change in the 1970s resulted inqualitative actual change, as documented above. The final outcome,however, was largely unpredicted. The conservative land use and transportpolicies in the city centre not only helped to preserve, as desired andexpected, its residential function. They were also, unexpectedly andunwillingly, a factor in its later gentrification and the development of aburgeoning tourist and leisure industry there (Terhorst and van de Ven,2003). Furthermore, constraints on city centre development indirectlyhelped shift the focus of economic activity in the region towards theemerging centres in the periphery and the suburbs, not an entirely unfore-seen but certainly at the time not even a deliberate policy goal. As far asmobility is concerned, while the outcome of the policy turn in the 1970swithin the city (fewer cars, more bikes, and a more liveable public space)was by and large an explicit goal, the related development of a diffused,multi-centred urban region where the car dominates and adequate publictransport is lagging behind was not.

Evolutionary urban transportation planning? 301

Policy Implications

While still exploratory and necessarily limited in scope, the analysis abovedoes provide some evidence that the Amsterdam land use and transportsystem has changed in an evolutionary, complex fashion, in the sensedefined by the first set of hypotheses. Periods of incremental change havebeen followed by periods of radical change, path dependency has played adecisive (in this chapter only superficially explored) role, and transitionphases have been characterized by both possibilities to affect the outcomeand inability to predict it. Let us now move to the second set of hypothe-ses. These posit that because the systems behaved in an evolutionary,complex fashion, successful policies needed to:

Build upon the unique set of opportunities and constraints for changedetermined by a specific historical development path and local combi-nation of factors.

The fact that successful policies (that is, policies that have achieved theirdeclared goals) have such characteristics is seen as verification of thishypothesis.

Successful policies in Amsterdam in the period under examination haveexplicitly or implicitly recognized the specificity of the city, that is, the exis-tence of path dependency. This seems true in phases of both incrementaland radical change. Again, the discussion will be limited here to the mor-phological aspects. The repeated failure of attempts at radical transfor-mation of Amsterdam city centre and the success of more conservativeland use and transport policies there are the clearest example (see Boxes13.1 and 13.2).

It is also important, however, to underline here that the acknowledge-ment of path dependency by no means implies that radical change is bydefinition impossible, or should not be striven for. The urban and regionalstructure that has ultimately emerged in Amsterdam is profoundly differentfrom the pre-existing one (see Figure 13.1). In the case of the conservativepolicies for the city centre, Terhorst and van de Ven (2003) even contendthat it is has been precisely this ‘freezing’ of the built environment therethat, by limiting the scope for large-scale, coordinated transformation, hasparadoxically given the market unprecedented freedom to re-shape landuses on the micro level, in often surprising ways. The system-wide effects ofthe implementation of relatively undisruptive, but still very far-reachinginfrastructure developments as the expansion of the motorway system andinter-regional railway links on the urban fringe and in the region areanother case in point.

302 Planning

In both examples above the actual impacts – and particularly the emer-gence of a strongly polycentric urban system – differed greatly from any-thing most were able to imagine at the time. Actually applying policies (thatis, experimenting) seems to have been necessary to realize the full impli-cations. The crucial question for change-minded policy makers seems thento be the following. How to identify interventions that are able to disruptthe present functioning of the system as little as possible while affecting itsfuture development as much as possible? At the same time, the limits to thepredictability of the outcome should also make us aware of the need toincrease the resilience and adaptability of the system, which leads us to thenext two hypotheses.

Increase the resilience of the system, that is, its ability to keep function-ing in the face of unexpected change. This seems especially important forthe shape of transportation networks, as this is relatively difficult/slow tochange.

The fact that successful policies were policies that have proved effective inqualitatively different contexts (that is, both before and after trend-breakingpoints) is seen as evidence for this hypothesis. There are several examples ofresilient interventions in the Amsterdam case. It is especially the shape ofthe infrastructure networks (not their function!) that seems to have had thischaracteristic. The combination of motorway and railway radials and tan-gents has proved able to support a wide variety of developments across thewhole period, including shifting foci of economic and social activity,different transport technologies, and the two major policy transitionsdescribed in Boxes 13.1 and 13.2. The necessity of robust choices withrespect to network morphology seems all the more crucial as theAmsterdam case also shows how alterations in its basic shape are extremelydifficult to implement (think of the failed attempts at infrastructure pene-tration into the historic city centre, and conversely of the positive role oflong-standing land reservations in its expansion). An intriguing policy ques-tion follows. Is it possible to identify in advance such resilient interventions,and – if so – how? I shall get back on this in the conclusions.

Increase the adaptability of the system, that is, its ability to react to un-expected change. This seems especially important for land use regula-tions and mobility management measures, as these are relativelyeasy/fast to change.

The fact that, in order to be effective, policies that were not resilient in thesense discussed in the previous section needed to be adapted, is seen as

Evolutionary urban transportation planning? 303

proof for this hypothesis. The Amsterdam case also shows several examplesof policy adaptation, particularly as far as land use policy and the mobil-ity management side of transport policy are concerned. The most poignantexample seems, once again, the radical change of course of transport andland use policies in the 1970s (see Box 13.1). Such adaptation has been anessential condition for the development of the new, quite successful policymix that – at least as far as the historic city is concerned – has shown to beviable up to the present day (whether this will also hold for the future is, ofcourse, a different matter). Policy change proved, however, all but a naturalprocess. A decade of intense conflict including violent riots and a ‘policytrauma’ that is still felt in the city was needed to achieve it. In many ways,the existing plans and planning institutions appear to have been, at least in-itially, hampering rather than promoting adaptation. A somewhat differentpicture is offered by the land use policy transition of the early 1990s (seeBox 13.2). Here, also, the capability to adapt has been an essential con-dition for the development of what most now see as a much more effectivestrategy for dealing with the reality of a multi-modal, multi-centred urbanregion. However, in contrast to the 1970s, the transition seems to haveoccurred in a much less traumatic manner (perhaps a sign that the planningsystem has become more adaptable?). The policy question is thus whetherand how such policy transitions could be made easier, or how the adapt-ability of policies could be increased. In other words: how can policies bemade more responsive to (unexpected) reactions from the society at large?But also: how can this be done without reducing too much the just as nec-essary stability of the policy context, that is, its resilience? This issue willalso be addressed in the conclusions.

4. DISCUSSION AND CONCLUSIONS

The central contention of this chapter is that an urban transport and landuse system capable of supporting economic and socio-demographic changeis also one capable of continuing to function in the face of change, that is,it must be a resilient system. Second, an urban transport and land usesystem capable of supporting economic and social change must be able tochange itself in response to change in the socio-economic environment,that is, it must also be an adaptable system. The Amsterdam case showsboth the workings of resilience and adaptability, and specific ways (that is,ways that take account of path dependency) of achieving them. Theresilience of the system is perhaps best shown by a transport network mor-phology (the combination of radial and tangential links, both road andrail) that has provided a relatively stable base for the radical shift from a

304 Planning

monocentric to a polycentric urban structure. The adaptability of thesystem is perhaps best shown by the (ultimately) successful re-shaping ofthe policy course, particularly in terms of land use and mobility manage-ment, in response to systemic crises. There is a link between the two. Theresilience of the transport network morphology has been a condition forthe adaptability of land use and mobility management strategies, becauseat all times it has allowed a choice between different land use and mobilitymanagement strategies.

However, the Amsterdam case also shows the limits of a purely rationalapproach (in the sense of ‘rational choice’, as in Simon, 1957, 1969; Marchand Simon, 1958) to achieving resilience and adaptability. The presentnetwork morphology is the result of a very long chain of decisions andactions, often unconsciously or unwillingly contributing to the final result.The land use and mobility management policy mix also emerged after aprotracted period of conflicts and contradictions, and many effects werenot anticipated. These limits to predictability are by no means specific ofthe Amsterdam context. In Britain for instance, the expectation that, fol-lowing the development of new radial and tangential roads, growth wouldstill take place in the city centre was long undisputed, while no model hadanticipated inner-city decline or massive decentralization (Banister, 2002).

In terms of decision making, the approach emerging from theAmsterdam case seems to contain elements of both the incremental model(Lindblom, 1959, 1968; Braybrooke and Lindblom, 1970) and the rationalmodel (Simon, 1957, 1969; March and Simon, 1958). It is incremental inthat it points to the role of the existing, historically grown situation inshaping the discussion around problems and solutions. It is rational in itsattempts at drawing implications from the awareness of the long-termimplications of decisions, particularly as they might affect the very scopefor choice at a later stage. This characterization by no means implies thatsuch a decision-making model was deliberately pursued in Amsterdam.Rather, it emerged through an often-painful process of trial and error. Theinteresting question is, of course, what more conscious efforts in this direc-tion would deliver, in Amsterdam and elsewhere.

This preliminary, exploratory analysis points to both research and policychallenges. As far as research is concerned, there is a need to gain bothgreater depth and greater breadth. Greater depth is most notably needed inorder to more fully capture the dynamics of transition phases. In particu-lar, the complex interplay of path dependency (of both facts and ideas) andunpredictability, and of possibilities of and limits to influencing theoutcome need to be understood better. Most notably, greater breadth isneeded in order to test the applicability to other geographical contexts ofthe proposed characterization of developments and of ways of achieving

Evolutionary urban transportation planning? 305

resilience and adaptability. As far as policy is concerned, the practicalimplications of an urban transportation planning also geared at increasingresilience and adaptability of the system should be further elaboratedupon. In addition, the complementarities with other, forecasting-basedapproaches should be explored.

What sort of urban transportation planning would this lead to? A refer-ence to Christensen’s (1985; see also Gifford, 2003) classic characterizationof uncertainty in planning can help make a first step. According toChristensen (Figure 13.2a) planning problems and approaches can be char-acterized in terms of the uncertainty about goals and the means to achievethem (or ‘technology’). The term ‘technology’ is used here in the broad senseof ‘means to achieve goals’. In this respect a transportation system is a tech-nology, but also a parking regime, or a marketing campaign. Furthermore,the term is inclusive of the economic, social and cultural institutions thatidentify the context in which a technology is developed and applied.Different sorts of uncertainty require different planning approaches. Whengoals are not agreed and the technology is unknown there is ‘chaos’. InChristensen’s interpretation these are untreatable planning issues, anduncertainty needs somehow to be reduced in order to proceed. When feas-ible, this should certainly be the preferred option. However, in many (evenif by no means all) situations uncertainty seems not reducible. What to dothen? Abstracting from the discussion of the Amsterdam case, Figure 13.2bsketches a possible approach. The starting point is the observation that evenwhen goals are not agreed a distinction can be made between goals that areindependent of the future technological context (as ‘promoting the growthof the urban economy’) and goals that are not (as ‘promoting the growth ofa specific economic sector in a specific location’). Analogously, even whenthe technology is unknown a distinction can be made between a technologythat only has the potential to serve limited goals (as a transportation systemconnecting a limited number of places in a limited number of ways) and atechnology that has the potential to serve more goals (such as a transpor-tation system connecting more places in more ways). When goals are bothnot agreed and dependent on a specific future technological context andtechnologies are both unknown and can only serve limited goals, optionsshould be kept open, thus preserving the adaptability of the system. On thecontrary, not agreed goals that are independent of the technological contextand unknown technologies that can serve many goals are, at least poten-tially, robust goals and technologies and should, with reference toChristensen’s characterization, be bargained over and/or experimentedwith. Because of the limits to predictability, only actual bargaining andexperimentation – or ‘policy experiments’ – will tell how true this potentialis. If this does prove to be the case, policies should be brought further

306 Planning

Evolutionary urban transportation planning? 307

Source: Christensen (1985).

Figure 13.2a Coping with uncertainty in planning

Agreed kn

own

unkn

own

Tec

hnol

ogy

Not agreed

Goal

1. Programming 2. Bargaining

3. Experimentation 4. Chaos

?

Source: Author’s own work.

Figure 13.2b Coping with irreducible uncertainty in planning (or ‘chaos’)

Not agreed goal

Experimentbargain

Keep optionsopen

Can

ser

vesi

ngle

goa

lsC

an s

erve

mul

tipl

e go

als

Robust goal,redfine technology

Unk

now

n te

chno

logy

Independent of thetechnological context

Dependent on thetechnological context

Robust technology,redefine goal

towards implementation, as they are likely to improve the resilience of thesystem. If not, the need to keep options open will be reintroduced. It isthrough such a recursive, exploratory process that the system can gain bothresilience (by means of robust measures) and adaptability (by means ofkeeping options open).

There is, however, a caveat. The above suggests that in cases of irre-ducible uncertainty and insufficient robustness, options should always beleft open. However, it is easy to think of situations where action might stillbe desirable (think of the development of an innovative transportationsystem with highly uncertain, but potentially highly rewarding impacts).The approach sketched above could still be useful. First, it will point to theneed to keep exploring ways of increasing the resilience and adaptability ofthe action (perhaps the innovative transportation system can be brokendown into smaller components and realized in a more incremental, experi-mental way?). Second, when further redesign is not deemed possible, it willmake explicit that decision makers are taking a risk with an unpredictableoutcome. In political (rather than technical) terms this can still be accept-able (or even desirable, as taking risks has always been considered a hall-mark of leadership). Even in this last case, however, allowance should bemade for learning, that is, to treat implementation as much as possible as a‘policy experiment’.

Economic and socio-demographic changes shape urban transport andland uses, but the latter provide in their turn a still essential physicalsupport to the former. In the face of rising complexity and persistinguncertainty about the future, planners should devote more energy tounderstanding the evolutionary, complex nature of change in urbanland use and transport systems, and, accordingly, to finding ways of pro-moting their resilience and adaptability. This would complement other,forecasting-based approaches, and allow transport providers to developand transport users to choose between different ways of moving around,both in the shorter and, most importantly, the longer term. The latterappears all the more urgent in the face of real uncertainty about the futureviability of the presently dominating transport solutions and a tendencynot to recognize this by those taking decisions. In this respect, the classicdefinition of sustainability proposed in the Bruntlandt report (WorldCommission on Environment and Development, 1987) still provides apoignant evaluation criterion. How does a particular transport and landuse policy affect the possibility of future generations making their ownmobility choices? An exploratory attitude seems essential, as the answerwill be different in different contexts, and contexts will keep changing,unpredictably.

308 Planning

REFERENCES

Allen, P.M. (1997), Cities and Regions as Self-Organizing Systems: Models ofComplexity, Amsterdam: Gordon & Breach.

ARCAM (Amsterdam Centre for Architecture) (1988), Boom Town Amsterdam,Amsterdam: ARCAM.

Banister, D. (2002), Transport Planning, 2nd edn, London and New York: Spon Press.Batten, D.F. (2001), ‘Complex landscapes of spatial interaction’, Annals of Regional

Science, 35(1): 81–111.Bertolini, L. and T. Spit (1998), Cities on Rails, London: Spon Press.Boschma, R., K. Frenken and J.G. Lambooy (2002), Evolutionaire Economie. Een

Inleiding, Bussum: Coutinho.Brand, A. (2002), ‘Het stedelijke veld in opkomst. De transformatie van de stad in

Nederland gedurende de tweede helft van de twintigste eeuw’, PhD dissertation,University of Amsterdam.

Braybrooke, D. and C. Lindblom (1970), A Strategy of Decision, New York: FreePress.

Bruhèze, A.A. de la and F. Veraart (1999), Fietsverkeer in Praktijk en Beleid in deTwintigste Eeuw, Den Haag: Ministerie van Verkeer en Waterstaat.

Christensen, K.S. (1985), ‘Coping with uncertainty in planning’, APA Journal,Winter: 63–73.

Dienst Ruimtelijke Ordening Amsterdam (2003a), Stadsplan Amsterdam.Toekomstvisies op de ruimtelijke ontwikkeling van de stad. 1928–2003, Rotterdam:Nai Uitgevers.

Dienst Ruimtelijke Ordening Amsterdam (2003b), Structuurplan Amsterdam.Kiezen voor stedelijkheid, Amsterdam: Dienst Ruimtelijke Ordening Amsterdam.

Dijkstra, C., M. Retisma and A. Rommerts (1999), Atlas Amsterdam, Bussum: Thoth.Dimitriou, H. (1992), Urban Transport Planning: A Developmental Approach,

London and New York: Routledge.Dosi, G. and R.R. Nelson (1994), ‘An introduction to evolutionary theories in eco-

nomics’, Journal of Evolutionary Economics, 4: 153–72.Gemeente Amsterdam (1996), De Zuidas Amsterdam, Amsterdam: Gemeente

Amsterdam.Gifford, J.L. (2003), Flexible Urban Transportation, Oxford: Elsevier Science.Honig, R. (1996), ‘Railinfrastructuur in Amsterdam. Trein, metro, sneltram’, in

planAmsterdam, 2(9/10): 1–26.Jacobs, J. (1961), The Death and Life of Great American Cities, New York: Random

House.Jansen, A. (2003), ‘Co-evolutie van infrastructuur en verstedelijkingsstructuur’,

Master’s Thesis, University of Amsterdam.Kenworthy, J. and F. Laube (2005), ‘An international comparative perspective on

sustainable transport in European cities’, European Spatial Research and Policy,12(1): 11–50.

le Clercq, F. (2002), ‘Planologie en mobiliteit in Amsterdam’, in H. Knippenbergand M. van Schendelen (eds), Alles heeft zijn plaats. 125 jaar Geografie enPlanologie aan de Universiteit van Amsterdam, 1877–2002, Amsterdam: Askant,pp. 399–419.

Lindblom, C. (1959), ‘The science of muddling through’, Public AdministrationReview, 19: 79–88.

Lindblom, C. (1968), The Policy-Making Process, Englewood Cliff, NJ: Prentice-Hall.

Evolutionary urban transportation planning? 309

March, J. and H. Simon (1958), Organizations, New York: Wiley.Meyer, M.D. and E.J. Miller (2001), Urban Transportation Planning, 2nd edn, New

York: McGraw-Hill.Nelson, R. and S. Winter (1982), Evolutionary Theory of Economic Change,

Cambridge, MA: Harvard University Press.Poelstra, H. (2003), ‘Eerst infrastructuur, dan beleid’, in Dienst Ruimtelijk Ordening,

Gemeente Amsterdam (ed.), Stadsplan Amsterdam. 1928–2003, Rotterdam: Naiuitgevers, pp. 118–29.

Portugali, J. (1999), Self-Organization and the City, Berlin: Springer.Regionaal Orgaan Amsterdam (2000), Evaluatie Regionaal Verkeer en Vervoerplan,

Amsterdam: Regionaal Orgaan Amsterdam.Regionaal Orgaan Amsterdam (2004), Regionaal Verkeer en Vervoerplan,

Amsterdam: Regionaal Orgaan Amsterdam.Rommerts, A. (1997), ‘De dynamische regio. De ruimtelijke opbouw van de regio

Amsterdam’, in planAmsterdam, 3(8/9): 1–33.Simon, H. (1957), ‘A behavioral model of rational choice’, in H. Simon (ed.),

Models of Man, New York: Wiley, pp. 241–60.Simon, H. (1969), Sciences of the Artificial, Cambridge, MA: MIT Press.Szejnwald Brown, H., P.J. Vergragt, K. Green and L. Bechicchi (2004), ‘Bounded

socio-technical experiments (BSTEs): higher order learning for transitionstowards sustainable mobility’, in B. Elzen, F.W. Geels and K. Green (eds), SystemInnovation and the Transition to Sustainability, Cheltenham, UK andNorthampton, MA, USA: Edward Elgar, pp. 191–219.

Terhorst, P. and J. van de Ven (2003), ‘The economic restructuring of the historiccity centre’, in S. Musterd and W. Salet (eds), Amsterdam Human Capital,Amsterdam: Amsterdam University Press, pp. 85–101.

van der Bergh, J. and D. Fetchenhauer (2001), Voorbij het Rationele Model:Evolutionaire Verklaringen van Gedrag en Sociaal-Economische Instituties, DenHaag: NWO/MaGW

Van den Heijden, K. (1996), Scenarios: The Art of Strategic Conversation,Chichester: John Wiley & Sons.

Van Nierop, I. (1993), ‘Verdichting rond Stations’, Master’s Thesis, University ofAmsterdam.

Wintershoven, L. (2000), Demografisch Eeuwboek Amsterdam. Ontwikkelingentussen 1900 en 2000, Amsterdam: Dienst Ruimtelijke Ordening Amsterdam.

World Commission on Environment and Development (1987), Our CommonFuture, Oxford: Oxford University Press.

310 Planning

absorptive capacityin firms 11and foreign spillovers 121–44and ICT adoption 258–9and innovation 11productivity growth see productivity

growthand technologies 128, 129

Accenture 52Acorn Computers 36–7, 42adaptationism

in evolutionary biology 207–8, 209institutional, Cambridge 43–4

agentsbounded rational 2, 6co-location of 10efficiency of 205interaction of 183, 210, 241, 249representative 1–2, 5see also networks

agglomerationautomobile industry 70, 86–8, 90automobile industry, US 7, 8, 14–15,

30, 70, 74–80creative destruction 2extraordinary 87–8and information spillovers 158and infrastructure 16–17and innovation 70mobile production factors 1, 2and patents 235and performance 70and production costs 89productive capacity in Europe 231radio industry 73, 74, 77, 86and shakeouts 88–9and spin-offs 30, 86–8television receiver industry 70, 71–4,

77, 79, 85–6, 89tire industry 84, 86–8, 90and wealth distribution 1, 18see also clusters

agglomeration economiesand automobile industry 86–8, 90location choice 8, 69–70, 79–80,

86–8, 90new firm location 8and related industries 9as spillover 88spin-offs 30and television industry 74, 79–80and tire industry 84, 86–8see also path dependency; spin-offs

Air France 51Alchian, A. 208Allergan 52Amadeus 51Amin, A. 1, 163Amsterdam see NetherlandsAndersson, C. 12Antonelli, C. 2, 182ARM 36–7, 42Arrow, K. 93, 152, 158, 161, 163Arthur, W.B. 2, 7, 8, 29–30, 180, 210Attaran, M. 204Audretsch, D. 163, 258, 262Audubon Society 99, 104, 105–6automobile industry

agglomeration 70, 86–8, 90and agglomeration economies 86–8,

90agglomeration in US 7, 8, 14–15, 30,

70, 74–80geographic structure, evolution of

76–90heterogeneity 79incumbents as source of competence

85–6knowledge diffusion 8, 9, 16knowledge, tacit 79, 80mergers 78networks 8pre-entry experience 79related industries, influence of 88

311

Index

seeding industries 76–7, 79shakeouts 88–9spillovers 88spin-offs 77–8, 79, 85, 86, 87–8, 90subcontracting 78, 79, 88technological progress 77UK 8–9, 16

Aventis 52Aydalot, P. and D. Keeble 6Aztema, O. and J. Weltevreden 258

Baldwin, J.R. and W.M. Brown 203–4,206, 212, 213–14, 215, 218

Baptista, R. 163Barabasi, A. and R. Albert 10, 11–12,

234Barrat, A. 12Basu, S. and D. Weil 123Battese, G. and T. Coelli 126, 127Becker, M. 18Begg, I. 203Bell, M. 162, 175Bertolini, Luca 279–310biotechnology research, Cambridge 7,

27, 31, 32, 33, 34, 39–41Birke, Daniel 180–200Black, D. and V. Henderson 203Bonaccorsi, Andrea 256–76Borgatti, S. 167, 182, 240Boschma, Ron A. 1–24, 30, 162, 163,

166, 203, 212, 281Bottazzi, G. 2Brakman, S. 1, 204Brenner, T. 2Breschi, S. 2, 10, 18, 28, 29, 88, 162,

182, 235, 236, 240, 241Brundtland Report 308Buenstorf, G. and S. Klepper 70, 80,

81, 82–3, 84, 87, 89Buick/General Motors 78Burt, R. 10business angels 60

Cadillac 78Cambridge

academic entrepreneurship 28Acorn Computers 36–7, 42bioinformatics 41biotechnology research 7, 27, 31, 32,

33, 34, 39–41

chemical engineering department 37clustering 30, 31–41customers, international 29, 30Defense Advanced Research Agency

(DARPA) 44Element-14 36entrepreneurship 43–4entry barriers, low 35firm turnover 34Greater Cambridge Partnership 44Human Genome project 41ICT 35–41, 43industrial ink jet printing 38–9Innovation Centre 43instrumentation sector 34–5internationalisation 41–2IP rights 36, 37–8, 39, 41, 43Medical Research Council 41Network Computer 36networks 28, 39, 43–4new business models 42new entrants 32, 35–7, 39, 41, 42, 43outsourcing 29, 39scientific instrumentation 32–5Small Business Innovation Research

(SBIR) 44spin-offs 28–9spin-outs 35–41, 42survival rates 32, 35technical design consultancies 37–8,

42technology transfer 43university links 35, 39, 40US investment 41–2, 43, 44venture capital 41–2

Cambridge Antibody Technologies39–41

Cambridge Consultants Ltd (CCL)37–8

Cambridge Display Technology andPlastic Logic 39

Cambridge Instruments 34Cambridge Scientific Instruments 34Camison, C. 161, 164, 172Canada, ICT regional disparities 257Caniëls, M. 2, 12Cantner, U. 9, 183Cantwell, J. and S. Iammarino 18Capello, R. and A. Faggian 161, 164,

165, 174

312 Index

Carrington, P. 10Carroll, G. and M. Hannan 9Castells, M. 3, 14CDT 42Celltech 41Chandra, S. 204, 210, 213Chang, H.-J. 15Chile, wine clusters analysis 166–76China, ICT regional disparities 258Chiroscience 41Christensen, K. 306Chrysler 79Cird/Galderma 52CIS 42clusters

Cambridge 30, 31–41embeddedness in 164, 165–6, 168,

171–4as endogenous process 28–30geographic 69heterogeneous performance, effects

on 164–6heterogeneous performance and

networks 161–79and innovation 162, 163, 172local supply chain benefits 28–9network analysis 10, 11, 240–45and patents 235spatial, evolutionary models 29specialised 5, 30and spin-offs 29–30US 30, 70wine in Chile 166–76see also agglomeration

Cohen, W. and D. Levinthal 122, 128,165, 258

Coleman, J. 147, 151collective action, and technological

change 95–6collective invention 162comparative advantage theory 206competition

and co-location 29competitive advantage 163cost 7–8in Europe 235and exiting firms 7international 9, 34, 74and regional development 204, 210,

257

and spatial concentration 9US 74

complexitycomplex knowledge transfer 148–53informational and knowledge

diffusion 147–60urban transportation planning 280,

281, 308congestion 6, 70, 223connectivity 11, 12, 14, 15, 62, 64, 65

see also networksconsumers

and bounded rationality 11co-locating with 28and market networks 180networks 11peer effect 180preference changes 94and social status 180

convergence 12, 122, 123, 257Cooke, P. 11Cordis/Zeneca 52Cowan, R. 11, 180, 183Crang, P. 113creative destruction 2

DARPA (Defense Advanced ResearchAgency) 44

data envelopment analysis 125, 126Davis, G.F. 93, 94, 95, 147De Jong, M. 6DEC/Compaq 52defence spending 45deregulation 94developing countries, technological

progress 127, 257Diamond Rubber 82DiMaggio, P. and W. Powell 114disk drive industry 90Dissart, J. 204diversity

evolutionary potential 12, 205, 206,207–8, 209–11

in firms 208Herfindahl index 213–14and innovation 15, 207, 210in new industries 77optimality and stability 208–9and path dependency 210portfolio theory 211–13

Index 313

and productivity levels 123radio industry 73, 86regional development 13, 16,

204–23, 205as risk-minimizing strategy 210, 211and selection 205, 207–8, 210Sophia-Antipolis science park 50,

52, 60, 62stability and growth 206–13and survival 208tire industry 82, 83US 203–29US county business patterns

employment data 213–23Domino Printing Services 38, 39Dopfer, K. 19Dosi, G. 2, 19, 165, 207, 281dotcom revolution 56, 59–60, 62

see also ICTDow Chemical 52Duranton, G. and D. Puga 203, 211, 213duration models 8

Echointeractive 62economic geography 93, 206, 208economic growth

decline and vested interests 12efficiency improvements 13and firm success 7inputs growth 13product life cycles 8, 12and regional development 204,

206–13and routines 204Sophia-Antipolis science park 6–7,

15, 16, 17, 55–6and spillovers 13uneven 2, 5urban 14and variety 13, 204see also productivity growth

Edison Electric Institute 105, 106efficiency

adaptive 205, 207–10dynamic and static 15and specialization 212

electrical power industry, US see USpower industry

Ellison, G. and E. Glaeser 69, 87, 270Elmjet 39

employment levelsCambridge 31–2, 33Sophia-Antipolis science park 49,

51, 52–3, 55, 56, 59US county business patterns 213–23see also unemployment

entrepreneurial opportunityconstruction 93–120

diagnostic framing 100–102legal and regulative structure 97–8motivational framing and resource

mobilization 104–7prognostic framing 100, 102–4and social movements 95–7, 100, 104US power industry see US power

industryentrepreneurship

academic 28Cambridge 43–4and clusters see clustersco-evolutionary process of 6and collective action 95–7competent, importance of 15and corporate form, problems with

96demand side 94endogenous nature of 6and evolutionary economic

geography 5–7and geographical proximity 5high-tech 6and productivity growth see

productivity growthregional, uneven rates of 5Sophia-Antipolis science park 53,

54, 56, 60, 62environmental movement, US 99, 101,

103, 105–7, 109–11Enzymatics 41EPO co-patenting applications 231–3,

235–6, 238, 242–7, 249–50Erasmus student flows 231–3, 236–7,

238–40, 243, 246–7, 250Eriksson 45Essletzbichler, Jürgen 2, 5, 203–29ETSI (European Telecom Standard

Institute) 56, 62Europe

agglomeration and productivecapacity 231

314 Index

competitiveness 235EPO co-patenting applications

231–3, 235–6, 238, 242–7,249–50

Erasmus student flows 231–3, 236–7,238–40, 243, 246–7, 250

Fifth Framework Programme (5FP)235

income levels 231inter-regional knowledge flows

230–55Internet hyperlinks 231, 232, 233–4,

238, 239, 240, 242, 243, 246–7,248

knowledge flows network analysis240–45

Lisbon 2000 European Council 231,236

Maastricht Treaty (Article 130G)235

mobile networks pricing strategy 181

R&D facilities from extra-Europeanfirms 55

research networks 231, 232, 234–5,238–40, 242, 243, 246–7, 249

Single European Act 235Sophia-Antipolis science park see

Sophia-Antipolis science parkevolutionary biology

adaptationism 207–8, 209survival in 207

evolutionary economic geographyapplications of 1–24, 162, 163–4case-study research 3and entrepreneurship 5–7and institutions 2macro levels 3, 4, 12, 13meso levels 3, 4, 7, 163, 164, 165methodology 3micro levels 4, 163, 164, 165network analysis 10–12, 240–45new industries see new industriespath dependency 2, 6, 18spatial concentration 9survival analysis 9territorial differences 2see also geographical distance;

geographical proximity;institutional economic

geography; new economicgeography

evolutionary economicsdiversity and selection 12, 205, 206,

207–8, 209–11and innovation 5, 182–3and knowledge diffusion 183regional development 15and routines 18, 207–8social network analysis 11–12,

182–3and social network theory 182–3spatial clusters 29urban transportation planning see

urban transport planningevolutionary growth theory 18

Farrell, J. and G. Saloner 180FDI 121, 246Feldman, M.P. 163, 203, 232Firestone 80, 81, 82firms

absorptive capacity and innovation11

in agglomerated regions seeagglomeration

clusters see clustersco-location 29core competencies 5diversity in 208economic behaviour, variety in 282economic growth and success 7embeddedness of 164, 165–6, 168,

171–4evolutionary growth theory 18, 165,

281–2exiting 7, 8, 57, 74founder history and location 5, 6, 8heterogeneous 5, 11, 129, 161–79high-tech 6, 10innovation in see innovationinternal resources 165, 174knowledge diffusion see knowledge

diffusionlocation behaviour 5, 6, 28, 29migration 13and networks see networksnew entrants see new entrantsnon-local relationships 11outsourcing 6, 29

Index 315

performance and geographicalproximity 162–4, 174–5

productivity levels see productivitylevels

in related industries 84–5relational proximity 161, 162–4relocation 5, 6routines in 5, 8size and spin-offs 7spatial concentration 7, 8, 9supply chain externalities 28technology sharing 128

Fleming, Lee 147–60Fligstein, N. 93, 94, 95, 97footwear industry 11, 89–90Foray, D. 15Ford Motor Co. 75, 78, 79France

engineering sector 16infrastructure 17, 44, 50science-industry relationships 60Sophia-Antipolis science park 6–7,

15, 16, 17, 44, 48–66university Internet hyperlinks 234see also Europe

free-riders, and collective action 94Freeman, C. 10, 12Frenken, Koen 1–24, 30, 149, 162, 208,

211, 212Friedland, R. and R. Alford 95Friends of the Earth 105Fujita, M. 1, 211

Galliano, D. and P. Roux 258Garnsey, Elizabeth 7, 27–47General Motors 75, 79geographic

attractiveness 29clusters 69

geographical distance 5, 163, 231–2gravity equations 246–51see also evolutionary economic

geographygeographical proximity

and entrepreneurship 5and knowledge diffusion 10, 14, 18,

152–3, 155–7, 161, 162–4networks and 10and performance 162–4, 174–5social boundaries and 152, 155, 157

see also evolutionary economicgeography

Germanyenvironmental sector 16synthetic dye industry 10university Internet hyperlinks 234see also Europe

Gertler, M.S. 1ghost towns 212Giampietro, M. and K. Mayumi 206,

207, 211Gifford, J. 279, 306Giugni, M. 95, 114Giuliani, Elisa 11, 161–79Glaeser, E. 69, 87, 204, 270Goodrich 80, 81, 82, 87–8Goodyear 80, 81Grabher, G. 12, 210Granovetter, M. 10, 18, 114, 152, 164,

210gravity equations 246–51Guimerá, R. and L. Amaral 12Gulati, R. 166

Hagedoorn, J. 10Hall, P. and A. Markusen 6Hannan, M. 9Hansatech 39Harris, C. 93Harvey, D. 203, 210Heffernan, Paul 27–47Henderson, J. 203, 204heterogeneity

accumulation process in Sophia-Antipolis science park 51, 52–4,56

automobile industry 79effects on clusters performance

164–6firms 5, 11, 129, 161–79in industry and labour productivity

139knowledge bases 165performance of clusters and

networks 161–79routines 7, 8in routines 7, 8Sophia-Antipolis science park 51,

52–4, 56spatial in ICT adoption 268

316 Index

and technical efficiency 129tire industry 84in urban transportation planning

279Hewlett Packard 34–5high-tech

region, Cambridge 6, 15, 17, 27–47SMEs 57, 58–60, 62–4see also ICT

Hodgson, G. 1, 205, 207Hohenberg, P. and L. Lees 14Holling, C.S. 205, 206, 208Hudson, R. 203Human Genome project 41

ICTand absorptive capacity 258–9Cambridge 35–41, 43domain names as proxy for adoption

259–64dotcom revolution 56, 59–60, 62high-tech entrepreneurship 6infrastructure 14Internet hyperlinks 231–4, 238–40,

242–3, 246–8local digital divide 257–9and market characteristics 259new technologies, territorial

adoption of 256–76regional disparities 257, 258regional disparities, US 257Sophia-Antipolis science park 51–2,

53–5, 57, 59–61, 63–5spatial econometric approach to

inequalities 256–76spatial heterogeneity in 268territorial adoption, econometric

models of 264–9see also high-tech

incumbents as source of competence85–6

India, ICT 257Indonesia

labour productivity analysis 121–44see also productivity growth

industryde-concentration of 7dynamics 4, 7–10geographic structure, evolution of 90life-cycle model 7

organisational ecology 9self-reinforcing process 9see also firmsinfrastructureSophia-Antipolis science park 50,

51, 54, 55, 63, 64UK 17, 305innovationand agglomeration 70and clusters 162, 163, 172and collective action 96and diversity 15, 207, 210and evolutionary economics 5,

182–3and firm’s absorptive capacity 11and industrial dynamics 94Innovation Centre, Cambridge 43network analysis 10–12, 183, 240–45and new sectors 10patents see patentspost-entry 9process 7–8and productivity growth 123, 124–5as search process 148, 149Sophia-Antipolis science park 49,

54, 56, 57, 60–65spatial distribution 262and technology levels 123in UK 41see also spillovers; patents; R&D

INRIA 56institutional economic geography

applications of 1, 93, 113–14case-study research 1–2see also evolutionary economic

geographyinstitutions

co-evolutionary process 9–10and collective action 95–6, 100and environmental shocks 94frameworks 5–6new 10reform of 6–7rigidities 12and spatial evolution 9–10theory 5

intellectual property 36, 37–8, 39, 41,43, 64

interactive learning, Sophia-Antipolisscience park 52, 56, 57, 60, 61, 62–3

Index 317

international trade theory, gravitymodel 14

internationalisation, Sophia-Antipolisscience park 50–51, 55, 56, 57–8

Ireland, ICT 257Italy

cooperative banks in rural areas 97domain name registrations 260–64footwear industry 11ICT regional disparities 258patents 267, 269territorial ICT adoption 264–9Third Italy 203university Internet hyperlinks 234wine clusters analysis 162, 166–76see also Europe

Iwai, K. 208

Jacob, Jojo 121–44Jacobs, J. 13, 204, 206, 212, 281Jaffe, A.B. 10, 16, 93, 161, 163, 232,

235, 250, 258, 270–71Japan, semiconductor technology 74Jessop, B. 203Jovanovic, B. 208

Katz, M. and C. Shapiro 180Kauffman, S. 149–50, 158Keller, W. 271Kitson, M. 203Klepper, Steven 2, 6, 7, 8, 9, 30, 69–92,

128–9, 139knowledge

causal ambiguity 148codified 148–9, 158, 232–3, 236tacit 5, 148, 158, 164, 165, 232, 233,

236knowledge bases, heterogeneous 165knowledge diffusion

in automobile industry 8, 9, 16and central hubs 14and clusters see clusterscomplex knowledge transfer 148–53complexity and access to template

150–51, 156–7and economic growth 13EPO co-patenting applications

231–3, 235–6, 238–40, 243–7,249–50

Erasmus student flows 231–3,236–40, 243, 246–7, 250

and evolutionary economics 183face-to-face interaction 106, 232,

236, 242, 248, 249and geographical proximity 10, 14,

18, 152–3, 155–7, 161, 162–4global knowledge 14, 18gravity equations 246–51and informational complexity

147–60inter-regional in Europe 230–55Internet hyperlinks 231–4, 238–40,

242–3, 246–8knowledge receipt as search 148–50location choice 29network analysis, Europe 240–45and network structure 10, 183and new entrants 83and relational proximity 9, 161,

162–4research networks 231, 232, 234–5,

238social boundaries 10, 151–3, 155–8,

163–4and technological communities 152,

153, 155, 157, 166tire industry 88types of 232–7US utility patents analysis 153–8see also spillovers

Kogut, B. and U. Zander 6, 18, 93–4,147, 148

Kort, R. 203Krackhardt, D. 181, 193Krugman, P. 1, 2, 28, 29, 69, 161, 204,

211, 232, 235

labourmigration 13mobility 10, 39skilled 9, 29

labour productivitystochastic frontier analysis 121–44

Lambooy, J. 2, 5, 15, 162, 212large company effects 45laser industry 90Lazerson, M. and G. Lorenzoni 161,

164learning-by-doing 5, 128–9

318 Index

learning, interactive 52, 56, 57, 60, 61,62–3

Lee, Brandon 93–120Leitner, H. and E. Sheppard 203Levins, R. and R. Lewontin 205Levinthal, D.A. 16, 165, 258licensing 42Lisbon 2000 European Council 231,

236Lissoni, F. 2, 10, 18, 28, 29, 88, 162,

182, 235, 236, 241localization economies 212location choice

agglomeration economies 8, 69–70,79–80, 86–8, 90

co-location 28–9heterogeneity 8knowledge diffusion 29localities growth model 12and spillovers 27, 28value chain considerations 28

lock-in, spatial 8, 12, 210, 282Los, Bart 121–44Lounsbury, D. and M. Ventresca 95Lovering, J. 203Lovins, Amory 101–3Lybertysurf 62

Maastricht Treaty (Article 130G) 235McAdam, D. 94, 97McCarthy, J. 97, 104Maggioni, Mario A. 2, 230–55Malizia, E. and S. Ke 214Markusen, A. 6, 176Marshall, A. 69, 93, 158, 161, 162–3,

164, 204Martin, R. 1, 2, 18, 93, 113, 114, 162,

203, 211, 231Maruyama, M. 29Maskell, P. 5, 10, 161, 162, 163, 164Matutinovic, I. 209, 210Metcalfe, J.S. 15, 182, 205, 207, 208mobile telecommunications industry

operator choice criteria 187–9, 193–6UK 183–98

Molina-Morales, F. and M. Martinez-Fernandez 161, 164, 165

Moore, G. and K. Davis 90Moulton, B. 170MS-DOS 36

multinationals 14, 18, 45Sophia-Antipolis science park 57–8see also oligopolies

Neary, J. 211Nelson, R.R. 2, 9, 15, 113, 122, 205,

207, 208and S.G. Winter 2, 5, 15, 19, 148,

150, 165, 207, 281neoclassicism

and economic geography 1, 204and price differentials 5

NetherlandsAmsterdam land use policy 285–90,

294–5, 296–7, 299, 300, 301Amsterdam urban transportation

planning 283, 284–308infrastructure 17Internet adoption 258see also Europe

networksagent interaction 183, 210, 241,

249aggregation 13–14analysis 10–12, 240–45Cambridge 28, 39, 43–4cities 14cluster firms’ heterogeneous

performance 161–79and clusters 10, 11, 240–45consumer 11dynamics model 11–12economics of 180–200externalities 210, 230and geographical proximity 10global knowledge 14, 18hub-and-spoke 12indirect effects 181infrastructure 12and innovation 10–12, 183, 240–45inter-city 13–14inter-regional 13–14and knowledge diffusion 10, 183see also knowledge diffusionlocal 8market, and consumers 180mobile telephony see mobile

telecommunications industryand multinationals 14preferential attachment 12

Index 319

quadratic assignment procedure 181,193

research (Europe) 231, 232, 234–5,238–40, 242, 243, 246–7, 249

small-world 183social network theory see social

network analysisSophia-Antipolis science park 51, 62transportation 14see also agents; connectivity

new economic geographydevelopment of 1diversity and economic growth 204geographical distance see

geographical distancegeographical proximity see

geographical proximitymethodology 1, 2see also evolutionary economic

geography; geographicaldistance; geographicalproximity

new entrantsand agglomeration economies 8bounded rationality 8Cambridge 32, 35–7, 39, 41, 42, 43and diversity 212and knowledge diffusion 83location choice 8Sophia-Antipolis science park 49,

54, 56, 59–60spatial distribution of 5, 9survival analysis 5, 7, 8, 9, 32, 35

new growth theory 13new industries

automobile industry see automobileindustry

and collective action 94diversification 77evolution of geographic structure in

69–92outsourcing 74radio industry agglomeration 73, 74,

77, 86seeding industries 76–7television receiver industry

agglomeration 70, 71–4, 77,85–6, 89

tire industry see tire industrynew sectors

co-evolution of 9–10and regional policy 16

new technologies, territorial adoptionof see ICT

Nottingham University BusinessSchool social network survey181–2, 185–98

Nuvolari, A. 149, 162

Odisei 62Oerlemans, L. and M. Meeus 164off-shoring 6Olds Motor Works 78, 79, 87–8oligopolies 70, 80

see also multinationalsOlivetti 36organic food industry 96organizations

and collective action 95–6, 105and consultancy use 96development of multi-locational 18environmental shocks 94membership 152routines 281–2theory 93, 94–5

outsourcing 6, 29, 39, 74Overman, H. 1

Paci, R. and S. Usai 235Pack, H. 122, 123, 141patents

and agglomeration 235and clusters 235EPO co-patenting applications

231–3, 235–6, 238, 242–7,249–50

and networks 28see also innovation: R&D

path dependencyand cumulative process 30and diversity 210evolutionary economic geography 2,

6, 18technology 180urban transportation planning

283–4, 298–300, 302see also agglomeration economies;

spin-offsPearson coefficients 173, 232, 239, 240,

268

320 Index

Penrose, E. 165Perez, C. 10, 12, 15performance, and agglomeration 70Perrow, C. 95Philips 34Pinch, S. 161Piore, M. and C. Sabel 93, 161Piscitello, Lucia 256–76planning

framework programmes, EU 235and regional development 206urban transportation see urban

transportation planningPlouraboue, F. 11Polanyi, M. 147, 148policy

evaluation 3, 4freedom 15, 16implications 3, 4, 14–17

Porter, M. 5, 28, 161, 163, 204portfolio theory 13, 311–13Portugal

entrepreneurial start-ups 83ICT regional disparities 258

Powell, W. 10, 114, 174Poyhonen, P. 246Pred, A. 14price differentials 5–6product life-cycle hypothesis 8, 12product standardisation 7–8product variety 215production costs, and agglomeration

89production location see location choiceproductivity growth

accumulation theories 122assimilation theories 122–3, 124capital deepening (creating

potential) 124and convergence 12, 122, 123, 257data envelopment analysis 125, 126decomposition analysis 139–40and diversity 123estimation method 126–7frontier and inefficiency estimation

131–9source identification 124–6stochastic analysis 121–44see also economic growth

proximity

geographical 5, 10, 152, 155, 157,175

relational 163, 175social 155, 157, 163, 175spatial 270

Pumain, D. 12, 14Pye (Philips) 34Pyke, F. 161

Quéré, Michel 48–66Quigley, J. 204

R&DEurope, facilities from extra-

European firms 55and new entrant survival 9research networks 231, 232, 234–5,

238Sophia-Antipolis science park 50,

51, 52–3, 54, 55, 56, 57–8, 65spillovers (Indonesia) 125–6, 127–8,

130–39see also innovation; patents

Rabellotti, R. and H. Schmitz 161, 164radio industry agglomeration 73, 74,

77, 86Rammel, C. and van den Bergh, J. 206,

207, 208, 209Rao, H. 93, 94, 95, 97, 100Reed, R. and R. DeFillippi 147regional development

co-evolutionary process of 6and competition 204, 210, 257and convergence 12, 122, 123, 257diversification 13, 204–23diversity 13, 16, 204–23, 205and economic growth 204, 206–13and economic survival 205evolutionary 15external shocks in demand 13and firm success 7gravity equations 246–51and ICT see ICTinfrastructure provision 16–17inter-regional knowledge flows in

Europe 230–55inter-regional networks 13–14and new sectors 16and planning decisions 206portfolio theory 13, 311–13

Index 321

renewability 12revolutionary 15spatial autocorrelation econometrics

13spatial lock-in 8and specialization 210–11stability 206–13US 203–29US county business patterns

employment data 213–23variety in 12–13, 16, 204, 212see also individual industries

relational proximity 175and knowledge diffusion 9, 161,

162–4renewable energy technology, US

power industry 10, 95, 98–100,102–7, 108

resources, creation of new 15Respublica 62Ricardo, D. 206Rigby, D. 2, 5, 207, 211Rivkin, Jan W. 147–60Robert-Nicoud, F. 211Rodan, S. and C. Galunic 166Rohlfs, J. 180Rohm & Haas 52Romanelli, E. and C. Schoonhoven 93,

95Romer, P. 93Rossi, Cristina 256–76routines

disruption of 94and economic growth 204and evolutionary economics 18,

207–8heterogeneous 7, 8organizational 281–2variety in 5, 6

Sapir, A. 230Saviotti, P.P. 13, 207Saxenian, A. 6, 35Schneiberg, M. 95, 96, 97, 98, 114Schoening, N. and L. Sweeney 206Schumpeter, J. 2, 18, 94, 128, 148science parks

reverse 53–4Sophia-Antipolis see Sophia-

Antipolis science park

Scott, A. 164, 203Scott, R. 95, 101, 113, 114selection

and diversity 205, 207–8, 210spatial differences 5

shakeouts 88–9Sheppard, E. 203, 211Siemens 45Sierra Club 99, 103, 104–5, 106–12Simmie, J. 27Sine, Wesley 93–120Small world 42SMEs

Small Business Innovation Research(SBIR), Cambridge 44

Sophia-Antipolis science park 56,57, 58–60, 62–3, 64

Smith, E. 204social boundaries

and geographical proximity 152,155, 157

knowledge diffusion 10, 151–3,155–8, 163–4

social movement organizations(SMOs) 104, 109, 114–15

social movement theory 93, 94–5,114

social network analysis 180–200and evolutionary economics 11–12,

182–3and knowledge diffusion 147, 151–3,

155network statistics 189–91regression results 191–6social status and consumers 180

social proximity 155, 157, 163–4, 175Solomon, S. 11Sophia-Antipolis science park

academic incubators 60business angels 60collaboration, local 62competitiveness 55, 56diversification 50, 52, 60, 62employment levels 49, 51, 52–3, 55,

56, 59entrepreneurial initiatives 53, 54, 56,

60, 62governance of 49, 50–51, 54–60, 63GSM technologies 65historical characteristics 49–54, 63

322 Index

ICT 51–2, 53–5, 57, 59–61, 63–5infrastructure 50, 51, 54, 55, 63, 64innovation 49, 54, 56, 57, 60–65intellectual property rights 64interactive learning 52, 56, 57, 60,

61, 62–3internationalisation 50–51, 55, 56,

57–8life sciences 51, 52–3, 54, 55networking 51, 62PhD training 54, 55, 56, 62–3policy implications 60–65R&D 50, 51, 52–3, 54, 55, 56, 57–8,

65relocations 62as reverse science park 53–4SMEs 56, 57, 58–60, 62–3, 64spin-offs, local 59–60, 62–3telecom equipment providers 62, 63,

64and University of Nice 52, 53–4, 60venture capital 60

Sorenson, Olav 6, 9, 89, 94, 147–60Spain

ICT regional disparities 258university Internet hyperlinks 234

spatialclusters 29concentration 7, 8, 9, 223econometric approach to inequalities

in ICT 256–76heterogeneity in ICT adoption 268lock-in 8, 12, 210, 282new entrants distribution 5, 9proximity 270systems 12–14

Spearman correlations 239, 240specialization 12, 15, 29, 203, 206,

210–11, 212, 221Sophia-Antipolis science park 51

spilloversabsorptive capacity and foreign

121–44and agglomeration 158agglomeration economies 88automobile industry 88and economic growth 13free-riders 28high-tech centres 27and location choice 27, 28

R&D (Indonesia) 125–6, 127–8,130–39

stochastic frontier analysis 121–44technology 128–9tire industry 88utility patents analysis for 153–8variety underlying 13in wine production 11see also knowledge diffusion

spin-offsagglomeration economies 30agglomeration inducement 30,

86–8automobile industry 77–8, 79, 85,

86, 87–8, 90Cambridge 28–9and clusters 29–30disk drive industry 90and firm size 7inherited routines 7laser industry 90model 7, 8–9Sophia-Antipolis science park

59–60, 62–3television industry 86, 87, 88tire industry 82–3, 84, 85, 86–8, 90in UK 6, 16see also agglomeration economies;

path dependencyspin-outs 30

Cambridge 35–41, 42Stam, E. 5, 6, 18start-ups see new entrantsStern, R. and S. Barley 95Stinchcombe, A. 94, 95stochastic frontier analysis, labour

productivity see productivitygrowth

Storper, M. 2, 113Strang, D. and E. Bradburn 97structural change 12Stuart, T. and O. Sorenson 6, 9, 94,

98Suchman, M. 98, 100–101sunk costs 12survival

and diversity 208–9in evolutionary biology 207new entrants 5, 7, 8, 9, 32, 35

sustainability

Index 323

Sophia-Antipolis science park 51,53, 54–5, 56, 63, 64

urban transportation planning 308Swaminathan, A. and J. Wade 94Swann, P. and M. Prevezer 2

Taylor, P. 14technical efficiency, and heterogeneity

129technological

communities 152, 153, 155, 157, 166congruence 123, 130–31proximity 130, 153

technological change 34, 94and collective action 95–6

technologiesand absorptive capacity 128, 129capital-intensive 123communities 152, 153FDI 121path dependency 180and plant age 128–9recombination 16rise and fall of 2similar, between enterprises 128spillovers 128–9technology transfer 90variety 204

Teece, D. 5, 123Télémécanique/Schneider 51television industry

agglomeration 70, 71–4, 77, 79,85–6, 89

and agglomeration economies 74,79–80

spin-offs 86, 87, 88Thalès 51The Technology Partnership (TTP) 38Thrift, N. 1, 113Tilly, C. 104Tinbergen, J. 14, 246tire industry

agglomeration 84, 86–8, 90and agglomeration economies 84,

86–8diversification 82, 83geographic structure, evolution of

76–90heterogeneity 84input markets 88

labour costs 84location of branch plants 83–4related industries, influence of 88seeding industries 84shakeouts 88–9spillovers 88spin-offs 82–3, 84, 85, 86–8, 90trade unions 84transportation costs 84

trade associations 84, 94transportation

network morphology 283, 301, 302and urban networks 14see also urban transportation

planning

Uberti, T. Erika 230–55UK

automobile sector 8–9, 16Cambridge high-tech region see

Cambridgeinfrastructure 17, 305innovation in 41investment levels 41mobile telecommunications industry

183–98new sector co-evolution 10Oxford 44related industries 9retail banking industry 10specialist labour markets 6spin-offs 6, 16standardised credit rating 41university Internet hyperlinks 234venture capital 37, 41

Ulanowicz, R. 206uncertainty 15, 16, 17

in urban transportation planning279–80, 283, 284, 306–8

unemployment 204, 206structural 13see also employment levels

UniCam 34Union of Concerned Scientists 99, 103Uniroyal 80universities

academic entrepreneurship 28Cambridge see CambridgeErasmus student flows 231, 232, 233,

236–7

324 Index

Internet hyperlinks 234manufacturing modernisation

initiatives, US 35PhD training, Sophia-Antipolis

science park 54, 55, 56, 62–3University of Nice 52, 53–4, 60see also Sophia-Antipolis science

parkurban economic growth 14urban transportation planning

Amsterdam 283, 284–308complexity of 280, 281, 308evolutionary approach 279–310heterogeneity in 279path dependency 283–4, 298–300,

302sustainability 308system resilience 303–4, 308transportation network morphology

283, 301, 302uncertainty in 279–80, 283, 284,

306–8variety in 280, 303

urbanization economies 212US

automobile industry agglomeration7, 8, 14–15, 30, 70, 74–80

clustering 30, 70competition, international 74county business patterns

employment data 213–23diversity, stability and regional

growth 203–29electrical power industry see

electrical power industry, USenvironmental movement 99, 101,

103, 105–7, 109–11fire insurance industry 96, 98government-university

manufacturing modernisationinitiatives 35

ICT regional disparities 257institutional changes, differential 10investment in Cambridge high-tech

industries 41–2, 43, 44National Energy Act (NEA) 100oil crises 99, 101, 107outsourcing 74power industry 98–107Prohibition period and breweries 97

Public Utility Regulatory PoliciesAct (PURPA) 100

radio industry agglomeration 73, 74railroad industry, Massacheusetts 97renewable energy technology 10,

98–100, 102–7, 108Route 128 45semiconductor technology 74Shockley and Fairchild spin-outs 30Silicon Valley 7, 34–5, 45, 69, 90,

203and Sophia-Antipolis science park

50–51television receiver industry

agglomeration 70, 71–4, 79, 80,85

tire industry agglomeration see tireindustry

utility patents analysis for spillovers153–8

US power industry 98–112diagnostic framing 100–102environmental movement 99, 101,

103, 105–7, 109–11hypothesis data and methods 107–12motivational framing and resource

mobilization 104–7prognostic framing 100, 102–4renewable energy technology 10, 95,

98–100, 102–7, 108Uzzi, B. 10, 183

van den Bergh J. 206, 207, 208, 209,281

Van Dijk, M. 123, 141van Wissen, L. 9variety

in economic behaviour of firms 282and economic development 13, 204labour productivity 138product 215in regional development 12–13, 16,

204, 212in routines 5in spillovers 13of strategy in science parks 49, 57,

62strategy in Sophia-Antipolis science

park 49, 57, 62technological 204

Index 325

in urban transportation planning280, 303

venture capitalCambridge 41–2Sophia-Antipolis science park 60UK 37, 41

Videojet 39Vrba, E. and S. Gould 205Vromen, J. 208

wage differentials 70, 89Wagner, J. and S. Deller 208Wassermann, S. and K. Faust 10, 162,

240Watts, D. 10, 241wealth distribution 1, 18Weber, A. 93

Wellcome 52Weltevreden, J. 258Werker, C. and S. Athreye 2Wernerfelt, B. 172Whitley, R. 1wine production

and clusters 162, 166–76and spillovers 11

Winter, S.G. 2, 5, 15, 19, 148, 150, 165,207, 281

Xaar 39

Zaheer, A. and G. Bell 161, 164, 165,172

Zander, U. 6, 18, 93–4, 147, 148Zucker, L. 28, 93

326 Index