Applied Evolutionary Economics and Economic Geography

Transcription

1 Applied Evolutionary Economics and Economic Geography

2

3 Applied Evolutionary Economics and Economic Geography Edited by Koen Frenken Utrecht University, The Netherlands Edward Elgar Cheltenham, UK Northampton, MA, USA

4 Koen Frenken 2007 All rights reserved. No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical or photocopying, recording, or otherwise without the prior permission of the publisher. Published by Edward Elgar Publishing Limited Glensanda House Montpellier Parade Cheltenham Glos GL50 1UA UK Edward Elgar Publishing, Inc. William Pratt House 9 Dewey Court Northampton Massachusetts USA A catalogue record for this book is 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 Koen Frenken. p. cm. Summary: The volume Applied Evolutionary Economics and Economic Geography is the fourth book published by Edward Elgar on applied evolutionary economics and stems from the fourth European Meeting on Applied Evolutionary Economics (EMAEE) held in Utrecht, May, The present volume Applied Evolutionary Economics and Economic Geography aims to advance empirical methodologies in evolutionary economics, this time with a special 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. HB97.3E dc ISBN Printed and bound in Great Britain by MPG Books Ltd, Bodmin, Cornwall

5 Contents List of figures List of tables List of boxes List of contributors Preface vii ix xi xii xiv 1. Introduction: applications of evolutionary economic geography 1 Ron A. Boschma and Koen Frenken PART I ENTREPRENEURSHIP 2. The Cambridge high-tech cluster: an evolutionary perspective 27 Elizabeth Garnsey and Paul Heffernan 3. Sophia-Antipolis as a reverse science park: from exogenous to endogenous development 48 Michel Quéré PART II INDUSTRIAL DYNAMICS 4. The evolution of geographic structure in new industries 69 Steven Klepper 5. Constructing entrepreneurial opportunity: environmental movements and the transformation of regional regulatory regimes 93 Brandon Lee and Wesley Sine 6. Absorptive capacity and foreign spillovers: a stochastic frontier approach 121 Jojo Jacob and Bart Los PART III NETWORK ANALYSIS 7. Informational complexity and the flow of knowledge across social boundaries 147 Olav Sorenson, Jan W. Rivkin and Lee Fleming v

6 vi Contents 8. Networks and heterogeneous performance of cluster firms 161 Elisa Giuliani 9. Social networks and the economics of networks 180 Daniel Birke PART IV SPATIAL SYSTEMS 10. Diversity, stability and regional growth in the United States, Jürgen Essletzbichler 11. Inter-regional knowledge flows in Europe: an econometric analysis 230 Mario A. Maggioni and T. Erika Uberti 12. Explaining the territorial adoption of new technologies: a spatial econometric approach 256 Andrea Bonaccorsi, Lucia Piscitello and Cristina Rossi PART V PLANNING 13. Evolutionary urban transportation planning? An exploration 279 Luca Bertolini Index 311

7 Figures 1.1 Evolutionary economic geography applied at different levels of aggregation High-tech firms in Cambridgeshire, Changing sectoral growth patterns among Cambridge high-tech firms Firm turnover (turbulence) in the Cambridge high-tech sector Comparison of survival rates for cohorts of Cambridge high-tech firms Companies with founders from Cambridge University engineering departments New firms started by founders and employees of Acorn Computers Industrial inkjet printing spin-outs originating in Cambridge Biotech firms originating in 12 Cambridge University departments Cumulative number of organisations and employment Cumulative number of organisations and employment (ICT) Cumulative number of organisations and employment (life sciences) Size distribution of ICT firms The geography of ICT activities Entry, exit and number of producers in the television industry, Percentage of television producers in New York, Chicago and Los Angeles, Entry, exit and number of producers in the automobile industry, Percentage of automobile producers in the Detroit area, Entry, exit and number of producers in the tire industry, Percentage of tire producers in Ohio, vii

8 viii Figures 6.1 Labour productivity growth decomposition Landscape without interdependence Landscape with maximal interdependence 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 Number of subscribers in the UK Development of subscriber market shares Distribution of distances between nodes Interaction network of students Volatility, diversity, growth Relationship between volatility and growth/diversity Co-patent network, , including Oberbayern, Darmstadt, Düsseldorf and Île de France Co-patent network, , excluding Oberbayern, Darmstadt, Düsseldorf and Île de France Distribution of ICT adoption and added value per inhabitant across Italian provinces: the Three Italies ICT adoption and added value per employee, univariate LISA Changes in the built-up area and infrastructure in the Amsterdam region, a Coping with uncertainty in planning b Coping with irreducible uncertainty in planning (or chaos ) 307

9 Tables 1.1 Two types of regional innovation policy Summary statistics and correlations for state-avoided cost analysis GEE model predicting state-avoided costs a Summary statistics of 17 5-digit ISIC industries: levels lb Summary statistics of 17 5-digit ISIC industries: average annual growth rates Stochastic frontier estimates for 17 5-digit industries Decomposition of productivity growth, A.1 Industrial classification Rare events logit models of the likelihood of a focal patent receiving a citation from a future patent Firm characteristics by cluster Collection of key variables Descriptive statistics and correlation matrix Probit estimations with marginal effects Gender and nationality of respondents Frequencies for choice criteria Do you know which operator your friends/family/ partner use? Local network density Mobile phone operators and nationality Determinants of choosing the same operator Predicted probabilities of using the same operator Friendship determinants Predicted probabilities of calling each other Correlates of volatility BEA area rankings by stability, diversity and growth Basic statistics of dependent and independent variables Correlation coefficients between dependent and independent variables Determinants of volatility Pearson and Spearman correlations between knowledge flow variables QAP correlation between knowledge flow variables 240 ix

10 x Tables 11.3 Network analysis indices of knowledge flow structures Gravity equation for knowledge flows ICT adoption: descriptive statistics ICT adoption in macro areas Ranking of Italian provinces by ICT adoption and per capita income in Spatial dependence tests for the dependent variable Specification of dependent and independent variables Correlation matrix Standard OLS models Spatial lag and spatial error models Overview of different domains of change in the Amsterdam urban region,

11 Boxes 13.1 The late 1960s and early 1970s: a transport and land use policy transition dissected The late 1980s and early 1990s: a land use policy transition dissected 294 xi

12 Contributors Luca Bertolini Amsterdam Institute for Metropolitan and International Development Studies (AMIDSt), University of Amsterdam, The Netherlands. Daniel Birke Nottingham University Business School, University of Nottingham, 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 College London, 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 for Manufacturing, University of Cambridge, UK. Elisa Giuliani Science Policy Research Unit (SPRU), University of Sussex, Brighton, UK. Paul Heffernan Centre for Technology Management, Institute for Manufacturing, University of Cambridge, UK. Jojo Jacob Eindhoven Centre for Innovation Studies (ECIS), Eindhoven University 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, Cornell University, Ithaca, NY, USA. Bart Los University of Groningen, The Netherlands. xii

13 Contributors xiii 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 di Milano, Italy. Michel Quéré Centre National de la Recherche Scientifique Groupe de Recherche 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, Cornell University, Ithaca, NY, USA. Olav Sorenson Rotman School of Management, Toronto, Canada. T. Erika Uberti Department of International Economics, Development and Institutions (DISEIS) and Faculty of Political Science, Catholic University, Milan, Italy.

14 Preface The volume Applied Evolutionary Economics and Economic Geography is the fourth book published by Edward Elgar on applied evolutionary economics, and stems from the fourth European Meeting on Applied Evolutionary Economics (EMAEE) held in Utrecht, May The first volume edited by Paolo Saviotti (Applied Evolutionary Economics, 2003) was the founding volume. The second volume edited by John Foster and Werner Hölzl (Applied Evolutionary Economics and Complex Systems, 2004) dealt with empirical applications of complex systems theory. The third volume edited by Andreas Pyka and Horst Hanusch (Applied Evolutionary Economics and the Knowledge Based Economy, 2006) discussed the role of knowledge in the modern economy from an evolutionary perspective. As with previous volumes, this one also aims to advance empirical methodologies in evolutionary economics, but with a special emphasis on geography. We have seen a growing, mutual interest between economic geography and evolutionary economics in recent times. In particular, after the 2001 Schumpeter Prize winning work by Steven Klepper on the spatial evolution of industries and the innovative applications of social network analysis by Stefano Breschi, Francisco Lissoni, Elisa Giuliani and others, evolutionary scholars have become interested in the geography of industries and networks. These two meso levels of analysis will be central in the book. New developments on the interface between geography and evolutionary economics also take place at the micro level, in particular, the geography of entrepreneurship, and at the macro level, in particular, the dynamics of spatial systems as reflected in the uneven growth of cities, regions and nations. The micro and macro levels will supplement the meso level in the book. The book thus 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 introductory chapter and ends with a policy chapter on planning. Many people have contributed to this book. First, I would like to thank the authors who have all written excellent chapters. Their efforts show the value added of an evolutionary approach in economic geography. Second, I would like to thank the 125 participants of the fourth EMAEE held in Utrecht. The conference theme being geography, innovation and networks, the event was unique in bringing together economists and geographers to xiv

15 Preface xv exchange ideas and advance our common understanding of economic geography. Third, I would like to thank the members of the scientific committee for their review work and the members of the local organising committee for preparing the conference. Special thanks go to Anet for her excellent planning of the conference, to Jesse for the IT support, and to Siebren for his personal assistance. Finally, I gratefully acknowledge the financial support of the Urban and Regional research centre Utrecht (URU), the Netherlands Graduate School of Housing and Urban Research (NETHUR), the Royal Netherlands Academy of Arts and Sciences (KNAW), the Netherlands Organisation for Scientific Research (NWO), and the Utrecht municipality. Koen Frenken Utrecht, March 2006

16

17 1. Introduction: applications of evolutionary economic geography Ron A. Boschma and Koen Frenken 1. INTRODUCTION Economic geography is the field of study that deals with the uneven distribution of economic activities in space. Two conflicting theories are currently influential in the field: institutional economic geography and the new economic geography. Institutional economic geography is dominated by scholars with a geography background and is akin to institutional economics (Hodgson, 1998). At the risk of oversimplification, institutional economic geography argues that the uneven distribution of wealth across territories is primarily related to differences in institutions (Whitley, 1992; Gertler, 1995; Martin, 2000). The new economic geography has been developed by neoclassical economists (Krugman, 1991; Fujita et al., 1999; Brakman et al., 2001), who view uneven distributions of economic activity as the outcome of universal processes of agglomeration driven by mobile production factors. Recent debates between geographers and economists have been fierce and with little progress (for example, Martin, 1999; Amin and Thrift, 2000; Overman, 2004). The lack of cross-fertilisation between the two disciplines can be understood from two incommensurabilities between institutional and neoclassical economics (Boschma and Frenken, 2006). First, institutional economic geography and new economic geography differ in methodology. Institutional economic geographers tend to dismiss a priori the use of formal modelling. Instead, they apply inductive, often, casestudy research, emphasising the local specificity of real places. By contrast, the new economic geography approaches the matter deductively using formal models based on neutral space, representative agents and equilibrium analysis. Proponents of the latter approach do not value, or even reject altogether, case-study research. Second, the two theories differ in core assumptions regarding economic behaviour. The new economic geography aims to explain geographical patterns in economic activity from utility-maximising actions of individual agents. By contrast, institutional scholars start from the 1

18 2 Introduction 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 geographically localised practices, which implies that localities ( real places ) are the 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 to economic explanations, and their study should therefore be best left to the sociologists, as Krugman once put it (Martin, 1999: 75). Evolutionary economic geography can be considered a third approach in economic geography. Evolutionary economists argue that the explanation to why something exists intimately rests on how it became what it is (Dosi, 1997: 1531). Rather than focusing on universal mobility processes underlying agglomeration (neoclassical) or the uniqueness of institutions in specific territories (institutional), an evolutionary economic geography views the economy as an evolutionary process that unfolds in space and time. In doing so, it focuses on the path-dependent dynamics underlying uneven economic development in space (Martin and Sunley, 2006). In particular, it analyses the geography of firm dynamics (such as the geography of entrepreneurship, innovation and extinction) and the rise and fall of technologies, industries, networks and institutions in different localities. In this view, uneven economic development requires an understanding of the Schumpeterian process of creative destruction at different levels of spatial aggregation (cities, regions, nations, continents). Even though evolutionary economics goes back at least to the seminal contribution by Nelson and Winter (1982), evolutionary approaches to economic geography are fairly recent (Arthur, 1994; Swann and Prevezer, 1996; Boschma, 1997; Rigby and Essletzbichler, 1997; Storper, 1997; Boschma and Lambooy, 1999; Antonelli, 2000; Caniëls, 2000; Klepper, 2001; Maggioni, 2002; Breschi and Lissoni, 2003; Bottazzi et al., 2004; Brenner, 2004; Werker and Athreye, 2004; Boschma and Wenting, 2005; Essletzbichler and Rigby, 2005; Martin and Sunley, 2006). The difference between evolutionary economic geography and both new and institutional economic geography can be summarised as follows (Boschma and Frenken, 2006). An evolutionary approach to economic geography is different from new economic geography in that it attempts to go beyond the heroic assumptions about economic agents and the reduction of geography to transportation costs. At the same time, evolutionary economic geography also differs from institutional economic geography in that an evolutionary approach explains territorial differences not primarily by referring to different institutions, but from differences in the history of firms and industries residing in a territory. An evolutionary analysis may well take into account the role of institutions though, but in a co-evolutionary perspective (Nelson, 1995).

19 Introduction 3 Methodologically, evolutionary economic geography differs from both institutional and new economic geography in that it combines all research methodologies: case-study research, surveys, econometrics, theoretical modelling exercises and policy evaluation can, in principle, all be based on evolutionary theorising. The present volume, Applied Evolutionary Economics and Economic Geography, aims to further develop an evolutionary economic geography. It does so by bringing together a selected group of excellent scholars coming from business studies, economics, geography, planning and organisational sociology. All contributors share an interest in explaining the uneven distribution of economic activities in space and the historical processes that have produced these patterns. The heterogeneity in backgrounds was overcome by a common understanding of the evolutionary nature of spatial processes. The end result is a volume of 13 chapters on various topics organised under the headings of entrepreneurship, industrial dynamics, network analysis, spatial systems and planning. The volume also reflects the variety of research methodologies characterising applied evolutionary economics, including case-study research (Garnsey and Heffernan, Chapter 2; Quéré, Chapter 3; Lee and Sine, Chapter 5; Bertolini, Chapter 13), duration models (Klepper, Chapter 4), data envelopment analysis (Jacob and Los, Chapter 6), complexity theory (Sorenson et 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., Chapter 12) 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 economic geography primarily fall under four categories: firm, industry, network and spatial systems. Their scheme also underlies the structure of the book with the various chapters being organised under one of these four headings. Following Figure 1.1, the categories follow from aggregating firms to their relevant meso levels of the industry in which they compete and the networks in which they exchange commodities and share knowledge. Aggregating in turn the meso levels to the macro level, one obtains the macro level of spatial systems. Following this scheme, localities in spatial systems, be it cities, regions or countries, can be characterised by their sector composition and their position in spatial networks, and structural changes herein over time (Castells, 1996).

20 4 Introduction Policy (Chapter 13) Macro Spatial systems (Chapters 10 12) Meso Industrial dynamics (Chapters 4 6) Network analysis (Chapters 7 9) Micro Entrepreneurship (Chapters 2 3) Introduction (Chapter 1) Source: Adapted from: Boschma and Frenken (2006, p. 293). Figure 1.1 Evolutionary economic geography applied at different levels of aggregation

21 Introduction 5 Entrepreneurship We consider evolutionary economic geography to involve a synthesis of evolutionary economics and economic geography. Following evolutionary economics, our starting point is the firm, which competes on the basis of its routines and core competences that are built up over time (Nelson and Winter, 1982). Organisational routines and core competences consist for a large part of learning-by-doing and tacit knowledge, which are hard to codify and difficult to imitate by other firms (Teece et al., 1997; Maskell, 2001). Consequently, organisations are heterogeneous in their routines, and persistently so (Klepper, Chapter 4; Giuliani, Chapter 8). Models can thus no longer rely on assuming a representative agent, but have to account for heterogeneous firms. This variety provides the fuel for selection processes, which causes some firms to prosper and grow and others to decline and possibly exit. From this evolutionary process of firm dynamics based on competition, innovation and selection, an emergent spatial pattern of economic activity arises. This evolving economic landscape, as reflected by spatial heterogeneity in firms routines, can be understood as the joint outcome of geographical proximity (enhancing innovation and imitation) on the one hand, and spatial differences in selection conditions on the other (Boschma and 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 space and time is driven by entry of new firms, exit of incumbent firms and relocation of incumbent firms. Through this process, new routines are being diffused in space. From an evolutionary perspective, one does not analyse new firm location solely as the outcome of rational decisions directed by price differentials, as in neoclassical theory, or in terms of comparing institutional frameworks in different areas, as in institutional theory. Rather, one is interested in the history of the founder and key employees of a new venture to account for routines transferred from a previous activity, and how that affects their survival. And, to understand uneven rates of regional entrepreneurship and entrepreneurial success, one is interested in the spatial distribution of resources required to start up a new business. As entrepreneurs require resources (capital, labour, networks, knowledge) to start new ventures, and resources tend to concentrate in space, as in urban areas (Hoover and Vernon, 1959) or specialised clusters (Porter, 2000), the probability of starting a new venture can also be made dependent on territorial conditions. This is not to say that price differentials (the neoclassical view) and place-specific institutions (the institutional view) do not matter. Rather, prices and institutions only condition the range of possible

22 6 Introduction economic behaviours and their locations, while the actual behaviours are determined by the path-dependent history of actors involved in particular territorial settings (Boschma and Frenken, 2006). The core concept of path dependence can also be fruitfully applied to firm location. Location decisions by firms are heavily constrained by the past. For example, many firms just start at locations where the founder lives, due to bounded rationality, or because the founder is socially embedded in local networks, and it is well known that most spin-offs locate near the parent firm (Cooper and Dunkelberg, 1987; Klepper, 2001). In either case, previous decisions taken in the past determine the location decision of a new firm. Path dependence also affects the probability of relocation as firms are expected to display a considerable degree of locational inertia. The probability of relocation decreases over time as a firm develops a stable set of relations with suppliers and customers and sunk costs accumulate in situ (Stam, 2003). Of course, even though path dependence constrains relocation of the firm, one can expect the firm to outsource parts of the production to low-wage locations, in particular, activities that rely less on the organisational 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 entrepreneurship (Hall and Markusen, 1985; De Jong, 1987; Aydalot and Keeble, 1988; Saxenian, 1994; Stuart and Sorenson, 2003). New high-tech firms are commonly thought to fuel employment growth and regional economic development. In the present volume, we focus on two exceptional European regions that have been successful in fostering high-tech entrepreneurship in information and communication technology (ICT). The two cases concern Cambridge, UK and Sophia-Antipolis near Nice. The development of Cambridge as a high-tech region can be understood as resulting from an endogenous evolutionary process of entrepreneurs setting up business and hereby improving the conditions for new ventures to occur (Garnsey and Heffernan, Chapter 2). The endogenous process encompassed the founding of companies by members of the university, spin-offs, the rise of local suppliers and the emergence of specialist labour markets. This process, however, has not been entirely automatic. Once congestion became problematic and university regulations were perceived unfavourable for entrepreneurship, collective action resulted in institutional reform. Thus, the history of the Cambridge region illustrates both the endogenous nature of entrepreneurship and the co-evolutionary process of entrepreneurship, regional development and institutional change. Another example of successful regional development is the science park of Sophia- Antipolis. However, its development was far from endogenous. Rather the

23 process was triggered by the presence of a few large companies, a favourable living environment and a visionary man (Quéré, Chapter 3). Interestingly, the process transformed from being triggered by external factors into a more endogenous process from the early 1990s onwards. The endogenous nature of the more recent history is evidenced by the fact that even though some larger firms left the park in the early 1990s to go to larger agglomerations such as Paris and London, employees decided not to leave the region, but to start their own ventures instead. In this particular case, it is the employee rather than the firm that shows locational inertia. Thus, the two cases of Cambridge and Sophia- Antipolis are different yet equally successful in the creation of new high-tech firms (see also, Garnsey and Longhi, 2004). Industrial Dynamics Introduction 7 Starting from the firm, the first meso level of aggregation that is specifically important in evolutionary economic geography is the industry level. In this context, the main phenomenon to be explained is the process of spatial concentration or de-concentration of an industry over time. Arthur (1994) developed two simple evolutionary models of spatial concentration by spin-off and by agglomeration economies (see also, Boschma and Frenken, 2003). In the spin-off model an industry comes into being as a Polya process of firms giving birth to firms giving birth to firms and so on. This process is known to have played an important role in the rapid growth and spatial concentration of several industries, including the concentration of the US automobile industry in the Detroit area (Klepper, 2001), the ICT sector in Silicon 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 lifecycle model, which synthesises five assumptions: routines are heterogeneous; spin-offs inherit the routines of parent firms; more successful firms grow faster; larger firms produce more spin-offs; and worse-performing firms are forced to exit due to competition. The first four mechanisms ensure that the region that hosts early, experienced and successful entrants will come to dominate the industry. In contrast to Arthur s spin-off model, this truly concerns a process of inheritance in which the experience of parent firms is inherited by spin-offs with a positive impact on their survival rates. The fifth mechanism of cost competition at the sector level asymmetrically affects regions, causing the region hosting the less successful firms to decline, leaving the region hosting the successful companies to dominate the industry. Typically, cost competition becomes fierce only after an industry has developed for a number of years, that is, after product standardisation has taken place and innovation shifts to process innovation

24 8 Introduction 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 routines are heterogeneous and unevenly spread. The predictions of the model can be tested econometrically in a relatively straightforward way using duration models (Klepper, 2001, 2002). Arthur s (1994) second model of agglomeration economies assumes that new firms start up rather than spin off from incumbent firms. The location choice of a new firm can therefore not be automatically determined by the location of the parent company: the location of the firm becomes a choice decision. Arthur assumes that each firm has a locational preference for one particular region. While Arthur is far from explicit on this matter, this heterogeneity in preferences can stem from bounded rationality yet may also be given an empirical meaning: start-ups typically locate their business in the region where the founder lives and/or held previous employment. Agglomeration economies arising from spatial concentration of firms operating in the same industry, cause the industry to concentrate in one single region even though the individual firms have different individual preferences. The reason is that once one region has attracted slightly more entrants than other regions, a critical threshold is passed, and suddenly all firms 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 easily distinguishable from the outcomes of the agglomeration economies model. We have, indeed, two different explanations for the same phenomenon of spatial concentration of an industry. As spin-off dynamics and agglomeration economies may well contribute to spatial concentration simultaneously, the challenge for empirical research is to disentangle both processes so as to assess their presence and importance. One out of the few studies that have attempted to do so is Klepper s (2001) study of the US automobile industry. In his econometric analysis, he included a dummy for being located in the Detroit area. The dummy showed no positive effect on the survival of firms, which suggests that agglomeration economies were not present. The use of a Detroit control variable, however, can be questioned, since a subset of firms within the Detroit area may have benefited from each other s presence through local networks (Giuliani, Chapter 8) or firms may have benefited from knowledge spillovers over a longer distance (Jacob and Los, Chapter 6). Despite this shortcoming, the result by Klepper (2001) strongly suggests that the concentration of the US automobile industry in Detroit can be attributed mainly to the self-reinforcing dynamics of successful firms creating successful spin-offs, and so on. A study by Boschma and Wenting (2005) on the spatial evolution of the British automobile sector came to similar conclusions regarding the

25 Introduction 9 self-reinforcing nature of spin-off dynamics, which, in the British case, led to a concentration in the Birmingham Coventry area. However, Boschma and Wenting also accounted for the presence of related industries (such as coach and cycle making) in a region as a potential source of agglomeration economies, which was shown to have a positive effect on the survival rate of firms. Thus, the local presence of related industries appeared to be beneficial due to, for example, knowledge spillovers and skilled labour, yet the local presence of a high number of firms operating in the same industry turned out to be harmful due to increased competition, lowering the survival chances of new entrants. Another recent elaboration on Klepper s model is by Cantner et al. (2005), whose methodology using instrumental variable estimation allows for post-entry innovation. In doing so, the survival probabilities are not only dependent on initial conditions of entrants, but also on the research and development activities they undertake during their lifetime. These contributions suggest that survival analysis is a promising research methodology in evolutionary economic geography. Importantly, in an evolutionary context, spatial concentration (or its absence) is not only an outcome of a process of industrial evolution, but also affects an industry s further evolution. This recursive relationship is central in another empirical tradition in industrial dynamics known as organisational ecology or firm demography (Hannan et al., 1995; Carroll and Hannan, 2000; Stuart and Sorenson, 2003; van Wissen, 2004). First, geographical concentration of industrial activities can generate positive feedbacks on entry rather than performance. This means that an industry can become concentrated through a self-reinforcing process of entry triggering more entry. Second, geographical concentration of firms increases the level of competition and makes entry less likely. This negative feedback set limits to spatial concentration. Typically, positive feedbacks operate at the start of an industry life cycle, while negative feedback takes over after a certain threshold of spatial concentration is passed. Interestingly, the two processes causing positive and negative feedbacks may well operate at different spatial scales depending on the type of industry (Jacob and Los, Chapter 6). In industries where demand is local and knowledge spillovers more global, one expects negative feedbacks to operate at a lower spatial level than positive feedbacks, resulting in a more even spatial distribution (Hannan et al., 1995). However, in markets where competition is global, but knowledge spillovers rather local, the reverse may well be the case. Institutions also affect the spatial evolution of industries. From an evolutionary perspective, the question is not so much whether particular institutions triggered the development of a particular industry in a certain region, but rather how institutions have co-evolved with the emergence of a new sector (Nelson, 1995). The co-evolutionary perspective is important

26 10 Introduction because it acknowledges that innovations leading to new sectors often require the restructuring of old institutions and the establishment of new institutions (Freeman and Perez, 1988). Examples of the co-evolution of new sectors and institutions are the rise of the synthetic dye industry in the second half of the nineteenth century in Germany (Murmann, 2003) and the evolution of the UK retail banking industry from the 1840s to the 1990s (Consoli, 2005). In their study of the spatial diffusion of the renewable energy technology, Lee and Sine (Chapter 5) also emphasise the differential institutional changes occurring in different American states. Network Analysis Networks provide another unit of analysis. Unlike the competitive nature of industrial dynamics, network relationships are less competitive and of a more complementary nature. One important aspect of networks in evolutionary economic geography is that these act as vehicles for knowledge spillovers. A key research question is then to determine whether knowledge diffusion and innovation is more a matter of being in the right place, in the right network, or in both (Boschma and Ter Wal, 2006). Social network analysis provides a rich toolbox for the analysis of the structure and evolution of networks (Wasserman and Faust, 1994; Carrington et al., 2005). What is more, there is a lot of interest in theorising about networks and network formation starting from the pioneering work by Granovetter (1973) and Burt (1982) to more recent, but already classic contributions of Watts and Strogatz (1998) and Barabasi and Albert (1999). In evolutionary economics, interest in networks stems primarily from the increasing importance of networks among high-technology firms (Hagedoorn, 1993; Powell et al., 1996), while geographical studies have shown the role of networking in clusters (Uzzi, 1996; Maskell and Malmberg, 1999). The central question has been whether agents profit from simply being co-located or whether network relationships are required to carry these knowledge flows. A related question is whether geographical proximity facilitates the formation of network links. An innovative study by Breschi and Lissoni (2003) found that, using co-inventor data to indicate social networks and patent citations to indicate knowledge flows, geographical localisation of knowledge spillovers can be largely attributed to social networks and labour mobility. This study shows considerable progress over the study by Jaffe et al. (1993), who treated geographical space as a black box. The Breschi Lissoni study suggests that geographical proximity is neither a necessary nor a sufficient condition for knowledge spillovers to occur. Rather, knowledge diffuses through social networks, which are dense between proximate actors, but also span across the globe.

27 Introduction 11 Network analysis between firms in specialised clusters is another field in which social network analysis can be fruitfully applied. Using survey data, Giuliani (2005) has been able to map the business and knowledge networks among wine producers in three different clusters. She found that the the distribution of connectivity is much more skewed in knowledge than in business networks, which suggests that only a few central firms profit from knowledge spillovers. This hypothesis has been put to the test in a followup study presented in this volume (Giuliani, Chapter 8), in which it is shown that a firm s centrality in knowledge networks is indeed positively affecting innovative performance, even after controlling for heterogeneity in internal competencies. A recent study by Boschma and Ter Wal (2006) on a footwear district in southern Italy tends to suggest that the absorptive capacity of firms is indirectly related to their innovative performance, through having non-local instead of local relationships. That is, the higher the absorptive capacity of a district firm, the better it is connected to organisations outside the district, which, in turn, impacts positively on their innovative performance. These studies show that social network analysis is a powerful tool in analysing the geography and structure of knowledge networks and the effect of a firm s network position in these networks on its performance. In a similar fashion, the concept of regional innovation systems (Cooke et al., 1998) can be operationalised empirically more systematically by mapping the various network relations of actors that are part of the regional system with other actors within and outside the regional system. Evolutionary theorising has also argued that, due to bounded rationality, consumers also rely on personal networks. As a result, certain decisions by central actors can propagate through the network, leading many consumers 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 the geographical nature of such personal networks, can also be explored empirically using social network analysis. A nice example of such an approach is the study by Birke (Chapter 9), who conducted a survey among students asking them about their personal networks and their choice of mobile telephone operator so as to analyse the effect of personal networks on the choice of operator. Hitherto, the use of social network analysis in evolutionary economics has been almost exclusively static. A future challenge is to understand the spatial evolution of networks. This requires longitudinal data and methods to analyse the dynamics of networks over time. An influential theoretical model of network dynamics is the model by Barabasi and Albert (1999). In this model, a network grows as new nodes connect to a network. Nodes are assumed to attach themselves to other nodes with a probability proportional

28 12 Introduction to the latter s connectivity. This principle is known as preferential attachment, which means that a new node prefers to link with a well-connected node so as to profit from its connectivity. Well-connected nodes will then tend to become even more connected, while peripheral nodes in the network will tend to remain peripheral. The resulting distribution of connectivity will be extremely skewed (scale free). Which of the nodes becomes the central node is path dependent, and thus unpredictable, although early entrants will have a much higher probability of becoming central than later entrants. The stochastic logic underlying the Barabasi Albert model of network formation has also been applied to the spatial evolution of networks where new nodes can occur anywhere in space, and connections between nodes are made dependent on both geographical space (negatively) and preferential attachment (positively). The resulting topology and spatial organisation of a network 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 still been rather limited. Spatial Systems Aggregating sectors and networks to the macro level of spatial systems, one obtains a model of the growth of localities (cities, regions, countries), as depending on their sectoral composition and global network position, and the structural changes herein occurring over time. The sectoral logic underlying the evolution of spatial systems is better known as the process of structural change (Freeman and Perez, 1988; Boschma, 1997, 2004). Cities and regions that are capable of generating new sectors with new product life cycles will experience growth, while cities and regions that are locked into earlier specialisations with mature life cycles will experience decline. Importantly, there is no automatic economic or political mechanism to ensure that cities or regions will successfully renew themselves. Rather, one expects localities in most instances to experience decline after periods of growth due to vested interests, institutional rigidities and sunk costs associated with previous specialisations (Grabher, 1993). There are, however, still very few systematic evolutionary studies on convergence and divergence at different spatial scales (for example, Pumain and Moriconi- Ebrard, 1997; Caniëls, 2000). This can be partly understood from the demanding data requirements for systematic analysis of long-term dynamics, especially if one is interested in analyses at subnational levels. A particularly popularly topic in economic geography concerns the role of variety in regional growth. Economic theory has long been focused on

29 Introduction 13 explaining economic growth by a combination of growth in inputs and efficiency improvements (Solow, 1957). The underlying qualitative nature of economic development, in terms of the variety of sectors or the variety of technologies, has been addressed only rarely. One can distinguish three types of relationships between variety and economic development (Frenken et al., 2005, 2006). The first approach centres on variety, knowledge spillovers and growth, which has become a central theme in what is called new growth theory. It has been argued that, apart from spillovers occurring between firms within a sector, spillovers also occur between sectors, which are commonly referred to as Jacobs externalities, after Jacobs (1969). A second way to relate variety to regional economic development is to view variety as a portfolio strategy to protect a region from external shocks in demand (Essletzbichler, Chapter 10). In this context, one also speaks of regional diversification analogous to corporate diversification as a risk-spreading strategy. A third type of relationship between variety and economic development concerns the long-term effect of variety on the economic system. An economy that does not increase the variety of sectors over time, will suffer from structural unemployment, and will ultimately stagnate. In this view, the development of new technologies and sectors in an economy is required to absorb labour that has become redundant in existing sectors (Pasinetti, 1981, 1993; Saviotti and Pyka, 2004). This process underlying long-term growth has major geographical implications, when new sectors emerge in other areas than the ones where old sectors are located. This would imply that labour becomes redundant primarily in areas where the old sectors are concentrated, while new employment is primarily created in new areas. This imbalance may be counteracted by labour migration from old to new areas and by firm migration in the opposite direction. Although many empirical studies have analysed the effects of variety on regional growth in the past decade or so, some methodological issues in empirical research remain. First, the measurement of variety is not trivial. For example, one would like to distinguish between related variety underlying spillovers and unrelated variety underlying the portfolio effects (Frenken et al., 2006; Essletzbichler, Chapter 10). Second, explaining regional phenomena requires a careful econometric specification so as to allow different effects to take place at different spatial levels of aggregation. For example, the rate of regional growth or the rate of regional information technology (IT) adoption can be made dependent on the rate of growth in neighbouring regions through the use of spatial autocorrelation econometrics (Essletzbichler, Chapter 10; Bonaccorsi et al., Chapter 12). The network perspective also lends itself for aggregation to the macro level. By aggregating networks between firms to the locations of these firms, one obtains inter-city and inter-regional networks. The underlying concept

30 14 Introduction of network cities has become very common among geographers (Pred, 1977; Hohenberg and Lees, 1995; Castells, 1996). The central idea underlying the concept of network cities holds that connectivity contributes both to urban economic growth and to urban inequalities. Examples of empirical studies that map urban networks include networks based on the ties between headquarters and subsidiaries of multinational organisations (Taylor, 2001; Alderson and Beckfield, 2004), on transportation networks (Matsumoto, 2004) or IT infrastructure (Moss and Townsend, 2000). In these views, cities can develop a more central network position by attracting corporate headquarters or functioning as transportation or IT hubs. The concept of inter-city networks can also be applied to inter-regional networks, as the contribution by Maggioni and Uberti (Chapter 11) shows. Regions acting as central hubs in the development and diffusion of knowledge will be more central in these networks, while other regions will stay more peripheral. Network position is thus expected to affect regional growth, as central hubs will receive more, and more relevant, knowledge spillovers. Using Tinbergen s (1962) gravity model from international trade theory, one can also analyse to what extent geographical distance affects the strength of knowledge flows between any two regions. This question has also been taken up by Maggioni and Uberti (Chapter 11). As for the study of firm networks, the dynamic analysis of urban and regional networks is still in its infancy. Understanding the structure of a network at one moment in time requires an understanding of the evolutionary process that has given rise to such structures. An interesting research avenue is to analyse the determinants of changes in network structures in a spatial system. For example, does the accession of Eastern European countries reorganise the hierarchy in the European city system? And, historically, can we relate the rise and fall of cities to their changing positions in global knowledge networks around emerging technologies and infrastructures (Pumain, 1997)? 3. POLICY The contributions in the present volume focus on understanding spatial phenomena from an evolutionary perspective. General policy implications are often hard to draw, if only because evolutionary theorising leaves room for small events to have long-lasting effects. Some may even go a step further to suggest that evolutionary analysis often shows the limited potential of policy makers to truly influence long-term geographical patterns of economic growth. For example, Klepper s (Chapter 4) conclusion that the US automobile industry became concentrated in Detroit for accidental

31 Introduction 15 reasons, suggests that efforts to attract new industries to a particular city or region have a low probability of success. What matters most is to have competent entrepreneurs, the presence and actions of which are hard to influence by policy. Similarly, the success story of Sophia Antipolis (Quéré, Chapter 3) suggests that its success is unique and difficult to copy. The process of regional development was set in motion by external factors such as climate, the presence of multinationals, the international airport, and one visionary man. And, in the case of Cambridge, regional development was fuelled by its excellent university as well as by the benefits of the Greater London area at just one hour from Cambridge (Garnsey and Heffernan, Chapter 2). Even if policy implications of evolutionary economics are inherently difficult to derive, a growing number of evolutionary economists are trying to 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 neoclassical economics is to be replaced by dynamic efficiency (Nelson and Winter, 1982). In other words, one is not only interested in the allocation of scarce resources present today, but also in the opportunities to create new resources in the future. In the context of economic geography, the question becomes how to design policies that promote dynamic efficiency at urban and regional levels. Boschma (2005) distinguished between two types of regional policy: evolutionary and revolutionary (Table 1.1). Evolutionary regional policy takes the specific local context and industrial structure as the starting point. It is a fine-tuning policy that aims to strengthen the connectivity between the elements of the regional system. In these circumstances, local policy makers have few degrees of freedom, yet are more likely to be successful as long as their actions are localised, that is, focused on reproducing and strengthening the existing structures. In other words, the local environment Table 1.1 Two types of regional innovation policy Evolutionary type of policy Location-specific policy Fine-tuning Strengthening existing connectivity Benefiting from specialisation Few degrees of freedom Less uncertainty Revolutionary type of policy Generic policy Restructuring of institutional framework Stimulating new connections Stimulating diversity More degrees of freedom More uncertainty Source: Adapted from Boschma (2005).

32 16 Introduction determines to a large extent available options and probable outcomes of regional policy. The goal of a revolutionary regional policy, by contrast, is the restructuring of the social and institutional framework by constructing new regional systems, increasing diversity and a high degree of openness regarding the inflow of labour, capital and knowledge. In these circumstances, local policy makers have more degrees of freedom, but at the cost of a higher degree of uncertainty regarding the actual outcome of regional policy making and its success. Since path dependence is less relevant, it is less meaningful to account for the location-specific context as a starting point for regional policy. Radically new trajectories of industrial development build on generic conditions, because the existing actors and institutional environment are unlikely to provide the specific stimuli. The case of Sophia Antipolis seems to be a good example of such a development. The paradox of regional policy holds that it can be very effective and successful in conserving economic activity by means of evolutionary policies, yet it has difficulty triggering, or even opposes new economic activity necessary for long-term development. Note, however, that evolutionary and revolutionary policies are not mutually exclusive. One can pursue finetuning policies in existing sectors while improving the generic conditions for revolutionary change to take place. However, such a two-goal policy requires careful policy making, because policies designed for one goal may in practice hamper the achievement of the other one. A way to combine both objectives is to enhance the creation of new industrial trajectories, be it new technologies or new sectors, by means of building upon the existing competence base of firms, employers and employees in the region. Radical innovations often stem from the (quite unexpected) recombination of existing technologies in entirely new ways (Levinthal, 1998). A famous example has been the rise of an environmental sector after the decline of the mining industry in the Ruhr area. A broad engineering base in the Emilia Romagna region provided a fertile ground for the emergence of a broad range of industries such as ceramics, food packaging, robotics, car manufacturing and agricultural machinery during the post-war period (Boschma, 2004). Another example is the birth of the automobile industry in the Coventry Birmingham area in England, which was partly determined by the strong presence of the bicycle and carriage industry (Boschma and Wenting, 2005). This policy captures the importance of creating related variety in a region, which broadens a region s sectoral base, while fostering knowledge spillovers between the sectors (Frenken et al., 2005, 2006). Another domain of policy, which is of crucial importance for urban and regional economic growth, is infrastructure provision. The growth of agglomerations is limited by the capacity and quality of its infrastructure

33 Introduction 17 networks. For this reason, successful regional policy always requires a complementary transportation infrastructure policy. Again, Sophia Antipolis serves as a successful example (Quéré, Chapter 3), while Cambridge suffered precisely from a mismatch between its economic development and infrastructure provision (Garnsey and Heffernan, Chapter 2). Adopting an evolutionary approach to transportation planning in the agglomeration of Amsterdam, Bertolini (Chapter 13) attempts to derive some general guidelines for planning. Given the inherent and irreducible uncertainty about the future regional development and landuse claims, urban transportation systems should be capable of resilience, that is, still function properly in the face of change. At the same time, if necessary, the system must also be responsive to change, that is, it must be adaptable. In transport systems, resilience is best shown by the network morphology and multi-modality, while adaptability is foremost a property of the policy system. The link between the two is important: in the case of Amsterdam, the resilience of the transport network morphology has been a condition for the adaptability of land use and mobility management policies, because it allowed a choice at all times between substantially different policy courses. 4. DISCUSSION Using the micro meso macro scheme in Figure 1.1 as a framework, we have discussed various applications of evolutionary economics in the field of economic geography. The common denominator in these approaches is to view spatial structures as the outcome of historical processes, and as conditioning but not fully determining economic behaviour. The explicit historical nature of evolutionary analysis, however, poses demanding requirements for empirical research. One needs to collect time-series data of evolving populations, be it from technologies, sectors, networks, cities or regions, and to apply appropriate methodologies to analyse the data collected. The contributions by Klepper (Chapter 4), Jacob and Los (Chapter 6) and Essletzbichler (Chapter 10) are fine examples of the use of econometric techniques applied to time-series data. However, other methodologies are also available to fruitfully apply evolutionary economics. For example, case-study research, combining written and oral sources, can provide 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 also be approached from an evolutionary perspective, for example, by deriving

34 18 Introduction 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-series data were available. Apart from data limitations and methodological challenges ahead, there are still a number of conceptual weaknesses that hamper the application of evolutionary economics to economic geography: for example, the concept of routines still needs to be refined (Becker, 2004), and their role in the development of multi-locational organisations is still quite unclear (Stam, 2003, 2006); and the evolutionary theory of the firm has little to say about multinational organisations, exceptions aside (Kogut and Zander, 1993; Cantwell and Iammarino, 2003). Another key concept in evolutionary economics is path dependence. Yet, its fruitful application in economic geography is still surrounded by a number of unsolved issues (Martin and Sunley, 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 knowledge spillovers, the mechanisms underlying such spillovers are still poorly understood as well as to what extent these mechanisms are sector and/or region specific. Furthermore, the importance of knowledge spillovers may be specific for the geographical distance over which they occur. The more important information flows typically stem from more distant locations, a geographical principle that might reflect the strength of weak ties (Granovetter, 1973). However, research that takes into account global spillovers is still scarce (Jaffe and Trajtenberg, 1999). In this light, the contributions by Jacob and Los (Chapter 6) and Maggioni and Uberti (Chapter 11) are especially important. The big question regarding the unequal distribution of wealth among nations needs to be addressed more often and more systematically. An evolutionary economic geography may provide a new understanding of core periphery patterns at different spatial scales as evolutionary outcomes of path-dependent dynamics. Such an approach would combine the Schumpeterian analysis of structure change with the spatial process of agglomeration and global networking. However, evolutionary growth theory (as does growth theory more generally) still lacks an explicit spatial structure. A challenge ahead is to transform evolutionary growth theory into a theory explaining the evolution of uneven distribution of economic activities in space. In all, recent research, including the chapters in this volume, has shown the value added of an evolutionary approach in economic geography. An evolutionary economic geography aims to improve our theoretical and empirical understanding of the economy as an evolutionary process that

35 Introduction 19 unfolds in space and time. Starting from the seminal contribution by Nelson and Winter (1982) and its theoretical elaborations in subsequent works (Dosi et al., 1988; Dopfer, 2005), a number of frameworks are being developed that specifically deal with geographical issues, including location theory and entrepreneurship, the spatial evolution of sectors, the geography of social networks, the evolution of spatial systems, and urban and regional planning. Methodologically, a variety of approaches are being pursued ranging from case-study research and social network analysis to duration models and spatial econometrics. Theoretically coherent and methodologically open, an evolutionary perspective is helpful in understanding the specific histories of firms and regions using a framework that is less restrictive than the neoclassical paradigm, yet more generally applicable than the institutionalist approach. It is time to take geography seriously in applied evolutionary economics. BIBLIOGRAPHY Abernathy W.J. and Utterback, J. (1978), Patterns of industrial innovation, Technology Review, 50: Alderson, A.S. and Beckfield, J. (2004), Power and position in the world city system, American Journal of Sociology, 109: Amin, A. and Thrift, N. (2000), What kind of economic theory for what kind of economic geography?, Antipode, 32(1): 4 9. Andersson, C., Frenken, K. and Hellervik, A. (2006), A complex network approach to urban growth, Environment and Planning A, 38: Andersson, C., Hellervik, A., Lindgren, K., Hagson, A. and Tornberg, J. (2003), Urban economy as a scale-free network, Physical Review E, 68(3): Part 2. Antonelli, C. (2000), Collective knowledge communication and innovation: the evidence of technological districts, Regional Studies, 34(6): Arthur, W.B. (1994), Increasing Returns and Path Dependence in the Economy, Ann Arbor, MI: University of Michigan Press. Aydalot, P. and Keeble, D. (eds) (1988), High Technology Industry and Innovative Environments: The European Experience, London and New York: Routledge. Barabasi, A.L. and Albert, R. (1999), Emergence of scaling in random networks, Science, 286(5439): Barrat, A., Barthelemy, M. and Vespignani, A. (2005), The effects of spatial constraints on the evolution of weighted complex networks, Journal of Statistical Mechanics, Art. No. P Becker, M.C. (2004), Organizational routines. A review of the literature, Industrial and Corporate Change, 13(4): Boschma, R.A. (1997), New industries and windows of locational opportunity. A long-term analysis of Belgium, Erdkunde, 51: Boschma, R.A. (2004), Competitiveness of regions from an evolutionary perspective, Regional Studies, 38(9): Boschma, R.A. (2005), Rethinking regional innovation policy. The making and breaking of regional history, in G. Fuchs and P. Shapira (eds), Rethinking

36 20 Introduction Regional Innovation and Change: Path Dependency or Regional Breakthrough?, Dordrecht: Springer, pp Boschma, R.A. and Frenken, K. (2003), Evolutionary economics and industry location, Review of Regional Research, 23: Boschma, R.A. and Frenken, K. (2006), Why is economic geography not an evolutionary science? Towards an evolutionary economic geography, Journal of Economic Geography, 6(3): Boschma, R.A. and Lambooy, J.G. (1999), Evolutionary economics and economic geography, Journal of Evolutionary Economics, 9: Boschma, R.A. and Ter Wal, A.L.J. (2006), Knowledge networks and innovative performance in an industrial district. The case of a footwear district in the south of Italy, Industry and Innovation, forthcoming. Boschma, R.A. and Wenting, R. (2005), The spatial evolution of the British automobile industry, Papers in Evolutionary Economic Geography 05.04, econ.geog.uu.nl. Bottazzi, G., Dosi, G., Fagiolo, G. and Secchi, A. (2004), Sectoral and geographical specificities in the spatial structure of economic activities, Laboratory of Economics and Management (LEM) Working Paper , sssup.it/. Brakman, S., Garretsen, H. and Marrewijk, C. van (2001), An Introduction to Geographical 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 local innovation systems: a critical survey, Industrial and Corporate Change, 10(4): Breschi, S. and Lissoni, F. (2003), Mobility and social networks: localised knowledge spillovers revisited, Centre of Research on Innovation and Internationalization (CESPRI) Working Paper 142, Burt, R.S. (1982), Toward a Structural Theory of Action: Network Models of Social Structure, Perception, and Action, New York: Academic Press. Caniëls, M. (2000), Knowledge Spillovers and Economic Growth. Regional Growth Differentials across Europe, Cheltenham, UK and Northampton, MA, USA: Edward Elgar. Cantner, U. Dreßler, K. and Krüger, J.J. (2005), Knowledge compensation in the German automobile industry, Jenaer Schriften zur Wirtschaftswissenschaft, 11/ Cantwell, J.A. and Iammarino, S. (2003), Multinational Corporations and European Regional Systems of Innovation, London: Routledge. Carrington, P.J., Scott, J. and Wasserman, S. (eds) (2005), Models and Methods in Social Network Analysis, Cambridge: Cambridge University Press. Carroll, G.R. and Hannan, M.T. (2000), The Demography of Corporations and Industries, 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 historical perspective, Oxford Development Studies, 31(1): Consoli, D. (2005), The dynamics of technological change in UK retail banking services. An evolutionary perspective, Research Policy, 34: Cooke, P., Uranga, M.G. and Extebarria, G. (1998), Regional innovation systems: an evolutionary perspective, Environment and Planning A, 30:

37 Introduction 21 Cooper, A.C. and Dunkelberg, W.C. (1987), Entrepreneurial research: old questions, new answers and methodological issues, American Journal of Small Business, 11: Cowan, R., Cowan, W. and Swann, G.M.P. (1997), A model of demand with interactions among consumers, International Journal of Industrial Organization, 15(6), De Jong, M.W. (1987), New Economic Activities and Regional Dynamics, Netherlands Geographical Studies no. 38, Amsterdam: Royal Netherlands Geographical Society. Dopfer, K. (ed.) (2005), Evolutionary Principles of Economics, Cambridge: Cambridge University Press. Dosi, G. (1997), Opportunities, incentives and the collective patterns of technological change, Economic Journal, 107(444): 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 geography of technology evolution, Tijdschrift voor Economische en Sociale Geografie, 96(1): Foray, D. (1997), The dynamic implications of increasing returns: Technological change and path dependent inefficiency, International Journal of Industrial Organization, 15(6): Freeman, C. and Perez, C. (1988), Structural crisis of adjustment, business cycles and investment behaviour, in Dosi et al. (eds), pp Frenken, K., van Oort, F.G., Verburg, T. and Boschma, R.A. (2005), Variety and regional economic growth in the Netherlands, Papers in Evolutionary Economic Geography 05.02, Frenken, K., van Oort, F.G. and Verburg, T. (2006), Related variety, unrelated variety 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: the evolution of high tech Cambridge and Sophia-Antipolis, International Journal of Technology Management, 28(3 6): Gertler, M.S. (1995), Being there : proximity, organization, and culture in the development and adoption of advanced manufacturing technologies, Economic Geography, 71: Giuliani, E. (2005), The structure of cluster knowledge networks: uneven and selective, not pervasive and collective, Danish Research Unit for Industrial Dynamics (DRUID) Working Paper , Grabher, G. (1993), The weakness of strong ties: the lock-in of regional development in the Ruhr area, in G. Grabher (ed.), The Embedded Firm: On the Socio- Economics of Interfirm Relations, London and New York: Routledge, pp Granovetter, M. (1973), The strength of weak ties, American Journal of Sociology, 78(6): Guimerà, R. and Amaral, L.A.N. (2004), Modelling the world-wide airport network, European Physical Journal B, 38(2): Hagedoorn, J. (1993), Understanding the rationale of strategic technology, Strategic Management Journal, 14(5): Hall, P.G. and Markusen, A.R. (eds) (1985), Silicon Landscapes, Boston, MA: Allen & Unwin.

38 22 Introduction Hannan, M.T., Carroll, G.R., Dundon, E.A. and Torres, J.C. (1995), Organizational evolution in a multinational context: entries of automobile manufacturers in Belgium, Britain, France, Germany, and Italy, American Sociological Review, 60(4): Hodgson, G.M. (1998), The approach of institutional economics, Journal of Economic Literature, 36(1): Hohenberg, P.M. and Lees, L.H. (1995), The Making of Urban Europe , Cambridge, MA: Harvard University Press. Hoover, E.M. and Vernon, R. (1959), Anatomy of a Metropolis: The Changing Distribution 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: Jaffe, A.B., Trajtenberg, M. and Henderson, R. (1993), Geographic localization of knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics, 108(3): Keeble D., Lawson, C., Moore, B. and Wilkinson, F. (1999), Collective learning processes, networking and institutional thickness in the Cambridge region, Regional Studies, 33(4): Klepper, S. (2001), The evolution of the U.S. automobile industry and Detroit as its 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. automobile industry, Industrial and Corporate Change, 11(4): Kogut, B. and Zander, U. (1993), Knowledge of the firm and the evolutionary theory of the multinational corporation, Journal of International Business Studies, 24: Krugman, P.R. (1991), Increasing returns and economic geography, Journal of Political Economy, 99(3): Lambooy, J.G. and Boschma, R.A. (2001), Evolutionary economics and regional policy, Annals of Regional Science, 35: Levinthal, D.A. (1998), The slow pace of rapid technological change: Gradualism and punctuation in technological change, Industrial and Corporate Change, 7: 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 critical reflections, Cambridge Journal of Economics, 23(1): Martin, R. (2000), Institutional approaches in economic geography, in E. Sheppard and T.J. Barnes (eds), A Companion to Economic Geography, Oxford and Malden, MA: Blackwell, pp Martin, R. and Sunley, P. (2006), Path dependence and regional economic evolution, Papers in Evolutionary Economic Geography 06.06, geog. uu.nl. Maskell, P. (2001), The firm in economic geography, Economic Geography, 77(4): Maskell, P. and Malmberg, A. (1999), Localised learning and industrial competitiveness, Cambridge Journal of Economics, 23(2): Matsumoto, H. (2004), International urban systems and air passenger and cargo flows: some calculations, Journal of Air Transport Management, 10(4):

39 Introduction 23 Metcalfe, J.S. (1995), Technology systems and technology policy in an evolutionary framework, Cambridge Journal of Economics, 19: Moss, M.L. and Townsend, A.M. (2000), The Internet backbone and the American metropolis, The Information Society, 16: Murmann, J.P. (2003), Knowledge and Competitive Advantage: The Co-evolution of Firms, Technology, and National Institutions, Cambridge: Cambridge University Press. Nelson, R.R. (1995), Co-evolution of industry structure, technology and supporting institutions, and the making of comparative advantage, International Journal of the Economics of Business, 2(2): Nelson, R.R. (1999), The sources of industrial leadership: a perspective on industrial policy, De Economist, 147: 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: Pasinetti, L.L. (1981), Structural Change and Economic Growth, Cambridge: Cambridge University Press. Pasinetti, L.L. (1993), Structural Economic Dynamics, Cambridge: Cambridge University Press. Perez, C. and Soete, L. (1988), Catching up in technology: entry barriers and windows of opportunity, in Dosi et al. (eds), pp Phelps, N.A. (2004), Clusters, dispersion and the spaces in between. For an economic geography of the banal, Urban Studies, 41(5 6): Plouraboue, F., Steyer, A. and Zimmerman, J.-B. (1998), Learning induced criticality in consumers adoption pattern: a neural network approach, Economics of Innovation and New Technology, 6: Porter, M.E. (2000), Location, competition, and economic development: local clusters in a global economy, Economic Development Quarterly, 14: Powell, W.W., Koput, K.W. and SmithDoerr, L. (1996), Interorganizational collaboration and the locus of innovation: networks of learning in biotechnology, Administrative Science Quarterly, 41(1): Pred, A. (1977), City-Systems in Advanced Economies, London: Hutchinson. Pumain, D. and Moriconi-Ebrard, F. (1997), City size distributions and metropolisation, Geojournal, 43(4): Rigby, D.L. and Essletzbichler, J. (1997), Evolution, process variety, and regional trajectories of technological change in US manufacturing, Economic Geography, 73(3): Saviotti, P.P. and Pyka, A. (2004), Economic development by the creation of new sectors, Journal of Evolutionary Economics, 14(1): Saxenian, A. (1994), Regional Advantage, Cambridge, MA: Harvard University Press. Solomon, S., Gerard, W., de Arcangelis, L., Jan, N. and Stauffer, D. (2000), Social percolation models, Physica A, 277: Solow, R.M. (1957), Technical change and the aggregate production function, Review of Economics and Statistics, 39: Stam, E. (2003), Why butterflies don t leave: locational evolution of evolving enterprises, Dissertation, Utrecht University. Stam, E. (2006), A process model of locational change: implications for an evolutionary economic geography, in A. Pyka and H. Hanusch (eds), Applied

40 24 Introduction Evolutionary Economics and the Knowledge-Based Economy, Cheltenham UK and Northampton, MA, USA: Edward Elgar, pp Storper, M. (1997), The Regional World: Territorial Development in a Global Economy, London: Guilford Press. Stuart, T. and Sorenson, O. (2003), The geography of opportunity: spatial heterogeneity in founding rates and the performance of biotechnology firms, Research Policy, 32(2): Swann, P. and Prevezer, M. (1996), A comparison of the dynamics of industrial clustering in computing and biotechnology, Research Policy, 25: Taylor, P.J. (2001), Specification of the world city network, Geographical Analysis, 33: Teece, D., Pisano, G. and Shuen, A. (1997), Dynamic capabilities and strategic management, Strategic Management Journal, 18(7): Tinbergen, J. (1962), Shaping the World Economy: Suggestions for an International Economic Policy, New York: 20th Century Fund. Uzzi, B. (1996), The sources and consequences of embeddedness for the economic performance of organizations: the network effect, American Sociological Review, 61(4): Vernon, R. (1969), International investment and international trade in the product life-cycle, Quarterly Journal of Economics, 80: van Wissen, L. (2004), A spatial interpretation of the density dependence model in industrial demography, Small Business Economics, 22(3 4): Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods and Applications, Cambridge: Cambridge University Press. Watts, D.J. and Strogatz, S.H. (1998), Collective dynamics of small-world networks, Nature, 393: Werker, C. and Athreye, S. (2004), Marshall s disciples: knowledge and innovation driving regional economic development and growth, Journal of Evolutionary Economics, 14: Whitley, R. (1992), Business Systems in East Asia: Firms, Markets and Societies, London: Sage.

41 PART I Entrepreneurship

42

43 2. The Cambridge high-tech cluster: an evolutionary perspective Elizabeth Garnsey and Paul Heffernan 1. INTRODUCTION New knowledge-based firms emerge and grow around centres of learning and research. A process akin to ecological succession occurs as resources in the local science base are converted into and attract business activity, giving rise to a richer, more diverse economic habitat. To gain a better understanding of these developments we need to examine how processes of change operate over time. Cambridge provides an exemplar of endogenous formation of a high-tech cluster through spin-off,agglomeration and institutional adaptation. The importance of the strong science base at Cambridge is universally acknowledged, but the evolutionary micro processes through which its influence was exerted require further elucidation. Positive externalities providing incentives to firms to cluster in an area are not necessarily present at the time of the emergence of a new cluster of activity. Explanations in terms of measurable externalities beg the question of the evolutionary processes which gave rise to them. In other high-tech centres (Simmie et al., 2004), spillover effects have resulted from government spending on infrastructure, from large company investments, from metropolitan structures and defence spending on information technology (IT). These influences were absent in the case of Cambridge. After identifying the conceptual building blocks, we then examine indices of the growth of high-tech clusters in the Cambridge area. Underlying these aggregate trends are self-reinforcing mechanisms involving business spinouts and networks of knowledge diffusion. Case studies illustrating these processes, in the clustering of scientific instrumentation, information and communication technology (ICT), technical design consultancies and biotechnology, further develop the thesis. Global linkages of Cambridge firms and the local process of institutional transformation in the region, resulting from collective action of local firms, reveal how global and local connections interact. 27

44 28 Entrepreneurship 2. CLUSTERING AS AN ENDOGENOUS PROCESS While traditional approaches in economic geography explained the location of industry in terms of exogenous factors such as demand or international trade effects, more recently endogenous factors have been identified as drivers of local clustering. In particular, spillover effects are said to shape location decisions where firms derive benefit from local investments, the cost of which they do not themselves bear (Breschi and Lissoni, 2001). Value chain considerations shaping location decisions of firms are sometimes subsumed under spillover and sometimes treated as an alternative externality affecting location decisions. From an evolutionary perspective, the most striking feature of Cambridge as a high-tech centre is the way in which participants transformed the area, creating the very externalities that are assumed to be attributes of the area in leading accounts. Although local firms undoubtedly benefited from the university infrastructure, particularly ventures incubated within university departments, the concept of knowledge spillovers does not fully explain spin-out activity. University scientists can create their own firms to capture returns from their intellectual capital; they do not have to allow commercial firms to reap the gains from their intellectual capital, as is implied by the free-rider connotations of spillover (Zucker et al., 1998). Patents offer means of attracting capital through the prospects they offer for appropriating returns. However, the formal and informal relationships that these scientists develop with commercial organisations extend well beyond the market relations identified by Zucker et al. Through the operation of their networks, academic entrepreneurs have been among those creating a more favourable local business environment. The other endogenous determinant of clustering recognised in the literature, relates to local supply chain benefits. This approach sees clusters as geographic concentrations of interconnected companies and their specialised suppliers, service providers and firms in related industries, together with associated institutions (Porter, 1990). Supply chain benefits stem from co-locating with suppliers and customers, with a consequent lowering of transport and coordination costs, while co-location may also provide information about new input technologies and potential products (Krugman, 1991). Co-locating with customers provides information about their market needs, and may offer economies of scale to suppliers. But supply chain effects are not spillovers in that they do not offer free-rider gains (Zucker et al., 1998). Firms must invest time and effort in internalising supply chain externalities through deliberate measures in order to benefit from labour availability, to reach price agreements, to establish partnerships or make procurement arrangements. Firms in competition

45 The Cambridge high-tech cluster 29 with each other will only co-locate if they can achieve such benefits and overcome any disadvantages in having nearby competitors. Information on supply chain arrangements that work well is diffused over time on a cumulative basis. The customers of Cambridge high-tech companies are international rather than local, and relatively few of them have local suppliers in their specialised production chain. However, they do benefit from wider value chain externalities (Krugman, 1991). We shall see that high-tech firms in the Cambridge area currently make use of value chain complements or substitutes for the firms internal activities, in particular by outsourcing to local legal and business services. These have been attracted to the area by the presence of high-tech firms, illustrating the endogenous dimension of the transformation of the area. For knowledge-intensive firms, access to specialised 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 the locality, but a labour market of experienced specialised professionals, another outcome of cumulative processes. Explanations of local clustering in terms of positive externalities have largely adopted cross-sectional methodologies to identify the determinants of clustering (Breschi and Lissoni, 2001). This method is unable to explain why clusters emerged in some places and not in others before such time 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 factors favouring the growth of a specific location, a finding anticipated by Maruyama s work on positive feedback (Maruyama, 1963). The first is a dynamic model in which firms sequentially choose locations on the basis of agglomeration economies, including skilled labour markets, specialised suppliers and knowledge spillovers. Arthur showed that local concentrations of firms in a given industry are neither fully determined by geography for example, local resource endowments, transport potential and firm requirements nor entirely a matter of historical chance, as assumed in competing theories of industrial location. Instead, both geographic attractiveness and accidental historical order of choice come into the equation, with various outcomes following from upper limits to agglomeration benefits. A second model by Arthur (1994) is based on industry formation through spin-off (or spin-out 1 ). Clustering can be explained by a sequential process in which the probability of a region producing spin-offs at time t 1 is dependent on the relative number of firms that located at time t.by drawing randomly, at each time t, one firm that produces a spin-off, an evolving spatial distribution of firms in an industry is simulated. Typically,

46 30 Entrepreneurship the resulting spatial distribution is highly skewed, because some regions will by chance have a relatively high number of spin-offs early on and, subsequently, these will produce further spin-offs. The outcome of the spinoff model is dependent on the assumption that spin-offs locate in the same region as the parent company, since without this assumption one would obtain a random location model, but the assumption that spin-offs locate near 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 the San Francisco Bay area where spin-outs from Shockley and Fairchild were celebrated (Cooper, 1971) as well as in a study of clustering of the US automobile industry in Detroit (Klepper, 2002). Arthur s approach is illuminating in exploring the interplay of chance and cumulative process manifest in path dependence (Boschma and Frenken, 2003). In particular it highlights the mechanisms of spin-offs and of agglomeration economies which are likely to operate simultaneously. The contribution of these models lies in demonstrating the principle that even for regions equivalent in terms of institutions and endowments, the incidence of specialist clusters will be highly uneven because it occurs through chance events that set in motion a self-amplifying process in which success breeds success. Arthur s abstract models are designed to exclude a variety of other causal factors influencing the emergence of local activity, among which are culture, institutions and endowments. In the following analysis, we observe that the endowment represented by Cambridge University provided critical conditions for the emergence of local industry through clustering. This endowment can be viewed either as systematic or as a chance initial condition, depending on whether the geography of the university s location or the historical accident of its location is emphasised. Exogenous factors were also at work in the form of international demand for high-tech products and services to which Cambridge companies responded. But as Arthur s model predicts, endogenous spin-off and agglomeration economies have played the major role in the emergence of high-tech industry in Cambridge. In this chapter, clustering is related above all to the inter-generational spin-out process in Cambridge. The firms in question are connected locally by mobile people and knowledge to a greater extent than by supply relations, and operate in value chains that have global reach. We show that the cumulative, path-dependent nature of local value creation and appropriation processes is the source of their locational impact, to an extent ignored by the static methodologies of literature that neglect evolutionary processes.

47 The Cambridge high-tech cluster THE EMERGENCE OF THE CAMBRIDGE HIGH- TECH CLUSTER Cambridgeshire is located northeast of London; the county extends 1300 square miles and had a population of about at the turn of the millennium. Local firmsare active in IT, advanced electronics and engineering, materials, instrumentation, biotechnology and research services, developing the emerging and early diffusing technologies characteristic of high-tech sectors (Butchart, 1986). The development of high-tech industry in the Cambridge region has been rather exceptional, also labelled the Cambridge phenomenon (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 jobs in the Cambridge high-tech sector. A consultants report had identified extensive spin-out activity (Segal et al., 1985). By the end of the century there were more than 1200 technologyrelated firms (depending on definitions and area); these firms employed people, approximately 10 per cent of the total Cambridgeshire workforce (Figure ). By 2000, the combined turnover of technology-based enterprises was over 3.5 billion sterling (Heffernan and Garnsey, 2002). 3 The recession of the early 1990s reduced employment in the larger, older firms in electronics and instrumentation, but new firms and specialisations arose, restoring technology-employment totals. Manufacturing declined as Firms Source: figures based on Garnsey and Cannon-Brookes (1993); 1984 and 1986 figures interpolated; data since 1988 derived from Cambridgeshire County Council Employment database, as described in Heffernan and Garnsey (2002), extended to 2004 and excluding Peterborough. The data differ from those used by PACEC (2003), which examined high-tech activity over the Greater Cambridge Partnership area, a wider area than Cambridgeshire. Figure 2.1 High-tech firms in Cambridgeshire,

48 32 Entrepreneurship a proportion of high-tech activity as service activities increased, as in the UK economy at large. Figure 2.2 shows the changing distribution of firms and employment across a selection of high-tech sectors (Butchart, 1986). Software firms were the largest single group, reflecting both extensive IT expertise and low barriers to entry. Instrumentation, and research and development (R&D) were also important sectors, the latter being the largest employer. Biotechnology-related firms accounted for 15 per cent of firms and 29 per cent of employment by Absolute numbers of Cambridge high-tech manufacturing firms were on the increase until the 2004 survey (in contrast with figures for the same sectors in the UK). High-tech employment in the area in and around Cambridge 4 fell by about 5 per cent between 2002 and 2004, but the number of firms remained unchanged in the face of the technology slump and collapse in market valuations that followed the boom of the 1990s. The Cambridge phenomenon has grown mainly through the entry of new enterprises, most of these remaining small. The mean size of firms in 2004 was 33, less than that in 1988 (38), though median size increased slightly from 6.5 to 7. It must be noted, however, that a skewed firm-size distribution is universal and applies over time as well as nationally and regionally; only a small proportion of any cohort of firms grow to become major employers over time (Storey, 1994). It requires a large pool of start-ups to generate a few major companies. Firms of small size dominate in high-tech Cambridge to a lesser extent than is found across the UK as a whole as seen when relevant sectors are matched (Heffernan and Garnsey, 2002). The continued expansion of high-tech activity depended, however, on the entry of new firms exceeding exit rates. Between 1989 and 2004 the mean number of new high-tech firms registered by the local authorities in Cambridge each 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 2.4 shows that survival rates for Cambridge technology-based firms were consistently higher than the regional and national averages, and compared well with survival rates reported in other studies (see Kirchoff, 1994; Slatter, 1992; Storey, 1994). This graph suggests that Cambridge firms had superior capabilities compared to the UK average. 4. CASE STUDIES OF CAMBRIDGE CLUSTERS Scientific Instrumentation The earliest Cambridge high-tech cluster was of scientific instrumentation firms, a response to the rise of the science departments of Cambridge

49 2004 Number of firms Employment Source: Calculated from County Council records. Figure 2.2 Changing sectoral growth patterns among Cambridge high-tech firms Telecommunications Electrical engineering IT hardware IT services IT software Research and development Biotech Instrumentation 33

50 34 Entrepreneurship New firms Lost firms Net effect Source: Calculated from County Council records. Figure 2.3 Firm turnover (turbulence) in the Cambridge high-tech sector University in the nineteenth century. By the 1980s this cluster included Cambridge Instruments, UniCam and other Pye (Philips) sites employing several thousand in the area. Large numbers of managers and technicians in 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 manufacturing firms, proved unable to adapt to rapid technological change and international competition (Koepp, 2002). They were acquired and downsized in a series of retrenchments and new, much smaller firms were formed by former managers and technicians. By the late 1990s all the older instrumentation companies had been acquired and de-merged. The instrumentation sector now includes about 100 firms in the Cambridge region with very diverse products. Some newer ones have responded to international opportunities by addressing expanding markets for automating biotechnology research. The presence of biotech networks in the area has been a factor in the recognition of lab automation opportunities and in access to relevant competence in the life sciences. But overall, the capability of the area in instrumentation is less than could be expected from the concentration of science laboratories. 5 There was a failure of established firms to maintain a leading role in the area such as Hewlett Packard achieved in

51 The Cambridge high-tech cluster % % Firms remaining 100% 90% 80% 70% 60% 50% 1989/90 cohort 1991/92 cohort 1993/94 cohort 1995/96 cohort 1997/98 cohort UK East 1995 UK % Years after start-up Note: ONS rates are based on value-added tax (VAT) registrations and deregistrations which 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 Cambridge high-tech firms (with regional and national averages) Silicon Valley (Saxenian, 1994) or to be renewed through government university manufacturing modernisation initiatives such as were undertaken elsewhere in the US (Best and Forrant, 2000). ICT The information technology sector has shown the highest rates of new firm formation, reflecting demand for IT products and services and low barriers to entry. There has been no official audit of companies founded by members of the University of Cambridge. Only firms in which the university took out equity are registered, a very small proportion of the total originating in the university. New evidence shows that over a hundred companies were founded by members of the university computer science department, about 80 per cent located in the UK and about 50 per cent in the Cambridge area. 6 IT companies have also originated from other departments, above all the engineering department, which has been the source of a diverse set of new technologies and firms mainly located in the area (Figure 2.5). As with most populations of new firms, only a small proportion of departmental spin-outs have grown to above average size,

52 36 Entrepreneurship Vivamer 2002 Chem Eng Splashpower 2002 Cool Analgesia Alphamosaic 2001 Cirocco CCL 1960 Blue Technologies 2001 Chem Eng 1980 Cambridge Semiconductors Delcam 1968 Jumpleads 2000 Vulcan Machines 2000 ChyGwyn 1999 Cambridge Flow Solutions 1999 CAM 3D 1999 CRISP 1999 Transversal 1998 Spark Ltd 1998 CMIL 1998 Sintefex Audio Lab 1997 Richard Marshall Ltd 1996 Softsound 1995 Acq Cambridge University Engineering Department Ionotec 1995 Failed Acquired Univ IP/ equity Still in existance in 2002 Shearline Precision Engineering 1973 TopExpress 1979 Acq Adder Technology 1984 Synoptics 1985 Entropic 1995 Acq HRM Software 1986 Qudos 1986 failed Acq Cambustion 1987 JMEC 1989 Asymptote 1989 CEDAR Audio 1989 PCME 1991 Biorobotics 1992 Acq Granta Design 1993 Figure 2.5 Companies with founders from Cambridge University engineering departments but several firms founded by former students have become leading companies in their sector, including, for example, Autonomy. Several hardware pioneers did not survive the maturation of the microcomputer industry. But Acorn Computers, founded in 1979, illustrates the influence that an originating firm can have on its progeny. The founders of Acorn Computers, Hermann Hauser and Chris Curry, drew on expertise from the University Computer Lab to develop an innovative microcomputer, the production of which was subcontracted outside the area. Acorn s explosive growth was stemmed by a sudden slump in consumer demand in 1984 and overcommitment to suppliers. The company only survived through acquisition by Olivetti. Acorn s proprietary operating system came under competition from the new industry standard, MS-DOS. Under Olivetti s ownership, Acorn attempted to pioneer the Network Computer in an international strategic partnership that fell through, the Network Computer being a product before its time. Acorn was wound up in 1999 so as to realise for Olivetti the value of its shares in ARM, and to create the new spin-out, Element-14. Acorn had funded the development of ARM s technology before spin-out and was its first customer. Acorn Computers

53 The Cambridge high-tech cluster 37 Real VNC, 2002 Ubisense (Ubiquitous Systems), 2002 Level 5 Networks (Cambridge Internetworking), 2002 Cambridge Broadband, 2000 (Cambridge Network Ltd, 1998) Adaptive Broadband, 1998 E*Trade UK, 1998 Amadeus Capital Partners, 1997 NetChannel Ltd, 1996 NetProducts Ltd, 1996 IPV (Telemedia Systems), 1995 Electronic Share Information Ltd, 1993 Advanced Displays SynGenix, 1992 Icora Semiconductor, 2003 Orbis 1978 ACORN Founders Pre ABC, 1988 IXI Ltd 1990 EO Inc. (with AT&T) 1995 Vocalis, Pogo Mobile Solutions, 2002 Commtag, 2000 Element 14, 1999 Xemplar Education, 1996 ncipher, 1996 STNC, 1993 ANT, 1993 ARM, 1990 GIS, 1985 Clearswift (Net Tel Computer Systems),1982 IQ Bio, 1981 Qudos Technology Ltd, 1985 Harlequin Ltd, 1986 ATM 1993, VIRATA Figure 2.6 New firms started by founders and employees of Acorn Computers was a learning organisation for the whole area, providing experience to many local entrepreneurs and managers. The range and depth of competence developed at Acorn made it possible for former members to start large numbers of local spin-outs (Figure 2.6). A further 20 firms were founded by employees leaving ARM to start a business. 7 Technical Design Consultancies A cluster unique to Cambridge is made up of technical design consultancies working with leading international clients. These have become repositories of a wide range of competence, technical and managerial. They engage in prototype and small-scale production as well as advisory consultancy. The pioneer, Cambridge Consultants Ltd (CCL), founded by former members of the chemical engineering department in 1960, attempted to establish a manufacturing unit in their early days. Lack of competence in manufacturing led to a cash flow crisis and acquisition by A.D. Little. Thereafter the company specialised in technical services, a sector in which firms can minimise the sunk costs that often undermine UK ventures. Figure 2.7, below, shows the spin-outs from CCL. When employees wished to use CCL s intellectual property (IP) to develop a product, CCL helped

54 38 Entrepreneurship them with seed capital, taking out equity in such ventures (Auton and Biddle, 2001). Former CCL employees later spun out other design consultancies, via another consultancy with a local office. One of these, The Technology Partnership (TTP), has been the source of several further spinouts. TTP s sales were under 20 million in 2001, but with its combined spin-outs, sales were over 80 million.8 Industrial Ink Jet Printing Largely spinning out of CCL is a cluster of Cambridge firms that have achieved global expansion on the basis of a common set of technologies. Figure 2.7 depicts the genealogy of seven local industrial ink jet printing companies (Imaje is located in France and Willett in Corby). By around 2000 the ink jet printing firms employed over 3000 directly and their custom provided further jobs in PCB and precision engineering firms in the regions. They were dominant in international markets for non-impact product identification, a smaller market than the major market for small business and desktop ink jet printing. Their revenues were estimated at 2000 Inca Digital Printers 1990 Xennia Xaar Cambridge Consultants Limited Linx Willett Biodot Domino Printing Sciences Elmjet Imaje Figure 2.7 Industrial inkjet printing spin-outs originating in Cambridge

55 500 million in The production chain is international; ink jet printing (IJP) firms source jewels from Switzerland, pumps from the US and precision components from many other sources, but they also share some local suppliers. The firms do not formally collaborate, or regularly supply each other, but interviewees report frequent informal interaction and linkages resulting from their common origins and the mobility of staff. IJP customer companies in the area have helped local PCB and precision engineering suppliers to upgrade their performance; these contractors were then able to help other customers in the area to upgrade their products and production process. Local IJP clients remain an important source of custom for local suppliers, but subassemblies came to be sourced internationally as the industry matured. The role of leading firms in production networks is confirmed by the experience of Domino Printing Sciences, which grew to over 1000 employees and had a particular impact on suppliers in eastern England such as Hansatech in King s Lynn. The industrial IJP firms continue to provide a local labour market skilled in relevant competences. When ink jet technologies were adopted by new entrants developing advanced materials such as light emitting polymers (Cambridge Display Technology and Plastic Logic) they were able to hire professional staff with experience in the local IJP cluster, demonstrating the role of job mobility in the diffusion of competence in the area. However, at technician and operator levels, IJP firms found it difficult to recruit skilled personnel at competitive rates, and it was partly for this reason that Videojet (the acquirer of Elmjet) moved out of the Cambridge area, while Xaar s manufacturing function was relocated as a result of a merger. Biotechnology The Cambridge high-tech cluster 39 The biotech cluster clearly illustrates the interaction of endogenous and exogenous influences. There are about 110 biotech companies in the Cambridge area, 40 or so indigenous spin-outs from university departments and local research institutes, the others being local start-ups or external entrants. Because biotech has extensive applications, these companies embody a wide range of competence in diverse life sciences, instrumentation, chemistry and computing. Twelve different university departments were the source of the biotech companies recognised by the university as spin-outs because they embody university IP and equity. The nautilus form of Figure 2.8 illustrates the increasing incidence of biotech spin-out activity over time. Among the 42 firms shown in Figure 2.8, there were 20 in biopharm therapeutics and three in biopharm diagnostics, all of these engaged in R&D activities. Only one firm, Cambridge Antibody Technologies, was

56 40 Entrepreneurship Notes: 1. Celltech, Cambridge Drug 29. Cambridge Biotechnology, 2. Affinity Chromatography, Discovery, Kudos, Akubio, Cantab Pharmaceuticals, 17. AdproTech, Purely Proteins, Abcam, Genepta, Cambridge Antibody 19. Sense Proteomic, 33. Smart Holograms, 2002 Technology, Daniolabs, Cambridge Sensors, 20. Paradigm 35. Blue Gnome, Therapeutics, 36. Vivamer, BioRobotics, Diagnostics for 7. Microbial, Solexa, 1998 the Real World, Hexagen, Cambridge 38. Ionscope, Cambridge Microbial 39. Ampika, 2003 Combinatorial, Technologies, Cambridge Lab on Chip, Clinical & Metris Therapeutics, Biomedical 41. Protein Logic, Computing, 42. Novexin, Cambridge 1999 Biotransforms, 24. De Novo 1997 Pharmaceuticals, 12. RiboTargets, Biotica Technology, 25. Astex 1997 Technology, Cambridge 26. Diversys, 2000 Bioclinical, 27. Avidis, Cool Analgesia, 2001 Source: Research Services, Cambridge University. Figure 2.8 Biotech firms originating in 12 Cambridge University departments

57 The Cambridge high-tech cluster 41 in integrated drug discovery and production. The remaining spin-outs were supporting drug discovery through bioinformatics (10) and specialist services, devices and supplies (8). Ownership changes have been considerable, with over a quarter of the biotech spin-out ventures being acquired or merged. Local biopharm ventures together with attracted firms are sufficiently numerous to provide a local labour market in commercial life sciences. As the cluster has matured, the professional networks formed with companies elsewhere and around local career structures have become increasingly significant. 10 The strength of life sciences in the university and location of the Medical Research Council and other research institutes in the area ensured that as market applications for biotech emerged, Cambridge would be a leading centre. This was confirmed when a number of biotech incubators were established in the late 1990s and the Human Genome project was sited locally. Spin-outs were attracted from big pharmaceuticals and from other universities. Case histories reflect complex alliances and reconfigurations. For example, Enzymatics, originally financed by British Sugar, was closed when its funding was terminated, but its IP became the endowment of a local spin-out, Chiroscience, which later merged with Celltech, sold to a Belgian pharmaceutical company in INTERNATIONALISATION AND INSTITUTIONAL ADAPTATION The case studies all show that the emergence of high-tech clusters has been largely an endogenous process driven by spin-outs and emerging agglomeration advantages. Two accompanying processes that have further nurtured the clustering processes deserve further attention. First, the increasing international linkages of Cambridge firms and the transformation of institutions such that they have become more supportive for the further development of high-tech clusters. Internationalisation New firms in the area struggled to obtain investment, reflecting the shortterm focus of capital markets and rates of return that are higher in other more cartelised, less innovative activity in the UK economy. The introduction of standardised credit rating by British banks did not favour innovative businesses. Until the late 1990s, venture capital in the area consisted in three funds investing in about five ventures each among all Cambridge high-tech companies. Financial conditions improved when high returns on US tech-based

58 42 Entrepreneurship investments changed the investment climate at the turn of the century. By the late 1990s, several hundred million pounds worth of venture capital was under management in the area through new venture capital funds seeking to invest in technology-based firms in Cambridge and beyond. Further seedcapital initiatives by the new government in favour of the commercialisation of science-based technology programmes were undertaken. A few highprofile Cambridge firms realised the financial advantage of their booming share prices. Cambridge companies survived the Internet boom and crash better than elsewhere because of the diversity of applications and markets. The attraction of international firms to the area has been a relatively new development (Garnsey and Longhi, 2004). But the ability of the area to attract inward investment had earlier been demonstrated by international acquisition of local enterprises. A 1993 study showed that the most promising Cambridge high-tech enterprises were prone to acquisition, reducing the pool of companies with prospects for independent growth (Shah and Garnsey, 1993). This pattern has continued with the acquisition of a range of promising young Cambridge companies by US corporations in particular. There is case-study evidence (for example, Acorn, CIS, Smallworld and BioRobotics) that acquisition reduces innovative potential by subjecting the acquired unit to corporate strategic priorities. A counter-effect is that inward investment through acquisition stimulates further spin-out activity in the area. Although high-tech firms in Cambridge are linked locally by common origins of their members and local expertise, their production networks are more international than local. Cambridge high-tech firms are highly export oriented; even the technical design consultancies report that the majority of their clients are international. Export performance was outstanding even before the high-tech boom. The need to establish relations with customers overseas from start-up stage makes considerable demands on young Cambridge companies; those engaged in micro-global efforts cannot take the easier route of building initial relationships with domestic customer and supplier companies which is open to many US start-ups. The need for new firms to build the foundations for growth from the outset through a structure of alliances was the lesson drawn by local firms from earlier experience in the area. New business models developed by Cambridge firms involve licensing and establishing close linkages with manufacturing partner firms globally. Alliance capabilities are needed to reach multiple markets and are facilitated by international networks developed by local entrepreneurs. Cambridge enterprises that include ARM and CDT have demonstrated that licensing can secure substantial returns provided that the technologies are well protected and have multiple market applications.

59 Institutional Adaptation The Cambridge high-tech cluster 43 The Cambridge University long operated as a community of scholars made up of colleges and departments rather than as a conventionally managed organisation; the central administration was minimal until after 2000 and did not have the means or inclination to manage technology transfer centrally. From 1986, British universities had rights to IP in work funded by the research councils. The Cambridge University was unusual in 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 members from developing commercial applications or starting new businesses so long as they carried out their teaching and research duties. In professional communities moral pressures can be brought to bear on individuals who fail to conform to norms and expectations. Cambridge academics are embedded in a network of obligations, collegial and departmental in a university with a strong research and teaching ethos. Entrepreneurial academics who undertook business activities over and above their academic duties and used the revenues to fund research students and departmental research were able to alter a climate of opinion that had earlier been hostile to business activity by academics. The transformation of Cambridge from intellectual retreat to high-tech centre took place initially as enterprising IT experts associated with the university detected and responded to growing economic demand for information technology. By the 1970s, expertise in science and technology developed at Cambridge University was diffusing from research into business activity. By 1995, the Science Park housed over 70 firms and 3500 jobs. The Innovation Centre founded by St John s College became the nodal centre of the Cambridge Phenomenon. 11 Local and central government were unsupportive of business expansion in Cambridge. But the authorities were unable to stem the expansion in technology enterprise. After years of spontaneous expansion, Cambridge high-tech companies faced problems they could not solve on an individual basis. Inadequate public transport and shortages of housing and skilled technical labour had become significant constraints on growth and had not been addressed by government. Informal linkages were rapidly mobilised to 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 US markets and funding for Cambridge high-tech firms, drawing on contacts forged by entrepreneurs who had strong US connections. Business networks in the Cambridge area extended their reach into the policy arena. Since 1997, Cambridge entrepreneurs have taken part in government policy

60 44 Entrepreneurship making. Local government officials and businesses formed the Greater Cambridge Partnership, recognising that a larger territory than the city and its villages was the appropriate unit for sustainable development. Whether Cambridge could or should be the hub of a new high-tech region remained controversial; strong anti-growth feeling and concern about the polarisation of the labour market coexisted with advocacy of economic potential. Working groups were set up to encourage related expansion in the region and beyond. Local recognition of the finance gap for firms developing emerging technologies and the extent to which US technology enterprise has benefited from much higher levels of government support, for example, through Defense Advanced Research Agency (DARPA) and Small Business Innovation Research and (SBIR) programmes, resulted in lobbying for a bill, sponsored by a Cambridge Member of Parliament, providing for more grant and procurement support on the US model to high-tech ventures in the UK. 6. CONCLUSION In summary, endogenous developments in Cambridge encompass the founding of companies by current and former members of the university, clustering stimulated by serial spin-outs from originator firms, the rise of local suppliers and, especially significant, the emergence of specialist labour markets. These developments depended on demand for high-tech output and exerted attraction effects through business services drawn to the area, through the implantation of international subsidiaries, inward investment via acquisition and the attraction of venture capital funds. Together these processes, endogenous and exogenous, contributed to the development of local competence and capabilities resulting in the formation and success of many new firms. While the process of transformation has many features in common with ecological succession in the natural world, this new economic habitat developed its own identity and local consciousness at a higher-order level. The creation of a favourable selection environment was the unintended outcome of many local decisions, but over time a sense of purpose was created in the local high-tech community whose champions set out deliberately to attract resources to the area. The need for such action points to the limitations of spontaneous expansion unsupported by regional or technology policy. It is difficult to isolate the effects of knowledge networks around a strong science base in places where many other influences have been at work. Co-determinants elsewhere include earlier industrial experience (Oxford), government spending on infrastructure (Sophia-Antipolis),

61 The Cambridge high-tech cluster 45 large company effects (Siemens in Munich, Eriksson in Stockholm), metropolitan influences (London, Paris) and defence spending and procurement (Silicon Valley, Route 128). But in the Cambridge area there was minimal impact from exogenous investment influences such as have contributed to local spillover effects for technology-based firms elsewhere. Thus Cambridge provides unique evidence of the economic potential of a knowledge-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 of Cambridge). In this usage the term university start-up company is used to refer to businesses 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 equity stake. We use the term spin-out in the inclusive sense long used in the research literature and official spin-out company for those started with university IP or equity. 2. Total employment , based on the 1997 Employment census, adjusted by the Cambridge County Council Research Group. The number of firms and levels of employment reported here are derived from an adjusted dataset, refined from that used by the County Council which employs a less restrictive definition of high-tech and counts operating units rather than firms. Alternative definitions of high-tech activity are used by PACEC (2003), based on the wider Greater Cambridge Partnership area, and by the Cambridge investment consultancy, Library House, for their commercial database. For continuity and comparability we used SIC categories rather than non-standardised categories 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 Council Employment database, as described in Heffernan and Garnsey (2002). For the purposes of this chapter, the data have been extended to 2004, but the area covered has been restricted to the City of Cambridge and the adjacent South Cambridgeshire area, which have a concentration of firms originating in the university. 5. For example, while manual DNA sequencing originated in Cambridge, automated gene sequencing equipment essential to the Human Genome project was developed by Advanced Biosystems in California. 6. Cambridge Ring Newsletter. University of Cambridge Computer Science Department, Robin Saxby, Chairman of ARM, Diebold Conference, LSE, 28 April 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 than university ties to the Cambridge biotech cluster from a study of recent spin-outs, but do not 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 research and technical staff of the university were involved in applied research supported by outside funding (Garnsey and Cannon-Brookes, 1993). Cambridge University scientists took part in international networks, as a result of which they were the first to recognise the benefits to academic science departments of having companies sited locally to help commercialise scientific inventions, provide research collaboration and jobs for graduates. Social networks in the university and city encouraged innovation, with social interaction

62 46 Entrepreneurship across professional groups through college and community connections. Individual managers of banks local branches became part of an emerging business community and were able to provide overdrafts on a discretionary basis. Individuals among college administrators championed the Science Park and Innovation Centre (Garnsey and Longhi, 2004). REFERENCES Arthur, W.B. (1994), Increasing Returns and Path Dependence in the Economy, Ann Arbor, MI: University of Michigan Press. Auton, P. and Biddle, H. (2001), Successful spin-outs, by design, Arthur D. Little Prism, Issue 1. Best, M.H. and Forrant, R. (2000), Regional industrial modernization programmes: two cases from Massachusetts, European Planning Studies, 8(2): Boschma, R.A. and Frenken, K. (2003), Evolutionary economics and industry location, Review of Regional Research, 23: Breschi, S. and Lissoni, F. (2001), Knowledge spillovers and local innovation systems: a critical survey, Industrial and Corporate Change, 10(4): 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: the Cambridge biotechnology cluster European Planning Studies, 11(7): Cooper, A. (1971), Spin-offs and technical entrepreneurship, IEEE Transactions on Engineering Management, 18(1): 2 6. Cooper, A.C. and Dunkelberg, W.C. (1987), Entrepreneurial research: old questions, new answers and methodological issues, American Journal of Small Business, 11: Garnsey, E. and Cannon-Brooke, A. (1993), The Cambridge phenomenon revisited: aggregate change among Cambridge high technology firms since 1985, Entrepreneurship and Regional Development, 5: Garnsey, E. and Longhi, C. (2004), Complex processes and innovative places: the evolution of high tech Cambridge and Sophia-Antipolis, International Journal of Technology Management, 28(3 6): Heffernan, P. and Garnsey, E. (2002), Technology and knowledge based business in the Cambridge Area: a review of the evidence, Centre for Technology Management Working Paper 2002/01, University of Cambridge. Keeble, D. (1989), High-technology industry and regional development in Britain: the case of the Cambridge phenomenon, Environment and Planning C, 7(2), Keeble, D., Lawson, C., Moore, B. and Wilkinson, F. (1999), Collective learning processes, networking and institutional thickness in the Cambridge region, Regional Studies, 33(4): Kirchoff, B.A. (1994), Entrepreneurship and Dynamic Capitalism: The Economics of Business Firm Formation and Growth, Westport, CT: Praeger. Klepper, S. (2001), Employee startups in high-tech industries, Industrial and Corporate Change, 10(3): Klepper, S. (2002), The evolution of the U.S. automobile industry and Detroit as its 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 and Entrepreneurship from Silicon Valley and Europe s Silicon Fen, Chichester: Wiley.

63 The Cambridge high-tech cluster 47 Krugman, P. (1991), Increasing returns and economic geography, Journal of Political Economy, 99(3): Maruyama, M. (1963), The second cybernetics: deviation-amplifying, mutual causal processes, American Scientist, 51: PACEC (2003), The Cambridge Phenomenon Fulfilling the Potential, Report for the Greater Cambridge Partnership, Public and Corporate Economic Consultants, Cambridge. Porter, M. (1990), The competitive advantage of nations, Harvard Business Review, March April: Saxenian, A. (1994), Regional Advantage: Culture and Competition in Silicon Valley and Route 128, Cambridge, MA and London: Harvard University Press. Segal Quince Wicksteed Ltd. (1985), The Cambridge Phenomenon: The Growth of High Technology Industry in a University Town, 2nd edn, 1990, Cambridge, UK. Shah, S. and Garnsey, E. (1993), The acquisition of high tech firms: evidence from Cambridge, Judge Institute Research Papers in Management, University of Cambridge. Simmie, J., Siino, C., Zuliani, J., Jalabert, G. and Strambach, S. (2004), Local innovation system governance and performance: a comparative analysis of Oxfordshire, Stuttgart and Toulouse, International Journal of Technology Management, 28 (3 6): 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):

64 3. Sophia-Antipolis as a reverse science park: from exogenous to endogenous development Michel Quéré 1. INTRODUCTION The Sophia-Antipolis science park is often presented in the media as a European model of science park development. There are obvious reasons for that, especially because of the historical background of the experiment. The park started from scratch in the 1970s and reached an impressive stage of accumulation whereby more than jobs are now in existence on site. We shall argue that the park constituted a unique experiment due to the fact it has to be considered as a reverse science park as the university and research institutions joined the park only at a later stage. A related feature of the park holds that for a long time its development has drawn on exogenous sources. Only recently have some developments become truly endogenous, 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 a more specific issue, which is the capability of that project to transform itself into a real science park project, encouraging and benefiting from local entrepreneurial initiatives. Section 4 deals with a general discussion about that transformation, with a specific insistence on governance issues and more in particular with a discussion about how innovation opportunities progressively emerge locally. Section 5 focuses on innovative behaviours arising from local interactive learning in information and communication technology (ICT) activities, in accordance with either the type of firms involved in these processes or internal external interactions. Section 5 also deals in more detail with the policy implications of that transition towards a science park functioning in ICT activities. Concluding remarks about the sustainability of the Sophia-Antipolis science park and local policy making will be provided in the final section. 48

65 Sophia-Antipolis as a reverse science park SOPHIA-ANTIPOLIS: MAIN HISTORICAL CHARACTERISTICS The usual definition of a science park requires the existence of complementary resources in a similar location, namely academic resources including training and research capabilities, and high-tech start-ups that either result from or interact with the local research environment. These background conditions are usually complemented by any kind of governance structure favouring the connecting dimension among those components. What characterises Europe with regard to the governance and regulation of science parks is the wide variety of mechanisms and the diversity of governance structures aimed at ensuring the development of experiments (Gaffard and Quéré, 1996; Quéré, 1998). The governance structures vary considerably among European countries, from university or even private structures in Anglo-Saxon types of regulation to public-owned offices as well as private/public specific organisations in countries such as France, Italy and/or Germany. The Sophia-Antipolis science park is an original example for discussing types of governance mechanisms appropriate to encouraging innovation and business opportunities from local learning and interactions among various economic actors. Partly this results from the genesis of the project, which is quite recent (35 years ago), and partly from the different developmental stages faced by the project. Figure 3.1 depicts the quantitative accumulation from the mid-1970s and it demonstrates the unique character of the project Legal entities Employment Cumulative number of legal Years entities Cumulative employment Source: Sophia Antipolis Enterprise Management Sophia Antipolis Côte d Azur (SAEM SACA), Sophia-Antipolis. Figure 3.1 Cumulative number of organisations and employment

66 50 Entrepreneurship Initially, Sophia-Antipolis was planned to be a city of science, culture and wisdom. The genesis of the science park was a purely private initiative, instigated by Pierre Laffitte. Laffitte was a member of the board of the ENSMP (École Nationale Supérieure des Mines de Paris), one of the socalled French Grandes Écoles, and what he had in mind was to establish a sort of community of scientists in the South of France benefiting from the Sun Belt effect. He targeted one of the few landscapes still available locally, which was a forest near the village of Valbonne, seven kilometres north of the city of Antibes. The enterprise started in 1969 and the first buildings were erected in However, the project collapsed financially; the infrastructure costs were extremely high and the private organisation involved was unable to survive the mismatch between costs and benefits within the first years of implementation. The initial project could have come to a halt at this point, since no real accumulation process had taken place when the private initiative was withdrawn. However, this has not been the case due to a transformation process from a private to a public governance structure. The local public authorities (the Conseil Général des Alpes Maritimes: CGAM) have retained the project with the aim of complementing the local economy with other types of activity that were thought to be compatible with the dominant character of tourism. The local environment (the Côte d Azur) traditionally depends on tourism activities and it was a central concern for the public authorities to diversify without damaging that dominant economic infrastructure. High-tech activities were thought to be acceptable in the sense that their relative image and perception in the public would not endanger tourist flows. From 1977, the Sophia-Antipolis experiment shifted from a purely private to a public initiative in which the CGAM has played a critical role. The latter empowers the local Chamber of Commerce to manage the experiment. This change of governance goes hand in hand with a change in both the concept and the content of the project: from a City of Science, Culture and Wisdom to the concept of an International Industrial Park. Local public authorities favoured the location of research and development (R&D) units of international firms, that is, they aimed to attract external resources that would diversify the economy of the Côte d Azur. A deliberate choice not to accept the location of manufacturing activity was made and the selective character of attracting R&D units was due to the search for compatibility with tourism. From that time, a change in public support was also important. Not only did the public authorities take over from the initial private initiative in order to ensure the physical infrastructures, but they also developed an active advertising strategy in order to promote the location at the international level and in the United States in particular. As

67 Sophia-Antipolis as a reverse science park 51 a result, a change in the scale of the experiment occured and a progressive accumulation 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 the project. In that period, from 1977 to the end of the 1980s, the park exhibited a spectacular accumulation process of external activities (see Figure 3.1). This particular (second) stage of development can be characterised by the following structural patterns: an influential resilience from the private (individual) initiative, even if local authorities have moved away from the initial concept to a standard international industrial park; a spectacular process of accumulation of external resources, mainly R&D units from large international firms; a site attractiveness mainly due to public infrastructures and utilities used 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 are concerned and, as a consequence, a very low level of local connections and interactive learning on site; and an important dependence of local units from exogenous (internationally wide) firm decision making. These structural characteristics induced an obvious fragility in the sustainability of the experiment. However, this stage can also be thought of as a necessary step to ensure and secure a local mass effect and the emergence of local interactions among the components of the park. The almost random accumulation process resulted in the progressive specialisation into two major activity areas: ICTs and life sciences. ICTs include software, telecom and image engineering activities but also electric, electronic and micro-electronic activities, automation, and signal and system engineering. This gave rise to a number of R&D activities that are quite large and diversified, resulting in potential opportunities for interactions. Moreover, this potential has been enhanced by the location of public research institutes working in the same knowledge domains. Figure 3.2 shows the accumulation process in the ICT activities during the 1990s. It especially shows how ICT activities have faced a transition phase and escaped from a relative quantitative stagnation to increase again from the mid-1990s onwards. By 2000, ICTs accounted for about 75 per cent of high-tech local employment. They originated from either large domestic companies such as Air-France, Thalès and Télémécanique/Schneider or large foreign ones such

68 52 Entrepreneurship Legal entities Years Employment Cumulative number of legal entities Cumulative employment Figure 3.2 Cumulative number of organisations and employment (ICT) as Amadeus, Accenture and DEC/Compaq. Both types of firms located their R&D activity in the park, which complemented the accumulation of public research resources locally. If ENSMP was already among the first actors locating in Sophia-Antipolis, research units from CNRS (Centre National de la Recherche Scientifique: National Centre for Scientific Research), INRIA (Institut National de Recherche en Informatique et Automatique) and the University of Nice-Sophia-Antipolis finally formed a critical mass that favoured the existence of business opportunities through endogenous interactive learning. The situation in life science activities is different, because no similar mass effect has been reached. No such eclectic accumulation as in ICT activities has been obtained and there are no more than 50 local firms involved in those activities. During its development, the park benefited from R&D units from firms such as Dow Chemical, Dow Corning, Wellcome, Allergan, Rohm & Haas, Aventis and others. These units were acting essentially in activities such as fine chemistry, pharmaceuticals and dermatology. In that respect, business opportunities stemming from the diversity of capabilities that have been accumulated locally are not easy to develop. Most of these local units are internally oriented, as they often have an exclusive customer located at the international level. As such, they are not very interactive with other actors in the park, since they function mostly as an internal service provider. Some of the units are more research oriented, such as Cird/Galderma and Cordis/Zeneca, but their local relationships are still very limited. Thus, within-park opportunities through interactive learning are not well developed. Figure 3.3 reports the quantitative accumulation in those sectors and demonstrates the lack of a significant mass effect in life sciences.

69 Sophia-Antipolis as a reverse science park Legal entities Employment Cumulative number of Years legal entities Cumulative employment Figure 3.3 Cumulative number of organisations and employment (life sciences) In sum, by the turn of the 1990s, the first stage of the experiment had resulted in an obvious quantitative success as about 700 firms involving about 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 accumulation in such a short time period has been found elsewhere. A second stage of the experiment began at the turn of the 1990s. At that time, there was a cyclical crisis in ICT activities, which provoked an increased instability and volatility in local R&D units. This endangered the sustainability of the project, and from 1990 to 1995, no real improvement in employment accumulation was possible. The quantitative observation expressed a transition phase in the local accumulation regime. On the one hand, no serious candidate to locate in Sophia-Antipolis was available, probably due to increasing competitive pressure from other European countries (see Section 3); on the other, some of the R&D units that were at the source of the success left the park, under pressure from external (international) headquarters and their related decision making on how to organise R&D and business units in Europe. Curiously, this business-cycle crisis has benefited the park as it has induced a change in the accumulation regime, from an exogenously driven process to a more endogenous process of entrepreneurial initiatives. The local process of accumulation was again sensitive from 1995 but is now much more the result of an internally driven process than one of externally driven accumulation of resources. Since the mid-1980s, the Sophia-Antipolis park has become a European model of a science park. This is somewhat misleading. Sophia-Antipolis should be properly thought of as a reverse science park, as the local university was

70 54 Entrepreneurship one of the last institutions to join the process of local development. The location of PhD training from the university system has occurred only since the mid-1980s, but has been extremely powerful in favouring the previous transition from an exogenous-towards an endogenous-driven process of growth. Beyond the university system, the progressive accumulation of research institutes, especially in ICT sectors, has also favoured the transition towards a more traditional type of science park development. From the mid-1990s, the major structural patterns characterising the economic working of the park can be characterised by: a larger influential role of the university of Nice-Sophia-Antipolis in that transition; a modest but steady improvement in the accumulation of academic and public research resources, beyond the local university; a substitution of large R&D units by more endogenous entrepreneurial initiatives (especially in ICT activities); and a more interactive process of localised learning allowing for exploring innovative business opportunities on a large scale from the local environment (the emergence of a local entrepreneurial climate and culture). From this rough historical description of the Sophia-Antipolis park, one can provide some insights about the local process of growth as well as about the 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 is exhibiting quite a large economic bio-diversity, which makes the experiment unique and distinctive. Furthermore, due to exogenous forces, the park could reach a critical mass in ICT activities, which appears to be a necessary condition for innovative opportunities to arise. However, such a mass effect cannot in itself be a sufficient condition for endogenous development 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 discussed in the next section. 3. THE GOVERNANCE OF A SCIENCE PARK: LESSONS FROM THE SOPHIA-ANTIPOLIS EXPERIMENT The governance of the Sophia-Antipolis science park has become a central issue in order to guarantee the sustainability of the project in the future. To

71 Sophia-Antipolis as a reverse science park 55 understand the context of policy making, we first need to understand the structural factors that have influenced the evolution and governance of the park. Among these factors are the Côte d Azur Sun Belt effects, which are as various as the existence of an attractive international airport, local infrastructures facilities and the sunny climate. But these factors also encompass more dedicated efforts such as an active advertising policy developed by local policy makers when they decided to follow up the project. The main cyclical factor was the need from extra-european firms to locate some of their R&D facilities in Europe. Basically, the European market was perceived to be so complicated for legal and governmental reasons that firms needed to benefit from some local R&D facilities, adjusting their product portfolios to the decentralised and complicated local legislations that characterised Europe at that time. As such, Sophia-Antipolis was a good candidate to adjust correspondingly with intra-european customers. In that respect, the quality of physical infrastructures on the Côte d Azur that was inherited from tourism activities has been a positive factor that has made that location competitive relative to other potential choices within Europe. Second, the mass effect in ICT activities has also been enhanced by academics and public research organisations. Due to the influential role of Pierre Laffitte, some training and research activities have been located in that area from the starting phase. However, training and research infrastructure has not experienced a similar growth trend as private R&D units. Nevertheless, the infrastructure has progressively supported quite a significant volume of employment currently involving about 2000 individuals and 5000 students on site. Part of this improvement is due to national decision making, which deliberately decided to locate some research units from the French CNRS or from other public research institutes including the aforementioned ENSMP, INRIA and ADEME (Agence de l Environnement et de Maîtrise de l Energie). Part of this improvement is due to an increasing interest from the University of Nice-Sophia-Antipolis in the park from the mid-1980s onwards, when the university decided to locate PhD training related to ICT and life science activities in the park. Third, the mass effect resulting from an active international advertising strategy and the accumulation of academic resources has provoked a significant change in the local growth regime. Both factors changed the driving force of local growth from an exogenous-driven process to a more endogenous one. The period of that transition regime corresponds to the early 1990s to the mid-1990s. At that time, a crisis occurred locally, primarily due to a cyclical downturn, leading to a decreasing attractiveness of the Sophia-Antipolis park for R&D units of large international firms and even to the departure of some of them from the park. The reason for the structural shock was on the one hand an influential effect of the European

72 56 Entrepreneurship Union (EU) integration process and, on the other, an improvement in knowledge about the EU market for local R&D units. In short, there was a lack of relative competitiveness with the Sophia-Antipolis park, as far as traditional 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 were not convinced and decided to stay on the French Côte d Azur for personal reasons. As such, they tried to engage in individual business by developing entrepreneurial initiatives locally. As a consequence, the transition enhanced local interactions and favoured the development of a local labour market for high-skilled capabilities. This improvement in the labour market for highly qualified personnel is to be thought of as a consequence of the success of the park in its capability to shift towards a more endogenous growth development, that is, to behave as a real science park. This evolution has also been favoured by the progressive accumulation of academics, public research personnel and PhD students on site. A stock of around 2000 individuals acting in academic and public research institutes added to a volume of 5000 students is part of the transition as it has been extremely useful in supporting the previous entrepreneurial initiatives. Some institutions have taken a particularly active role in the process; for example, INRIA has been especially active in launching entrepreneurial initiatives from internal research projects. Moreover, other institutions such as ETSI (European Telecom Standard Institute) have had a specific influence in attracting European firms in telecom activities and in contributing in a local distinctive European capability in that sector. Both institutions have been particularly helpful in favouring the local sustainability of the transition towards a more endogenous process of accumulation. Fourth, the most spectacular dimension of the transition lies in the birth of a huge number of high-tech very small firms, the results of complicated combinations of individuals from R&D units, PhD students and public research fellows. These combinations emerge locally almost as a reactive strategy to either not leaving the Côte d Azur, or being unable to find a suitable position in the public research system. However, new business opportunities from local interactive learning have been quantitatively spectacular. Moreover, it is not simply a business-cycle effect of the dotcom revolution that occurred around 2000; it is a real process of transformation of the park that leads us to question its ability to behave as a real science park and to discuss the transition towards an actual local innovation system. As a consequence, the governance of the project is at stake as the initial advertising policy that ensured its success has also been questioned. What kind of policy initiative is needed to ensure the success of the transformation process is now becoming a real issue.

73 Sophia-Antipolis as a reverse science park 57 Fifth, and as a consequence of the previous remark, the transition results in a new effect: a much higher volatility of local firms. As far as ICT activities are concerned, it is possible to identify a group of about 300 firms locally. 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, especially because ICT firms are still very small in size and do not depend on a local market base. Thus, a local mismatch between R&D units from large international firms and small and very small high-tech firms is still a problem, as far as a discussion about the existence (or not) of a real local system of innovation is concerned. Therefore, it would be interesting to discuss further the ability of firms to participate in the transformation of the park, especially because the variety of strategies to deal with and accompany that transformation is high. The ability of the territory to behave as a science park where innovation can occur from interactive learning among the local components can be questioned from different viewpoints. In what follows, the implications of the change towards a more endogenous accumulation process will be considered 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 distinguish between three main categories of firms that are developing innovative behaviour locally: (i) R&D units of large international firms; (ii) small and medium-sized enterprises (SMEs); and (iii) spin-offs/starts-ups. These categories are quite different and, as such, call for different kinds of governance 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 match between supply and demand. The supply was the availability of square metres and/or offices in an attractive area surrounding the Côte d Azur; the demand was the need of large international firms to adjust their product portfolio to specific requirements from European countries. As we have already indicated, the location of R&D units in Sophia-Antipolis has progressively been challenged by other European locations. At the turn of the 1990s, some R&D units located in Sophia-Antipolis considered alternative

74 58 Entrepreneurship locations in Europe Ireland or Scotland for clear advantages in terms of high-qualified labour costs; southern Germany or northern Italy when they wanted to improve interactions with their main or representative European customers; Paris or London when their EU location was dealing more extensively with financial and/or administrative matters. As a consequence, the relative advantages of Sophia-Antipolis progressively diminished and those R&D units either relocated elsewhere in Europe or downsized their local units in terms of employment. Note, however, that these firms had little involvement in local learning with other actors in the park, be it other firms or public research institutions. R&D units were mainly dedicated to intra-firm activity and used as an exclusive provider of internal services for other units of the company. As such, their embeddedness in the local economic environment was weak. Moreover, subcontracting activities were mostly non-local, too. Thus, these R&D units were fully dependent on external decision making that was also internationally driven and unaffected by local policy making. Therefore, even though these R&D units of international firms ensured the quantitative success of the Sophia-Antipolis park, they were not fundamental to local interactive learning. Nevertheless, two caveats apply. First, the external threat to relocate stemming from internal firm hierarchy had a positive effect on local interactions. Proving the existence of complementary resources locally was perceived as the best means of avoiding a departure from the park. Therefore, those R&D units developed localised interactions and learning as a sort of reactive strategy in order to enhance the local attractiveness and to avoid moving from Sophia-Antipolis. Second, a significant number of individuals concerned with the relocation process decided not to leave the South of France but to engage in new local independent businesses. They refused the related mobility to other parts of Europe and developed entrepreneurial initiatives locally. Moreover, some of the local R&D units (such as Lucent on site, but also Alcatel, IBM and Texas Instruments in the immediate vicinity) encouraged the transition themselves and favoured the establishment of intra-entrepreneurial support (such as in-house incubators). The transition to a more endogenous type of development is thus partly the result of a reactive strategy of R&D units of large firms that were not fully convinced about the wisdom of leaving the site. Yet, this reactive strategy was not pervasive enough to prevent 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 technological SMEs. There are very few of them, as the Côte d Azur has never

75 Sophia-Antipolis as a reverse science park 59 Legal entities Figure to to to to to and more Employment Size distribution of ICT firms been an industrial area. Most of the firms that are emerging locally do not reach the critical size seen as necessary for real SMEs. Figure 3.4 illustrates this argument by considering the size patterns of the mass effect reached in ICTs. Most of the ICT firms located in Sophia-Antipolis have fewer than 50 employees. Obviously, this characteristic makes these SMEs quite vulnerable locally. There are basically no local markets as the area does not benefit from an industrial infrastructure. As a consequence, the SMEs have difficulty developing and growing locally. They essentially depend on a very small number of customers and that factor makes their development unlikely to be locally sustainable in the long term. As a consequence, improving their growth sustainability means either that they expand in geographical terms and have to develop towards a market base that is no longer local (mainly Bordeaux, Toulouse, Montpellier and Marseille), or that they merge or make contractual agreements with larger firms that can be perceived as a guarantee to secure their development (especially in financial terms and in related firm investment). Local Spin-offs and Start-ups From the discussion about the two previous firm types, it is obvious that many firms created locally are partly the result of the crisis when R&D units from large firms decided to create local spin-offs. The latter are the source for the existence of technological SMEs that exploit market niches from their Sophia-Antipolis location in diverse activities such as technological expertise and computer services (system design, software, network engineering and so on). A strategic positioning of these SMEs is part of the dotcom revolution. They are not fully part of the impressive economic development of e-business applications, but they provide the technological

76 60 Entrepreneurship logistics to develop innovation in those activities. As such, these SMEs are more technology oriented and appear critical to ensure the technological improvement and related logistics to the so-called dotcom revolution. Therefore, these firms have not suffered so much from the recent dotcom crisis and have succeeded in surviving and even improving their technological expertise and capabilities despite the cyclical downturn. The existence of these SMEs has also been favoured by a general improvement in science industry relationships within the French context. In a country where the academic environment has been extremely isolated from industry, a real change has occurred during the last decades. Each university and public research institute is now trying to engage and develop connections with firms. This national transformation has obviously benefited the entrepreneurial climate in Sophia-Antipolis. The academic and public research systems have become more receptive to the idea of engaging with private actors and developing professional relationships with the latter (contractual terms, patents and licence agreements, and so on). The improvement in the professional character of negotiations related to the whole range of science industry relationships has mutually benefited local interactive learning and the development of business initiatives stemming from complementarities between private and public expertise, essentially, in the field of ICT activities (Quéré and Ravix, 1997). Another general trend is the development of intermediation through venture capital initiatives and the availability of a diversified intermediation supply. This concerns not only venture capital as traditionally defined, but also other intermediaries. The latter can be purely private (such as business angels), purely public (such as academic incubators established by the Law on Innovation from 1999), or semi-public. These changes have triggered many more interactive relations within the park and, more generally, the emergence of a local entrepreneurial climate and endogenous business opportunities. 5. THE GOVERNANCE OF THE TRANSFORMATION TOWARDS A SCIENCE PARK MODEL: POLICY IMPLICATIONS In order to behave as a real science park, the Sophia-Antipolis experiment has to encourage a transformation from exogenous to endogenous development. The necessary condition for such a transformation is the existence of a significant critical mass effect. It is obvious that the latter has been reached only in ICT activities. This is why the following discussion is centred on ICT and the way interactions occur and take place in a business environment, which is obviously larger than the frontiers of the

77 Sophia-Antipolis as a reverse science park 61 Local interactions Hardware activities Operators Equipment providers Applications development Software activities Global interactions Figure 3.5 The geography of ICT activities Sophia-Antipolis park itself. Figure 3.5 makes the point in a more synthetic manner. It offers a more comprehensive framework on how innovative behaviours through local interactive learning take place in a larger marketplace. The figure represents the distribution of ICT activities in a type and space framework. Of course, this is just a rough approximation of a stylised fact representation of those ICT activities. But it helps the discussion as the matrix between the type of activity and the importance of local versus non-local relationships provides a synthetic view about the dominant characteristics of innovative firms behaviours through interactive learning in the Sophia-Antipolis environment. Most of the telecom providers (operators) located in the park are not extremely active locally. Even France Telecom, which could be thought of as having a localised advantage, cannot be considered as a central actor in local innovative behaviours. These firms are essentially interested in absorbing 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 to reflect on the future of markets and services as well as to benefit from attractive human resources. However, their active participation in localised innovation processes is weak. They benefit from an image effect and they can also offer higher earnings to specific human resources that are revealed as useful to them from the working of the park. But this is a sort of predating strategy (both conscious and unconscious) that does not fully benefit the overall entrepreneurial climate.

78 62 Entrepreneurship Telecom equipment providers, however, have a pivotal role in the economic working of the park. On the one hand, they are participating in an industrial context which implies that they will be identified and act as global players. On the other, they are apparently more active in terms of local interactive learning than operators. In that respect, telecom equipment providers consider Sophia-Antipolis to be a technological platform where they can test technological opportunities thanks to local technological start-ups. Moreover, the location of specific institutes like ETSI and public research institutions is essential to them in order to secure innovative investment. In that respect, local collaborations are a means of exploring and defining new market opportunities. The latter include networking and connectivity interfaces, technical compatibility between different equipment and standards and so on. Such an experimental process is basically the result of the local accumulation of external resources that creates a diversified productive environment where diverse and various capabilities are possibly complementing each other. Some equipment suppliers have even developed in-house incubators in order to benefit from that variety of capabilities existing in the park. This gives to the park, at least for specific types of activities, a status of European experimental project enhancing the reputation of Sophia-Antipolis as one of the leading locations in Europe as far as particular telecom activities are concerned. Application and service providers represent the category of firms for which interactive learning is the most developed and possibly most beneficial. It is possible to distinguish two types of firms. First, SMEs that succeeded in taking an active part in the e-economy, that is, those that benefited from the dotcom revolution and, as such, enhance the Sophia- Antipolis image effect. However, success stories have been relatively few and probably did not contribute significantly to the local economy. In fact, many of these firms disappeared from the local environment when acquired by large external firms. As such, local entrepreneurship then benefits the rest of the world economy instead of benefiting the local economy. This is largely the case for firms such as Respublica, Lybertysurf, Echointeractive or Odisei. High-tech SMEs can also be obliged to connect more densely with their main customers and suppliers, which leads local entrepreneurial initiatives to relocate in other locations in order to secure their market and application bases. The second category of firms is science and technology-based SMEs that develop technological support and services on the application side of the so-called new economy. Here, the interesting aspect is that these firms are the result of local interactive learning among different components, including public research institutes, PhD student projects, spin-offs and so on. They are the most important result of the transition phase in the

79 Sophia-Antipolis as a reverse science park 63 accumulation process from an exogenous towards an endogenous type of development. A byproduct of these ventures is the higher local sustainability as they develop distinctive capabilities that would not be possible to reproduce easily from another location. Thus, the localised character of their innovativeness is the best guarantee for the sustainability of the Sophia-Antipolis park. Any departure from Sophia-Antipolis should result in high transition costs and in a risk for firms interested in benefiting from SME expertise. As a consequence, the second category of firms is the one that best expresses the transformation of Sophia-Antipolis towards an actual science park. Finally, it is important to stress that the number of such firms (about 20) is relatively small in compared with the whole set of resources 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 on interactive learning and resulting in new high-tech start-ups is quite important. It stressed that Sophia-Antipolis can be considered as a real science park but only for a subset of ICT activities for which all the necessary components for a science park have reached a sufficient mass effect to favour such endogenous initiatives. This observation brings us back to the park governance issue and to discuss in more detail what kind of policy making is appropriate to encourage and accelerate the transformation of the science park. What is particularly clear from the development of the park is that policy making has almost exclusively been concerned with the provision of physical infrastructures. This is of course a peculiar characteristic due to the historical background of the project (which started from scratch in the 1970s). The proactive character of policy making has basically been to advertise the project properly at the international level. However, such a policy has quite exclusively been based on promoting the local physical infrastructures. As a consequence, there is no real policy that has been concerned with the internal understanding of localised and interactive learning within the different components of the science park. Endogenous development has mainly been thought of as a natural process that should derive automatically from the accumulation mass effect. Obviously, it is not and the previous transformation depicted from ICT activities questions seriously the ability of policy making to encourage the economic sustainability of that experiment. Moreover, ensuring the accumulation of resources stemming from endogenous development is a difficult issue. On the one hand, it still has to face the intrinsic unstable character of international players (telecom operators and some equipment providers as well) for which external decision making is still the dominant regulation for their location in Sophia-Antipolis. As such,

80 64 Entrepreneurship these firms have a low level of sensitivity to any kind of local public decision making. On the other hand, endogenous SMEs are difficult to secure in the park as the economic climate is traditionally not a marketplace for these high-tech firms. With tourism still dominating the local economy, policy is still relatively unconcerned with high-tech activities. With regard to local policy making, three main issues can be addressed in relation to the sustainability of the economic transformation of Sophia- Antipolis and its transition towards a real science park. First, the role of physical infrastructures is not to be denied. If the park is now a leading place in Europe for some ICT activities, it is basically to be found in the choice of local policy makers to invest in a high-broadband telecom network based on a fibreoptic technology within the park by the 1980s. This choice has had a considerable impact on the site attractiveness from that time and has been crucial for ICT firms located in Europe to test for high volumes of information exchange and the related development of new applications. In other words, the quality of local infrastructures is still a necessary, if not a sufficient, condition for the economic sustainability of the park. However, what type of infrastructures are appropriate for the future of the park is probably something that should currently be elaborated within the framework of the park and in deeper connection with private actors. This is apparently not the case as the connections between private and public decision making are weak. The connectivity between private and public decision making within the framework of the park to define the physical infrastructures appropriate to its future development is a challenging issue for the sustainability of the park. Second, within the last decade, different forms of collective coordination among local private and public actors have emerged. Specific associations and thematic clubs have developed, most of them in a spontaneous way in order to favour local coordination and collective innovative projects on a local base. Some of them are in fact the result of a reactive strategy to deal with the risk of departure from the park. It was a means of bargaining more effectively with external decision making to justify the need to stay on site. This mechanism has been complemented by another trend, which is the emergence of a local market for some local SMEs within the park. Technology and business consulting firms, intellectual property rights and finance-oriented firms are parts of a logistical infrastructure to high-tech SMEs. Such consulting infrastructure is progressively transforming the Sophia-Antipolis park into a real marketplace. Both effects are part of a transformation process from an exogenous to an endogenous mode of development. However, both mechanisms are also disconnected from public policy concern. There is, however, a need for a more effective way to deal with the interplay between private and public characteristics of that

81 Sophia-Antipolis as a reverse science park 65 transformation. Policy makers can no longer consider interactive innovative learning as the result of a spontaneous order. They have to think about proper investment favouring the coordination of all the actors that are part of that shift towards a more endogenous type of development. Third, for a long time the Sophia-Antipolis park was perceived as an isolated island in the Côte d Azur environment. This is less and less true as the connectivity of firms located in the park to the environment is increasing. The improvement of localised innovative learning in ICT activities also implies complementing R&D with non-r&d activities (prototyping, designing and testing applications and so on). The latter require other capabilities and relationships with various sorts of markets. As far as local capabilities are available, this process can be expanded locally through all sorts of connections to other firms in the surrounding environment. To mention one example, one challenge is to connect the dominant local economic activity (tourism) to the park in a more developed perspective. Tourism is an interesting application area where ICT activities can be applied in various ways to improve innovation in that sector. The distinctive capabilities accumulated in the park in areas such as wireless and Global System for Mobile Communications (GSM) technologies can certainly be combined in original ways for tourism applications. However, this again requires from policy making a forecasting ability to deal properly with such challenges. Altogether, the critical issues show the difficulty of developing policies supporting the transition of the science park from an exogenously driven to an endogenously evolving economy. To most local policy makers, however, the dominant perception is still that the endogenous character of local development should be a natural result of the accumulation process. However, this is obviously not the case as endogenous development can be hampered by local constraints and global trends (Garnsey and Heffernan, this volume). 6. CONCLUDING REMARKS The Sophia-Antipolis experiment is unique. The main factor explaining the success of the project lies in the general advantages exhibited by the French Riviera that were well suited to large multinational companies for locating in Sophia-Antipolis either European administrative centres or R&D units. At that first stage of its development, Sophia-Antipolis was exogenously driven and the experiment exhibited a significant comparative advantage regarding other potential locations in Europe. However, the economic working of Sophia-Antipolis was vulnerable, because innovative activities

82 66 Entrepreneurship were not based on local relations. Nevertheless, the park has benefited from a large accumulation of external (international) resources that has been a necessary condition to ensure its quantitative success and to create the required conditions to shift to real science park dynamics. The latter occurred 20 years after the starting phase of the experiment and the situation seems to be improving in recent years. However, the transformation is still modest in quantitative terms in comparison with the overall accumulated activities in the location and does not really benefit from an active and appropriate support from local policy making. The transition, essentially limited to ICT activities, is to be thought of as a sort of spontaneous change, mostly due to incentives from individuals not to leave the enjoyable Côte d Azur environment, which is essentially the result of behaviours from private actors. As a consequence, the transition is still very fragile and should be central for policy making (Quéré, 1999). The latter should concentrate on understanding the transition and defining more precise intervention to encourage its sustainability. This is still thought to be necessary in 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 conditions 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 uncertain creation of an innovative milieu, in R. Ratti, A. Bramanti and R. Gordon (eds), The Dynamics of Innovative Regions, Aldershot: Ashgate, pp Quéré, M. (1997), Sophia-Antipolis as a local system of innovation, Economia & Lavoro, 3 4: 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: a discussion about innovation systems, in O. Fabel, F. Farina and L.F. Punzo (eds), European Economies in Transition, London: Macmillan, pp Quéré, M. and Ravix, J.L. (1997), Production de connaissance et institutions innovatrices: le chercheur-entrepreneur, Revue d Économie Industrielle, 79:

83 PART II Industrial Dynamics

84

85 4. The evolution of geographic structure in new industries Steven Klepper* 1. INTRODUCTION In recent years there has been a resurgence of interest among economists in questions related to geography. In part spurred by the clustering of economic activity in Silicon Valley, attention has focused on why certain industries 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 what would be expected merely on the basis of the chance location of a limited number of plants of unequal size. Their findings suggest that some degree of agglomeration is the norm, but the kind of extreme clustering present in Silicon Valley is the exception. While clustering in some industries can be explained by the uneven geographic distribution of a key input, instances of 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 has been the object of much theorizing in recent years. Testing of the new theories of geography, though, has lagged behind. The object of this chapter is to review recent evidence and theorizing on the evolution of a select group of new industries to probe the determinants of the geographic structure of industries. Modern theories of geography feature the influence of agglomeration economies on the location of producers. Such economies can derive from the sharing of inputs whose production involves increasing returns, labor market pooling that facilitates a better match between the needs of firms and the skills of workers, and spillovers of knowledge that are mediated by distance (see Marshall, 1920). Other mechanisms, such as firms locating closer to demanders to economize on transportation costs, can also induce agglomeration (Krugman, 1991). All of these benefits impart a selfreinforcing character to agglomerations. The more firms in an area then the greater agglomeration benefits, and the greater such benefits then the more firms will be drawn to an area and the better firms in the area will perform. 69

86 70 Industrial dynamics Congestion costs, in the form of higher land prices and compensating wage differentials, ultimately limit the extent of agglomerations. Until that limit is reached, though, all firms located in agglomerated areas benefit from the externalities resulting from their collective presence. If agglomeration economies are influential, it might be expected that industries would agglomerate around regions where successful early entrants located. Such regions would initially produce more output, employ more labor, and be subject to more innovation, all of which would contribute to agglomeration economies that would attract subsequent entrants and enhance the performance of firms located there. There has been little empirical investigation of the evolution of the geographic structure of new industries, but Klepper (2003) argues that this is not the way either the automobile or television receiver industries evolved. Both industries were initially characterized by a large number of producers and then experienced sharp shakeouts and evolved to be tight oligopolies. The automobile industry became famously agglomerated around Detroit, MI even though production in the Detroit area was initially negligible and early entry provided a decided competitive advantage (Klepper, 2001). Television producers were initially heavily concentrated in just three cities: New York, Chicago and Los Angeles. Even though early entry was also advantageous in TVs (Klepper, 2002a), over time New York and Los Angeles lost all their producers and Chicago did not grow, causing the industry to become more dispersed over time. Klepper (2003) advanced a hypothesis based on the ideas of organizational birth and heredity to explain the evolution of the geographic structure of both industries. Subsequently Buenstorf and Klepper (2005a, 2005b) explored the evolution of the geographic structure of the pneumatic tire industry, which was also famously agglomerated around a single city, Akron, Ohio. Similar to autos and television receivers, initially many firms produced tires and then the industry experienced a sharp shakeout and evolved to be a tight oligopoly. Unlike automobiles, the industry was concentrated around Akron from its outset, and over time the agglomeration of the industry there grew. Buenstorf and Klepper (2005a, 2005b) investigated the extent to which agglomeration economies influenced the location and performance of tire firms. They concluded that it was not primarily agglomeration economies but similar forces to those operating in the automobile industry that caused the industry to become so heavily agglomerated. We review the evolution of the market and geographic structure of the automobile, television receiver and tire industries in order to gain insights into the primary forces governing the agglomeration of industries. 1 We begin with television receivers, which is the easiest to understand and provides a useful backdrop for the automobile and tire industries. Next we

87 The evolution of geographic structure in new industries 71 consider the evolution of the automobile industry, followed by the evolution of the tire industry. We conclude with observations about the importance of organizational heredity and birth in shaping industry agglomeration. 2. TELEVISION RECEIVERS The annual number of entrants, exits, and producers of television receivers in the United States over the period based on listings in Television Factbook (Warren Publishing Company) is presented in Figure 4.1. A total of 177 firms entered the industry, most of them by Experimental television systems were introduced prior to the Second World War, but the war delayed the start of the industry until RCA and DuMont were the first firms to begin producing in Many firms followed soon after, reflecting the rapid growth in the sales of television receivers. Entry peaked in 1948 at 54 firms, and by 1955 entry was negligible. The number of firms rose from 1946 to 1949, reaching a peak of 105 in 1949, and then fell sharply. International competition, initially from Japan, began in the late 1960s when roughly 30 US-based producers were left in the industry. At that point RCA and Zenith were the top two US producers of television receivers, accounting for 39 per cent of US sales of black and white TVs and 48 per cent of the sales of color TVs, and the four-firm concentration Producers Entrants Exits Source: See Klepper (2003). Figure 4.1 Entry, exit and number of producers in the television industry,

88 72 Industrial dynamics 120% 100% 80% 60% % NY producers % Chicago producers % LA producers 40% 20% 0% Source: See Klepper (2003). Figure 4.2 Percentage of television producers in New York, Chicago and Los Angeles, ratios in black and white and color TVs were 61 and 65.5 per cent, respectively. International competition mounted over time and the number of US-based firms fell steadily. By 1989 only three US producers were left in the industry, all of which were destined to exit within a short period. Klepper (2003) analysed the location of the TV producers, which was heavily concentrated in three US cities: New York, Chicago and Los Angeles. Although these three cities accounted for only 15 per cent of the US population, 73 per cent of television producers entered in the three cities, with 44 per cent entering in New York, 15 per cent in Chicago and 14 per cent in Los Angeles. Figure 4.2 presents the annual percentage of television producers based in each of these three cities from 1946 to New York initially contained over 50 per cent of the television producers, but over time this percentage declined sharply. By 1970 New York s share had declined to 20 per cent, and by the end of the 1970s no firm was based in New York. Firms were slower to enter in Los Angeles, but by the mid- 1950s, 20 per cent of the producers were located there. Subsequently the share of producers in Los Angeles fell sharply, and by the mid-1970s no firm was based in Los Angeles either. Chicago accounted for around 25 per cent of television producers in the initial years of the industry. It maintained its share through about 1980, after which its share increased sharply as the number of firms dwindled from eight to three. Thus, at the start of the industry, television producers were heavily concentrated in three cities, but from the mid-1950s until 1980 the collective share of producers in the three cities declined from 70 to 25 per cent.

89 The evolution of geographic structure in new industries 73 The evolution of the location of the television producers was greatly influenced by the location of firms in the radio industry. Entry, exit and the location of radio producers was reconstructed from annual listings of radio producers in Thomas Register of American Manufacturers (Thomas Publishing Company; see also Klepper and Simons, 2000; Klepper, 2003). At the start of the television industry, 266 US firms produced radios. They were heavily concentrated in New York, Chicago and Los Angeles, which accounted for 33, 15 and 7 per cent, respectively, of all radio producers in Of the 177 television entrants, 58 or approximately one-third diversified from the radio industry, and nearly all began producing television receivers where they produced radios. Thus, it is not surprising that among the 58 diversifiers, 55 per cent of them located in New York, Chicago and Los Angeles, which mirrors the fraction of radio producers in the three cities. Perhaps more surprising is that among the remaining 119 entrants, 82 per cent also located in these same three cities, especially in New York and Los Angeles, which accounted for 53 and 18 per cent, respectively, of these entrants (Klepper, 2003). Klepper and Simons (2000) demonstrated that the radio producers that diversified into the television industry tended to be the largest and most experienced ones, and they tended to enter earlier than other entrants into the TV industry. They also found that the diversifiers from the radio industry had much lower hazards of exit at all ages than non-radio diversifiers, and among the radio diversifiers, the larger ones had much lower hazards at all ages. Indeed, 13 of the top 14 television producers over the history of the industry were diversifiers from the radio industry, and four of the top five television producers were among the top five radio producers as of 1940 (the other radio producer among the top five in 1940 was among the top 10 TV producers). The location of the leading radio producers was the dominant force shaping the location of TV producers in the long run and the evolution of the geographic structure of the industry. Only one of the top radio producers was located in New York, and it accounted for only 11 per cent of the sales of the top radio producers as of 1940, and Los Angeles had no leading radio producer as of With the leading radio producers ultimately dominating the television industry, New York and Los Angeles were destined to experience a sharp decline in their share of television producers as the industry proceeded through its shakeout. Chicago had five of the top 16 radio producers that jointly accounted for 38 per cent of the sales of the leading radio producers as of Correspondingly, Chicago had three of the top 10 television producers and maintained its share of television producers over time. The other leading radio producers were scattered throughout the Northeast and Midwest. Many of these firms,

90 74 Industrial dynamics including RCA, Philco and GE, became leading television producers. They survived much longer than other firms, and as a result the base location of television producers became increasingly dispersed throughout the Northeast and Midwest as the industry evolved. In a statistical analysis of firm hazard rates, New York and Los Angeles firms had higher hazards of exit than firms located elsewhere, but once the background and time of entry of firms was controlled, there were no significant differences in the hazard rates of firms by region (Klepper, 2003). International competition further contributed to a geographic dispersal of television production. While US firms maintained their base locations, they increasingly moved their operations into low-wage countries in order to counter foreign competition (Levy, 1981, pp ). But this did not head off their demise. They were behind the technological frontier, especially regarding the use of semiconductor technology. They lost much of their market share to Japanese firms that had pioneered the use of semiconductor components in radios and that were consistently ahead of the US firms in the use of semiconductor components in television receivers (La France, 1985). Television receivers illustrate a few themes that are pertinent to automobiles and tires. First, firms in related industries are important seeds for firms in a new industry, and their location is an important determinant of where entrants into the new industry locate. Second, the pre-entry experience 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 the industry over the long term. Consequently, over the long run the base location of the leading radio firms was the dominant influence on the location of television producers. Third, agglomeration economies were not a major factor shaping the base location of television producers. Two of the three regions where firms were concentrated declined over time, and regional differences in firm performance were largely due to differences in their preentry experience rather than any influence of the regions themselves. Finally, as the number of US firms declined, the leading firms increasingly moved their production into lower-wage areas, further dispersing production. 3. AUTOMOBILES The annual number of US entrants, exits and producers of automobiles from the start of the industry in 1895 through 1966 based on a list compiled by Smith (1968) is presented in Figure 4.3. Only firms that produced a nonnegligible number of automobiles are included, which encompassed 725 firms through 1966 (see Klepper, 2002a). In contrast to televisions, entry

91 The evolution of geographic structure in new industries Producers Entrants Exits Source: See Klepper (2003). Figure 4.3 Entry, exit and number of producers in the automobile industry, was initially low, reflecting the limited demand for automobiles when they were introduced. Subsequently entry grew, peaking at 87 firms in Entry remained high for the next three years and then declined to an average of 15 firms per year for the next 12 years, after which it became negligible. The number of firms peaked at 272 in 1909, after which it fell steadily despite average annual output growth of over 20 per cent during the next 15 years. By 1941 only nine firms were left in the industry. As of 1911, the top two firms in the industry, Ford and General Motors, accounted for 38 per cent of the output of automobiles. They increased their 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 jointly accounted for over 80 per cent of the output of the industry by the 1930s. The industry became famously agglomerated around Detroit, MI, but initially no firm produced a non-negligible number of automobiles in the Detroit area. Figure 4.4 reports the annual percentage of producers based in the Detroit area from the start of the industry through 1941, when only nine firms were left in the industry. 2 No producer was located in the Detroit area until 1901, when Olds Motor Works began production in Detroit and Lansing, MI. Olds was the first great firm in the industry. After Olds entry the percentage of automobile producers in the Detroit area steadily rose into the 1910s, when it peaked at over 20 per cent. It then fell back a little but rose again after 1920, exceeding 50 per cent by The firms based

92 76 Industrial dynamics % Detroit producers Source: See Klepper (2003). Figure 4.4 Percentage of automobile producers in the Detroit area, in the Detroit area were extraordinarily successful. By the mid-1910s they produced 13 of the 15 leading makes of automobiles, and over 60 per cent of automobiles were produced in Michigan, nearly all in the Detroit area. Although 69 producers entered the industry in 1895 to 1900 before any producer entered in the Detroit area, Detroit nonetheless became the capital of the US automobile industry by the mid-1910s, and it maintained its hegemony 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 only 18.6 per cent of the 725 entrants through 1966, followed by New York with 13.5 per cent and Ohio with 12.3 per cent. Thus, the top three states accounted collectively for 44.7 per cent of the entrants whereas the top three cities in televisions accounted for 73 per cent of the entrants. This largely reflects that the leading seeding industries for automobiles were considerably more dispersed geographically than the radio industry. Klepper (2001) identified firms that diversified into autos or were founded by an individual who headed a firm in another industry based on the listings in Smith (1968) and the brief histories of automobile firms in Kimes (1996). The industry from which the greatest number of these two types of entrants came was carriages & wagons. In a statistical analysis of the location of automobile entrants, Klepper (2003) found that states with more carriage & wagon production not only had more entrants originating from the

93 The evolution of geographic structure in new industries 77 carriage & wagon industry but also more of other types of entrants as well. Unlike radio producers, which were concentrated in three cities, the carriage & wagon industry was dispersed throughout the Northeast and Midwest, with the top three states accounting for only 32 per cent of all carriage & wagon producers. Moreover, among US states Michigan was ninth in terms of carriage & wagon producers and fourth in terms of value of carriage & wagon production. The second most important seeding industry for automobiles was bicycles, which was also dispersed throughout the Northeast and Midwest, and few bicycle firms were located in Michigan. Consequently, the geographic dispersion of entrants and the slow start of the industry around Detroit were predictable. Similar to televisions, automobile entrants that diversified from other industries, particularly carriages & wagons, bicycles and engines, had lower hazards of exit at all ages, as did new firms founded by individuals who headed firms in these industries (Klepper, 2001). But diversifiers were far less important in automobiles than televisions. Whereas 33 per cent of the entrants into the television industry diversified from the radio industry, only 16.6 per cent of entrants into the automobile industry were diversifiers from any industry (Klepper, 2003). In large part this reflects the novel technological challenges faced by automobile firms. Automobiles soon required precision manufacturing to produce interchangeable parts, manufacturing was done on an unprecedented scale, and technological progress was far more rapid than had occurred in carriages & wagons and other related industries. Consequently, experience in related industries was much less valuable in automobiles than in televisions. This opened up opportunities for new firms, especially firms with one or more founders that previously worked for an incumbent automobile firm, which are called spin-offs. Klepper (2001) identified the spin-off entrants and the firms their founders previously worked for, dubbed their parents, based on the brief firm histories in Kimes (1996). Approximately 20 per cent of entrants into automobiles were spin-offs, most of which were founded by top managers or heads of incumbent firms. At their peak in 1916, spin-offs accounted for 11 of the 15 leading makes of automobiles. Nearly all of these spin-offs descended from the leading automobile producers in the sense that their founders had worked for one of these firms (Klepper, 2001, 2005). Statistical analyses indicated that the annual likelihood of a firm having employees leave to start spin-offs was greater for better-performing firms, and on average better-performing firms had better-performing spin-offs (Klepper, 2001). One explanation for these patterns is that leading incumbent firms provided a superior venue for employees to learn about organizational best practices, especially top employees. Top firms were also magnets for talented individuals, which is another

94 78 Industrial dynamics possible reason why their spin-offs performed so well. Spin-offs formed for various reasons. Among the top firms, many of them arose from internal disputes about strategy and technology, reflecting control struggles that were common in the early years of the industry (Klepper, 2005). In large part because of the influence of Olds Motor Works, spin-offs played a key role in the concentration of the industry around Detroit. Olds Motor Works had been a successful producer of steam and gasoline engines before it entered the automobile industry. Its manufacturing and marketing experience enabled it to become the first firm to sell over 1000 automobiles in a year, selling more than 5000 by Olds subcontracted all of its parts, which involved orders of unprecedented size, to various local firms, providing its subcontractees with invaluable experience. Two of these firms were instrumental in the formation and success of Cadillac and Ford Motor Co., both of which were located in Detroit Another one of Olds subcontractors initially financed Buick, which was located in Flint, MI near Detroit. Buick was the cornerstone of the later merger that formed General Motors. This same contractor also co-founded another successful firm, Maxwell-Briscoe, which later evolved into Chrysler. Olds Motor Works, Cadillac, Ford Motor Co. and Buick/General Motors were among the most successful early automobile producers, and they collectively unleashed a spin-off juggernaut that propelled Detroit to become the automobile capital of the United States. They were the most prolific parents in the industry, reflecting the greater rate of spin-offs among the better firms. Olds had more descendants than any other firm in the industry, and in total 41 firms descended from Olds, Cadillac, Ford and Buick/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 11 of the 13 spin-offs that produced leading makes of automobiles after 1903, with each firm spawning at least two of these spin-offs. Consequently, by the mid-1910s nearly all the leading makes of automobiles were made by firms based in the Detroit area. With the leading makes accounting for over 80 per cent of the output of the industry, Detroit firms totally dominated the industry. Indeed, what distinguished Detroit was primarily its spin-offs, which accounted for 48 per cent of the entrants in Detroit versus only 15 per cent of the entrants elsewhere. Moreover, spin-offs in Detroit greatly outperformed spin-offs elsewhere, whereas the rest of the entrants in Detroit performed comparably to their counterparts elsewhere. In a statistical analysis, the superior performance of firms in Detroit was confined to its spin-offs, and their superior performance in turn was largely attributable to their superior heritage rather than being located in Detroit (Klepper, 2001).

95 The evolution of geographic structure in new industries 79 The leading firms remained based in Detroit, but over time they conducted more of their business outside of Detroit as they established branch assembly plants throughout the United States. It was much cheaper to ship parts rather than a finished car. Consequently, if a firm had sufficient output to accommodate multiple plants of minimum efficient size then it made sense to build branch assembly plants closer to the market. Ford, the largest producer in the 1910s, was the first to build branch assembly plants in the 1910s. It was followed by General Motors in the 1920s and later Chrysler and 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 automobiles as 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 for the new industry. Also like TVs, there was enormous heterogeneity in entrants in terms of their pre-entry experience that persistently affected their performance. The key difference between autos and TVs was that spin-offs were competitive with, if not superior to, diversifiers. This reflected both the limited relevance of prior industries to autos and possibly the distinctive opportunities within firms, particularly the leading firms, for high-level employees to learn valuable tacit organizational knowledge that they could apply to their own firms. With better firms having higher spin-off rates and better-performing spin-offs, the spin-off process effectively led to a build-up of firms and activity around the leading firms in the industry. This was especially potent in autos because of the location of four of the most successful early firms in one narrow region, fueling a great agglomeration of activity there. The four firms were connected through Olds, which was the catalyst for the agglomeration of the industry around Detroit. But the other three were important parents of spin-offs, and their creation near the industry leader added another random element to the agglomeration process that could help explain why agglomerations as extreme as autos are rare. Indeed, while Detroit was part of the manufacturing belt dating back to the 1860s, it was hardly the most likely place for the automobile industry to agglomerate. Its development was largely attributable to the influence of Olds Motor Works and the inherent randomness in the location of any one firm. Agglomeration economies from locating near other producers do not appear to have been a major factor in causing the industry to agglomerate around Detroit. Indeed, the establishment of branch assembly plants throughout the US by the leading firms is indicative of the potential disadvantages of the leading firms clustering in one area. It is notable that it took the industry leaders to exploit the advantages of assembling cars

96 80 Industrial dynamics outside of Detroit. This is indicative of the difficulty of imitating the leaders of the industry from afar, no doubt in part due to the tacit knowledge the leaders possessed. It is also suggestive of why the leading firms had more and better spin-offs their high-level employees had access to valuable tacit knowledge about how to structure their own firms. The evolution of the market structure of the automobile industry may have influenced the evolution of its geographic structure, though not directly. The TV industry also experienced a shakeout and evolved to be an oligopoly, yet its geographic structure evolved in an opposite direction to autos. Moreover, the hegemony of Detroit was established before the shakeout in the automobile industry began. It is possible, though, that the eventual drying up of entry that characterized the industry after 1925 eliminated a force that could have unseated the leaders and conceivably reduced the concentration of the industry around Detroit. 4. TIRES The annual number of entrants, exits and producers of tires in the United States over the period based on listings in Thomas Register of American Manufacturers is presented in Figure 4.5. With the initial demand for automobiles limited, the demand for tires was initially modest and entry started out low. Subsequently it grew for many years, peaking in the early 1920s before falling off sharply and becoming negligible by A total of 533 firms entered the industry through 1930, after which no significant firm entered. The number of firms peaked in 1922 at 278 and then went through a long shakeout despite robust output growth interrupted only by the Great Depression. Only 51 firms were left in the industry in 1940, and by 1970 only 24 firms were still in the industry. The industry evolved to be a tight oligopoly dominated by Goodyear, Goodrich, Firestone and U.S. Rubber (Uniroyal). Together these four firms accounted for over 53 per cent of the output of the industry in 1926, which they increased to over 70 per cent by 1933 and then subsequently maintained (Klepper, 2002a). Similar to automobiles, the tire industry became heavily concentrated around a single city, Akron, OH, located in the northeastern part of Ohio near Cleveland. Figure 4.6 reports the annual percentage of 1930 and earlier entrants that were located in Ohio from 1906 to For the first 25 years or so Ohio generally accounted for between 20 and 30 per cent of all producers, but after 1930 the percentage of firms in Ohio rose steadily and by 1959 it exceeded 50 per cent. Firms in Ohio, especially around Akron, were distinctly successful, and by 1935 over 65 per cent of the output of tires was produced in Ohio (Buenstorf and Klepper, 2005b). Much of this output was

97 The evolution of geographic structure in new industries Producers Entrants Exits Source: See Buenstorf and Klepper (2005b). Figure 4.5 Entry, exit and number of producers in the tire industry, % Ohio producers Source: See Klepper and Buenstorf (2005b). Figure 4.6 Percentage of tire producers in Ohio, produced by Goodyear, Goodrich and Firestone, all of which were based in Akron. But firms in northeastern Ohio also dominated the next cadre of firms. As of 1920, six of the next 20 largest firms were located in Akron, and four others were located nearby in northeastern Ohio (ibid.).

98 82 Industrial dynamics Like automobiles, Ohio had more tire entrants than any other state, but it accounted for only 24 per cent of all the entrants through 1930, followed by New York with 15 per cent, New Jersey with 14 per cent, Pennsylvania with 8 per cent and Illinois with 7 per cent. Klepper (2002) and Buenstorf and Klepper (2005b) identified the entrants into tires that diversified from another industry, which in most cases was the rubber industry. Similar to TVs and autos, in a statistical analysis Buenstorf and Klepper (2005a) found that states with more rubber producers had more tire entrants that were diversifiers and also more of other types of entrants, and within Ohio counties with more rubber producers had more diversifying entrants. Similar to Michigan and autos, Ohio was not the leading state in terms of rubber producers, but was fifth in 1890 with 3.5 per cent of US rubber producers. While diversifiers had lower hazards of exit on average than other types of entrants, diversifiers accounted for only 15.6 per cent of all entrants, similar to automobiles. In part, this reflects that automobile tires represented a considerable break from prior rubber products. Bicycle tires did not readily scale to automobiles, tire manufacturing was much more complex than other rubber products, and tires were subject to much more technological change than other rubber products. Thus, like automobiles opportunities existed for regions that were not well stocked with firms in related industries. Within Ohio, the most important rubber producer at the start of the tire industry was BF Goodrich, which was located in Akron, where the (limited number of) rubber producers in Ohio were concentrated. Goodrich was a leading bicycle tire producer and successful producer of other rubber products, and like Olds Motor Works it was an important catalyst for the industry in Akron. It produced the first pneumatic automobile tire in 1896 and immediately became one of the leading producers of tires. Goodrich was influential 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 from Goodrich. Goodrich produced Kelly-Springfield s initial carriage tire based on a patented design before Kelly-Springfield initiated the production of automobile tires in Akron in Goodrich also initially produced tires for Firestone after its entry in Akron in 1900 and then supplied Firestone with prepared rubber and fabric when it began producing its own tires in Finally, Goodyear was founded in 1898 by the son of one of the original financiers of Goodrich that subsequently also operated a rubber firm (Buenstorf and Klepper, 2005b). With five leading firms located in Akron early on, the stage was set for spin-offs to play a key role in the further development of the industry around Akron. Buenstorf and Klepper (2005a) traced the origin of the

99 The evolution of geographic structure in new industries firms that entered in the state of Ohio through Like Detroit, spinoffs accounted for a disproportionate share of the entrants that originated from the Akron area 58 per cent of the 36 entrants that originated in Summit County (the home of Akron) were spin-offs versus 35 per cent of the other entrants originating elsewhere in Ohio. Most of them were formed by high-level employees, similar to the automobile industry. Furthermore, the bulk of the spin-offs that originated in Ohio entered in either the same or a contiguous county to where their employer was located (ibid.). A statistical analysis of the rate at which employees left Ohio firms to form spin-offs revealed that the highest spin-off rate among Ohio producers occurred in the leading Akron firms, followed by the next tier of leading producers in Ohio (ibid.). In an analysis of firm performance (Buenstorf and Klepper, 2005b), firms located in the Akron area had lower hazards than firms located in the rest of Ohio and outside of Ohio. Similar to Detroit, the distinctive performance of the firms in the Akron area was confined to the spin-offs located there. Furthermore, among all spin-offs in Ohio, those that descended from the leading firms or the second tier of leading producers had lower hazard rates, suggesting that the superior performance of the Akron spin-offs was largely attributable to their heritage. Buenstorf and Klepper (2005a) traced where entrants in Ohio originated, which for diversifiers was where they previously produced, for spinoffs where their parents were located, and for start-ups where their founders previously worked. Not only did spin-offs tend to locate in or close to their county of origin, but so did diversifiers and other start-ups. In a statistical analysis of the county where entrants located given their county of origin, Buenstorf and Klepper (2005a) found that the number of tire producers and the population of an entrant s county of origin did not positively influence its likelihood of entering there. However, these same characteristics influenced whether an entrant located in a distant county. Figueiredo et al. (2002) found similar patterns for modern Portuguese entrepreneurial start-ups. One interpretation of these findings is that entrants have valuable knowledge about their home region, such as where to find labor, input suppliers, transportation and even sources of knowledge spillovers, but they lack this knowledge about other regions. Consequently, even if their home region is not well stocked with firms in their industry and other industries and local markets for labor, inputs and so on are thin, they still know where to secure their needs. Without this knowledge about other regions, they would be better off locating in regions with more firms in their industry and in other industries because such regions would have better developed local markets 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

100 84 Industrial dynamics their base location, similar to autos. In the 1920s, the leading tire producers established branch manufacturing plants throughout the United States to save on transportation and labor costs (Jeszeck, 1982). This intensified after 1935 due to increasing militancy on the part of the union representing tire workers (ibid.), causing the share of tire production in Ohio to decline. Similar to the automobile industry, it was the leading firms that were in the vanguard of exploiting the advantages of more remote areas. Their willingness to set up plants outside of Ohio is suggestive of the limited advantages of locating in Akron. Consistent with this, Akron was not 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. Firms from the rubber industry, the most closely related industry, were important seeds for tire entrants, but spin-offs were also significant competitors. There was great heterogeneity among entrants in terms of their pre-entry experience that persistently affected their performance. One firm was a key catalyst for activity around Akron, both through its effects on other early Akron producers and through the spin-offs that it and the other successful Akron producers disproportionately spawned. As a consequence, the industry became extremely agglomerated around an unlikely region, reflecting both the randomness in the location of any one firm and the unlikely combination of early firms in one narrow region that was critical to the extreme agglomeration of the industry there. Agglomeration economies did not appear to play a major role in the agglomeration of the industry around Akron, and branching by the leaders eventually reduced the agglomeration of the industry there. The evolution of the market structure of the industry, particularly the eventual drying up of entry after 1930, may have contributed to the geographic concentration of the industry, but this concentration was established well before the industry underwent a shakeout. 5. OBSERVATIONS Various themes emerge from the study of the three industries regarding the evolution of the geographic structure of new industries. The Location of Firms in Related Industries Influences Where Entrants Locate In all three industries, the location of firms in related industries influenced the location of entrants into the new industry. This was most apparent in

101 The evolution of geographic structure in new industries 85 televisions, where both diversifiers and other entrants concentrated in the three cities where the radio firms were clustered. In both autos and tires, regions with more firms in related industries also had more diversifiers and other entrants. But firms in related industries were more dispersed in autos and tires than TVs, so entrants were more dispersed in these two industries than TVs. The radio industry may also have had more influence on the location of TV producers than any related industry had on autos and tires because of the greater overlap between radios and TVs than any product had with either autos or tires. This was reflected in the much higher fraction of entrants that were diversifiers (from the radio industry) in TVs than autos and tires. The influence of related industries on entry into TVs, autos and tires suggests two points. First, firms need competence to compete in a new industry, and one source of that competence is experience in related industries. Indeed, the fact that diversifiers in all three industries had lower hazards of exit on average than other entrants suggests that experience in a related industry was an important source of competence in all three industries. Second, diversifying entrants do not venture far geographically from their roots, which also appears to have been the case for spin-offs and other startups. Buenstorf and Klepper s (2005a) findings concerning the location of Ohio tire entrants suggests that entrants locate close to their roots to exploit valuable local knowledge they possess based on their pre-entry experience. Consequently, an important determinant of regional entry into a new industry is the stock of local firms that could provide the competence needed 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 competence for a new industry, in autos and tires incumbent firms also appear to have been an important source of such competence, especially the leading incumbents. The leading firms had higher rates of spin-offs, and on average their spin-offs were better performers than spin-offs from lesser firms. Moreover, their spin-offs were certainly competitive with if not superior performers to diversifiers from related industries, suggesting that the leading incumbent firms were also an important source of competence for entrants. The superior performance of spin-offs from the leading firms could reflect that these organizations had more to pass down to offspring. Alternatively, it could reflect that better firms attracted better managerial talent and more talented individuals founded superior firms. Judging from the dominance of the industry by diversifiers from the radio industry, spin-offswere not competitive in TVs. Two factors may have been

102 86 Industrial dynamics at work. Radios and TVs overlapped considerably in terms of technology and marketing whereas autos and tires represented a greater break from past products. Consequently, diversifiers from radios may have had a greater advantage in TVs than any kind of diversifier had in autos and tires, limiting 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, including spin-offs. Klepper (2002a) developed a model of shakeouts in which earlier entrants have a head start in building up a market for their products, which enables them to apply their research and development over a larger output, providing them with a competitive advantage. Spin-offs naturally enter later because they require a gestation period, in the form of employees gaining experience in incumbent firms. Consequently, they will be at a greater disadvantage in markets in which demand initially grows rapidly, as occurred in TVs relative to autos and tires. Consistent with this, entry became negligible within 10 years of the start of the TV industry whereas it continued 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 (and other entrants) did not venture far from their geographic roots, so entry was greater around successful firms. The spin-offs of successful firms also performed better than other spin-offs and were competitive with, if not superior to, diversifiers from related industries in autos and tires. Consequently, over time activity built up around successful early producers, especially in Detroit and Akron, where successful early auto and tire producers were concentrated. Entry in Detroit and Akron was disproportionately composed of spinoffs and it was spin-offs in both regions that performed distinctly well, suggesting that the spin-off process alone can give rise to agglomerations. On the other hand, it is conceivable that the agglomerations in both Detroit and Akron were driven by agglomeration economies, which could give rise to the same patterns of entry being concentrated in agglomerated areas and firms in agglomerated areas performing better than firms elsewhere. But if agglomeration economies were the primary cause of the agglomerations in autos and tires, then entrants of all kinds should have been attracted to Detroit and Akron. Moreover, entrants of all types should have performed better in Detroit and Akron than their counterparts elsewhere. Yet in both areas entry was disproportionately composed of spin-offs and only spinoffs performed better than their counterparts elsewhere. Furthermore, nearly all the spin-offs in Detroit and Akron had parents located there and

103 The evolution of geographic structure in new industries 87 so were not drawn from other regions. Judging from Buenstorf and Klepper s (2005a) findings for tires, spin-offs (and other types of entrants) that originated from agglomerated regions were also no more likely to locate in their home region than spin-offs that originated elsewhere. Thus, agglomeration economies do not appear to have played a major role in fostering the agglomerations in either autos or tires. The TV industry is also instructive about the power of spin-offs versus agglomeration economies to generate agglomerations. Agglomeration economies would have been expected to operate as strongly in TVs as autos and tires. Yet despite the extraordinary concentration of entrants in three narrow areas, the TV industry became less agglomerated over time. What appears to have been missing was a spin-off process that generated firms that were competitive with the leading diversifiers. This further suggests that the key to the agglomerations of the auto and tire industries was the spin-off process and not agglomeration economies. The Spin-off Process Can Magnify an Early Cluster of Leading Firms into an Extraordinary Agglomeration Within the first 10 years of the auto and tire industries, most of the leading producers in autos and tires were present in Detroit and Akron. The setting was ripe for the spin-off process to magnify the initial cluster of leading firms in each region, and this is precisely what occurred. Consequently, over time the percentage of firms and activity around Detroit and Akron increased and both regions evolved to account for over 60 per cent of activity in their industries. While establishment-level data are not available to compute the Ellison Glaeser (1997) index of geographic concentration, conservative estimates would put both industries in the tail of manufacturing industries in terms of geographic concentration. Having so many early leaders in one narrow region is surely an uncommon event, which would explain Ellison and Glaeser s (1997) finding that agglomerations 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 autos and tires, the spin-off process would still be expected to cause activity to build around successful early firms. Activity would still agglomerate, but it would be dispersed across more regions than in autos and tires. This could explain Ellison and Glaeser s finding that activity in most manufacturing industries agglomerates to some degree, but generally much less so than in autos or tires. In both autos and tires, the early leaders did not all locate in Detroit and Akron by chance. Rather, one firm in each region Olds Motor Works in autos and BF Goodrich in tires played a key role in other successful

104 88 Industrial dynamics producers being located nearby. Both firms were fertile sources of spin-offs, but their influence was broader. Olds provided valuable experience to its local subcontractors, some of whom later entered or financed the ventures of others, while Goodrich supplied inputs and sometimes initially manufactured tires for local tire firms. In both industries, input markets were not yet well developed, and so expertise accumulated in one firm redounded to the benefit of others nearby. This is a form of spillover, but it is of somewhat different variety than the externalities featured in modern theories of agglomeration economies. It is restricted to a small number of connected firms 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 in autos and tires got established in an unexpected place. Neither Detroit nor Akron was particularly distinguished in terms of activity in related industries or anything else that would have fueled an agglomeration there. On the other hand, both regions were located in the so-called manufacturing belt and certainly had a healthy amount of activity in related industries. As chance would have it, both ended up with a diversifier from a related industry that became the first outstanding performer in its industry. This was the seed for the agglomerations that emerged in both regions. With a single firm having so much influence on the agglomeration process, chance can play a big role in whether and where an agglomeration gets established. Again, the TV industry is instructive about the circumstances in which chance can have such a big influence on the location of an industry. Without a strong spin-off process in TVs, no firm had the kind of influence on the location of the TV industry as either Olds or Goodrich. Consequently, firms ended up congregating where firms in the radio industry were concentrated, limiting 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, as the TV industry illustrates. Moreover, by the time the shakeouts began in autos and tires, Detroit and Akron were already well established as the centers of activity in their respective industries. Thus, the shakeouts in autos and tires do not seem to have played a critical role in the agglomeration of either industry. But characteristic of shakeouts is the drying up of entry, as occurred in all three industries after their shakeouts began. This removed a force that potentially could have undermined the agglomerations that formed in autos and tires. Therefore, indirectly the agglomerations in autos and tires might have been promoted by the shakeouts both industries experienced.

105 The evolution of geographic structure in new industries 89 Ultimately, though, the forces underlying the shakeouts in both autos and tires 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 from the actions of the leading firms, though, the economies were not so overwhelming as to preclude the establishment of branch plants by the largest firms in the industry. Thus, as both industries consolidated and the leaders took over an increasing share of the industry s output, the leading firms established branch plants, which they generally located away from their base locations to save on transportation and labor costs. With the leading firms generally based in Detroit and Akron, this eventually caused the agglomerations in autos and tires to decline over time. The same forces also led the TV industry to become more dispersed over time as firms moved some of their operations offshore to take advantage of lower labor costs. Dumais et al. (2002) found that in manufacturing industries branch plants generally are de-agglomerating in the sense that their location and use causes employment to move away from agglomerated areas over time. They found that the main force sustaining agglomerations was the greater longevity of plants in agglomerated areas. Again, the findings for autos, tires and TVs are instructive about the forces possibly at work. Judging from Buenstorf and Klepper s (2005a) findings for tires, firms did not choose their initial locations to minimize the costs of production but to exploit local knowledge they had accumulated through their pre-entry experience. Thus, when branch plants were established, it was natural to locate them away from where the firms were based, which was generally in agglomerated areas. Furthermore, agglomerations themselves can raise the cost of production such as by land prices being bid up, necessitating the payment of compensating wage differentials. In tires, the agglomeration of activity in Akron also no doubt facilitated the union organization of workers, and militancy on the part of the union contributed to firms setting up branch plants elsewhere. The greater longevity of plants in agglomerated areas could result from a spin-off process comparable to the one that operated in autos and tires. Firms in agglomerated areas would be better performers 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 is enough evidence from other industries to suggest that the forces at work in auto, tires and TVs are operative in other industries as well. Sorenson and Audia (2000) found that in the footwear industry, entry was more likely in agglomerated areas even though firms had higher hazards of exit in these

106 90 Industrial dynamics areas. They interpreted this as a reflection of the natural tendency of entry to concentrate near incumbents even in the absence of agglomeration economies, consistent with the spin-off process in autos and tires. Gordon Moore of Intel fame, along with his co-author, implicated spin-offs in the semiconductor industry as the primary basis for the agglomeration of activity in Silicon Valley (Moore and Davis, 2004). The analog to Olds and Goodrich was Fairchild, whose offspring were so numerous that they were dubbed Fairchildren. Moore and Davis (2004) have a particularly interesting discussion of why working in an incumbent semiconductor firm provided distinctive organizational knowledge that enabled high-level managers to form their own successful spin-offs. A few studies analyse the spin-off process in specific industries without linking it explicitly to geography, and their findings are also consistent with those for autos and tires. In both the disk drive (Franco and Filson, 2000; Agarwal et al., 2004) and laser industries (Klepper and Sleeper, 2005), spinoffs performed distinctively well. In both industries more successful firms had higher spin-off rates, which also appears to have been the case among semiconductor firms located in Silicon Valley (Brittain and Freeman, 1986). In disk drives, the spin-offs of more successful firms were also better performers, which appears to have been associated with an (involuntary) transfer 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 differ from the outset in terms of their competence based on their pre-entry experience. But what exactly does the pre-entry experience of firms provide them and how does this influence their performance not just initially, but for many years after entry? Spin-offs played a key role in the agglomeration of both autos and tires. Why do spin-offs occur, why are they more prevalent among the leading firms, and what drives the correlation between the performance of spin-offs and their parents? Under what conditions do successful early entrants galvanize other firms to form and prosper nearby? More generally, what are the mechanisms that influence the transmission of knowledge across firms in the same industry and between suppliers and producers, and how is this mediated by geographic distance? These are just some of the questions raised by the evolution of the three industries. Hopefully, further examination of the way the geographic structure of new industries evolves will shed light on these questions and on the fundamental drivers of agglomerations and the geographic structure of industries.

107 The evolution of geographic structure in new industries 91 NOTES * Helpful comments were provided by an anonymous referee. Support is gratefully acknowledged from the Economics Program of the National Science Foundation, Grant No. SES The review of the evolution of the market structure of the three industries is primarily based on Klepper (2002a). The review of the evolution of the geographic structure of the three 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 and survival, Academy of Management Journal, 47, 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 semiconductor industry, mimeo, University of California, Berkley, CA. Buenstorf, Guido and Steven Klepper (2005), Regional birth potential, agglomeration economies, and home bias in the location of domestic entrants, mimeo, Max Planck Institute, Jena. Buenstorf, Guido and Steven Klepper (2005b), Heritage and agglomeration: the Akron tire cluster revisited, mimeo, Max Planck Institute, Jena. Dumais, Guy, Glenn Ellison and Edward L. Glaeser (2002), Geographic concentration as a dynamic process, Review of Economics and Statistics, 84, Ellison, Glenn and Edward L. Glaeser (1997), Geographic concentration in U.S. manufacturing industries: a dartboard approach, Journal of Political Economy, 105, Figueiredo, Octavio, Paulo Guimaraes and Douglas Woodward (2002), Homefield advantage: location decisions of Portugese entrepreneurs, Journal of Urban Economics, 52, Franco, April M. and Darren Filson (2000), Knowledge diffusion through employee mobility, Federal Reserve Bank of Minneapolis, Staff Report 272. Jeszeck, Charles A. (1982), Plant dispersion and collective bargaining in the rubber tire industry, PhD dissertation, University of California, Berkeley. Kimes, Beverly R. (1996), Standard Catalog of American Cars, , 3rd edn, Iola, WI: Krause Publications. Klepper, Steven (2001), The evolution of the U.S. automobile industry and Detroit as its capital, mimeo, Carnegie Mellon University, Pittsburgh, PA. Klepper, Steven (2002a), Firm survival and the evolution of oligopoly, RAND Journal of Economics, 33, Klepper, Steven (2002b), The capabilities of new firms and the evolution of the U.S. automobile industry, Industrial and Corporate Change, 11, Klepper, Steven (2003), The geography of organizational knowledge, mimeo, Carnegie Mellon University, Pittsburgh, PA.

108 92 Industrial Dynamics Klepper, Steven (2005), The organizing and financing of innovative companies in the evolution of the U.S. automobile industry, in Naomi Lamoreaux and Kenneth Sokoloff (eds), The Financing of Innovation, Cambridge, MA: forthcoming. Klepper, Steven and Kenneth L. Simons (2000), Dominance by birthright: entry of prior radio producers and competitive ramifications in the U.S. television receiver industry, Strategic Management Journal, 21, Klepper, Steven and Sally Sleeper (2005), Entry by spinoffs, Management Science, 51, Krugman, Paul (1991), Increasing returns and economic geography, Journal of Political Economy, 99, La France, Vincent A. (1985), The United States television receiver industry, PhD dissertation, Pennsylvania State University, University Park, PA. Levy, Jonathan D. (1981), Diffusion of technology and patterns of international trade: the case of television receivers, PhD dissertation, Yale University, New Haven, 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 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, , American Journal of Sociology, 106, Thomas Publishing Company ( ), Thomas Register of American Manufacturers, New York: Thomas Publishing Company. Warren Publishing Company, Television Factbook, vols 5 59, Washington, DC: Warren Publishing Company.

109 5. Constructing entrepreneurial opportunity: environmental movements and the transformation of regional regulatory regimes Brandon Lee and Wesley Sine* 1. INTRODUCTION Past research on the geographic distribution of economic activity has focused primarily on specifying the conditions that sustain economic clusters 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 opportunity exists, theory must explain how information and resources for entrepreneurial activity come to be disproportionately massed in some places and at some times (Romanelli and Schoonhoven, 2001: 41). Economic geographers have begun to address this question by focusing attention on institutional geography those social, political and cultural contextual elements that enable, constrain, and refract economic development in spatially differentiated ways (Martin, 2000: 79). In this chapter, we advance the institutional geography agenda by investigating how differences in regional collective action affect state regulatory regimes and hence, opportunities for economic activity. Recent developments in organization theory that integrate social movement theory into accounts of the emergence, change and decline of institutions (Davis and Thompson, 1994; Fligstein, 1996; Rao et al., 2000; Davis et al., 2005) provide a fruitful avenue for understanding variation in the creation and existence of regional entrepreneurial opportunity. Research in both entrepreneurship and economic geography has largely neglected the role that social movements play in generating new opportunities for entrepreneurs. While a substantial body of work has outlined how the characteristics 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

110 94 Industrial dynamics 1999; Sorenson and Audia, 2000; Stuart and Sorenson, 2003), there is little research on how regional collective action influences geographic variation in the success of new industries and technologies. Increasingly, scholars have shifted attention away from the characteristics and abilities of the lone entrepreneur (McClelland, 1961) and toward the demand side of entrepreneurship (Thornton, 1999), with work centering on the impact of exogenous shocks or changes in the environment that stimulate opportunities for new types of economic activity and organizational forms (Schumpeter, 1934; Thornton, 1999; Sine and David, 2003). These environmental shocks or jolts provide a catalyst for action. Jolts disrupt the routines and practices of individual organizations (Meyer, 1982) and can precipitate institutional decline for entire industries and organizational fields (Tushman and Anderson, 1986; Greve, 1995; Hoffman, 1999). Technological change and innovation, deregulation, demographic shifts and changing consumer preferences can alter the relative value of information and resources in such a way as to catalyse new types of economic activity. While these exogenous shocks are often necessary for the creation of new opportunities, they are not sufficient in and of themselves. Extant research, however, reveals little about how these broader changes create and eliminate entrepreneurial opportunities and therefore this topic merits greater scholarly attention (Eckhardt and Shane, 2003). To address this gap, we draw upon research that suggests the creation of new forms (and opportunities for new forms, as we argue in this chapter) is not 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 entrepreneurial opportunity by examining how social actors engage in framing processes and the mobilization of resources (McAdam, 1996). By so doing, we clarify the mechanisms responsible for translating broader technological, economic and demographic disruptions into concrete entrepreneurial opportunities. Work in organization theory that employs the social movement framework generally does so in the context of industry participants seeking to change existing intra-industry arrangements or extra-industry constraints on their industry (for example, Fligstein, 1996; Davis and McAdam, 2000; Swaminathan and Wade, 2001). However, often, free-rider problems may impede collective action by industry participants and trade associations may not form until a later point in the industry life cycle (Rao, 2004). Taking a different approach, we suggest that broader social movements themselves, while potent motors for social change, also have the capacity to generate new markets and possibilities for entrepreneurial activity. In so doing, we answer calls to re-examine the linkages between organizational

111 Constructing entrepreneurial opportunity 95 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 movements are understudied (Giugni, 1998, 1999), scholarly consideration of their influence on markets and entrepreneurial activity is even more scant (but see Schneiberg, 2002; Lounsbury et al., 2003). To address this shortcoming, we examine how social movements influenced entrepreneurial opportunity creation in the US electrical power industry from 1978 to Prior to 1978, it was largely impossible for small, independent power producers using renewable energy technologies (RETs) to sell electricity to the US power grid. In 1978, federal legislation required incumbent utilities to purchase power from independent power producers. However, the federal government delegated all decision making and enforcement of the particulars of these regulations to state governments, opening the door for wide variation in the degree to which state-level regulation supported independent power producers. In this chapter, we use historical and quantitative analysis to examine the effect of state environmental movement membership on regional regulatory environments for RET entrepreneurs. By examining how social movements shape the creation of regulatory environments across US states, we account for geographic variation in entrepreneurial opportunity (Romanelli and Schoonhoven, 2001). In the next section, we outline how social movement organizations construct entrepreneurial opportunity in nascent industries. We then provide a brief history of the US electric power industry. This is followed by a historical narrative focused on how environmental social movements influenced state level incentive structures for entrepreneurs interested in developing ventures using alternative energy technologies. We then explain the data and methods employed and conclude with the analysis and a discussion of the results. 2. THEORY Social Movements and Entrepreneurial Opportunity Increased scholarly attention has focused on the role that collective action plays in the construction, change and devolution/destruction of institutions, technology regimes and new organizational forms (Garud and Van de Ven 1989; Davis and Thompson, 1994; Fligstein, 1996; Rao et al., 2000; Davis et al., 2005; Hargrave and Van de Ven, forthcoming). Because new forms, industries and technologies never emerge in empty social space,

112 96 Industrial dynamics processes of resource assembly, legitimation and integration require collective action (Sine and David, 2003). Change and innovation within a field or market generally comes from peripheral players that engage in collective action but do not have a stake in, or benefit from existing institutional configurations (Leblebici et al., 1991). Collective challenges to existing institutions or technologies have been increasingly theorized through a social movement lens to explain how collective efforts contest existing institutions and construct new institutions and organizational forms that embody shared interests and goals. For example, in their account of the emergence of the organic food industry, Lee and Lounsbury (2005) show how local coalitions of organic farmers collectively acted to distinguish organic from conventionally grown food by developing and refining a process theory of how food should be cultivated. By collectively organizing and embodying that theory in standards and certification procedures, organic farmers successfully created a new category of products termed organic. This collective work created a solid jurisdictional boundary around the organic food category that infused these products with value over and above similar products that were not labeled organic, allowing organic farmers and retailers to charge more for their products. In a similar vein, Schneiberg (2002) demonstrates how fire mutuals emerged as a new form to solve coordination problems that were not adequately addressed by either markets or hierarchies. These mutuals represented a means by which property owners and agriculturalists resisted economic centralization and drew upon anti-company politics and agrarian protest to forcefully articulate and instantiate an alternative model of organizing and managing risk in a new organizational form. Finally, David et al. (2005) show how in the early 1900s institutional entrepreneurs publicly identified particular problems associated with the emergence of the corporate form (Chandler, 1990). The dramatic changes in size and scope of firms resulted in a new set of challenges for both owners and managers that institutional entrepreneurs such as Arthur Little, Edwin Booz and James McKinsey argued were best addressed by external consultants well versed in contemporary social and natural science. These institutional entrepreneurs and the firms and industry associations they formed worked together to create and proselytize problem solution models that proposed using external consultants with established categories of expertise such as psychology or chemistry for solving organizational problems. Due to the collective efforts of these early pioneers, within less than a decade, management consulting became an established part of the business landscape. Drawing upon these studies, we suggest that localized collective action efforts can lead to the construction of regional entrepreneurial opportunities

113 by framing problems and solutions and constructing and utilizing mobilizing structures (McAdam, 1996). We define framing as a collective attribution process through which people can voice their grievances, specify the underlying logic of those grievances, and articulate and promulgate a solution (see Snow et al., 1986; Snow and Benford, 1988; Benford and Snow, 2000). Mobilizing structures are those formal and informal vehicles by which people engage in collective action (McCarthy, 1996). We argue that formal social movement organizations employ frames and mobilizing structures to create legal environments that facilitate particular types of entrepreneurial activities. Legal and Regulative Structure Constructing entrepreneurial opportunity 97 Weber (1978) explicated the state s role in fostering and shaping markets: Law can... function in such a manner that, in sociological terms, the prevailing norms controlling the operation of the coercive apparatus have such a structure as to induce, in their turn, the emergence of certain economic relations (667, emphasis added). Law and state regulation, through instrumental as well as normative and cultural cognitive means, influences the creation and subsequent expansion of new markets (Dobbin, 1994; Fligstein, 2001; Fligstein and Stone Sweet, 2002). In their study of the railroad industry in Massachusetts, Dobbin and Dowd (1997) found that material support from the state increased the number of foundings of railroad firms. Similarly, Lomi (1995) shows how direct policy efforts codified in law led to the diffusion of cooperative banks to rural areas in Italy where private stock banks were not willing to do business. Wade et al. (1998) demonstrate how state governments can curb particular economic practices. They find that during the Prohibition period in the United States, regulations at the state level were effective at barring in-state breweries, yet these regulations unintentionally spurred brewery foundings in adjacent states where these regulations did not exist. Over time, however, as the number of states with prohibition laws increased there was a decrease in brewery foundings in all states. These accounts, particularly the work by Wade and colleagues, demonstrate that policy governing economic activity has direct effects on organizational founding dynamics and that regulation can vary by region. However, we have little understanding of the origins and impetus for this type of regulation and what accounts for the variation in its adoption. Attempting to specify the mechanisms of legal structure diffusion, recent work (Schneiburg and Bartley, 2001; Strang and Bradburn, 2001; Ingram and Rao, 2004; Schneiberg and Soule, 2005) offers a more nuanced understanding of the origins, antecedents and factors that account for regulation

114 98 Industrial dynamics within a market. Schneiberg and Bartley (2001), for example, find that in the early US fire insurance industry, more robust regulation was likely to be enacted when marginal groups such as farmers and small businesses challenged big business in the political arena. Legal and regulatory structure within a region or locale can also serve as a mechanism that generates variation in agglomeration processes. Local institutions can structure access to resources and provide cognitive models that are important in the emergence of new firms (Suchman et al., 2001). While it is generally accepted that institutional forces work at a level that transcends regional boundaries, some of the most important ones operate more narrowly, within particular geographically or functionally bounded organizational communities (Suchman et al., 2001: 357). For example, in a 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 in their enforcement of non-compete clauses experience greater new venture formation. Further, they find that enforcement of those non-compete clauses lessens the effects of liquidity events on founding rates. This study supports Suchman et al. s (2001) claim that, Among the various institutional structures that might regulate the flow of resources and models to new firms, law occupies a particularly important place (362). In the following section, we draw upon historical and archival documentation to develop an analytical narrative that demonstrates the role that environmental social movements played in contesting existing methods of producing electricity. Through their collective efforts, they facilitated the construction of a regulatory environment that provided the necessary incentives 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 combination of oil, coal, large hydroelectric facilities, and to a lesser extent natural gas and nuclear technology to generate power. Other than a few small experimental facilities, utilities did not generate power using nonhydro renewable sources (US Department of Energy, 2001). This is not to say that there was no interest in renewable technologies. For example, the first wind turbine that produced electricity was built in 1888, and since that time a long line of fringe alternative power enthusiasts have promoted their respective technologies. However, prior to 1978, the electric utility industry consisted of vertically integrated utilities that generated and distributed electricity. Utilities largely rejected RETs because they were viewed as expensive and risky

115 Constructing entrepreneurial opportunity 99 when compared to the highly developed traditional generation technologies. Utilities controlled both power generation and distribution in particular regions. Renewable energy technologists could thereby only access regional wholesale and retail energy markets with the cooperation of the local utility. These utilities often resisted interconnection with independent generators because interconnection would likely increase their costs and utilities were not eager to distribute power generated by potential competitors. Since utilities could lock out these potential competitors by refusing interconnection, offering below market prices for independently produced power, and charging higher-than-average prices for back-up power (Hirsh, 1999: 81 3), there were no foundings of independent power generation facilities that sold electricity to the grid prior to This situation changed dramatically when in 1973 a Saudi oil embargo on the United States caused oil prices to more than double, reaching $25 a barrel in Further disruptions to oil supplies in 1978 pushed oil prices to $50 a barrel, five times the 1972 price of oil. Utilities attempted to reduce their reliance on oil by converting to expensive solid-fuel plants but despite those efforts, electric prices remained high. These price increases motivated policy makers to search for other ways to generate electricity that would decrease the country s dependence on foreign oil, thereby providing institutional entrepreneurs fertile opportunities to promote new technological agendas to these same policy makers (Sine and David, 2003). Environmental activists brought these agendas to national attention by questioning existing energy policies and practices. As early as the late 1960s, growing awareness of the harmful environmental effects of coal- and oilburning power plants and fear of nuclear power increased the hostility of many environmental groups toward electric utilities (Fenn, 1984: 51 2). Large, established environmental groups such as the Sierra Club, the Audubon Society, the Union of Concerned Scientists and others began to actively promote an energy conservation agenda that included increased use of renewable energy and more efficient use of energy from all sources (McCloskey, 1992; McLaughlin and Khawaja, 2000). Advocates of RETs argued that while these new technologies were relatively underdeveloped compared to coal- and oil-based technologies, they had several qualities that made them potentially better sources of power than conventional means. First, unlike coal, oil and gas, the process of generating power with RETs does not produce air or water pollution, making its environmental footprint smaller than that of large-scale hydroelectric plants. Moreover, unlike coal, the production of renewable energy does not require large mines or, as in the case of oil, run the risk of spills. Second, RETs are local sources of energy and thereby promote local jobs. Finally, given technological progress, RETs had the future potential of being priced similarly to

116 100 Industrial dynamics energy produced by traditional sources. However, like most claims about future technology progress, this last point was highly uncertain. 1 In reaction to the energy crisis and pressure from pro-conservation and renewable energy advocates, the National Energy Act (NEA) was passed in 1978 and section 210 of this law, the Public Utility Regulatory Policies Act (PURPA), allowed entrepreneurs to construct qualifying non-utility facilities free from the constraints of traditional utility regulation. Under section 210, utilities were required to interconnect with qualifying non-utility power plants (that is, qualifying facilities) and purchase power from qualifying facilities at utilities cost of generation (which came to be known in the industry as avoided cost ). Independent power plants qualified under section 210 if they used alternative energy resources such as wind, solar, biomass, garbage, wood, sewage sludge and other lower-grade fuels, or used cogeneration technology. Section 210 provided the legal structure at the national level that allowed alternative energy entrepreneurs to interconnect with and sell electricity to utilities. However, the Federal Energy Regulatory Commission (FERC) left the interpretation and enforcement of section 210 to state governments. Social Movements and the US Power Industry Social movements deinstitutionalize existing beliefs, norms and values embodied in extant forms and establish new forms that instantiate new beliefs, norms and values (Rao et al., 2000: 238). For social movements to be successful in these deinstitutionalization and institutionalization projects, movements must at the outset engage in the production and maintenance of meaning for constituents, antagonists and bystanders (Benford and Snow, 2000: 613). Producing and maintaining collective meaning (that is, collective action frames) includes three important processes: diagnostic, prognostic and motivational framing. Diagnostic framing involves creating a shared understanding of a problematic condition that is in need of change 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 for engaging in collective action (Benford and Snow, 2000). Beginning in the early 1970s, environmental movement organizations engaged directly in diagnostic, prognostic and motivational framing to advocate the use of renewable energy. Diagnostic framing Suchman (1995) argued that the recognition and articulation of a problem for which there is no adequate solution is the first step required for creating

117 Constructing entrepreneurial opportunity 101 change in taken-for-granted ways of doing something (Scott, 2001). The process of developing a description of, and detailed evidence about, a problem, its cause and its negative consequences, focuses the attention of the public and powerful actors on unsolved difficulties (Edelman et al., 1999; Sine and David, 2003). Unanswered problems provide a social location for new solutions and simultaneously delegitimate existing institutional arrangements that are not adequately addressing or solving the problems (Cohen et al., 1972). Educating relevant actors and the public about problems catalyses solution generation processes as relevant actors engage in sensemaking and collective problem solving (Mezias and Scarselletta, 1994; Weick, 1995). The public articulation and elaboration of problems creates opportunities for institutional entrepreneurs to persuasively argue for the implementation of new practices and attracts pools of alternative solutions. Prior to the 1970s, the US environmental movement took little notice of energy policy. However, after the publication of Rachel Carson s Silent Spring (1962), members of the environmental community engaged in framing coherent, consistent and salient critiques of the nation s dependence on fossil fuels and nuclear power. The core idea emanating from these critiques was that the growing level of energy consumption was not sustainable. The most extreme advocates believed that modern technological systems were not simply malfunctioning or manifesting inefficiencies; they were in some profound sense unable to preserve the environment, and the social and political structures that had produced those technologies could not 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 was environmentalist Amory Lovins. In a strategic and creative way, Lovins effectively reframed the energy problem. Instead of assuming an increased demand for energy, Lovins argued that people demand delivered services such as comfortable rooms, light, vehicular motion, food, tables, and other real things (Lovins, 1976: 78). Lovins s perspective suggested that the problem is not simply where to get more energy, of any kind, from any source, at any price, but rather how to supply just the amount, type and source of energy that will provide each desired service at least cost (Lovins and Lovins, 1974: 26). In sum, Lovins argued that the real energy problem lies 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 of supply, Lovins invoked the metaphor of someone who is unable to fill the

118 102 Industrial dynamics bathtub because the water keeps running out. Instead of thinking of the problem as having too small a water heater, the problem lies in not having a plug (Lovins and Lovins, 1974; Lovins, 1976). The energy problem gained greater salience as Lovins ascribed blame for energy shortages to shortsighted pricing policies. Instead of basing energy decisions on current average costs, as is common in the industry, Lovins advocated the use of long-run marginal costs, which would include capital costs associated with the construction of new centralized power plants. This was important because in many cases the cost of new capacity was substantially larger than past power facilities. He effectively argued that the legal infrastructure of the time essentially subsidized the true costs of nonrenewable energy by not accounting for externalities such as associated environmental degradation and not taking into account future costs of additional capacity. Prognostic framing Prognostic framing, or the generation of solutions, is closely associated with diagnostic framing (the articulation of problems) (Gerhards and Rucht, 1992). Lovins s critique of US energy policy and his reframing of the energy problem naturally dovetailed with the solutions he proposed. To remedy the problems he described, he proposed to do more with less energy through the use of energy-efficient technologies, allowing exactly the same output of goods and services. Another key characteristic of the solution included technologies that relied on renewable energy flows that matched user needs in both scale and quality (Lovins, 1976). By scale, Lovins suggested that large energy production facilities be utilized for large uses and small energy production technologies be used for small sources, such as a passive solar design to heat a home. By quality, he meant that expensive forms of energy, such as electricity, should be reserved for applications where electricity is appropriate and indispensable, such as for smelting, subways and other kinds of mechanical work, and not be used to perform small jobs such as domestic space heating. Implicit in his solution was the decentralization of energy systems to redress the supply/demand mismatch. Lovins s creative reframing of the energy problem and its associated solutions had an immediate and profound impact on the energy policy debate. He was attacked repeatedly for his position, yet responded articulately and rapidly to each criticism. One author noted, He presented a thorough critique of conventional thinking about energy policy and the most creative and sophisticated arguments in favor of solar energy to date. His analysis provided a rationale for a group of growing importance in the solar movement those who based their attachment to solar energy on beliefs

119 Constructing entrepreneurial opportunity 103 about the nature of environmental problems and the relationship of energy technologies to them. (Laird, 2001: 126) However, his work did not exclude more mainstream supporters, because it was grounded in economic efficiency, suggesting that his approach would be cheaper than the existing one. Furthermore, while Lovins s approach was consistent with ecological values, he argued repeatedly that it did not require them, suggesting that his approach could coexist with more conventional values like economic rationality (Laird, 2001). Lovins s diagnostic and prognostic framing of energy issues gained resonance because the frame was consistent with both environmental and cost benefit logics and therefore empirically credible (Benford and Snow, 2000). Consequently, this new energy frame gained acceptance from environmental groups that in turn promulgated and reinforced it through articles, 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 can be made a soft path alternative energy plan, one based on conservation and renewable sources. If adopted, it could solve our most pressing energy problems faster, cheaper and more cleanly than Carter s plan (Snyder, 1979: 4). The Union of Concerned Scientists (UCS) completed a study in 1980 that critiqued both the current system and the industry s nuclear alternative 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 energy shortages that will continue until mankind has learned to harness energy from renewable sources.... In the face of these mounting difficulties, a long-term strategy emphasizing major improvements in energy productivity (with an attendant reduction in energy growth) and a reliance on renewable energy derived from the sun [2] emerges as the clearest and most sensible solution to the great challenge we and future generations face.... We believe that the strategy outlined in this study can help ease the burden of transition to a benign and sustainable energy future for the United States. (UCS, 1980: 21) For the UCS, renewable energy provided a sensible solution to competing solutions such as nuclear power: On the other hand, we believe that most, if not all, of the major environmental and societal risks posed by a large-scale breeder future could be avoided in a well-planned solar energy economy. [3] A solar energy future would in some ways be simpler and in other ways more complex than a breeder future. While the individual solar technologies are, in themselves, much less complicated than breeder reactors, the design and integration of a comprehensive solar energy

120 104 Industrial dynamics system would involve a greater degree of sophistication than a system based entirely on breeders. However, the resulting system would be more flexible, less dangerous, better adjusted to the diverse energy needs of this complex industrial state, and thus a more stable basis for the nation s continued prosperity than any system based on nuclear fission. (Ibid.) Studies conducted by Friends of the Earth, the Audubon Society and the Sierra Club all provided a solid basis for critiquing existing technology and power generating sources and at the same time, advocated the use of alternative energy. For example, the UCS strongly advocated for wind, suggesting that it was the most viable, safe, benign and easily commercialized of all alternative energy technologies: Wind represents a large and nondepletable energy resource that can be utilized with minimal impact on the environment, producing no air and thermal pollution and requiring no water in its utilization. The simplicity of wind technology will allow for rapid deployment in comparison to many other energy technologies. Finally, the economic prospects of wind systems are quite promising. There are barriers as well to the widespread use of wind power, but none should prove insurmountable. (Ibid.: 145) Through effective framing, these social movement organizations provided meaning and momentum to efforts to change the structure of electricity production in the United States. They served as sounding boards, amplifying the importance and urgency of fostering the development of alternative energy sources. The development and subsequent promulgation of these frames became the basis for the mobilization of resources in support of RETs in Congress and among the rank-and-file membership of environmental groups. Motivational framing and resource mobilization Mobilizing structures are forms of organization available to social movement actors that include informal friendships, neighborhood and work settings that facilitate and structure collective action (Tilly, 1978), and formal organizations that embody and attempt to achieve the preferences and goals of a broader social movement (McCarthy and Zald, 1977). For the purposes of this chapter, we focus on formal social movement organizations (SMOs) as the key mobilizing structures used to recruit members, obtain and distribute material resources, disseminate information, provide protocols of action for members to follow, and facilitate a common cognitive framework among their members. For example, the Sierra Club served as a principal mobilizing structure for collective efforts to advocate RETs and develop supportive regulative structure. The elaborate organizational

121 Constructing entrepreneurial opportunity 105 structure of the Sierra Club (chapters in all 50 states) coupled with the flexibility to accommodate more informal and proactive activities in promoting RETs made it a potent organization for altering political landscapes. An article in a 1970 newsletter described the relationship between formal organizational structure and grassroots action in mobilization efforts: Perhaps the most important aspect of chapters and groups is that they are the Club s grassroots interface with the public. Some hold monthly membership meetings open to the public, with speakers and film programs. Others have organized speaker bureaus and presented programs to community groups. Many chapters publish excellent newsletters, which carry articles on local conservation campaigns, as well as scheduled activities. A few have opened community conservation centers, either with their own resources or in cooperation with other environmental organizations. Very often, it is at the group level that the nuts and bolts work of the conservation campaigns is carried out providing manpower to gather data, prepare lawsuits, testify at public hearings, influence public officials, and generate the enthusiasm and dedication that will carry to state and national levels, attracting support for our cause.... It is often at the group level that a conservation issue was first recognized, and then brought to the attention of the chapter and to the Board of Directors, possibly to become one of the club s priority campaigns nationally. The process is reversed, though, when appeals 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 in this complex structure, besides funds, public support and staff, is an informed, active membership. (Billings, 1971: 21) A bridging concept between mobilizing structures and framing processes is motivational framing (Benford and Snow, 2000), which provides the rationale for engaging in collective action. Important aspects of this type of framing include articulating the severity of the problem, the urgency for action, and the efficacy and propriety of a specific action (ibid.: 617). In the case of renewable energy, environmental groups provided a rationale for its members to support and advocate alternative energy sources by issuing a call to action, outlining the specific, urgent action to be taken, and emphasizing the consequences of inaction. For example, the Audubon Society, an organization whose primary focus had little to do with energy policy, provided 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 most economic and environmentally benign energy strategy. If organizations like Audubon and the many environmentally oriented individuals who make up the membership of such organizations are unable to communicate to their neighbors and governmental leaders both the merits and the urgency of the solar/efficiency approach, then the goals of this Plan and others like it will not be met. In that case, the energy growth point of view of organizations like the Edison Electric

122 106 Industrial dynamics Institute will dominate public policy debates, and the possibility of achieving economic 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 and describes the kind of political action to be taken (communicating with neighbors and government leaders), the need for urgency (goals must be met), and the effects of inaction (the Edison Electric Institute dominating public policy debates). Other types of activities prescribed by this environmental organization included increasing public awareness, scrutinizing state and local energy plans, lobbying at all levels of government, involving the community in studying local energy problems, and calling upon experts at universities to analyse the benefits of conservation for taxpayers (ibid.: 105). The Sierra Club constructed a similar mobilization frame and leveraged its formal and informal organizational structure to mobilize its members to promote regulatory change: The threat posed by President Carter s proposals is so great, and the need for action by every member so urgent that the board of directors has called for the mobilization of the Club s full resources for this Emergency Energy Campaign. Only a massive outpouring of grass-roots concern can transform the present political climate, encouraging Congress to drop the damaging proposals and enact more rational energy alternatives. Intensive organizing efforts have already been set in motion, and letter writing and media contacts have begun. All Sierra Club chapter and group leaders will be receiving regular updates as this campaign speeds along. (Snyder, 1979: 5) It is time for face-to-face mobilization. Conservationists must start meeting with their elected officials and candidates to tell them what they, as voters, expect and want. We need to stand up at rallies and ask, politely, but persistently, why Congress solved the energy crisis by putting most of the money into the least promising and most expensive technologies, the synthetic fuels.... If, in the next six months, every Sierra Club member would just once personally attend and participate in a political event, it would make a world of difference. (Coan and Pope, 1980: 11 13, 47) Environmental memberships readily accepted the mobilization frames promulgated by the leadership of environmental organizations and their call for political activism. In fact, comprehensive surveys of Sierra Club members reveal that their commitment reflected the directives, goals, and aspirations of the club s leadership. A survey in 1971 revealed significant commitment among the organization s rank-and-file membership manifest in their political activism: 60 per cent said that they sent their views on conservation matters to government officials at least once in the past year, with 15 per cent reporting nine or more communications; 40 per cent of the

123 Constructing entrepreneurial opportunity 107 members attended one or more political meetings or rallies in the last year and 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 of 1979, energy policy was at the fore for both the rank-and-file members and leaders of the Sierra Club. A second survey in 1979 revealed a membership highly committed to changing energy policy. The opening sentence of the story that introduced the results of the survey read: A new broadened Sierra Club motto might be in order. It might read:... To preserve, 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 development of 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 energy issues needed more effort on the part of the Club. Energy was also the third most popular issue that Club members showed a willingness to spend more time working on: 75 per cent of members expressed a strong interest in solar power, with the remainder expressing some or little interest (ibid.). These surveys demonstrate that the logics advocated by the Sierra Club leadership that framed conventional energy technology as an environmental problem and RET as a solution were accepted and enacted by the membership. In this section, we have provided qualitative evidence that suggests environmental movement organizations effectively framed energy problems and solutions, provided a rationale for action to their members, and marshaled resources through mobilizing structures to create favorable regulatory climates for independent power firms using RETs. Given the intensity and scope of activities engaged in by these social movement organizations, we hypothesize that higher levels of social movement activity in a state will increase the likelihood of a favorable regulatory environment for firms using RET. 4. DATA AND METHODS We test our hypothesis by examining the effects of Sierra Club membership on avoided cost rates. We draw on panel data that capture US state energy regulation from 1980 to Dependent Variable Our dependent variable is the average size of a state s avoided costs in a given year. Although the federal government loosely defined avoided cost

124 108 Industrial dynamics (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 some states, it was based on the cost of fuel of the most likely alternative while other states took into account capital costs. Calculations that took into account capital costs tended to be controversial because they varied tremendously depending on the type of technology used (that is, nuclear technology was more expensive than coal technology). Other sources of ambiguity included the choice of fuel type to use as the baseline to calculate the cost of generation. In other words, were public utility commissions to define avoided costs by the most expensive type of fuel used in a utility s portfolio or its least expensive? Further, should the public utility commissions assume high or low interest rates? Because of these ambiguities, there were substantial differences in how states calculated avoided costs and consequently, their amounts. Our data on avoided costs come from three sources: Solar Law Reporter (1981), Energy User News ( ), and Avoided Cost Quarterly ( ). Independent Variable We obtained state-level Sierra Club membership data. During the period of our study, the Sierra Club was one of the three largest environmental organizations in the United States (McCloskey, 1992). We chose the Sierra Club membership data due to the organization s size and status within the environmental community. We also acquired yearly state-level membership data from the UCS and found their membership distribution across US states to be highly correlated with that of the Sierra Club s (0.88). Membership data from the Audubon Society and several other large environmental organizations were not available because such information was not retained during the late 1970s and 1980s. Control Variables Our model controlled for the following state characteristics: per capita gross state product, change in gross state product, state population, congressional voting record on energy issues, 4 and net electricity imports. We also controlled for the availability of land with constant streams of highquality wind because wind turbines were the most common RET used by independent power entrepreneurs during our study period. Because the regulatory climate may be affected by the extent to which regulators are willing to monitor compliance and punish firms for not following formal regulations or informal norms, we also controlled for the activism of state

125 utility commissions. To assess commission activism, we measured the number of rate cases per utility, comprehensive audits per utility, comprehensive audits of private utilities, and total audits per utility. Audits provide a good measure of regulatory oversight because the purpose of audits is to verify the information given to the commission by utility companies. More audits conducted by a public utility commission are indicative of an involved and active state commission. Rate cases examine a utility s financial information to verify the justification for current or proposed rates. Utilities view audits and rate cases as highly disruptive and expensive. Excessive audits and rate cases can be viewed as either punitive measures or as a way of ensuring compliance with state regulatory policies. These data come from the National Association of Regulatory Utility Commissioners (NARUC) annual utility surveys. Model Specification and Estimation Constructing entrepreneurial opportunity 109 We measure the impact of environmental social movement organizations on the regulatory climate using avoided costs as the dependent variable. This variable indicates the extent to which a state s regulatory climate is supportive of alternative energy. We use panel data for each state-year between 1980 and The analysis begins in 1980 because this is the earliest date that states defined avoided costs. We employ a generalized estimating equation to predict the size of the avoided cost and corrected for common biases in the analysis of longitudinal data introduced by autocorrelation among residuals by assuming an unstructured error correlation. We used the STATA XTGEE command with the unstructured correlation option. An advantage of this model is that it does not assume that panels are uncorrelated. We used robust variance estimators in our analyses, reducing problems associated with heteroskedasticity and misspecification of the error structure (Allison, 1999). We also estimated the models using the ar 1 option, but found no significant difference in the results. We used a variance inflation factor test to ensure that our models were not significantly affected by multicollinearity and found that the factor scores in both models presented were less than 5, suggesting 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 the results of the generalized estimating equation predicting supportive regulatory climate. Several control variables significantly predict the amount of

126 Table 5.1 Summary statistics and correlations for state-avoided cost analysis Variable Mean S,D Avoided costs (cents/kwh) GSP per capita Class 3 & 4 wind availability State population (ln) Change in Gross State Product (GSP) 6. Net electricity imports (million kwh) 7. Rate cases per utility Comprehensive audits per utility 9. Comprehensive audits private utilities 10. Audits per utility Congressional voting record/ Environmental membership/1000 Note: All correlations greater than 0.11 are sig. at 0.05 level. 110

127 Constructing entrepreneurial opportunity 111 Table 5.2 GEE model predicting state-avoided costs Variable Avoided cost Model 1 Model 2 Control variables GSP per capita (47.371) (48.746) Class 3 & 4 wind availability 0.000* (0.000) (0.000) State population (ln) 0.429* 0.928** (0.174) (0.287) Change in GSP (0.017) (0.020) Net electricity imports (0.078) (0.082) Rate cases per utility (0.588) (0.640) Comprehensive audits per utility (1.429) (1.615) Comprehensive audits private utilities (0.057) (0.067) Audits per utility (0.044) (0.046) Congressional voting record/ ** 0.021** (0.007) (0.007) Independent variable Environmental membership/ * (0.255) Constant 9.164** ** (2.684) (3.074) Chi squared statistic Note: p 0.10; *p 0.05; **p Standard errors are in parentheses. the avoided cost in a given state. The availability of windy land is significant and negatively correlated with state-avoided costs. This is unusual because one would expect that state governments would create legal structures that take advantage of local natural resources. Instead, we find the opposite to be true: regulators are less likely to try and use incentives to motivate wind power entrepreneurship in locations with substantial endowments of windy land. This suggests that state governments use incentives to compensate for shortages of natural resources in areas where constituents value wind power. We also find that states with smaller populations and states

128 112 Industrial dynamics with congressional representatives who have a history of voting in favor of environmentally friendly energy policies are also more likely to have higher avoided costs. The results in Table 5.2 support our hypothesis; environmental group membership positively and significantly predicts avoided costs. An increase in Sierra Club membership of 1000 increases the size of a state s avoided costs by 0.63 cents per kwh. 6. DISCUSSION AND CONCLUSION In this chapter, we have shown how regional differences in social movement organization membership impacted the creation of supportive regulatory environments for renewable energy entrepreneurs. Prior to 1978, electric utilities dominated the production and distribution of electricity in the US. These organizations enjoyed regional monopolies and local legal structures that had evolved over four decades to accommodate the needs and interests of this organizational form. Thus, state legal structures were highly attuned to incumbent utilities and did not support other organizational forms such as independent power. For example, the retail price utilities charged customers for electricity was regulated and determined by public utilities commissions on a cost-plus basis which was defined as a utility s cost of generation plus a fair profit. However, states did not have a procedure for determining the wholesale price of renewable electricity generated by independent power plants, leaving it up to utilities to determine whether and at what price to purchase independent power. Moreover, the existing regulatory structure supported coal, natural gas and oil production through a variety of tax incentives and permitting systems. A similar supportive regulatory infrastructure did not exist for independent power companies using RETs. Transforming these regional regulatory environments to support a new organizational form (independent power generators using RET) was a necessary condition for entrepreneurial activity in this sector. Environmental movement organizations paved the way for the emergence of independent power using RETs by framing energy generation as not only an economic decision, but also a moral one. Environmental movement organizations promulgated this frame via their regional chapters. These chapters mobilized local members to promote renewable energy and directly engaged local constituents via educational programs, public relations and direct lobbying of state governments. By the early 1980s, over half of the more than members of the Sierra Club were dedicating some time each week to supporting this agenda. In regions where the Sierra Club had a sizeable membership, political landscapes that favored incumbent utilities

129 Constructing entrepreneurial opportunity 113 and their non-renewable technologies were transformed to support independent power producers using RETs. Our data demonstrate that much of the impact of national environmental organizations was quintessentially local. Although the public relations arms of the national organizations promulgated these new frames via the media nationwide, states support for these new organizational forms varied substantially. States with large and active Sierra Club or UCS chapters were significantly more likely to define avoided costs in such a way as to create opportunities for entrepreneurs using renewable energy technologies. Based on these findings, this chapter contributes to existing theory in four ways. First, unlike past work on industry emergence that treats regulatory 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 technical change, markets and institutions (Nelson, 1995). Regulation tends to foster inertia and path dependency and generally reflects and supports the interests and goals of dominant organizational forms (Fligstein, 1996, 2001). As such, dominant organizational forms defend and seek to preserve these favorable regulations. Consequently, such regulatory structures are difficult and expensive to change and because no single firm possesses the necessary resource base and scope to precipitate such substantive regulatory change, collective action is necessary. Regions with a strong presence of collective actors supportive of new categories of economic activity are more likely to see regulatory change that favors these new sectors. Building on research from the institutional tradition (Schneiberg and Bartley, 2001; Scott, 2001), our study shows that regulatory environments that supported renewable energy were more likely in regions with large environmental organization memberships such as the Sierra Club and the UCS. It is important to note that environmental groups did not endorse renewable energy because of its apparent economic efficiency. Nor was the emergence of RETs an instance of an inevitable outcome of technological change. Rather, environmental groups advocated RETs because they had a more benign effect on the environment a critical externality that the regulatory environment at that time did not take into account. These social movement organizations worked to create a set of rules and incentives that supported technical solutions aligned with ideologically-driven preferences for renewable energy and conservation. Second, our work advances an institutional approach to economic geography using an evolutionary perspective (Thrift and Olds, 1996; Crang, 1997; Martin, 2000; Sine and Lee, 2006). Storper and Walker note that socio-institutional structure is an essential underpinning of efficient capitalist production (1989: 5). Institutional theorists likewise contend that

130 114 Industrial dynamics economic activity is socially and institutionally situated and embedded in social and political structures (DiMaggio and Powell, 1983; Granovetter and Swedberg, 1992; Scott, 2001). The primary intent of an institutionalist approach to economic geography is to understand to what extent and through which means are the processes of geographically uneven capitalist economic development shaped and mediated by institutional structures (Martin, 2000: 79). However, anecdotal evidence and theorizing have not specified how supportive regulatory structure emerges at the regional and local level and what precise role it plays in regional economic development (ibid.: 88). By bringing the theoretical lens of social movements to bear on the question of spatial variation of economic opportunity, we provide new analytical leverage to these questions. Our research suggests the importance of considering how pre-existing regional differences in non-economic sectors can shape economic outcomes. In the case study presented in this chapter, social movements promulgated particular norms and values also constructed the cultural and regulatory infrastructure for entrepreneurs using RET. Third, we contribute to social movement literature by emphasizing the important local effects that movements have on regional economies. While past research on social movements has demonstrated how ideas diffuse across regions, there has been little research on how movements change local environments to foster new economic activity (but, see Schneiberg and Soule, 2005). For example, research has demonstrated how social movements such as the temperance movement grew and eventually influenced the ability of organizations to sell alcohol (Sandell, 2001), but we know little about how these activities influenced the rise of other sectors such as carbonated soda at the local level. Giugni (1999) has called for more empirical research that examines the relationship between social movements and social change by considering how the outcomes of social movements vary across different contexts. By emphasizing regional differences in social movement activity and its impact on the transformation of regulatory environments, this research provides evidence of the causal link between regional social movement activity and social and economic change. Finally, this case study demonstrates how existing mobilizing structures of social movement organizations can be leveraged to effectively advance new social agendas. In less than 10 years, environmental organizations effectively mobilized their members to help spread an obscure set of ideas about energy generation technologies throughout the US and embed it in both national and regional regulatory structures. It is precisely the ability of environmental organizations to redirect and broaden mobilizing efforts to include advocating RET through its organizational structure and preexisting repertoires of action that led to these dramatic regulatory changes

131 Constructing entrepreneurial opportunity 115 in such a short period of time. We find that years of promoting the protection of the environment led to the unanticipated consequence of a constituency ready to support renewable energy generation technologies. Once the national organization made the connection between environmental values and methods of energy generation, local chapters went to work creating fecund environments for this type of commercial enterprise. NOTES * This research was supported by the Initiative for Future Agriculture and Food Systems Grant no 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 by the fact that during our study the cost of power generation using RETs was always significantly higher than that of conventional technology, and many experts predicted that this would not change for decades. 2. Because the sun drives global climatic systems, energy derived from the sun includes solar thermal, photovoltaic, wind, biomass, ocean thermal energy conversion, tidal power, wave power, ocean current, and salinity gradient energy systems (UCS, 1980). 3. A solar economy refers to an economic system in which all energy is generated using renewable energy sources. 4. These data were obtained from the League of Conservation Voters which annually tracks the voting records of individual members of Congress on environmental-related legislation. 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: Arrow, K.J. (1962), The economic implications of learning by doing, Review of Economic Studies, 29: Benford, R.D. and D.A. Snow (2000), Framing processes and social movements: an overview and assessment, Annual Review of Sociology, 26: Billings, L. (1971), Sierra Club 1970: New chapters, groups, growth and change, Sierra Club Bulletin, 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 organizational choice, Administrative Science Quarterly, 17: Coombs, D. (1972), The club looks at itself, Sierra Club Bulletin, July/August, Crang, P. (1997), Cultural turns and the reconstitution of the economic, in R. Lee and J. Wills (eds), Geographies of Economies, London: Arnold, pp

132 116 Industrial dynamics David, R.J., W.D. Sine and H.A. Haveman (2005), Exploiting institutional change: entrepreneurship in the legitimation of new organizational forms, Paper presented at the Administrative Sciences Association of Canada Annual Meetings, Toronto, 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 Davis, G.F., D. McAdam, W.R Scott and M.N. Zald (2005), Social Movements and Organization Theory, Cambridge and New York: Cambridge University Press. Davis, G.F. and T.A. Thompson (1994), A social movement perspective on corporate control, Administrative Science Quarterly, 39: DiMaggio, P.J. and W.W. Powell (1983), The iron cage revisited: institutional isomorphism and collective rationality in organizational fields, American Sociological Review, 48: Dobbin, F. (1994), Forging Industrial Policy: The United States, Britain, and France in the Railway Age, Cambridge and New York: Cambridge University Press. Dobbin, F. and T.J. Dowd (1997), How policy shapes competition: early railroad foundings in Massachusetts, Administrative Science Quarterly, 42: Eckhardt, J. and S. Shane (2003), Opportunities and entrepreneurship, Journal of Management, 29: Edelman, L.B. (1990), Legal environments and organizational governance: the expansion of due process in the American workplace, American Journal of Sociology, 95: Edelman, L.B., C. Uggen and H. Erlanger (1999), The endogeneity of legal regulation: grievance procedures as rational myth, American Journal of Sociology, 105: Fenn, S. (1984), America s Electric Utilities: Under Siege and in Transition, New York: Praeger. Fligstein, N. (1996), Markets as politics: a political cultural approach to market institutions, American Sociological Review, 61: Fligstein, N. (2001), The Architecture of Markets: An Economic Sociology of Twentyfirst Century Capitalist Societies, Princeton, NJ: Princeton University Press. Fligstein, N. and A. Stone Sweet (2002), Constructing polities and markets: an institutionalist account of European integration, American Journal of Sociology, 107: Friedland, R. and R. Alford (1991), Bringing society back in: symbols, practices, and institutional contradictions, in W.W. Powell and P.J. DiMaggio (eds), The New Institutionalism in Organizational Analysis, Chicago: University of Chicago Press, pp Garud, R. and A.H. Van de Ven (1989), Innovation and the emergence of industries, in A.H. Van de Ven, H. Angle and M.S. Poole (eds), Research on the Management of Innovations, Cambridge, MA: Ballinger, pp Gerhards, J. and D. Rucht (1992), Mesomobilization: organizing and framing in two protest campaigns in West Germany, American Journal of Sociology, 98: Giugni, M.G. (1998), Was it worth the effort? The outcomes and consequences of social movements, Annual Review of Sociology, 98: Giugni, M.G. (1999), How social movements matter: past research, present problems, future developments, in M.G. Giugni, D. McAdam and C. Tilly (eds), How Social Movements Matter, Minneapolis, MN: University of Minnesota Press, pp. xiii xxxiii.

133 Constructing entrepreneurial opportunity 117 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: Hargrave, T.J. and A.H. Van de Ven (Forthcoming), A collective action model of institutional innovation, Academy of Management Review. Harris, C.D. (1954), The market as a factor in the location of industry in the United States, Annals, Association of American Geographers, 44: Hirsh, R. (1999), Power Loss: The Origins of Deregulation and Restructuring in the American Electric Utility System, Cambridge, MA: MIT Press. Hoffman, A.J. (1999), Institutional evolution and change: environmentalism and the US chemical industry, Academy of Management Journal, 42: Ingram, P. and H. Rao (2004), Store wars: the enactment and repeal of anti-chainstore legislation in America, American Journal of Sociology, 110: Jaffe, A.B., M. Trajtenberg and R. Henderson (1993), Geographic localization of knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics, 108: 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 change and the transformation of interorganizational fields: an organizational history of the U.S. radio broadcasting industry, Administrative Science Quarterly, 36: Lee, B.H. and M.D. Lounsbury (2005), Instituting industry governance: the organizational and legal bases of U.S. national organic food standards, Paper presented at the American Sociological Association annual meeting, Philadelphia, August. Lomi, A. (1995), The population ecology of organizational founding: location dependence and unobserved heterogeneity, Administrative Science Quarterly, 40: Lounsbury, M.D. and M. Ventresca (2002), Social structure and organizations revisited, in M.D. Lounsbury and M.J. Ventresca (eds), Research in the Sociology of Organizations, Vol. 19, New York: JAI/Elsevier, pp Lounsbury, M.D., M. Ventresca and P. Hirsch (2003), Social movements, field frames and industry emergence: a cultural political perspective on U.S. recycling, Socio-Economic Review, 1: Lovins, A.B. (1976), Energy strategy: The road not taken?, Foreign Affairs, 55: Lovins, A.B. and L.H. Lovins (1974), Energy solutions abound: just look around!, Business and Society Review, 82: 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 McAdam, D. (1996), Introduction: opportunities mobilizing structures and framing 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 University Press, pp McCarthy, J.D. (1996), Constraints and opportunities in adopting, adapting, and inventing, in D. McAdam, J.D. McCarthy and M.N. Zald (eds), Comparative

134 118 Industrial dynamics Perspectives on Social Movements: Political Opportunities, Mobilizing Structures, and Cultural Framings, Cambridge: Cambridge University Press, pp McCarthy, J.D. and M.N. Zald (1977), Resource mobilization and social movements: a partial theory, American Journal of Sociology, 82: 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 McLaughlin, P. and M. Khawaja (2000), The organizational dynamics of the U.S. environmental movement: legitimation, resource mobilization, and political opportunity, Rural Sociology, 65: Meyer, A.D. (1982), Adapting to environmental jolts, Administrative Science Quarterly, 27: Mezias, S.J. and M. Scarselletta (1994), Resolving financial reporting problems: an institutional analysis of the process, Administrative Science Quarterly, 39: National Audubon Society (1984), The Audubon Energy Plan,New York:National Audubon Society. Nelson, R.R. (1995), Co-evolution of industry structure, technology and supporting institutions, and the making of comparative advantage, International Journal of the Economics of Business, 2: Perrow, C. (1986), Complex Organizations: A Critical Essay, New York: Random House. Perrow, C. (2002), Organizing America: Wealth, Power, and the Origins of Corporate Capitalism, Princeton, NJ: Princeton University Press. Piore, M.J. and C.F. Sabel (1984), The Second Industrial Divide: Possibilities for Prosperity, New York: Basic Books. Rao, H. (2004), Institutional activism in the early American auto industry, Journal of Business Venturing, 19: Rao, H., C. Morrill and M.N. Zald (2000), Power plays: social movements, collective action and new organizational forms, in B.M. Staw and R.I. Sutton (eds), Research in Organizational Behavior, Vol. 22, New York: Elsevier, pp Romanelli, E. and C. Schoonhoven (2001), The local origins of new firms, in C. Schoonhoven and E. Romanelli (eds), The Entrepreneurship Dynamic: Origins of Entrepreneurship and the Evolution of Industries, Stanford, CA: Stanford University Press, pp Romer, P.M. (1986), Increasing returns and long-run growth, Journal of Political Economy, 94: Russo, M.V. (2001), Institutions, exchange relationships, and the emergence of new fields: regulatory policies and independent power production in America, , Administrative Science Quarterly, 46: Sandell, R. (2001), Organizational growth and ecological constraints: the growth of social movements in Sweden, 1881 to 1940, American Sociological Review, 66: Schneiberg, M. (2002), Organizational heterogeneity and the production of new forms: politics, social movements and mutual companies in American fire insurance, , in M.D. Lounsbury and M.J. Ventresca (eds), Research in the Sociology of Organizations, Vol.19, Oxford: JAI, pp Schneiberg, M. and T. Bartley (2001), Regulating American industries: markets, politics, and the institutional determinants of fire insurance regulation, American Journal of Sociology, 107:

135 Constructing entrepreneurial opportunity 119 Schneiberg, M. and S. Soule (2005), Institutionalization as a contested, multilevel process: the case of rate regulation in American fire insurance, in G. Davis, D. McAdam, W.R. Scott and M. Zald (eds), Social Movements and Organization Theory, Cambridge and New York: Cambridge University Press, pp Schumpeter, J.A. (1934), The Theory of Economic Development: An Inquiry into Profits, Capital, Credit, Interest, and the Business Cycle, Cambridge, MA: Harvard University 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: Sine, W.D. and B.H. Lee (2006), The winds of change: antecedents of entrepreneurial activity in the emerging wind power industry, Working paper, Cornell University. Snow, D.A. and R.D. Benford (1988), Ideology, frame resonance, and participant mobilization, in B. Klandermans, H. Kriesi and S. Tarrow (eds), From Structure to Action: Social Movement Participation Across Cultures, Greenwich, CT: JAI Press, pp Snow, D.A., E.B. Rochford, S.K. Worden and R.D. Benford (1986), Frame alignment processes, micromobilization, and movement participation, American Sociological Review, 51: 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 activity: geographic concentration of footwear production in the United States, , American Journal of Sociology, 106: Stern, R.N. and S.R. Barley (1996), Organizations and social systems: organization theory s neglected mandate, Administrative Science Quarterly, 41: Stinchcombe, A. (1965), Social structure and organizations, in J. March (ed.), Handbook of Organizations, Chicago: Rand McNally, pp 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 legitimating theory? Neoliberal discourse and HMO policy, , in J.L. Campbell and O. Pedersen (eds), The Rise of Neoliberalism and Institutional Analysis, Princeton, NJ: Princeton University Press, pp Stuart, T. and O. Sorenson (2003), Liquidity events and the geographic distribution of entrepreneurial activity, Administrative Science Quarterly, 48: Suchman, M.C. (1995), Managing legitimacy: strategic and institutional approaches, Academy of Management Review, 20(3): Suchman, M.C., D.J. Steward and C.A. Westfall (2001), The legal environment of entrepreneurship: observations on the legitimation of venture finance in Silicon Valley, in C. Schoonhoven and E. Romanelli (eds), The Entrepreneurship Dynamic: Origins of Entrepreneurship and the Evolution of Industries, Stanford, CA: Stanford University Press, pp Swaminathan, A. and J.B. Wade (2001), Social movement theory and the evolution of new organizational forms, in C. Schoonhoven and E. Romanelli (eds), The Entrepreneurship Dynamic: Origins of Entrepreneurship and the Evolution of Industries, Stanford, CA: Stanford University Press, pp

136 120 Industrial dynamics Thornton, P. (1999), The sociology of entrepreneurship, Annual Review of Sociology, 25: Thrift, N. and K. Olds (1996), Refiguring the economic in economic geography, Progress in Human Geography, 20: Tilly, C. (1978), From Mobilization to Revolution, Reading, MA: Addison-Wesley. Tushman, M.L. and P. Anderson (1986), Technological discontinuities and organizational environments, Administrative Science Quarterly, 31: 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, September. Utrup, K.A. (1979), How Sierra Club members see environmental issues, Sierra, March/April: Wade, J.B., A. Swaminathan and M.S. Saxon (1998), Normative and resource flow consequences of local regulations in the American brewing industry, , Administrative Science Quarterly, 43: 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:

137 6. Absorptive capacity and foreign spillovers: a stochastic frontier approach Jojo Jacob and Bart Los* 1. INTRODUCTION In the literature on economic growth in developing countries, international technology flows have gained growing attention. International technology can flow from the originating country to the receiving country in several ways. Among them, foreign direct investment (FDI) and trade in intermediate 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 neoclassical production function framework, in which the average response of the endogenous productivity variable to a change in one of the exogenous variables (such as the intensity of FDI and trade in intermediate inputs) is estimated by means of classical regression analysis. Although econometric scrutiny does not always confirm strong anecdotal evidence, the majority of studies do find significant positive impacts of international flows. 1 Most studies adopt an approach based on production functions, often by means of panel data regressions. One of the most prominent disadvantages is the impossibility of obtaining an understanding of the causes of observed heterogeneity. Rather, the focus of these studies is on representative behaviour (or, average behaviour ). Deviations from this behaviour are merely seen as realisations of a random noise process. If, for example, productivity performances show an increasing variance over time, the production function approach does not yield any insights, as the effect is just an increase in the variance of the stochastic random noise process. In this chapter, we borrow an interesting alternative approach from a subfield in econometrics called stochastic frontier analysis (SFA) to study labour productivity growth in Indonesian plants. 2 Typical techniques belonging to this subfield do not estimate average relationships between variables, but relationships for best-practice plants. Furthermore, they enable researchers to use plant characteristics as potential explanations of 121

138 122 Industrial dynamics the extent to which other plants performances fall below best practice. In the context of our chapter, this implies that we shall estimate relationships between inputs (capital, labour) and output (value added) for best-practice firms for several years, to get indications of the degree to which international technology spillovers affected productivity growth. Simultaneously, we shall link the underperformance of other firms to variables that relate to the evolutionary 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 the failure to fully assimilate international technology spillovers, and thereby to raise productivity to its potential level. 3 The organisation of the chapter is as follows. In Section 2, we shall briefly review some theories of productivity growth that are relevant for our empirical approach. Section 3 proposes our methodology. It deals with the appropriate technology accounting framework and discusses the way in which frontiers and distances to these frontiers are estimated. Section 4 is devoted to data issues. Special attention will be paid to the procedures we adopted 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 PRODUCTIVITY GROWTH Convergence (or its absence) of labour productivity levels has attracted a lot of attention, both from economic theorists and from more empirically oriented scholars. Although it is hard to classify theories in a field characterised by synthesis and hybridisation, roughly two categories of theories can be discerned. We follow Nelson and Pack (1999) in using the labels accumulation and assimilation theories. Accumulation theories basically assume that raising capital intensities (be it physical capital or human capital) automatically leads to labour productivity growth, although increasingly more investment is required for a given productivity gain. In this view, labour productivity growth is governed by movements along the production function of a given country, sector or firm under consideration. This perspective implies that accumulation of capital is the cause for growth in labour productivity. Assimilation theories challenge this view. Here, technology is seen as something that does not automatically and immediately flow across firms or countries. Instead, only firms or countries that have invested sufficiently in their absorptive capacities will be able to turn innovations developed elsewhere into productivity gains for themselves. In the view of assimilationists,

139 Absorptive capacity and foreign spillovers 123 policies to stimulate entrepreneurship and eagerness to learn have been much more important. Such a view on macroeceonomic performance can, with relatively minor modifications, be transferred to studies at firm or plant level. The resource-based view of the firm (see Teece, 2000, for example) stresses that long-run firm performance is mainly determined by learning capabilities. In this chapter, we shall differentiate between two barriers to attaining productivity levels attained by better-performing plants. The first type of barrier relates most strongly to issues mentioned above. Pack (1987) and Van Dijk (2005, Ch. 8) show that plants that are similar in terms of the types of machines installed attain widely varying productivity levels. 4 Apparently, learning and organisational capabilities are not identically distributed across plants, which shows up in different productivity figures for plants with more or less identical equipment installed. The second type of barrier is quite closely associated with what Abramovitz (1989) labelled technological incongruence. A similar idea has recently been proposed in the form of a formal model by Basu and Weil (1998). They defined technologies as particular combinations of inputs, or, in other words, capital labour ratios. New knowledge is only appropriate for a range of such technologies. Firms or countries will in the short run only be permitted to benefit from innovations if these relate to technologies that are comparable to the ones operated. 5 In the longer run, non-appropriate innovations can become appropriate if the firm or country invests to such an extent that it shifts its technology to a capital intensity level comparable to the innovating firm or country. 6 The predictions concerning convergence and divergence of productivity levels that follow from the Basu and Weil model are based on the assumption that more capital-intensive technologies allow producers to attain higher maximum levels of labour productivity. Los and Timmer (2005) showed that both types of barriers to catch-up play an important role in the empirics of macroeconomic growth. We adopt several parts of their methodology to investigate how innovations, changes in absorptive capacity and technologies operated contribute to the productivity growth experiences at the plant level in Indonesia s manufacturing sector. 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

140 124 Industrial dynamics upgrading which creates potential for spillovers. Next, we shall discuss the estimation methods required to arrive at quantifications of these effects. In the third subsection, we shall describe the empirical model. Identifying the Sources of Growth Los and Timmer (2005) decomposed labour productivity growths rates of a group of countries, between 1970 and 1990, into the effects of movements towards the frontier, or changes in technical efficiency (assimilation), movements of the frontier (innovation), and capital deepening (creating potential). The decomposition form itself was popularised by Kumar and Russell (2002), but Los and Timmer were the first to link their results to the theories discussed in the previous section. Our approach starts from a similar perspective. It is novel in the sense that it explicitly relates the decomposition results to observable characteristics of the plants. Figure 6.1 shows a plant s actual labour productivity levels y 0 and y 1,in an industry with production frontiers f 0 and f 1,for periods 0 and 1, respectively. The labour productivity change (y 1 /y 0 ) of this plant can be decomposed in the following way: y 1 y 0 y 1 y d y a y 0 y c y a y d y b 0.5 y b y a y d y c 0.5 (6.1) or (1 y T ) (1 y A ) (1 y C ) (1 y I ). (6.2) In the first term on the right-hand side (1 y A ), a value of y A larger than 0 indicates that the plant under consideration has increased its labour productivity for the technology operated. In other words, it indicates that the plant has been able to bring about an increased exploitation of technological potential as compared to the maximum productivity observed for the equipment operated. We call this the assimilation effect. 7 The second explanatory factor (1 y C ) indicates the changes in labour productivity due to increases in capital intensity alone. While a higher capital intensity in itself does not generate higher labour productivity, it can lead to an upward shift in the attainable or the target productivity levels, depending on the slope of the frontier. Therefore, a value greater than 0 for y C can be interpreted as creating potential. The third factor (1 y I ) points to the effect of localised technological change that results in the upward shift of the production frontier. Assuming that the plant s capital intensity remains

141 Absorptive capacity and foreign spillovers 125 Y/L y d F(1) y b y c F(0) y a y 1 * (1) y 0 *(0) C/L Figure 6.1 Labour productivity growth decomposition constant, a positive value for y I indicates that it has benefited from an increase in the maximum attainable labour productivity levels for the given technologies. We call this the innovation effect. Los and Timmer (2005) estimated the productivity frontiers for the beginning and end periods using data envelopment analysis (DEA). We follow a similar approach, but with the key difference that we derive the frontier labour productivity levels by means of stochastic frontier analysis (SFA). This change of method has advantages and drawbacks. The major drawback is that truly localised innovation cannot be modelled, as the estimated elasticity of foreign research and development (R&D) spillovers (the source of technological change) is the same across the full range of technologies. As a result, the shifts in the frontier labour productivity levels always amount to an identical proportional growth rate

142 126 Industrial dynamics 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 sensitive 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 quite a bit. In view of the sizeable measurement and reporting errors that are often found in plant-level surveys, especially in developing countries, we feel that the net advantage of SFA as compared to DEA is clearly positive. Estimation Method In recent years, a number of studies have employed stochastic frontier estimations for estimating and explaining inefficiencies of firms and plants in industries. Until recently, the standard approach was a two-stage estimation procedure, in which the production frontier is first estimated. In the second stage, the resulting inefficiencies (the vertical distances from the observed productivities to the estimated frontier) are regressed on firmspecific variables (see, for example, Pitt and Lee, 1981). 8 Estimation in the second stage, however, contradicts the assumption of identically distributed inefficiency effects that underlies the estimation of the stochastic frontier in the first stage. To overcome this methodological problem, several authors have suggested single-stage procedures for simultaneously estimating both the stochastic frontier and inefficiency functions. The Battese and Coelli (1995) model is one such approach. Consider the following production function for panel data: y it X it it, (6.3) where y it is the dependent variable corresponding to the ith plant and time t, X is a vector of explanatory variables, and it is the composite error term. It consists of a white noise error v it : v it ~ iid N(0, 2 v ) and u it. The two sets of disturbances are assumed to be independent. The u it s are non-negative random variables associated with technical inefficiencies, and are assumed to be independently (but not identically) distributed as truncations (at zero) of the N( it, 2 u ) distribution, with: it Z it, (6.4) in which Z is a vector of observable, non-stochastic explanatory variables associated with technical inefficiency, and is a vector of unknown coefficients.

143 Absorptive capacity and foreign spillovers 127 In this model, the maximum likelihood method is used for the simultaneous estimation of the parameters of the frontier and technical inefficiency models, that is, estimation of the values of the unknown parameters s, s, u2 and v2.we computed the estimates using the FRON- TIER software package (Coelli, 1996). Battese and Coelli (1995) also provide an expression for the conditional expectation of exp( u it ) given e it. The maximum likelihood estimation of this function is used to estimate the technical efficiency index of the ith firm at time t, based on expected values conditional on the observed values of the explanatory variables in X and Z. When the productivity frontier is expressed in logarithms, the technical efficiency index (TEI) can be expressed as follows: TEI it exp( u it ). (6.5) This index has a value between 0 and 1, with 0 (u it ) indicating the least efficient, and 1 (u it 0) the most efficient plants. Changes in TEI as defined in equation (6.5) denote a part of the actual shift 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 of technology-specific knowledge. It is that part of the assimilation effect which can be explained by the changes in the indicators of absorptive capacity, given their estimated coefficients from the SFA model. The remainder of the upward shift cannot be explained, and is calculated as the difference between the actual growth in labour productivity and the predicted growth in labour productivity derived from the SFA model. The Empirical Model We begin with a production frontier of the Cobb Douglas form, augmented to account for technological change. In most developing countries, and especially so in Indonesia, foreign technology is the key source of technological progress, because of the virtual absence of own technological efforts. We account for technology flows from abroad by constructing a measure of international R&D stock (IRD). The augmented production function is defined as follows: Y it AE it L it IRD t, (6.6) where, Y it is value added of plant i at time t, E is total energy use (as a proxy for capital goods use, as further explained below), L is total number of workers, and IRD is the international R&D stock representing the technology flows available to all plants in the industry (see the following section

144 128 Industrial dynamics for a fuller description of the variables). The variable IRD is interpreted to be 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 y it l it a (e it l it ) IRD t it, (6.7) where the lower-case symbols denote variables in logarithms. In the transformation of equation (6.6) to (6.7), we impose the assumption of constant returns 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 the inefficiency function, based on the procedure described in the previous subsection. Given that technology-embodied inputs have often shown to be an important channel of foreign technology diffusion, a plant s access to imports might be a good proxy of its access to foreign technology. Access to a source of technology does not, however, imply that the acquisition of technology is guaranteed. This is because technology is not entirely codified, and indeed often takes a highly tacit form (Polanyi, 1958). Therefore, the extent to which a plant is able to absorb knowledge related to new technologies, 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 in a plant s workforce. The ownership structure of a plant can also be a significant factor influencing the capacity to assimilate knowledge. A plant with foreign management control might be expected to run more productively than, for example, a non-professional, family-controlled enterprise. The foreign connection may enable the former to adapt itself much more quickly than the latter to global changes in technology, production relations and so on. The performance of enterprises as compared to other enterprises with similar technologies may also depend on its size. As noted by Tybout (2000), in many developing countries, the demand for manufactured products is skewed towards simple items which can be efficiently produced using cottage techniques. An opposite effect would be the operation of Schumpeterian dynamics that leads to greater learning efforts by large firms. This may result from scale economies, availability of internal resources in the presence of imperfect markets and/or uncertainties, synergies between technological, production, marketing and distribution activities, and so on. The empirical evidence, mostly pertaining to advanced economies, shows no consensus, 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

145 Absorptive capacity and foreign spillovers 129 (2002) argues, with increases in competitive conditions firms with greater experience have greater leeway in enhancing their capabilities. Keeping these considerations in mind, we consider the following absorptive capacity variables for the mean inefficiency model represented by equation (6.4): Z it : Access it ; LQual it ; Foreign it ; Age i ; Size it, where, Access represents access to technology spillover, defined as the share of imported intermediate inputs in total intermediate inputs; LQual stands for the quality of labour at a plant, defined as the share of non-production (white-collar) workers in total employment; Foreign represents the proportion of foreign ownership in a plant; Age is measured as the difference between the year of inception and the year of operation; and Size is defined as the logarithm of the total number of workers. A final aspect to consider is the influence on technical efficiency of factors observable only to the managers of a plant, which are not reflected in survey-based dataset like ours. As a result, such firm-specific effects (or heterogeneities) may be related to other regressors of the model, which may cause biased results. To overcome this problem, we adopt a specification that 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, backcast and Statistic Industri (SI), constructed by the Indonesian Bureau of Statistics (Badan Pusat Statistik, BPS) (see Appendix A of Jacob and Los, 2005, for a detailed description of the data sources, variables, cleaning processes and so on). The datasets cover all large and medium-sized plants in the manufacturing sector of the country, from 1975 to After applying cleaning procedures to account for duplications, reporting errors and data entry errors, we focus our analysis on industries defined at a low level of aggregation (5-digit classification). This allows us to investigate productivity for sets of plants with homogeneous activities. Since the panel data SFA approach is data intensive, we select 17 industries for which at least 10 plants are included in the data set (see Appendix Table 6A.1). The industries under investigation are quite diverse, which allows us to identify inter-industry differences in the importance of absorptive capacity for productivity performance. Studies aiming at explaining total factor productivity (TFP) growth are often hindered by immeasurable fluctuations in capacity utilisation.

146 130 Industrial dynamics Although we do not study TFP growth but labour productivity growth, we encounter similar problems: Our technologies are defined by capital intensities, that is, ratios between a (quasi-) fixed input and a much more variable input. We circumvent this problem by proxying a plant s capital use by its energy consumption, about which much information is available in our dataset. 11 This is also convenient from a practical point of view, since our data on energy use cover a much longer time span than those on investment on which we should base our capital stock estimates. As a determinant of labour quality, we do not have detailed information about skills in our database for a sufficiently long period. Therefore, we proxy differences in skill level of labour force across plants by differences in the share of non-production workers in the total workforce. Finally, we should describe how we estimate international R&D stocks that capture technology flows. Since Indonesian firms generally do not undertake any formal R&D activities themselves, it can safely be assumed that new technology must come from abroad (Hill, 1996). Our admittedly poor, but widely accepted assumption if suitable output indicators of innovation are not available, is that technology production is proportional to R&D expenditures. We have data on R&D expenditures by industry for 10 countries that together account for approximately 60 per cent of the imports to Indonesia and about 85 per cent of the total OECD R&D expenditure. The selection of this sample is justified because empirical evidence suggests that it is not the intensity of import per se that matters, but rather the distribution of the countries of origin. The more you import from highly R&D intensive countries, the larger the impact of foreign R&D (Lichtenberg and van Pottelsberghe de la Potterie, 1998, p. 1483). Apart from imports, technology purchase, technology collaboration and exports by Indonesian firms as well as foreign investment in the domestic market can all act as carriers of technology spillovers. To accommodate these different channels of technology flow,we do not weight foreign R&D stock by the volume of imports to Indonesia, which is a standard approach in the spillover literature. Instead, we consider solely the technological proximity between the foreign and domestic industries to weight the foreign R&D stock. The specific channels of foreign technology flows are introduced in the inefficiency function. In the first step, foreign R&D stock is weighted by an index of technological distance between the sector of origin and the sector of destination. We use a patent-based measure of technological distance between sectors derived 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 technological congruence between sectors in the advanced economy and Indonesia that are comparable in terms of their classification code. This

147 Absorptive capacity and foreign spillovers 131 weight accounts for inter-country differences between sectors. It captures the idea that an industry in a follower country benefits more from the global pool of technology, the greater its technological congruence with industries in advanced countries. The resulting international R&D stock at the industry level can be expressed in terms of the following equation: IRD j (t) c, k (RD ck P kj S cj )(t), (6.8) where IRD j is the international R&D stock resulting from technology flows available to all plants in the Indonesian industry j; RD ck is the R&D stock in sector k of partner country c; P kj is an element of the patent information flow matrix P, and it captures the flow of sector k s R&D efforts to sector j; and S cj is the technological congruence between sector j of Indonesia and the same sector of its partner country c. S cj is derived by comparing the input coefficient vectors for sector j in the two countries. It takes a value of 1 if sector j is perfectly similar in the two countries, and zero in the event of complete 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 of aggregation of 2, 3 and, in a few cases, 4-digit (ISIC Rev. 2), IRD j in the above equation corresponds to these levels. To generate IRD at the 5-digit level, we constructed similarity indices between the two sets of classifications, 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 in the SFA model in 17 5-digit-industry, plant-level samples. Table 6.1a shows the means and standard deviations across plants of the levels of the variables. Table 6.1b shows similar statistics of the average annual growth rates of the variables. The highest average labour productivity level and growth rate is found for drugs (35222). The lowest average values for these two variables are found for tobacco (31410) and clay tiles (36422), respectively. The former industry is also found to have the lowest average energy intensity, while plywood (33113) has the highest. Plants in garments (32210) recorded the fastest average growth in energy intensity, while those in crumb rubber (35523) recorded the slowest growth. Plants in paints (35210) could benefit from the biggest pool of relevant R&D done in foreign countries. The smallest pool is found for sawmills (33111). In other

148 Table 6.1a Summary statistics of 17 5-digit ISIC industries: levels Industry (ISIC) Plants Val/lab Egy/lab IRD Age Foreign Access Lqual Size (4.925) (1.026) (0.000) (7.624) (0.027) (0.044) (0.069) ( ) (6.495) (5.073) (0.000) (15.421) (0.134) (0.165) (0.160) ( ) (3.521) (0.661) (0.000) (10.542) (0.004) (0.063) ( ) (8.500) (0.983) (0.000) (14.033) (0.105) (0.065) ( ) (7.151) (32.014) (0.000) (9.159) (0.151) (0.136) (0.070) ( ) (2.867) (2.109) (0.000) (12.699) (0.144) (0.067) ( ) (5.964) (6.301) (0.000) (9.202) (0.214) (0.083) ( ) (6.830) (1.248) (0.000) (10.996) (0.132) (0.199) (0.066) ( ) (12.767) ( ) (0.000) (5.757) (0.197) (0.016) (0.071) ( ) (8.183) ( ) (0.000) (3.144) (0.139) (0.035) (0.060) ( ) (4.172) (4.811) (0.000) (10.588) (0.081) (0.060) ( ) (8.181) (5.118) (0.000) (16.939) (0.133) (0.121) ( ) 132

149 (30.403) (3.062) (0.000) (17.239) (0.262) (0.233) (0.153) ( ) (22.037) (5.229) (0.000) (10.031) (0.349) (0.243) (0.195) ( ) (16.738) ( ) (0.000) (3.843) (0.430) (0.000) (0.067) (85.367) (5.550) (5.669) (0.000) (4.998) (0.242) (0.058) ( ) (1.675) (1.673) (0.000) (15.973) (0.069) (0.071) (73.559) Notes: (i) Val/lab value added per employee; Egy/lab energy use per employee; Size number of employees; IRD international R&D stock in real 1990 purchasing power parity (PPP) US dollars (millions). Value added and energy are in thousands of real 1990 PPP US dollars. (ii) See main text for the definitions of the other variables. (iii) See Appendix Table 6A.1 for industry definitions. (iv) Means, standard deviations in brackets. 133

150 Table 6.1b Summary statistics of 17 5-digit ISIC industries: average annual growth rates Industry (ISIC) Plants Val/lab Egy/lab IRD Foreign Access Lqual Size * (0.160) (0.178) (0.000) (0.296) (0.078) (0.063) (0.211) (0.371) (0.000) (0.062) (0.583) (0.156) (0.071) * (0.365) (1.034) (0.000) (0.165) (0.103) (0.228) (2.950) (0.000) (0.488) (1.703) (0.057) (0.208) (0.287) (0.000) (0.044) (1.538) (0.110) (0.069) (0.207) (0.225) (0.000) (2.084) (0.142) (0.064) (0.513) (0.212) (0.000) (0.327) (0.086) (0.083) (0.260) (17.706) (0.000) (0.164) (2.822) (0.492) (0.114) (0.210) (0.330) (0.000) (0.005) (0.442) (0.126) (0.051) (0.156) (0.410) (0.000) (0.027) (3.295) (0.298) (0.092) (0.377) (0.152) (0.000) (4.609) (0.084) (0.075) (0.178) (0.256) (0.000) (0.764) (0.126) (0.065) 134

151 (0.226) (0.321) (0.000) (0.026) (0.882) (0.091) (0.053) (0.608) (0.216) (0.000) (0.036) (0.304) (0.079) (0.063) * (0.153) (0.174) (0.000) (0.039) (0.101) (0.033) (0.202) (0.340) (0.000) (0.336) (0.097) (0.098) (0.091) (0.246) (0.000) (0.292) (0.192) (0.046) Note: * The growth rate of the variable under consideration is positive in only one observation. For other notes, see Table 6.1a. 135

152 136 Industrial dynamics Table 6.2 Stochastic frontier estimates for 17 5-digit (ISIC Rev. 2) industries a Industry (ISIC Rev. 2) Production function Constant (1.765)* (19.375) (5.993) (0.998)* (5.495)* (6.428)* (11.056)* (3.900)* Egy/lab (0.039)* (0.040)* (0.025)* (0.086) (0.029)* (0.037)* (0.053)* (0.022)* IRD (0.114)* (0.225)* (0.382) (0.089)* (0.314)* (0.376)* (0.533)* (0.229)* (Mean) Inefficiency function Constant (0.406)* (18.938) (1.132)* (0.972) (0.786)* (1.286) (1.373) (0.718)* Age (0.120)* (0.144) (0.426)* (0.333) (0.180)* (0.232) (0.201) (0.083) Foreign (0.293) (0.364)* (11.460) (0.255)* Access (0.330)* (0.442) (4.324) (0.999) (0.149) (0.179) (0.485) (0.111)* Lqual (0.231)* (0.338)* (1.499) (0.997) (0.356)* (0.496) (0.685) (0.376) Size (0.093)* (0.116)* (0.183)* (0.214)* (0.074) (0.100) (0.114)* (0.056) Gamma (0.049)* (1.917) (0.107)* (0.220)* (0.537) (0.082)* (0.811) (0.447) Plants Observations 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. variables too, differences are noticeable. For example, the access-tospillover 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. Average foreign ownership grew fastest in garments (32210), while it declined in macaroni (31171) and crumb rubber (35523). 12 Another noticeable feature in both Tables 6.1a and 6.1b is the high standard deviations reported for many of the variables across plants in the industries considered. Quite often, the standard deviation is larger than the mean value. This

153 Absorptive capacity and foreign spillovers Production function (2.131) (8.177)* (1.025)* (1.031)* (4.554)* (1.464)* (4.848)* (1.521)* (15.799)* (0.052) (0.039)* (0.043) (0.035)* (0.049)* (0.031) (0.045)* (0.039)* (0.040)* (0.150)* (0.537) (0.071)* (0.056)* (0.237)* (0.077)* (0.288)* (0.090)* (0.850)* (Mean) Inefficiency function (0.691)* (0.582)* (0.818) (0.426)* (1.360)* (0.600)* (0.210)* (0.629)* (4.923) (0.171) (0.227) (0.240) (0.123) (0.260) (0.132)* (0.026)* (0.180)* (0.211)* (0.785) (3.456) (0.422)* (0.391) (0.428) (1.037) (2.197) (0.947) (0.257) (0.228) (0.116)* (1.029) (0.136) (0.418)* (0.862) (1.055)* (1.202) (0.401)* (0.673) (0.342)* (0.217)* (0.764)* (0.764) (0.223)* (0.185) (0.278) (0.085) (0.205)* (0.120)* (0.127) (0.143)* (0.144)* (0.000)* (0.179)* (0.128)* (0.044)* (0.095)* (0.021)* (0.000)* (0.027)* (0.875) phenomenon reflects the highly dual structure of the Indonesian manufacturing sector, described in detail by, among others, Hill (1996). Table 6.2 reports the SFA estimation results. For brevity, we do not report the estimation results for the plant dummies included in the inefficiency function. The results for the frontier production function show that the coefficients of both energy intensity, Egy/Lab, and the international R&D stock representing knowledge spillovers, IRD, have a positive sign in most industries. The estimated coefficients of energy intensities for the slopes of the productivity frontiers are generally fairly small, however, and even statistically insignificant at the 10 per cent level for four industries. This implies that it does not pay off very much for plants just to invest more, as is suggested by advocates of accumulationist theories. Consequently, accumulation alone cannot be considered as an important source of productivity growth in Indonesian manufacturing. 13

154 138 Industrial dynamics The sensitivity of the frontier to increases in foreign R&D appears to be prominent. Apparently, the best-practice plants in Indonesia reaped substantial fruits from technological spillovers from abroad. The coefficient for this 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 deviations from the best-practice technology. The estimate for the variance parameter (Gamma in Table 6.2) that corresponds to the estimated share of the inefficiency term in the variance of the composite error term has a positive sign in all industries, and is significant in most industries (12 industries). This should be considered as evidence for the idea that inefficiency effects contribute substantially to the variety of labour productivity levels indicated by positive standard deviations for this variable. A negative sign for the coefficient of a variable indicates a negative impact of that variable on inefficiency. Among all the variables, changes in labour quality (LQual) variable provide the most promising explanation for changes in comparison to best-practice performance. Its coefficient has a negative sign in most industries (11 out of 17), and is statistically significant in seven industries. Foreign ownership (Foreign) has a significantly negative coefficient only in three of the nine industries in which the shares held by foreign firms are positive in one or more establishments. For the remaining industries, changes in the degree of foreign ownership do not appear to matter for assimilation. It might well be that a linear specification of the inefficiency effects is not most appropriate here. Explorations to use multiple-regime econometrics (to identify critical values of the degree of foreign ownership), however, are beyond the scope of this chapter, if only because such analyses have hardly been attempted in the SFA branch of econometrics. We argued before that access to spillover (Access) is likely to exert a major influence on the technical efficiency of plants. However, this variable yielded a negative coefficient in only eight industries, with statistical significance limited just to three industries. One reason for this result could be the narrowness of our measure of access to spillover as it does not consider the import of capital goods. Second, the import intensity of intermediate inputs use is rather low in most industries as may be required to generate sufficient within-plant variations. An additional, and probably the most important, cause of very few significant results is the huge measurement errors that characterise datasets like ours. Although, we did clean the data extensively, it is unlikely that this has removed all measurement errors. The Age variable demonstrated a favourable impact on assimilation in a majority of the industries. A negative sign for its coefficient in 11 out of a total number of 17 industries appear to suggest that a plant s ability to

155 Absorptive capacity and foreign spillovers 139 assimilate knowledge spillovers from similar best-practice plants increases with its experience. As argued by Klepper (2002), under competitive pressures, firms with greater experience are better positioned to enhance their capabilities. Our period of analysis covers the export-oriented phase hence, the more competitive phase of industrialisation in Indonesia. We may therefore conclude that firms which have been in operation for a longer period of time have been more successful in enhancing their technological and managerial capabilities, and therefore in meeting the challenges of increased competition. The final variable to be discussed is Size, which displays considerable inter-industry variations in its influence. It had an adverse impact on assimilation in a majority of the industries (a positive coefficient in 11, with statistical significance in five industries). Of the six remaining industries where its influence has been favourable, in five cases the coefficients displayed statistical significance. The discussion so far has revealed that changes in deviations from best practice (due to plant-specific factors) are a significant determinant of a plant s labour productivity growth. In the following subsection we extend these results by decomposing labour productivity growth into that resulting from the shifts in the frontier of an industry s technology, capital deepening and efficiency gains (or losses). We provide theoretically grounded interpretations for the contributions of these three factors to average labour productivity growth in an industry. Results for the Decomposition Analysis We decomposed plant-level labour productivity growth during the period in each of the 17 industries under consideration. We then calculated their industry average using the geometric average of the initial and final year employment shares of plants as the weight. 14 Table 6.3 shows the decompositions of average labour productivity growth rates during the period. The first column shows the compound annual average growth in labour productivity and the second column, the period growth rate of productivity. The remaining columns represent the contribution of the four components to (period) labour productivity growth. All industries experienced a positive growth in labour productivity during this period, with important inter-industry variations as may be expected of a heterogeneous group of industries. In a majority of the industries, we see that productivity growth resulted from a combination of assimilation (unexplained) and innovation. While the former was the main contributor in most of the industries, the latter was the leading contributor in industries like paints & varnishes, plastics and rubber. Explained assimilation was a major

156 140 Industrial dynamics Table 6.3 Decomposition of productivity growth, Industry Annual Period Contribution to productivity growth (ISIC growth growth Explained Unexplained Innovation Potential Rev. 2) (%) (%) assimilation assimilation Note: See Appendix Table 6A.1 for industry definitions. 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 all industries. This result is mainly due to the flat shapes of the estimated frontiers. Increasing capital intensity hardly contributes to a higher potential labour productivity. In other words, learning or assimilation potentials remained more or less stagnant. 6. CONCLUSIONS AND FUTURE RESEARCH In this chapter, we proposed an SFA approach to study labour productivity growth in the Indonesian manufacturing sector (in the late 1980s and early 1990s). We focused specifically on the effects of inflows of foreign technology and the role of absorptive capacity changes on plant-level differences in productivity growth. Our main findings are that foreign R&D played a significant role in moving the productivity frontier in an upward direction. Hence, the inflow

157 Absorptive capacity and foreign spillovers 141 of technology was an important source of productivity growth for bestpractice plants. The creation of spillover potential (by investing to use more capital-intensive technologies) barely contributed to labour productivity growth, since relations between capital intensity and labour productivity were almost non-existent for best-practice plants. As a consequence, shifting to higher capital intensities hardly implied more potential spillovers to benefit from. This finding is clearly at odds with the major assumptions underlying the accumulationist theories of growth. Assimilation (movements towards the frontier) played an important role. We could distinguish between two kinds of assimilation effects. The first type could be explained by our absorptive capacity indicators, such as labour quality and degree of foreign ownership. For many industries, we found estimation results that underline the importance of building absorptive capacity for assimilating knowledge from best-practice firms that operate similar technology. In a quantitative sense, however, these effects were often dwarfed by the second kind of assimilation effects. These unexplained assimilation effects were very big and dominated the composite effect. The importance of unexplained assimilation is worrisome on the one hand, in the sense that we cannot explain much. On the other hand, it confirms our feeling that much heterogeneity of plants is not captured by survey-based datasets. Our absorptive capacity indicators are rough ones, and are subject to considerable measurement error. In our view, more indepth case studies (like Pack, 1987, and Van Dijk, 2005) offer much better opportunities for assessing the importance of foreign technology and differences in absorptive capacity. Studies like ours can play a useful role in investigating to what extent case-study results can be generalised. In that sense, the next steps along the lines of the present chapter could be to probe further into the inter-industry differences concerning the estimation results and to link such differences to differences in the types of technologies used. Industrial taxonomies, for example, as proposed by Pavitt (1984) might constitute an interesting and worthwhile starting point in this respect. NOTES * A first version of this chapter was presented at the European Meeting on Applied Evolutionary Economics 2005 (Utrecht, The Netherlands). Useful comments by participants and, in particular, Koen Frenken and Fabio Montobbio are gratefully acknowledged. 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 technologyassimilation model of Nelson and Pack (1999).

158 142 Industrial dynamics 4. Pack (1987) studied the performance of textile plants in Kenya, the Philippines and the UK. Van Dijk (2005) focused on the productivity levels of paper-making plants in Indonesia and Finland. 5. Basu and Weil (1998) illustrate this concept by arguing that new knowledge pertaining to the very capital-intensive maglev trains in Japan will not be useful to transporters in Bangladesh using very capital-extensive bullock cart technologies. 6. Atkinson and Stiglitz (1969) introduced the concept of localised learning by doing by which they suggested that firms improve the productivity of a particular mix of capital and labour over time. Basu and Weil (1998) extended this notion by emphasising the importance of localized knowledge spillovers. 7. Below, we shall argue that our estimation framework allows us to decompose assimilation effects into explained assimilation (explained by means of absorptive capacity indicators) 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 professional associations and so on. To investigate the learning effect of scale, we include the variable plant size in our inefficiency function. 10. We shall limit our analysis to the period due to the better quality of the data since 1985 and to the crisis of This approach dates back to the seminal neoclassical macroeconomic growth accounting 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-level variables (see Section 4). 13. It should be noted, however, that this result might be partly due to the fact that new equipment 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 decomposition equation were transformed, by taking their logarithms, into additive components. REFERENCES Abramovitz, M. (1989), Thinking About Growth, Cambridge: Cambridge University Press. Atkinson, A.B. and Stiglitz, J.E. (1969), A new view of technological change, Economic Journal, 79, Basu, S. and Weil, D.N. (1998), Appropriate technology and growth, Quarterly Journal of Economics, 113, Battese, G.E. and Coelli, T.J. (1995), A model for technical inefficiency effects in a stochastic frontier production function for panel data, Empirical Economics, 20, Coelli, T.J. (1996), A guide to FRONTIER Version 4.1: a computer program for stochastic frontier production and cost function estimation, CEPA (Centre for Efficiency and Productivity Analysis) Working Paper, No. 7, University of New England, Department of Econometrics, University of New England, Armidale. Coelli, T.J., Rao, D.S.P. and Battese, G.E. (1998), An Introduction to Efficiency and Productivity Analysis, Boston, MA: Kluwer Academic. Cohen, W. and Levinthal, D. (1990), Absorptive capacity: a new perspective on learning and innovation, Administrative Science Quarterly, 35, Evenson, R. and Westphal, L. (1995), Technological change and technology strat-

159 Absorptive capacity and foreign spillovers 143 egy, in T.N. Srinivasan and J. Behrman (eds), Handbook of Development Economics, Vol. 3, Amsterdam: North-Holland, Hill, H. (1996), The Indonesian Economy since 1966: Southeast Asia s Emerging Giant, Cambridge: Cambridge University Press. Jacob, J. and Los, B. (2005), The impact of international technology spillover and absorptive capacity on productivity growth in Indonesian manufacturing firms, Paper presented at the European Meeting on Applied Evolutionary Economics (EMAEE), Utrecht, May Jorgenson, D.W. and Griliches, Z. (1967), The explanation of productivity change, Review of Economic Studies, 34, Keller, W. (2004), International technology diffusion, Journal of Economic Literature, 62, Klepper, S. (2002), The capabilities of new firms and the evolution of the US automobile industry, Industrial and Corporate Change, 11(4), Kumar, S. and Russell, R.R. (2002), Technological change, technological catch-up and capital deepening: relative contributions to growth and convergence, American Economic Review, 92, 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), International R&D spillovers: a comment, European Economic Review, 42(8), Los, B. and Timmer, M. (2005), The appropriate technology explanation of productivity growth differentials: an empirical approach, Journal of Development Economics, 77, Marsili, O. (2001), The Anatomy and Evolution of Industries: Technological Change and 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, Pack, H. (1987), Productivity, Technology, and Industrial Development: A Case Study in Textiles, Oxford and New York: Oxford University Press. Pavitt, K. (1984), Sectoral patterns of technical change: towards a taxonomy and a theory, Research Policy, 13, Pitt, M.M. and Lee, L.F. (1981), The measurement and sources of technical efficiency in the Indonesian weaving industry, Journal of Development Economics, 9, 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, 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 technology flow matrices, Weltwirtschaftliches Archiv, 133, Wang, H.-J. (2003), A stochastic frontier analysis of financing constraints on investment: the case of financial liberalization in Taiwan, Journal of Business and Economic Statistics, 21,

160 144 Industrial dynamics APPENDIX 6A Table 6A.1 Industrial classification No. Industry ISIC Rev. 2 1 Macaroni, spaghetti, noodles etc Bakery products Dried tobacco and processed tobacco Clove cigarettes Weaving mills except gunny and other sacks Made-up textiles except wearing apparel Knitting mills Wearing apparel made of textiles (garments) Sawmills Plywood Furniture and fixtures, mainly wood Printing, publishing and allied industries Paints, varnishes and lacquers Drugs and medicines Crumb rubber Plastics, bags, containers Clay tiles 36422

161 PART III Network Analysis

162

163 7. Informational complexity and the flow of knowledge across social boundaries Olav Sorenson, Jan W. Rivkin and Lee Fleming 1. INTRODUCTION Scholars from a variety of backgrounds economists, sociologists, strategists and students of technology management have sought a better understanding of why some knowledge disperses widely while other knowledge does not. In this quest, some researchers have focused on the characteristics of the knowledge itself (for example, Polanyi, 1966; Reed and DeFillippi, 1990; Zander and Kogut, 1995) while others have emphasized the social networks that constrain and enable the flow of knowledge (for example, Coleman et al., 1957; Davis and Greve, 1997). This chapter examines the interplay between these two factors. Specifically, we consider how the complexity of knowledge and the density of social relations jointly influence the movement of knowledge. Imagine a social network composed of patches of dense connections with sparse interstices between them. The dense patches might reflect firms, for instance, or geographic regions or technical communities. When does knowledge diffuse within these dense patches circumscribed by social boundaries but not beyond them? Synthesizing social network theory with a view of knowledge transfer as a search process, we argue that knowledge inequality across social boundaries should reach its peak when the underlying knowledge is of moderate complexity. 1 To test this hypothesis, we analyse patent data and compare citation rates across three types of social boundaries: within versus outside the firm, geographically near to versus far from the inventor, and internal versus external to the technological class. In all three cases, the disparity in knowledge diffusion across these borders is greatest for knowledge of an intermediate level of complexity. 147

164 148 Network analysis 2. THE TRANSFER OF COMPLEX KNOWLEDGE Our hypotheses build on two themes in the literature. First, the transfer of knowledge from one party to another typically requires effort on the part of the recipient to fill gaps in the transmitted knowledge and to correct transmission errors. The acquisition of knowledge therefore is best seen not as the receipt of a complete, well-packaged gift, but rather as a search process. Second, social networks, and consequently the social boundaries that 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 ingredients might include both physical components and processes. The recipe further explains how to combine these components and processes in what order, in which proportions, under what circumstances to achieve a desired end. Viewing knowledge as a recipe leads naturally to thinking of innovation as a search for new recipes. Following a long tradition beginning at least as early as Schumpeter (1939) we explicitly model innovation as a search process; inventors explore the space of possible combinations of ingredients (that is, recipes) for new and better alternatives. In discussing this process, we adopt the idea of a fitness landscape as a metaphor for the characteristics of the search space. Innovators search these landscapes for peaks and plateaus, which correspond to good recipes, useful inventions and profitable strategies. Once a useful innovation has been discovered, transferring its recipe, even between cooperative actors, can fail for at least two reasons. First, the recipient usually does not fully understand the original recipe, as a result of imperfections in the transfer process. He/she must therefore begin a search for the missing information and to correct the errors in his/her (imperfect) copy of the recipe. Second, the local ingredients and the experience of the recipient rarely match those of the sender; recipients may therefore need to adapt the original recipe to their own context. Knowledge recipients do not act as passive beneficiaries; they actively search, recreate and build upon the original recipes. In this process, the transfer of certain types of recipes is particularly difficult. For instance, knowledge characterized by causal ambiguity (Lippman and Rumelt, 1982), a high degree of tacitness (Polanyi, 1966), or difficult codification (Zander and Kogut, 1995) may resist transfer because any communication of such recipes proceeds only with many and large gaps. Our focus, however, is on the informational complexity of transferred

165 Informational complexity and the flow of knowledge across boundaries 149 knowledge. We consider a piece of knowledge complex if it comprises many elements that interact richly (Simon, 1962), and we pay special attention to the intensity of interdependence among the ingredients in a recipe. To connect the degree of informational complexity to the characteristics of 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 these components interact in determining the fitness of a particular configuration of components. In our context, N is the number of ingredients used in a recipe, and K is the richness of the interactions between those ingredients. To understand better the way in which the model relates interdependence to the search process, consider two examples with N 3. Figure 7.1 depicts a fitness landscape for a recipe with no interdependence between its components. Each vertex represents a different potential configuration; the arrows connecting them show paths towards higher fitness levels. When K 0, Kauffman randomly assigns a fitness from the uniform unit distribution to each value (0 or 1) of each element. The overall fitness value for a particular configuration is the average of each element s fitness contribution. As one can see, 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. The value of the fitness contribution for each component then depends not just on its value but also on the values of two other components. Each component therefore can contribute any of eight (2 2 2) different fitness levels. Kauffman again randomly assigns these values from the uniform unit 123 w 1 w 2 w 3 W N w i i 1 N 100 (0.40) 000 (0.53) 010 (0.57) 110 (0.43) 101 (0.63) 001 (0.77) 111 (0.67) 011 (0.80) 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 a uniform [0,1] distribution. Figure 7.1 Landscape without interdependence (N 3, K 0)

166 150 Network analysis 123 w 1 w 2 w 3 W N w i i N 100 (0.37) 000 (0.63) 010 (0.40) 110 (0.40) 001 (0.30) 101 (0.87) 111 (0.63) 011 (0.47) 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) distribution. Even in this simple example, one can see that interdependence complicates search; depending on where one begins, an agent using a simple hill-climbing algorithm could arrive at either the global maximum (101) or a local one (000). Complex knowledge resists transfer by making it difficult for a recipient to fill transmission gaps. On the landscapes depicted, a gap is equivalent to not knowing the correct value for the global optimum of one of the three elements. Interdependence produces two effects that undermine the recipient s attempts to regenerate the original recipe (that is, identify the optimum). First, small errors in transmission cause large problems when ingredients cross-couple in a rich manner. Second, interdependence leads to a proliferation of local peaks. These peaks undermine improvement through incremental search because changing any single element degrades the quality of the outcome (Kauffman, 1993). As a result, searchers frequently 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 crucially on access to the original success, which serves as a template (Nelson and Winter, 1982: ; Winter, 1995). For reasons considered below, individuals vary in their degree of access to the template. Superior access facilitates the receipt of knowledge by allowing the recipient to commence search with fewer errors and by permitting him/her to solicit advice from

167 Informational complexity and the flow of knowledge across boundaries 151 the source during the search process. Consider two actors, both attempting to assimilate a valuable piece of knowledge but who differ in their access to the template. The first has superior, though still imperfect, access to and understanding of the original, successful recipe. The second has far poorer access. How valuable is the first actor s superior access to the template during the search process? We contend that the value depends on the complexity of the knowledge being transferred. Suppose first that the ingredients used in the recipe do not interact (that is, K 0). In this situation, the first actor s access to the template does not produce a persistent advantage. Through routine, incremental search, the second actor can reconstruct the recipe. Few local peaks threaten to trap the poorly informed recipient. As a result, both actors eventually fare equally well; search on the part of a recipient can easily substitute for highfidelity transmission. Next consider knowledge with an intermediate degree of interdependence. Local peaks now appear, but they remain relatively few in number. The wellinformed actor begins his/her search near, but not precisely at, the original optimal set of ingredients. Through incremental search, he/she can find the proper combination of ingredients. The second actor, who begins his/her search farther from the target and receives less guidance about the direction in which to explore, more likely becomes ensnared on some local peak, away from and inferior to the original success. Here superior template access gives the 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 incremental search will likely build on the original knowledge with any success. The first actor s superior access to the template thus has little value beyond the second s inferior access. Taken together, these arguments imply that the advantage of superior but imperfect access to the template reaches its peak at moderate levels of interdependence between knowledge components (Rivkin, 2001, develops this 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 not link actors at random. Rather, sociologists have consistently noted and demonstrated that networks concentrate within the boundaries of communities and organizations. Our study tests the salience of three types of

168 152 Network analysis social boundaries organizational memberships, geographic regions and technological communities in structuring social networks, and concomitantly influencing the flow of knowledge. Consider organizational boundaries first. A firm attempting to replicate and build on its own prior success has better access to its knowledge than would an outside imitator, both because fellow members of an organization share codes, specialized languages and beliefs that facilitate high-fidelity transmission (Arrow, 1974) and because strong interpersonal ties and dense social networks inside a firm provide access to the template (Granovetter, 1985). As argued above, the value of this access peaks for transmitting knowledge of intermediate complexity: Hypothesis 1 The advantage in receiving and applying knowledge that members of the same firm have over members of different firms reaches its 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, country or state) as the innovator also have superior access to the template. The geographic concentration of social relations reflects a variety of factors: the greater 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 knowledge have better access to it: Hypothesis 2 A nearby knowledge recipient s advantage in receiving and applying knowledge over a distant recipient peaks for knowledge of intermediate interdependence. To the extent that networks localize geographically, even within a firm, organizations find it difficult to diffuse knowledge beyond its point of origin. Within a firm, then, we expect simple knowledge to spread broadly and highly complex knowledge to remain isolated within a single team or department. Knowledge of moderate complexity, however, should spread within a firm to the edges of a facility or locale, but not to geographically distant installations: Hypothesis 3 Within a firm, a nearby knowledge recipient s advantage in receiving and applying knowledge over a distant recipient reaches its maximum for knowledge of intermediate interdependence.

169 Informational complexity and the flow of knowledge across boundaries 153 An analogous argument applies to technological communities (also called communities of practice, defined in terms of cognitive proximity). Actors who work in the same technological domain as an inventor have superior access to the template. Universities, trade associations, professional societies, industry consortia and work experience foster dense social connections within such technological communities. These communities also develop common knowledge and communal languages that can facilitate knowledge transfer. Membership within a common technological community thus engenders superior access to the template, which should have its greatest impact when the target knowledge displays moderate interdependence: Hypothesis 4 The advantage in receiving and applying knowledge that a member of a technological community has over a non-member of the community reaches its maximum with knowledge of intermediate interdependence. 3. EMPIRICAL CORROBORATION To test these hypotheses, we analysed prior art citations to all US utility patents granted in May and June of 1990 (n ). The data came from 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 citation as evidence of knowledge diffusion. Our statistical approach involves estimating the likelihood that a focal patent receives a citation from a future patent as a function of several factors: the interdependence of the knowledge underlying the focal patent, the status of the citing patent s inventor as an insider or outsider on some dimension with respect to the focal patent, the interaction of interdependence and insider/outsider status, and a set of control variables. The results of the estimation allow us to examine how the likelihood of insider citation compares to the likelihood of outsider citation and, crucially, whether the gap between the two probabilities peaks when the focal patent embodies moderately interdependent knowledge. Our unit of analysis is a patent dyad, one patent issued in May or June of 1990 and one issued later that may or may not cite the first. Hence our approach conceptually follows that of other studies of the likelihood of tie formation in this case, the likelihood that a future patent builds on the knowledge embodied in one of our focal patents. Specifically, our analysis follows Sorenson and Stuart (2001) in adopting a case-control approach to analysing the formation of ties. We begin by including all cases of future

170 154 Network analysis patents, from July 1990 to June 1996, that cite any of our focal patents: in total. Since these citations occur, the dependent variable for these cases takes a value of 1 to denote a realized citation. In addition, we pair each of the focal patents with four future patents that do not cite it (but that could have), with the dependent variable set to zero. From this set of , we restrict our analysis to the dyads in which both inventors reside in the United States, leaving us with a set of dyads. Interdependence For each dyad, we measure the complexity of the knowledge in the focal patent, k,by observing the historical difficulty of recombining the elements that constitute it (Fleming and Sorenson, 2001). Though the metric involves intensive calculation, the intuition behind it is simple: a technology whose components have, in the past, been mixed and matched readily with a wide variety of other components has exhibited few sensitive interdependencies and receives a low value of k. The measure takes the subclasses identified in a patent as proxies for the underlying components (see Fleming and Sorenson, 2004, for survey-based validation of the measure). We compute k in two stages. Equation (7.1) details our calculation of the ease of recombination, or inverse of interdependence, for each subclass i used in patent j.we first identify every use of subclass i on patents from 1980 to The denominator is simply the tally of the number of patents with a classification in subclass i. To compute the numerator, we count the number of different subclasses appearing with subclass i on previous patents. Hence, our measure increases as a particular subclass combines with a wider variety of technologies, controlling for the total number of applications. This term captures the ease of combining a particular technology. Ease of recombination of subclass i E i Count of subclasses previously combined with subclass i. Count of previous patents in subclass i (7.1) To create our measure of interdependence for an entire patent, we invert the average of the ease of recombination scores for the subclasses to which it belongs (equation 7.2). Count of subclasses on patent j Interdependence of patent j K j. E i j i (7.2)

171 Informational complexity and the flow of knowledge across boundaries 155 Social Boundaries Three variables capture the insider/outsider status of the potential citing inventor with respect to the holder of a focal patent. The variables reflect membership in organizational, regional and technical communities. Same assignee is set to one if two patents in a dyad share a common owner and is zero otherwise. Geographic proximity is equal to the natural log of the distance in miles between the first inventors listed on the two patents in a dyad multiplied by negative one (so that we expect larger values to increase the likelihood of citation). Same class is set to one if two patents belong to the same primary technological class a proxy for shared membership in a community and is zero otherwise. In all three cases, we test our hypotheses by interacting k and its square with the proxy for the density of social networks whether due to firm boundaries, geographic proximity, or technological similarity. The benefits of social proximity should peak for inventions of moderate complexity. The regressions also include several control variables. Subclass overlap is the number of subclasses that the two patents in the dyad have in common divided by the number of subclass memberships for the (potentially) citing patent. An activity control estimates the typical number of citations received by a patent in the same technological areas as the focal patent (see Fleming and Sorenson, 2001). Recent technology is the average reference number of the patents listed as prior art, a measure of the closeness of the patent to the technological frontier. We also include counts of the number of backward patent citations and backward non-patent citations, the number of class memberships, and the number of subclass memberships for the focal patent. We report robust standard errors and correct for potential bias in logistic regression of rare events (King and Zeng, 2001). 4. RESULTS The results appear in Table 7.1. Model 1 tests hypothesis 1 by interacting k and k 2 with same assignee. As expected, membership within the same firm produces the greatest diffusion advantage over outsiders for knowledge of intermediate complexity, as evidenced by the positive coefficient on k same assignee and the negative coefficient on k 2 same assignee. The interactions between geographic distance and interdependence in model 2 tests hypothesis 2, again showing strong support. Model 3 tests hypothesis 3 by re-estimating model 2, but only for dyads where both patents belong to the same firm. In essence it asks: does geography still matter for

172 156 Network analysis Table 7.1 Rare events logit models of the likelihood of a focal patent receiving a citation from a future patent Model 1 Model 2 Model 3 Model 4 Model 5 Only same assignee k (0.302) (0.307) (0.359) (0.321) (0.362) k (0.082) (0.074) (0.117) (0.098) (0.107) k same assignee (0.515) (0.555) k 2 same assignee (0.203) (0.273) k ln (dist) (0.131) (0.213) (0.139) k 2 ln (dist) (0.044) (0.074) (0.066) k same class (1.146) (1.032) k 2 same class (0.363) (0.330) Same assignee (0.280) (0.276) (0.281) (0.278) ln (dist) (0.031) (0.031) (0.066) (0.030) (0.029) Same class (0.306) (0.307) (0.394) (0.302) (0.299) Subclass overlap (0.316) (0.317) (0.349) (0.316) (0.314) Activity control (0.287) (0.287) (0.248) (0.388) (0.289) Recent technology (0.171) (0.170) (0.309) (0.165) (0.151) Backward patent citations (0.013) (0.013) (0.008) (0.014) (0.014) Backward non patent (0.006) (0.006) (0.037) (0.005) (0.005) citations Number of classes (0.140) (0.140) (0.249) (0.138) (0.137) Number of subclasses (0.045) (0.045) (0.048) (0.044) (0.044) Constant (0.714) (0.703) (1.208) (0.684) (0.675) Log-likelihood N Note: p 0.05; p 0.01; p Model 3 includes only those dyads for which same assignee 1.

173 Informational complexity and the flow of knowledge across boundaries 157 knowledge diffusion within firms? In support of hypothesis 3, the results reveal that even within firm boundaries, social networks influence the flow of knowledge, with the greatest disparity between local diffusion and distant diffusion arising for knowledge of moderate interdependence. Model 4 tests the salience of technological communities. Once again, the estimates show strong support; the impact of technological community membership on citation probability peaks for intermediate k. Model 5 includes all three measures of social proximity simultaneously and shows that each has an independent and significant effect when estimated together, in support of hypotheses 1, 2 and 4. As expected, the value of superior access to the template reaches a maximum for knowledge of moderate interdependence, regardless of whether the superior access comes from organizational membership, geographic proximity, or technological community membership. In addition to being significant, the effects have substantial economic import. For simple or highly complex knowledge, the insider has no greater likelihood than the outsider of attaining and building on the knowledge in a focal patent. For knowledge of moderate complexity at the gap-maximizing levels of k, a firm insider is 218 per cent more likely than an outsider to transfer knowledge effectively; an inventor located in the same zip code is 160 per cent more likely to absorb the knowledge in a region than one at the average distance ( 665 miles); and a technological insider is 238 per cent as likely as a technological outsider to build on knowledge in the class. 5. DISCUSSION Our analyses considered the impact of superior access to some original knowledge on the likelihood of diffusion as a function of knowledge complexity, 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 to both classes of would-be recipients. For knowledge whose ingredients display a moderate degree of interdependence, however, superior but imperfect access to the template translates into better knowledge reproduction. Thus in our patent data, the largest gap between the ability of a close recipient to build on prior knowledge relative to the ability of a distant recipient arises when the cited patent involves moderate interdependence. Our findings speak to the question, when does inequality of knowledge arise across social borders? One might initially suspect that highly complex knowledge, the most difficult to reproduce, would create the greatest

174 158 Network analysis inequality. But this intuition ignores the fact that inequality in its sharpest form requires some diffusion: to create the most inequity across social boundaries, knowledge must creep up to the edge of the thick patch of connections in which it originated but not beyond. This phenomenon, we have argued, most likely occurs for moderately complex knowledge. Thus, for example, one might expect industries based on moderately complex knowledge to display especially wide intra-industry dispersion in long-run financial returns. Our argument may also shed light on a conundrum of the literature on economic geography. Explanations for agglomeration based on information spillovers assume that membership in a local community allows firms to benefit from the knowledge developed by other firms in the region, but that firms outside the region are excluded from these benefits (for example, Marshall, 1890; Arrow, 1962). What type of knowledge would have such a characteristic? The literature to date has focused on tacit knowledge, but typically uses the term simply to refer to uncodified (as opposed to uncodifiable) knowledge (an endogenous outcome of firms decisions to invest in codification; Brökel, 2005). Our results, building on Kauffman s NK model, point to a different (presumably more exogenous) dimension: informational complexity. Industries that rely on moderately complex knowledge might be especially likely to display geographic agglomeration (for empirical corroboration, see Sorenson, 2004). Our empirical results come from patent data alone, but the basic logic of our hypotheses applies to knowledge in general, not just the knowledge underlying inventions (see Wolter, 2006, for a model based on interdependence in production). Hence, future research might usefully examine these dynamics across a wide range of applications including organizational learning, the diffusion of management practices, knowledge management, and the sustainability of knowledge-based competitive advantage. NOTE 1. This version describes the intuition underlying our theoretical model and reports novel empirical results. For those interested in a more detailed description of the theoretical model, see Rivkin (2001) and Sorenson et al. (2004). REFERENCES Arrow, Kenneth J. (1962), The economic implications of learning by doing, Review of Economic Studies, 29:

175 Informational complexity and the flow of knowledge across boundaries 159 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: 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: Davis, Gerald F. and Henrich R. Greve (1997), Corporate elite networks and governance changes in the 1980s, American Journal of Sociology, 103: 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: Fleming, Lee and Olav Sorenson (2004), Science as a map in technological search, Strategic Management Journal, 25: Frenken, Koen and Alessandro Nuvolari (2004), The early development of the steam engine: an evolutionary interpretation using complexity theory, Industrial and Corporate Change, 13: Granovetter, Mark S. (1985), Economic action and social structure: the problem of embeddedness, American Journal of Sociology, 91: Hägerstrand, Torsten ([1953] 1967), Innovation Diffusion as a Spatial Process, Chicago: University of Chicago. Hall, Bronwyn H., Adam B. Jaffe and Manuel Trajtenberg (2001), The NBER 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: Oxford University. King, Gary and Langche Zeng (2001), Logistic regression in rare events data, Political Analysis, 9: Lippman, Steve and Richard Rumelt (1982), Uncertain imitability: an analysis of interfirm differences in efficiency under competition, Bell Journal of Economics, 13: 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 imitation, and sustainable competitive advantage, Academy of Management Review, 15: Rivkin, Jan W. (2001), Reproducing knowledge: replication without imitation at moderate complexity, Organization Science, 12: Schumpeter, Joseph (1939), Business Cycles, New York: McGraw-Hill. Simon, Herbert A. (1962), The architecture of complexity, Proceedings of the American Philosophical Association, 106: Sorenson, Olav (2004), Social networks, informational complexity and industrial geography, in D. Fornahl, C. Zellner and D. Audretsch (eds), The Role of Labour Mobility and Informal Networks for Knowledge Transfer, Berlin: Springer-Verlag, pp

176 160 Network analysis Sorenson, Olav, Jan W. Rivkin and Lee Fleming (2004), Complexity, networks and knowledge flow, Harvard Business School Working Paper No , Cambridge, MA. Sorenson, Olav and Toby E. Stuart (2001), Syndication networks and the spatial diffusion of venture capital investments, American Journal of Sociology, 106: Winter, Sidney G. (1995), Four Rs of profitability: rents, resources, routines, and replication, in C. Montgomery (ed.), Resource-based and Evolutionary Theories of the Firm: Towards a Synthesis, Boston, MA: Kluwer, pp Wolter, Kerstin (2006), Divide and conquer? The role of governance for the adaptability of industrial districts, Advances in Complex Systems, forthcoming. Zander, Udo and Bruce Kogut (1995), Knowledge and the speed of transfer and imitation of organizational capabilities: an empirical test, Organization Science, 6: Zipf, George K. (1949), Human Behavior and the Principle of Least Effort, Reading, MA: Addison-Wesley.

177 8. Networks and heterogeneous performance of cluster firms Elisa Giuliani* 1. INTRODUCTION This chapter explores the relationship existing among the heterogeneous nature of firms in industrial clusters, their structural position in local networks and their performance. Following the rising interest for spatially agglomerated 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 empirically that the performance of firms in clusters is related to firm-level knowledge endowments and their position in the knowledge network. A starting argument of this chapter is that of questioning the widely accepted view that knowledge is diffused in clusters in a rather pervasive and unstructured way, and that this is what affects the enhanced performance of cluster firms as compared to isolated ones. Most economists and economic geographers share this view. On the one hand, in fact, economists stress the public nature of knowledge (Arrow, 1962) and argue that geography facilitates inter-firm learning and innovation because of localized knowledge spillovers (for example, Jaffe et al., 1993); on the other, recent work done by economic geographers argues that it is not geography per se that matters for innovation, but it is a common institutional endowment and firms relational proximity (later defined), which facilitate the diffusion of knowledge and enhance collective learning in clusters (for example, Maskell and Malmberg, 1999; Capello and Faggian, 2005). A reason for this is the often presumed co-occurrence of firms business interactions and knowledge flows 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 perform differently (Lazerson and Lorenzoni, 1999; Rabellotti and Schmitz, 1999; Camison, 2004; Molina-Morales and Martinez-Fernandez, 2004; Zaheer and Bell, 2005). In line with this, some have expressed their conceptual 161

178 162 Network analysis discontent about the pervasive and unstructured view of clusters innovation (see, for example, Breschi and Lissoni, 2001), and others have examined how key notions of evolutionary economics may be incorporated into economic geography (Boschma and Lambooy, 1999; Boschma and Frenken, 2006). In this vein, some have pointed out the need to understand the heterogeneity of cluster firms performance and the characteristics of a cluster 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 wine clusters Colline Pisane and Bolgheri/Val di Cornia in Italy and Colchagua Valley in Chile. It shows that firms perform differently within clusters and that such differences are due to both firm knowledge bases and to their degree of embeddedness in the local knowledge network. In contrast, interfirm relational proximity is a less powerful factor affecting firm performance. The chapter concludes by drawing implications for the concept of clusters in economic geography. 2. GEOGRAPHY, RELATIONAL PROXIMITY AND THE DIFFUSION OF KNOWLEDGE: IMPLICATIONS FOR FIRM PERFORMANCE The process of knowledge diffusion and generation in clusters of firms has traditionally been based on different reinterpretations of the Marshallian, externality-driven, world of industrial districts. Several empirical studies have 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, and children learn many of them, unconsciously.... Good work is rightly appreciated, inventions and improvements in machinery, in processes and the general organisation of the business have their merits promptly discussed: if one man starts a new idea, it is taken up by others and combined with suggestions of their own; thus it becomes the source of further new ideas. (Marshall, 1920, p. 225; emphasis added) He envisaged two mechanisms by which knowledge spillovers were generated, the first one was through the embodied capabilities acquired by workers being part of the district and the second one was the sharing of ideas among businessmen, a process which Allen (1983) has defined collective invention (see also Nuvolari, 2004). In Industry and Trade (1919), Marshall defined this as industrial atmosphere, highlighting its sticky nature:

179 Networks and heterogeneous performance of cluster firms 163 But an industry which does not use massive material, and needs skill that cannot be 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 not easily 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 idiosyncratic meso characteristic of districts. Several scholars have advanced in this field and have elaborated on the original Marshallian ideas. My argument here is that most of the studies undertaken in this direction have taken a meso perspective to analyse learning, innovation and performance in clustered firms. As such, they have given less emphasis to the micro, and to how the micro can affect the meso. Among these studies, I shall consider here only the most influential contributions of both economists and economic geographers. As anticipated in the introduction, the economists view is that knowledge spillovers, which are by definition a public good (Arrow, 1962), tend to be highly localized (Jaffe, 1989; Jaffe et al., 1993), a property that links conceptually geography and innovation. Within this stream of studies, robust empirical evidence has shown that a relationship exists between spatial 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 believe that geography matters for innovation and for competitiveness (for example, OECD, 2001). As an example, in his work on industrial clusters and nations competitive advantage, Porter (1998) connects the processes of learning and innovation in clusters to the Marshallian atmosphere concept, stating that the information flow, visibility, and mutual reinforcement within such a locale give meaning to Alfred Marshall s insightful observation that in some places an industry is in the air (p. 156). He notes moreover that more important, however, is the influence of geographic concentration on improvement and innovation (p. 157), since proximity increases the speed of information flow within the national industry and the rate at which innovations diffuse (p. 157). This view supports the idea that geography matters for innovation and, implicitly, for economic performance. Some economic geographers seem to have moved beyond that. It has been argued that geographic proximity per se is not sufficient to generate learning, and that other forms of proximity are required for inter-firm learning and innovation to occur (Boschma, 2005). Among these, great emphasis is given to the role of social proximity also known as relational proximity (for example, Maskell and Malmberg, 1999; Amin and Cohendet, 2004).

180 164 Network analysis Industrial clusters, being a spatially localized set of economic activities, are in fact envisaged as embedded economies (Granovetter, 1985) where social relationships, such as friendship and kinship, are entangled with productive ones. More specifically, relational proximity is believed to favour the formation of relational capital, defined as a sort of productive thickening based on market and cooperative inter-firm relationships (Scott, 1998). The relational capital, favouring the interaction of productive agents and the diffusion of tacit knowledge (Howells, 2002) is finally said to be the substratum of collective learning (Capello, 1999). Relational proximity and embeddedness are thus considered to operate at the meso level and perform several functions in the context of innovation (Oerlemans and Meeus, 2005), favouring firm performance accordingly. In a recent study on several district areas in Italy, Capello and Faggian (2005) find support for the importance of relational capital in fostering the innovative performance of firms. They therefore argue that [R]egional economists are... correct in underlining that not only are intra-firm characteristics crucial for innovation, but also (and maybe most of all) that the location of firms in an area where the local labour market and the tight links with 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 on explaining why firms that are part of an industrial cluster tend to perform better than isolated ones. An underlying, implicit assumption, is that firms that are part of the same industrial cluster will benefit more or less equally from the presence of external economies at the local level (Marshall, 1920) and, more specifically, from a common geographical, sectoral and relational proximity. However, an increasing number of studies have recently started to highlight that, in spite of a general homogeneity of meso conditions in clusters, firms perform differently (Lazerson and Lorenzoni, 1999; Rabellotti and Schmitz, 1999; Camison, 2004; Molina-Morales and Martinez-Fernandez, 2004; Zaheer and Bell, 2005). 3. WHAT AFFECTS HETEROGENEOUS PERFORMANCE IN CLUSTER FIRMS? Understanding the factors that lead to heterogeneous firm performance in industrial clusters requires that the focus of analysis shifts from the meso to the micro. As suggested by Lazerson and Lorenzoni (1999) and more recently by Maskell (2001b), individual firms are the key actors in the development of territorial clusters. In line with this, my argument here is that it

181 Networks and heterogeneous performance of cluster firms 165 is the characteristics of firms, and their inherent heterogeneity, that generate (or inhibit) the conditions at the meso level that ultimately enhance cluster firm performance. Thus, it is a micro to meso perspective that this chapter takes, rather than the meso to micro one commonly found in the cluster literature. Accordingly, the performance of firms should be explored by considering the interplay between their internal resources and the external, meso conditions present in the cluster. The relationship between firms internal resources and performance has already 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 heterogeneous knowledge bases. By knowledge base I mean here the set of information inputs, knowledge and capabilities that inventors draw on when looking for innovative solutions (Dosi, 1988, p. 1126). Knowledge is seen as residing in firms skilled knowledge workers, who embody tacit capabilities. At the same time, knowledge is not merely the sum of each individual 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, as an organized entity, and it is not reducible to what any single individual knows, or even to any simple aggregation of the various competences and capabilities of all the various individuals, equipments, and installations of the firm. The knowledge base is, moreover, considered here as the result of a process of cumulative learning, which is inherently imperfect, complex and path dependent (Dosi, 1997) and which delivers persistent heterogeneity between the firms in the economic system and, understandably, in a cluster. Such heterogeneity, in turn, deepens the uniqueness of resources deployed by firms and explains different growth rates and performance (Penrose, 1959). It is reasonable, moreover, that firms that have stronger knowledge bases will perform better than others in the cluster, as they will have easier access to external knowledge and rejuvenate their internal capabilities accordingly (Cohen and Levinthal, 1990). This chapter will explore the following research question: how does the heterogeneity in firm knowledge bases relate to their performance? The important issue here is, however, not simply whether micro-level conditions affect performance, but what is their interaction with the external environment in the cluster. Innovation rarely occurs in isolation, and, as emphasized by most of the cluster literature, the degree to which firms are embedded in local networks influences their performance (Molina-Morales and Martinez-Fernandez, 2004; Capello and Faggian, 2005; Zaheer and Bell, 2005). Being relationally embedded in a cluster means that firms interact

182 166 Network analysis frequently on business-related matters. For example, if entrepreneurs are members of the same local consortium they will meet at local events and discuss their productive activities. Similarly, if two firms share machinery, their technical employees will meet and discuss their appropriate use. All these interactions generate a trustworthy environment in the cluster, which may facilitate the sharing of information and knowledge, thus enhancing the overall firm capabilities to innovate. In a previous study on wine clusters (Giuliani, 2006), I have shown that business interactions of this type occur in a pervasive way, resembling what Marshall called industrial atmosphere. This means that firms show a rather homogeneous behaviour in interacting with 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 of business interactions, their performance will be homogeneously distributed. The interaction for business-related matters is only one of the several informal networks formed by firms in clusters (Boschma, 2005). In fact, different types of networks are likely to carry different informational content and they may affect firm performance differently (Gulati, 1998; Rodan and Galunic, 2004). In Giuliani (2006), I disentangle the knowledge network, based on the transfer of knowledge for the solution of technical problems, from the overall network of business interactions (Figure 8.1). The structural properties of the knowledge network suggest that it is built on a more selective basis, if compared to the network of business interactions. This means that knowledge diffuses in clusters in a less pervasive and serendipitous way than is commonly envisaged by the economists and economic geographers. This property may have important implications on the distribution of firm performance in clusters. Given its selectivity, if firm performance is affected more by the degree of embeddedness in the knowledge network than in the network of business interactions, it is reasonable to 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 three wine clusters in Italy and Chile, namely: Colline Pisane (CP), Bolgheri/Val di Cornia (BVC) and Colchagua Valley (CV). The analysis has required careful data collection through interviews. Interviews were carried out with the skilled workers (that is, oenologists or agronomists) and the survey was directed to producers of fine wines. This analysis includes only horizontal relationships among firms that operate as wine producers, whereas vertical

183 Networks and heterogeneous performance of cluster firms 167 (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 Note: BI stands for business interaction ; KN for knowledge ; CP for the Colline Pisane cluster; BVC for the Bolgheri/Val di Cornia cluster; CV for the Colchagua Valley cluster. Source: Giuliani (2006) (based on UCINET-Netdraw; Borgatti et al., 2002). Figure 8.1 Types of networks

184 168 Network analysis Table 8.1 Firm characteristics by cluster Characteristics of firms by: Cluster CP CV BVC (N 32) (N 32) (N 41) (a) Size (employees) Small (1 19) Medium (20 99) Large ( 100) (b) Ownership Domestic Foreign (c) Organization structure 1. Part of a group, vertically integrated firms 2. Part of a group, 13 vertically disintegrated firms 3. Independent, vertically integrated 4. Other (e.g., cooperatives) 9 (d) Year of localization Up to 1970s s s s Note: The numbers refer to percentages within the respective cluster. linkages are not explored here. Data were gathered using the universe of fine wine producers populating the three clusters, 1 32 in CP, 41 in BVC and 32 in CV, making a total of 105 firms. Further information about the population of firms is reported in Table 8.1. Apart from general background and contextual information, the interviews were designed to obtain information that would permit the development of quantitative indicators in three key areas: (i) the knowledge base of firms; (ii) the degree of embeddedness of firms in the network of business interactions; and (iii) the degree of embeddedness of firms in the network of knowledge. These are summarized in Table 8.2. Econometric Estimation The analysis is based on an econometric estimation, using a Probit model with marginal effects. The estimations are carried out on the aggregate data

185 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 with training, human resources and R&D. Correspondingly, the structured interviews sought detailed information about: (i) the number of technically qualified personnel 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 been employed (number of firms), and (iv) the intensity and nature of the firms experimentation activities (experimentation intensity) an appropriate proxy for knowledge creation efforts, since information about expenditure on formal R&D would have been both too narrowly defined and too difficult to obtain systematically. 2. Network of business interactions In the questionnaire-based interview, relational data were collected through a roster recall method: each firm was presented with a complete list (roster) of the other firms in the cluster, and was asked the question reported below: With which of the cluster firms mentioned in the roster do you interact for business 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 was presented a complete list (roster) of the other firms in the cluster, and was asked the questions reported below: i. If you are in a critical situation and need technical advice, to which of the local firms mentioned in the roster do you turn? [Please rate the importance you attach to the knowledge linkage established with each of the firms according to its persistence and quality, on the basis of the following scale: 0 none; 1 low; 2 medium; 3 high]. ii. Which of the following firms do you think have benefited from technical support from this firm? [Please indicate the importance you attach to the knowledge linkage established with each of the firms according to its persistence and quality, on the 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 of 0 to 3. A value of 0 is given when no linkage is formed. A value of 1 corresponds to an occasional knowledge linkage with limited content in terms of quality of the knowledge flow, whereas a value of 3 corresponds to a persistent knowledge linkage that carries finegrained knowledge. It is worth remembering here that this study was designed to collect cross-sectional data only. Therefore, the relational data gathered through the above-mentioned roster studies refer to the very recent past in which the interviews have been carried out (maximum of two years).

186 170 Network analysis obtained by pooling together the three clusters variables. Since data are coming from three different geographical clusters of firms, I applied the Moulton method (Moulton, 1990) to control for the possibility that the random disturbances in the regression are correlated within each cluster. Dependent Variable Performance at the firm level is measured here by an indicator of the quality achieved by each firm s wines. The quality of worldwide wines is annually assessed and rated by international panels of experts and published in several specialized wine journals (for example, Wine Spectator, Decanter, Wine Enthusiast and Robert Parker s Guide). Having a wine rated by any of these international specialized journals is, first, an acknowledgement of the qualitative properties of the wines, and second, a very powerful marketing device for a firm, since experts ratings strongly influence market prices (Nerlove, 1995; Combris et al., 1997, 2000; Landon and Smith, 1997). The performance of a firm is therefore seen here as its capacity to develop new wines, which are valued as quality wines by international experts. The indicator adopted in this chapter is drawn from one of the above international wine journals: Wine Spectator. 2 This journal wine rating is based on the quality assessment of an international panel of expert oenologists, who review more than wines each year in blind tastings. After tasting, oenologists assign a score to each wine brand according to a 100- point scale, ranging from 100, when the wine is of outstanding quality, to 50 when it is of poor quality. 3 Certain information is listed with each rated wine: the wine vintage, the wine area and the market price. The indicator used here (RATING) is valued 1 when any of the firm s wines has been assigned at least 70 points in years , the minimum threshold for a wine to be considered of drinkable quality and to be recommended by the journal. It is valued 0 otherwise. A lag of two years is allowed between the year in which the interviews were carried out and the vintage of the most recent rated wines. It is worth noting that the majority of the wines tasted are submitted to Wine Spectator by the wineries or their US importers. Additionally, the journal spends substantial effort in buying and reviewing wines that are not submitted, at all price levels. Accordingly, a firm s wines may not be rated for three main reasons. First, due to a selection bias, firms with poorquality achievements will have little incentive to send their wines to the journal for assessment. Second, US importers will not recommend and signal wines to Wine Spectator when they consider them of poor quality. Third, Wine Spectator itself selects out all the wineries producing very poor-quality wines. These considerations suggest therefore that firms

187 whose wines are not rated tend to be poor performers. The same applies for firms whose wines are rated but are assigned less than 70 points. Using Wine Spectator as a unique source of information, however, may pose some robustness problems. Even if it is highly unlikely that a successful winery will not be spotted by the journal, it is still possible that some wineries are overlooked. In order to control for that, I correlated RATING and two other indicators drawn from Wine Spectator s ratings the sum of scores per planted vineyard hectares (SSH) and the average price of rated wines 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 perceived by its members. 4 I find strong correlations between the perceived performance and the three Wine Spectator indicators: RATING (0.65**), SSH (0.65**) and PRICE (0.62**). These results suggest that the performance indicator RATING is robust enough to measure the quality of wines achieved by cluster firms. Independent Variables Networks and heterogeneous performance of cluster firms 171 Firm knowledge base (KB) The knowledge base of the firm is measured by extracting a factor from the principal component analysis of the four variables listed in Table 8.2 (Point 1). The factor explains more than 75 per cent of variance and it has been calculated 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 linkages for business matters with other firms in the cluster. The existence of a business interaction is mapped by the question in Table 8.2 (Point 2). The degree centrality of the network of business interactions (BI_DC (j) ) is considered here a proxy of firms the relational embeddedness in the cluster. It is measured by the extent to which an actor j is central in a network on the basis of the ties that it has directly established with other i actors of the network ( (x (ji) )). This measure uses undirected dichotomous data. The value has been normalized by its theoretical maximum (g 1), where g is the number of firms in each cluster. BI_DC ( j) (x (ji)) g 1. Embeddedness in the knowledge network (KN_DC) This variable measures the degree to which a firm is central in the knowledge 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

188 172 Network analysis degree centrality (KN_DC). This measure uses undirected dichotomous data. The value has been normalized by its theoretical maximum (g 1), where g is the number of firms in each cluster: Control Variables KN_DC (j) (x ( ji) ) g 1. I control here for the following firm-level variables that are commonly associated with performance: the size of firms, measured by the log of employees (LEMP), the age of the firm (AGE), measured as the number of years since the start of operations until The ownership (OWN) which is 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 Table 8.1, firms in the clusters have four different types of organizational structures: ORG1 corresponds to firms that are part of a national group and perform all phases of the production process within the cluster; ORG2 refers to firms that are also part of a national group but perform only part of the production process, usually grape-growing, within the cluster; ORG3 refers to firms that are independently owned and that perform all production 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 and Table 8.4 the results of the Probit estimation. Model 1 in Table 8.4 includes only the control variables, showing that only size is positively related to the likelihood of a firm being rated by Wine Spectator and therefore with its performance. Model 2 shows that the value of firm knowledge base is strongly and positively related to performance, a result that is in line with several other recent contributions (for example, Camison, 2004; Zaheer and Bell, 2005). This is explained by the fact that firms that have better-educated or moreexperienced knowledge workers (Drucker, 1993), and that carry higher internal experimentation intensity, are more likely to exploit knowledge for the generation of successful innovations (March, 1991). This in turn drives a firm to achieve higher performances (Wernerfelt, 1984). Model 3 provides support for the view that firms that interact for business matters are more likely to be good performers. This result is consistent with the fact that being co-located in the same industrial cluster, and being

189 Table 8.3 Descriptive statistics and correlation matrix Mean Std dev. RATING LEMP AGE OWNER ORG1 ORG2 ORG3 ORG4 KB BI_DC KN_DC RATING LEMP *** 1.00 AGE OWNER * ORG * 0.34*** *** 1.00 ORG *** 0.23*** ORG *** 0.42*** *** 0.75*** 0.44*** 1.00 ORG *** 1.00 KB *** 0.66*** *** 0.51*** *** BI_DC *** 0.22*** *** 1.00 KN_DC *** 0.43*** *** 0.17* 0.35*** *** 0.55*** 1.00 Note: * Significant at 10%; ** Significant at 5%; *** Significant at 1%. Based on Pearson coefficients. 173

190 174 Network analysis Table 8.4 Probit estimations with marginal effects Dependent Model 1 Model 2 Model 3 Model 4 Model 5 variables Control Knowledge Knowledge Knowledge Knowledge variables base only base with base with base with only BI_DC KN_DC BI_DC and KN_DC df/dx (s.e.) df/dx (s.e.) df/dx (s.e.) df/dx (s.e.) df/dx (s.e.) LEMP (0.052)** (0.070) (0.052) (0.071) (0.770) AGE (0.000)* (0.000)*** (0.000)** (0.000)*** (0.000)*** OWNER (0.099) (0.082) (0.138) (0.084) (0.082) ORG3 a (0.225) (0.335) (0.231) (0.334) (0.335) KB (0.018)*** (9.296)*** (0.017)*** (0.018)*** BI_DC (0.003)** (0.001) KN_DC (0.005)*** (0.006)*** N Log pseudo likelihood Pseudo R Note: * Significant at 10%; ** Significant at 5%; *** Significant at 1%. a ORG1, ORG2 and ORG4 have been dropped due to collinearity. embedded in the local network, may facilitate access to relevant information or may ease inter-firm transactions. In this case, firms may benefit from several types of externalities and enhance their performance accordingly. This evidence seems to support the view that both intra-firm resources and relational proximity matter for firm performance (Capello and Faggian, 2005). However, no evidence is found that the latter matters more than the former. Model 4 finds a strong and positive relationship between the degree of firms embeddedness in the knowledge network and their performance. This is in line with most of organizational sociology s literature (for example, Powell et al., 1996; Smith-Doerr and Powell, 2003), since the access to external sources of knowledge for the solution of internal problems favours innovation and enhances firm performance. The interesting result here is that the coefficient of KN_DC is higher in Model 4 than the

191 Networks and heterogeneous performance of cluster firms 175 coefficient of BI_DC in Model 3. Furthermore, when both BI_DC and KN_DC are considered in the estimation (Model 5), BI_DC ceases to be significant, while both KB and KN_DC persist in being positive and strongly significant. 5 The results of this study indicate that both firm internal capabilities, and the degree to which firms are embedded in local networks, matter for their performance. However, it is relevant to note that it is a specific type of network that affects performance most: the knowledge network. It is reasonable to argue that this is connected to the structural properties of this network. As illustrated by Giuliani (2006), the knowledge network is formed on a selective rather than pervasive, collective basis. This is due to two factors: first, firms with stronger knowledge bases have more to transfer and are understandably more likely to be targeted by other firms for technical advice. Second, firms with stronger knowledge bases will seek technical advice from equally advanced firms, thus targeting firms with strong knowledge bases in their search for external knowledge (Giuliani and Bell, 2005). On the basis of this, communities of knowledge are formed in clusters by a restricted group of equally advanced firms. An implication of selectivity is that the knowledge that is circulated within the community is likely to be of valuable content, which in turn enriches the knowledge base of the member firms and, consequently, their performance. It is therefore reasonable to argue that this evidence casts doubts on the importance of meso-level conditions, such as geographical, sectoral and relational proximity, for performance. More convincingly, it seems that knowledge endowments affect performance both directly and indirectly through the generation of a local knowledge network, which serves to enhance individual firms capabilities. 6. CONCLUSION This chapter has attempted to understand the factors that influence firm performance in industrial clusters. Taking an evolutionary approach to this domain of studies, it has shown that the heterogeneous distribution of firm knowledge bases is related to their performance, both directly and indirectly through the participation in the local knowledge community. Following up on a previous study (Giuliani, 2006), this novel empirical evidence shows that, in spite of pervasive business interactions, the performance of firms is unevenly distributed in the clusters. The econometric estimations provide support for this and show that firm performance in clusters depends on their internal capabilities (that is, knowledge bases) and their capability of being connected to the local knowledge network. More importantly, this empirical evidence seems consistent with the fact that

192 176 Network analysis similar meso characteristics that is, the geographic and relational proximity of firms constitute the substratum neither for collectively shared knowledge flows (Giuliani, 2006). On this basis, two considerations can be raised. First, one should be extremely careful in associating the concept of industrial clusters with enhanced performance and competitiveness, even when firms are geographically and relationally proximate. Instead, more rigorous studies should be carried out in the future that analyse the interplay between firms, the cluster knowledge network, and performance. Second, as recently suggested by Markusen (2003), more rigorous analysis in regional studies will provide better indications for policy makers. Indeed, this study supports the view that cluster performance is more likely to be enhanced by strengthening firms knowledge bases rather than by pooling firms together 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 a previous version of this chapter. Thanks go also to Marcelo Lorca Navarro, Cristian Diaz Bravo, Erica Nardi and Elena Bartoli for their support during fieldwork. Financial support by the UK Economic and Social Research Council (PTA ) is also gratefully acknowledged. 1. The lists of firms are drawn from official sources: the S.A.G. (Servicio Agricola y Ganadero) 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 the web and coverage (countries, vintages, wine areas). 3. The rating is based on the following criteria: Classic: a great wine; Outstanding: superior character and style; Very good: wine with special qualities; Good: a solid, well-made wine; Average; drinkable wine that may have minor flaws; Below average; drinkable but not recommended; Poor; undrinkable, not recommended. 4. The questionnaire included a question asking the respondents to name the firms in their cluster that they perceived as having achieved high performance in terms of quality of wines 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 given more robust or insightful econometric estimations. This model is applied in other forthcoming works by the author. REFERENCES Allen, R.C. (1983), Collective invention, Journal of Economic Behaviour and Organization, 4: 1 24.

193 Networks and heterogeneous performance of cluster firms 177 Amin, A. and Cohendet, P. (2004), Architectures of Knowledge: Firms, Capabilities and Communities, Oxford: Oxford University Press. Arrow, K.J. (1962), Economic welfare and the allocation of resources for invention, in R.R. Nelson (ed.), The Rate and Direction of Inventive Activity, Princeton, NJ: Princeton University, pp Audretsch, D. and Feldman, M.P. (1996), R&D spillovers and the geography of innovation and production, American Economic Review, 86: Baptista, R. (2000), Do innovations diffuse faster within geographical clusters?, International Journal of Industrial Organization, 18: Barney, J.B. (1991), Firm resources and sustained competitive advantage, Journal of Management, 17: Bell, M. and Albu, M. (1999), Knowledge systems and technological dynamism in industrial clusters in developing countries, World Development, 27: 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, Regional Studies, 39(1): Boschma, R.A. and Frenken, K. (2006), Why is economic geography not an evolutionary science? Towards an evolutionary economic geography, Journal of Economic Geography, 6(3): Boschma, R.A. and Lambooy, J.G. (1999), Evolutionary economics and economic geography, Journal of Evolutionary Economics, 9: Breschi, S. and Lissoni, F. (2001), Knowledge spillovers and Local Innovation Systems: a critical survey, Industrial and Corporate Change, 10: Camison, C. (2004), Shared, competitive and comparative advantages: a competencebased view of industrial-district competitiveness, Environment and Planning, A36: Capello, R. (1999), Spatial transfer of knowledge in high technology milieux: learning versus collective learning processes, Regional Studies, 33: Capello, R. and Faggian, A. (2005), Collective learning and relational capital in local innovation processes, Regional Studies, 39(1): Cohen, W.M. and Levinthal, D.A. (1990), Absorptive capacity a new perspective on learning and innovation, Administrative Science Quarterly, 35(1): Combris, P., Lecocq, S. and Visser, M. (1997), Estimation of hedonic price equation for Bordeaux wine: does quality matter?, Economic Journal, 107: Combris, P., Lecocq, S. and Visser, M. (2000), Estimation of hedonic price equation for Burgundy wine, Applied Economics, 32: Dosi, G. (1988), Sources, procedures and microeconomic effects of innovation, Journal of Economic Literature, 26: Dosi, G. (1997), Opportunities, incentives and the collective patterns of technological change, Economic Journal, 107: Drucker, P. (1993), Post Capitalist Society, New York: Harper Business. Feldman, M.P. (1999), The new economics of innovation, spillovers and agglomeration: a review of empirical studies, The Economics of Innovation and New Technology, 8: Giuliani, E. (2006), The selective nature of knowledge networks in clusters: evidence from the wine industry, Journal of Economic Geography, in press. Giuliani, E. and Bell, M. (2005), The micro-determinants of meso-level learning and innovation: evidence from a Chilean wine cluster, Research Policy, 34(1):

194 178 Network analysis Granovetter, M. (1985), Economic action and social structure: the problem of embeddedness, American Journal of Sociology, 91: Grant, R.M. (1996), Toward a knowledge-based theory of the firm, Strategic Management Journal, 17: Gulati, R. (1998), Alliances and networks, Strategic Management Journal, 19: Howells, J.R.L. (2002), Tacit knowledge, innovation and economic geography, Urban Studies, 39(5 6): Jaffe, A.B. (1989), Real effects of academic research, American Economic Review, 79: Jaffe, A.B., Trajtenberg, M. and Henderson, R. (1993), Geographic localization of knowledge spillovers as evidence from patent citations, Quarterly Journal of Economics, 188: 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: Lazerson, M.H. and Lorenzoni, G. (1999), The firms that feed industrial districts: a return to the Italian source, Industrial and Corporate Change, 8: March, J.G. (1991), Exploration and exploitation in organizational learning, Organization Science, 2(1): Markusen, A. (2003), Fuzzy concepts, scanty evidence, policy distance: the case for rigour and policy relevance in critical regional studies, Regional Studies, 37: 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): Maskell, P. (2001a), Towards a knowledge-based theory of the geographical cluster, Industrial and Corporate Change, 10: Maskell, P. (2001b), The firm in economic geography, Economic Geography, 77(4): Maskell, P. and Malmberg, A. (1999), Localised learning and industrial competitiveness, Cambridge Journal of Economics, 23(2), Molina-Morales, F.X. and Martinez-Fernandez, M.T. (2004), How much difference is there between industrial district firms? A net value creation approach, Research Policy, 33: Moulton, B.R. (1990), An illustration of a pitfall in estimating the effects of aggregate variables on micro units, Review of Economics and Statistics, 72(2): 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: Nuvolari, A. (2004), Collective invention during the British Industrial Revolution: the case of the Cornish pumping engine, Cambridge Journal of Economics, 28(2): Oerlemans, L.G. and Meeus, M.T.H. (2005), Do organizational and spatial proximity impact on firm performance?, Regional Studies, 39(1): Organization for Economic Cooperation and Development (OECD) (2000), Local partnership, clusters and SME globalisation, Conference for Ministers responsible for SMEs and Industry Ministers, Bologna, Italy, June.

195 Networks and heterogeneous performance of cluster firms 179 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 districts to knowledge clusters : a model of knowledge dissemination and competitive advantage in industrial agglomerations, Journal of Economic Geography, 3: Piore, M.J. and Sabel, C.F. (1984), The Second Industrial Divide: Possibility for Prosperity, New York: Basic Books. Porter, M. (1998), On Competition, Cambridge, MA: Harvard Business Review Series. Porter, M. (1990), The Competitive Advantage of Nations, London: Macmillan. Powell, W.W., Koput, K.W. and Smith-Doerr, L. (1996), Interorganizational collaboration and the locus of innovation: networks of learning in biotechnology, Administrative Science Quarterly, 41: Pyke, F., Becattini, G. and Sengenberger, W. (1990), Industrial Districts and Interfirm Co-operation in Italy, Geneva: International Institute for Labour Studies. Rabellotti, R. and Schmitz, H. (1999), The internal heterogeneity of industrial districts in Italy, Brazil and Mexico, Regional Studies, 33(2): Rodan, S. and Galunic, C. (2004), More than network structure: how knowledge heterogeneity influences managerial performance and innovativeness, Strategic Management Journal, 25: Scott, A.J. (1998), Regions and the World Economy: The Coming Shape of Global Production, Competition and Political Order, Oxford and New York: Oxford University Press. Smith-Doerr, L. and Powell, W.W. (2003), Networks and economic life, in N. Smelser and Swedberg, R. (eds), The Handbook of Economic Sociology, London and Princeton, NJ: Russell Sage Foundation and Princeton University Press. UNCTAD (2001), World Development Report: Promoting Linkages, Geneva, New York: United Nations Conference on Trade and Development. UNIDO (2001), Development of Clusters and Networks of SMEs, Vienna: United Nations Industrial Development Organization. Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods and Applications, Cambridge: Cambridge University Press. Wernerfelt, B. (1984), A resource-based view of the firm, Strategic Management Journal, 5: Zaheer, A. and Bell, G.G. (2005), Benefiting from network position: firm capabilities, structural holes and performance, Strategic Management Journal, 26:

196 9. Social networks and the economics of networks Daniel Birke* 1. INTRODUCTION The work by Arthur (1989) and the other literature on path dependence are prominent in many evolutionary arguments. Path dependence arises in increasing return technologies and describes the property that for many technologies, historically small events can have a profound and lasting impact on the development path of technologies. It is well understood that technologies with increasing returns can exhibit multiple equilibria and that these equilibria can often be local rather than global optima. Causes for increasing returns can be learning and network effects. The models of path dependence, however, have been based on rather simple assumptions about the choice behaviour of consumers in markets with network effects. The empirical question remains: how do consumers choose between rival products in a market with network effects? Consumers interact with other consumers in a variety of ways. Information about products and services is often spread by word of mouth and individuals are more likely to choose a product about which they have heard from a friend or which they have tried out with the help of a friend. In many cases, consumers also try to associate themselves with their peer group by consuming similar goods, try to imitate consumption behaviour of groups which they regard as having higher social status and try to distinguish themselves from groups with lower social status (see Cowan et al., 1997). Network effects have been treated in the literature mainly as an only positive peer effect, where every individual is a peer of everyone else, that is, where this peer effect is anonymous. Here, we analyse the impact of peer groups on choice behaviour applied to the case of mobile telephony. After the seminal article of Rohlfs (1974), and the influential papers of Farrell and Saloner (1985) and Katz and Shapiro (1985), there has been a plethora of theoretical studies into the nature of network effects and by now network effects theory has reached a rather mature state. The literature on network effects typically distinguishes between two types of such 180

197 Social networks and the economics of networks 181 effects: direct and indirect. Direct network effects refer to the case where users directly benefit from other users of the same network. In mobile telecommunications, a direct network effect arises when a user can call a larger set of other users. Indirect network effects, on the other hand, arise because bigger networks support a larger range of complementary products and services. In second generation mobile networks, indirect network effects are only of second-order significance, but they will play an increasing role after the introduction of third generation networks, where usage will be heavily influenced by the availability of data services. Mobile networks are highly compatible with each other from a technical point of view. Users typically do not experience any quality difference between calls made to the same network and calls made to other networks. Network effects are mainly induced by network operators through higher prices for calls to other networks (off-net calls) than for calls to the same network (on-net calls). This pricing strategy is pursued by operators in most European countries. In a previous paper (Birke and Swann, 2006), we have shown that choice of mobile phone operators is strongly coordinated within households and that this effect is far more important for operator choice than the effect of overall network size. This is interesting in the context of the network effects literature, as an often assumed equivalence between direct and indirect network effects hinges on the assumption that only overall network size matters and not who is on the network. Whereas this assumption seems tenable for markets with indirect network effects, it is doubtful for markets with direct network effects. In these markets, consumers are interested primarily in which of their interaction partners uses the same technology and rather less in the overall number of users. This chapter looks directly at operator choice in a social network. For this purpose, we conducted a survey of a class of undergraduate students at Nottingham University Business School, asking them to identify their social network and filling in a questionnaire about their mobile phone usage. We thus were able to obtain a relatively well-bounded network. Obviously, students do have a social network apart from their university class, but the results show that interaction between students was strong within class and that students coordinated operator choice within this social network and in particular among students sharing geographical origin. Network data analysis exhibits two distinctive features that have to be addressed by the researcher. First, observations are not independent of each other. We shall try to overcome this problem of structural autocorrelation by using a technique called quadratic assignment procedure for permutationbased estimation of standard errors (see Krackhardt, 1988). Second, observations often are not drawn from a random sample. In our case, the student sample is a convenience sample and we certainly cannot claim to be close to

198 182 Network analysis a random sample. This limits the generalisability of the results, but, as we shall discuss later, it also yields some interesting results that we would not have obtained with a random sampling approach. This chapter is organised as follows. Section 2 discusses social network analysis and its usefulness for researchers interested in evolutionary economics. Section 3 gives a brief introduction to the UK mobile telecommunications market. Section 4 outlines the approach taken with the survey. It presents descriptive analysis of students attitudes towards mobile telecommunications and a description of the social network within the class. This is followed by a graphical and statistical analysis. Section 5 discusses the results. 2. SOCIAL NETWORK ANALYSIS AND EVOLUTIONARY ECONOMICS Research on social networks (and indeed networks in general) has increased rapidly in the last two decades and is undertaken in a variety of different disciplines (see Borgatti and Foster, 2003). Social network analysis (SNA) has its root in sociology and in its early years has been widely used to analyse interaction between individuals. These interactions can take on a variety of forms, from friendship over business to sexual relationships. However, SNA is not limited to social networks, but can be applied to the analysis of networks in general. Economics has been slow to make use of advances in social network analysis and only recently have such studies been conducted in higher numbers and especially in fields associated with evolutionary economics (see, for example, Breschi and Lissoni, 2004; Giuliani, this volume; Maggioni and Uberti, this volume). Why is social network analysis particularly helpful for researchers in the tradition of evolutionary economics? SNA is inherently relational, contextual and systemic, which is similar to how many researchers associated with evolutionary economics see the world and their discipline. Evolutionary economics emphasises the role that diversity of people or organisations plays for the understanding of economic processes (see Metcalfe, 1998). Furthermore, constraints in learning and in the selection process make evolutionary economics much more inclined to structural theories of economic interaction than mainstream economics, where most of the interactions occur in an (anonymous) marketplace. Both, learning and selection environment are argued to be highly localised both socially and geographically (see Antonelli, 1995) and one main reason for this localisation is the underlying structural patterns of interaction. A central theme of evolutionary economics is the importance of innovation and a rich variety of academic papers focus on the innovation

199 Social networks and the economics of networks 183 process. Closely intertwined with this topic is the role of knowledge and how knowledge is acquired by firms. It is generally recognised that a high percentage of knowledge is not readily codifiable and is transmitted through direct interaction between individuals. The paper by Cowan and Jonard (2004), for example, is work along these lines; it looks at the relationship between network structure and diffusion performance in a knowledge barter process and is based on the idea of small-world networks. In general, there is high interest among researchers in the evolutionary tradition in using agent-based simulations to capture economic phenomena and here again in particular in innovation research (see, for example, the discussion in Frenken (2006) and the references therein). A common theme to a lot of this work is a rejection of the standard assumption of uniform agents interacting with other agents through an anonymous marketplace. Network structure plays an important role as an underlying assumption of how agents interact or as a key parameter that is varied and analysed by the researcher. However, for simplicity the literature often assumes completely regular network structures or easy network generation mechanisms and it is therefore important to confront these models with empirical findings. In the recent past, there has also been an increasing interest in empirical work examining and using network structure. Cantner and Graf (2005) employ SNA to describe and analyse the evolution of the innovator network in the town of Jena, Germany and Giuliani s chapter in this volume looks at knowledge networks in the Italian and Chilean wine industry. Other works, like Murmann s (2003) book on the economic history of the dye industry, have used social network analysis in a descriptive way to portray the academic industrial dye knowledge network in the nineteenth century. Likewise, there is a growing literature in sociology about how economic transactions are embedded in social relationships (see Uzzi, 1996). Friendship ties with competitors, for example, have been found to improve the performance of hotels in Sydney due to enhanced collaboration, mitigated competition and better information exchange (see Ingram and Roberts, 2000). These networks can further be used for the effective social enforcement of rules and norms of doing business. Contrary to these contributions, this chapter will focus exclusively on the demand side of the market an area that remains scarcely studied. 3. THE MOBILE TELECOMMUNICATIONS INDUSTRY IN THE UK In the United Kingdom, there are four main Global System for Mobile Communications (GSM) operators: Vodafone, O 2,T-Mobile and Orange

200 184 Network analysis Figure Q Q Q Q Q Q Q Q Q Q Q Number of subscribers in the UK (in millions) No. of subscribers and a purely third generation operator: 3. O 2 (Cellnet) and Vodafone started operation in 1985 with analogue mobile networks. 1 There had been relatively slow growth until the entry of T-Mobile (One-to-One) and Orange after 1993 introduced stronger competition to the market. However, the market really took off with the widespread use of prepaid cards, which made mobile telephony attractive for the mass market and especially for low-usage consumers. Although a first prepaid tariff was launched by Vodafone in September 1996, prepaid usage became popular only after mid As Figure 9.1 shows, this led to a period of rapid expansion in the number of subscribers, which lasted roughly until early/mid After the burst of the stock market bubble in mid-2001, operators cleaned up their customer base of inactive consumers (note the short decline in Figure 9.1) and have since continued to grow, but this time more gradually and with a stronger focus on increasing the average revenue per user (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 penetration rate of over 85 per cent, future revenue growth has to come from an increased ARPU rather than from a bigger customer base. In May 2000, the four GSM operators and 3 were awarded licences for third generation Universal Mobile Telephone System (UMTS) networks for about 4 billion each. Many expect that 3G networks will further boost revenues in the mobile telecommunications market. In 2003, 3 introduced the first third generation network in the UK. After a slow start, the company now has over 3 million users. The other companies recently followed with their own third generation networks.

201 Social networks and the economics of networks % 35% 30% 25% 20% 15% 10% Figure 9.2 Q Q Q Q Q Q Q Q Q Q Q Q Development of subscriber market shares Vodafone O 2 T-Mobile Orange Especially interesting for our analysis is the development of market shares (see Figure 9.2). At the end of 1998, the market was dominated by the incumbent operators O 2 and Vodafone, which together accounted for almost 70 per cent of the market. However, by the beginning of 2001 this lead had dissipated and subscriber market shares were levelled. Today, the market is about equally split among the four GSM operators. 2 The ability of T-Mobile and Orange to catch up with the incumbent operators is somewhat unique to the UK market and is different from, for example, the German market in which the two biggest operators (T-Mobile and Vodafone) still control about 80 per cent of the market and have reported stable market shares in recent years. With strong network effects present in the market, this catch-up by T-Mobile and Orange is surprising, as network effects result in a strong tendency towards higher market concentration. It could be argued that the development in the UK market is due to the high compatibility between networks. However, in Birke and Swann (2006) we showed that network effects do play an important role in the adoption of mobile telephones and in operator choice. 4. COORDINATION OF CHOICE OF MOBILE PHONE OPERATORS The Survey To analyse how consumers coordinate their choice of mobile phone operator with their social network, we looked at a network from a class of undergraduate students. We expected mobile phones to be a rather important tool of interaction for this group and the network to be rather well

202 186 Network analysis bounded with a high percentage of intra-network interaction. The sample consists of students from a second-year undergraduate course called Economics of Organisation B, which was held at Nottingham University Business School in spring Most students from this course study for a three-year degree and have been studying together in a variety of other courses for about 18 months. Although most students can be expected to use their mobile phone to interact with many people outside the class, a reasonably regular interaction between the students can be assumed. The questionnaire on which the survey is based consists of two parts. In the first part, we asked students for some demographic details and about their attitudes and behaviour with regard to mobile phones. Information on operator choice and demographic variables like gender and nationality are used as input for the regression analysis. The other answers are mainly used to gain insights into how the respondents use their mobile phones and for a first descriptive analysis. In the second part of the questionnaire, students were handed out a list of course participants and were asked to identify themselves and the people they communicate with. The exact wording of the question was Please tick the people that you call. 3 Thus, we are able to build up a network of the relevant communication relations within the class. Both parts took about 10 minutes to fill in and were distributed and collected 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) of whom 158 were identified as students from the course list. Of the remaining 13 students, three respondents indicated that their name was not on the list and another 10 did not identify themselves. To every student for whom we did not receive an identified questionnaire (which mostly included students who missed class), we sent out a reminder and consequently received 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. For most of the descriptive statistics, all 175 responses are used and for all analysis relating to social networks only the subsample of 159 students is used. All of the students are undergraduate students and are almost of the same age. There is about an even number of male (48 per cent) and female (52 per cent) students among the respondents. Importantly for our analysis, the share of foreign students is rather high, with only slightly over half of the students originating from the UK and with another 8 per cent coming from other European countries (see Table 9.1). Chinese students are the second biggest group in the course (22 per cent). The high number of foreign students is interesting for us, as we expect that students who come to the UK primarily to study would have a social

203 Social networks and the economics of networks 187 Table 9.1 Gender and nationality of respondents British Other Chinese Other Africans The Total European Asian Americas Male (65.2%) (64.3%) (15.4%) (23.5%) (50%) (48%) Female (34.8%) (35.7%) (84.6%) (76.5%) (50%) (100%) (52%) network that revolves more around their university class than that of British students. As can also be seen from Table 9.1 the majority of Chinese and other Asian students are female, while the majority of English and other European students are male. Criteria for Choice of Mobile Phone Operator A first option to analyse why the respondents chose their operators is to ask them directly about their choice criteria. Table 9.2 gives the response frequencies for a number of criteria. Quality, special offers, cost of calls and operator choice of friends and family all seem to be important. The obvious drawbacks of directly asking respondents about their choice criteria are the difficulties of comparing the relative importance of the different factors. Furthermore, it is not always clear whether the given answers are the actual reasons for choosing an operator or whether it is an ex post rationalisation of the choice process. Quality of the network, for example, is named as an important criterion by most respondents. However, the quality of the four GSM networks in the UK is roughly equivalent in terms of most quality characteristics such as network coverage, international roaming and customer service. 4 We might rather measure the general importance that the respondents attribute to quality when choosing a product than the particular importance for mobile phone networks. Of course, a precondition for consumers coordinating their operator choice is some knowledge about what operators their peers use. This is far from given, as mobile networks are hard to identify from telephone numbers. In the UK, there are several hundred prefixes associated with the different networks. 5 This is contrary to, for example, Germany where there are only 23 prefixes, which makes it far easier to identify people who are using the same operator especially in the earlier years of mobile phone adoption when not all of these prefixes were used. In general, operator identification from telephone numbers is easier in smaller countries where fewer different prefixes are needed to cover all subscribers.

204 188 Network analysis Table 9.2 Frequencies for choice criteria Strongly Agree Neither Dis- Strongly Don t agree nor agree disagree know (1) (2) (3) (4) (5) Quality of the network (e.g. network coverage, roaming possibilities etc.) Special offer Cost of calls, text messages in general It is cheaper, because my friends/family use the same network Cost of handset Handsets available from this operator More services available (games etc.) Good customer service 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%) Information on who else is using the same network therefore has to be obtained by other channels in the UK. It seems likely that the availability of this information is directly linked to the closeness of two individuals in the social network. This information could then be obtained either through direct conversation or through identifying the operator from a mobile phone. The latter requires that the name or logo of the operator is conspicuously placed on the mobile phone, which is only the case for cobranded mobiles. In recent years, operators try to raise the awareness of their brand and increasingly place their logos on mobile phones next to the logos of the mobile phone manufacturer, or mobile phones exclusively carry the brand of the network operator. According to Table 9.3, respondents claim to know the mobile network operator for a high percentage of their peers. Especially the operators

205 used by family members and partners are known by the large majority. The lower figure of knowledge for operators used by friends is an indicator that operator coordination between friends might be lower than within households. Network Statistics Social networks and the economics of networks 189 Besides an analysis of individual respondents, we can also analyse the characteristics of the social network as a whole. A first measure characterising the overall network is its density. For directed graphs, it is calculated as 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 directed network of N people. In our case, N is the number of students that filled in the roster of the questionnaire, which is 159. In total, 815 different relations were identified. However, 195 relations are to non-respondents in the class and only 620 are to respondents, resulting in a density of g 620/(159*158) At the individual level this network density stems from 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 have only a limited capacity to communicate with other people. Furthermore, our measure of Who do you call? can be assumed to refer to rather close contacts compared, for example, with the question Who do you know?. Network density can also be measured at an individual level, that is, one can calculate the network density of each individual s network. If individual x has n friends, then this local network density measure gives the percentage of ties present between the n friends. Individual 1, for example, has six friends. Those six friends have 28 ties between each other, out of a possible t n(n 1) 6*5 30 ties, resulting in a local network density of Table 9.4 gives the frequencies of local network density for different numbers of friends. The local networks observed in the course are rather tight cliques with an average local density of Rather dense cliques should favour a coordination of operator choice. Table 9.4 also shows that local network density decreases with the number of friends. Clearly, larger networks are less closely interconnected. If every node in a graph is connected to all other nodes via a path, then the graph is said to be connected and consists of one single component. If there are disconnected subsets, several components exist. The number of components can be calculated in two different ways. First, one can calculate the number of weak components, that is, two persons are in the same component, if one of them can reach the other. Second, strong components require that both persons can reach each other. In undirected or

206 190 Network analysis Table 9.4 Local network density No. of friends Mean No. of observations Local network density 18% 16% 14% 12% 10% 8% 6% 4% 2% 0% Figure 9.3 Distribution of distances between nodes symmetrical networks, these measures coincide, but in our case, we have a directed network and consequently we get two measures. Calculating weak components, we end up with three components in the network (plus two students who only nominated students who did not respond). The largest component consists of the large majority of students altogether 146 students are part of this component (92 per cent). The next largest component consists of nine students and then there is a third group of two students who nominated each other, but did not nominate anyone else (and were not nominated by other students). The result is an indication that the class network can really be seen as one network. The average distance between two reachable nodes in the network is 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* , pairs can reach each one another (72 per cent). The distribution of distances between these pairs is displayed in Figure 9.3. The distances are roughly normally distributed and the distribution has a mode of 6. Although the large majority of nodes belong to the same component, the average path length is rather long and has a considerable variance, which

207 already hints at some clustering in the network, which can be further analysed using graphical techniques. Graphical Analysis of Social Network Social networks can very usefully be analysed by graphical representations of these networks, in particular in the case of medium-sized networks with a couple of hundred nodes. 6 Figure 9.4 depicts the social network within the class of students, based on their communication pattern. It is a directed graph and arrows depict the direction of the nominations from the roster. The graph was created using a spring embedding algorithm from UCI-NET, which is based on the idea of simulating the social network graph as a system of mass particles. Nodes are the mass particles and the edges are springs between the particles, while the algorithm tries to minimise the energy of this system. Some form of clustering immediately becomes obvious. First, shapes of the objects, depicting nationalities, are highly clustered. Chinese students for example (up triangles) almost exclusively communicate with other Chinese students. At the bottom right of the graph, there is a group of Asian students who even form a distinct component and have only communication links within the group. This is the second largest component from the previous section. Two Spanish students form the third largest component, which can also be found at the bottom right of the graph. Finally, there are two isolates at the upper left. Second, the graph shows a clustering of shades, which are depicting main operator chosen. 7 This clustering of shades clearly occurs along nationality lines. Chinese students are in the majority using Vodafone and similar tendencies can be observed for other nationalities. However, there also seems to be a coordination of operators within nationalities, that is, also within nationalities students that call each other tend to use the same mobile phone operator. The strong correlation of operator choice within nationalities also becomes clear by a cross-tabulation of nationality and operator used (see Table 9.5). Almost 80 per cent of Chinese students choose Vodafone. Similarly, almost half of the British students opted for an O 2 mobile, whereas six out of 10 Africans use T-Mobile. Regression Results Social networks and the economics of networks 191 The original data of communications patterns are organised in a square matrix of 159 rows and columns with 1 s indicating a communication relationship and 0 s indicating the absence of a communication relationship. As is common for network data, diagonal values are not allowed. For a

208 Shapes British Other European Chinese Other Asians Africans The Americas Figure 9.4 Interaction network of students Key Vodafone Virgin Orange Other operators 3 T-Mobile O2 192

209 Social networks and the economics of networks 193 Table 9.5 Mobile phone operators and nationality British Other Chinese Other Africans The Total European Asian Americas (11.1%) (23.1%) (13.5%) (17.7%) (20%) (13.5%) O (46. 7%) (38.5%) (2.7%) (29.4%) (33.3%) (31.8%) Orange (18.9%) (15.4%) (5.4%) (5.9%) (20%) (14.1%) T-Mobile (6.7%) (60%) (7.1%) Virgin (2.2%) (33.3%) (1.8%) Vodafone (14.4%) (23.1%) (78.4%) (47.1%) (33.3%) (31.8%) Total regression analysis, this matrix is transformed into dyadic relationships (relationships between two nodes). We therefore get N(N 1) different dyads. When using network data, the assumption of ordinary least squares (OLS) and logit models that observations are independent fails. Observations are clearly not independent as there are at least N 1 dyads involving every individual. This correlation between observations involving the same nodes stems, for example, from the fact that it is far more likely to have the same operator as your friends if you use an operator with a high market share in the network. This would result in a positive correlation between observations from the same row or column and consequently, while parameter estimates are unbiased, estimated p-values overstate the significance level. One possibility to adjust for incorrect standard errors is the quadratic assignment procedure (QAP) as proposed by Krackhardt (1987, and 1988) for social network data. The idea of QAP is to permute rows and columns of the original data matrix for the dependent variable and then to re-estimate the original regression model. This process is reiterated to obtain an empirical sampling distribution from which QAP p-values are calculated. Details of the estimation procedure used are described in Birke and Swann (2005). We are estimating a logit model with same_operator as the dependent variable. This variable takes on the value 1 if two students use the same operator and 0 otherwise. There are four independent variables that are constructed in a similar way: same_nation (respondents of the dyad have the same nationality/come from the same group of nations as defined

210 194 Network analysis above), friend 8 (respondents call each other on their mobile phone), same_sex (nodes have the same gender) and same_payment (respondents use the same type of payment: contract versus pre-paid). Table 9.6 shows the results from a logit estimation of the model with QAP p-values. Same_nation, friend and same_sex are highly significant and show the expected sign, confirming the graphical analysis from Figure 9.4. Two respondents of the same nationality, who are friends and of the same sex are significantly more likely to use the same operator. Same_nation and friend have a particularly high significance level and in fact no permutation resulted in a parameter estimate higher than the observed values from the original regression. Same_sex is still significant at the 5 per cent level, but the 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 a number of constellations. For two respondents who are friends, of the same nationality and of the same gender, there is an almost 50 per cent probability that they also use the same operator. On the other hand, for two respondents who don t call each other, are not of the same nationality and not of the same gender, this probability is only 18 per cent. Most of this variation is due to the friend and same_nation parameters. The model from Table 9.6 excludes the operator 3, because of its different pricing structure. Analysing the same model for different operators yields a positive coefficient for the friend parameter for all operators, but 3. 9 This is further support for our hypothesis that network Table 9.6 Determinants of choosing the same operator Parameter estimates QAP p-values same_nation p friend p same_sex p same_payment p Constant p Table 9.7 Predicted probabilities of using the same operator Not same nationality Same nationality Not friends Not same sex Same sex Friends Not same sex Same sex

211 Social networks and the economics of networks 195 effects are the reason for consumers coordinating their operator choice. Operator 3 is the only UK operator that typically does not induce network effects, but rather offers packages of calling time regardless of the network to which calls are made. The incentive for 3 users to coordinate with their peers is therefore lower. This can also clearly be seen in Figure 9.4, where 3 users are evenly distributed over the graph. The results are also contrary to the argument that learning effects or word of mouth might be the prime cause of this coordination. The third generation network and handsets of 3 are arguably more difficult to master than other mobile phones and we would expect a coordination of operator choice for 3 if these effects were strong. The correlation of operator choice within nationalities is particularly interesting and this might have several causes. All UK operators also operate networks in a number of other countries; sometimes under the same brand, sometimes under different brands. Non-UK students might have simply continued to use the same operator they already used in their home country. However, concentration of operators worldwide is far lower than in the market for mobile phone handsets. Furthermore, most students come from countries where these operators do not have a network, as most operators have a rather European focus. This coordination of operators within nationalities might be due to common unobserved characteristics and attitudes of respondents with the same background or it could be a coordination mechanism. We therefore regress friend on same_nation and same_sex. Both having the same nationality and being of the same sex are important predictors of friendship between two respondents (see Table 9.8). Students from the same nation and from the same sex interact far more frequently. Both parameters are highly significant, but the coefficient of same_nation is about three times bigger than for same_sex. Like for operator choice, we can again calculate the predicted probabilities for different constellations (see Table 9.9). The predicted probability of an interaction between two respondents is generally rather low, but for two respondents from the same nationality and the same sex this probability is 10 times higher than for two respondents of different nationalities and different gender. Table 9.8 Friendship determinants Parameter estimates QAP p-values same_nation same_sex Constant

212 196 Network analysis Table 9.9 Predicted probabilities of calling each other Not same nationality Same nationality Not same sex Same sex The group of students for which coordination is strongest are Chinese students who in the large majority used Vodafone. To the best of our knowledge, at the time of the survey 10 there was no special tariff offered by Vodafone targeting Chinese students (like, for example, cheap calls to China) and Vodafone does not have an own network in the PRC, which students might have used prior to their study in England. 11 Asking fellow students why Chinese students choose Vodafone as their operator, they replied that other Chinese students told them on arrival that all Chinese students use Vodafone and that they should also use it, if they want other people to call them. This has afterwards also been confirmed by other Chinese students and by students from other nationalities. 12 If nationality is a strong determinant of friendship, it is a good choice to use the same network as other people from the same nation in order to reduce the number of off-net calls. Furthermore, even when accounting for this effect, friends are still more likely to choose the same operator. 5. DISCUSSION We have shown that consumers coordinate their choice of mobile phone operators not only within households, but also in their wider social network. We further found that this depends on the price difference between on- and off-net calls induced by most operators. Like the results from Birke and Swann (2006), this is further evidence that in markets with direct network effects it matters to the consumer who is on the same network. In this respect, direct network effects are different from indirect network effects, where only the total number of users matters. As discussed earlier, the sample on which this chapter is based is far from random and it is therefore difficult to generalise the findings to, say, the British population. The high percentage of foreign students might have favoured the results to a certain extent. However, we can also observe a strong coordination of operators among British students. Furthermore, it can be assumed that a significant part of a student s communication takes place outside the Economics of Organisation B class. Results from the survey (and common sense) suggest that, for example, calls to the partner

213 Social networks and the economics of networks 197 are a significant share of all calls. Consequently our results might rather understate the extent of coordination between close friends. In further research using calling records from land lines, we have found that the large majority of calls are made to three to five parties, but that these interaction partners vary over time. If this also holds for mobile phones then, at any one point in time, it should be relatively straightforward to coordinate operator choice with these peers. If geographical origin of students is an important predictor of friendship, then these geographical patterns might be even more stable than the friendships existing at the time of choosing the operator (for international students this typically is the time of arrival to the UK). The results pertaining to differences in interaction within and between students from different nationalities are interesting in their own right. First, they show that not only is geographic distance an important factor in many diffusion and coordination processes but also geographic origin. Second, the analysis gives a powerful illustration of how individual choice can be constrained by network structure. Even if coordination of consumption choices in markets with global network effects is important, different operators (or technologies) can easily coexist with one another due to local network effects, as long as global network effects are not too strong. Finally, we can speculate that Chinese students overwhelmingly using Vodafone might be a case of path dependence where a relatively small past event made some Chinese students choose Vodafone as their operator and where it is now beneficial for new arrivals to choose Vodafone simply because the large majority of other Chinese students use the same operator. The empirical analysis in this chapter focuses on one specific point in time and therefore cannot capture dynamics in the system. A longitudinal analysis based on social network data is even harder to achieve than for traditional datasets, due to the more severe consequences of missing values and the related problems with sampling new participants. However, this analysis could in theory be done relatively easily with access to electronic data on mobile calling patterns between individuals. This would also directly relate the frequency and cost of interactions with network choice. By using electronic calling records, it would be possible to get a far bigger sample size, which in theory could include all subscribers to a particular network. The obvious difficulty here is the confidentiality of this data and the resulting reluctance of companies to grant access to it. This chapter has demonstrated the fruitfulness of the use of social network analysis techniques for the analysis of economic problems. One drawback of SNA is the need for network data which is often not readily available. With the digitalisation of many aspects of society, much more

214 198 Network analysis network data are easily accessible in electronic forms to researchers nowadays. However, there remains a considerable number of research areas that require the collection of primary data by the researcher, which is an approach traditionally seldom pursued by economists. Social network analysis enables a more realistic and rich modelling of many economic decision-making processes and therefore in many cases justifies the additional efforts in data collection. Although this should be beneficial to economics as a whole, as we have argued, this could be particularly interesting for work in the tradition of evolutionary economics. NOTES * I am grateful to Peter Swann, David Paton, Robin Cowan, seminar participants at Chimera, University of Essex, Nottingham University Business School and participants at the EMAEE conference in Utrecht, The Netherlands for helpful comments. I would also like to acknowledge financial support from the University of Nottingham Business School and the ESRC. The analysis has been conducted using the econometrics package STATA, SAS and the social network software UCI-NET and PAJEK. 1. See Valletti and Cave (1998) for an analysis of the UK market from 1985 to 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 customers generate a higher ARPU. 3. For simplicity, we shall call respondents who communicate with each other friends in the 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 are ported (Q3 2004). 6. See Freeman (2005) for an overview of graphical representations for social network analysis. 7. Some of the respondents had up to three different mobile phone operators. Most of them indicated a main operator used; for three respondents the main operator was decided randomly 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 assume that if a tie is not reciprocated, this is because the respondent forgot to nominate the interaction partner and not because it is a true one-way relationship. This is a standard assumption of social network analysis. The results are virtually the same when using the non-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. O 2 has taken the lead here and there is anecdotal evidence that some Chinese students are switching to O 2 to profit from these discounts. 11. Vodafone has a minor stake in China Mobile, but it is a rather small stake (approximately 3.27 per cent) and is most likely not known to the average consumer. 12. One of the comments I received from international seminar and conference participants was that they encountered similar coordination mechanisms when they moved abroad.

215 REFERENCES Social networks and the economics of networks 199 Antonelli, C. (1995), The Economics of Localized Technological Change and Industrial Dynamics, Dordrecht etc.: Kluwer Academic Publishers. Arthur, W.B. (1989), Competing technologies, increasing returns, and lock-in by historical events, Economic Journal, 99(394), Birke, D. and Swann, G.M.P. (2005), Social networks and choice of mobile phone operator, Nottingham University Business School: Industrial Economics Division Occasional Paper Series, No Birke, D. and Swann, G.M.P. (2006), Network effects in mobile telecommunications an empirical analysis, Journal of Evolutionary Economics, 16(1): Borgatti, S.P. and Foster, P.C. (2003), The network paradigm in organizational research: a review and topology, Journal of Management, 29(6), Breschi, S. and Lissoni, F. (2004), Knowledge networks from patent data: methodological issues and research targets, in Glänzel, W., Moed, H. and Schmoch, U. (eds), Handbook of Quantitative Science and Technology Research, Dordrecht: Kluwer, pp Cantner, U. and Graf, H. (2005), The network of innovators in Jena: an application of social network analysis, Paper presented at the 4th European Meeting on Applied Evolutionary Economics (EMAEE), May, Utrecht, The Netherlands. Cowan, R., Cowan, W. and Swann, G.M.P. (1997), A model of demand with interactions among consumers, International Journal of Industrial Organization, 15(6): Cowan, R. and Jonard, N. (2004), Network structure and the diffusion of knowledge, Journal of Economic Dynamics & Control, 28(8): Farrell, J. and Saloner, G. (1985), Standardization, compatibility, and innovation, RAND Journal of Economics, 16(1): Freeman, L.C. (2005), Graphic techniques for exploring social network data, in Carrington, P.J., Scott, J. and Wasserman, S. (eds), Models and Methods in Social Network Analysis, Cambridge: Cambridge University Press, pp Frenken, K. (2006), Technological innovation and complexity theory, Economics of Innovation and New Technology, 15(2): Ingram, P. and Roberts, P.W. (2000), Friendships among competitors in the Sydney hotel industry, American Journal of Sociology, 106(2): Katz, M.L. and Shapiro, C. (1985), Network externalities, competition, and compatibility, American Economic Review, 75(3): Krackhardt, D. (1987), OAP partialling as a test of spuriousness, Social Networks, 9(2): Krackhardt, D. (1988), Predicting with networks: nonparametric multiple regression analysis of dyadic data, Social Networks, 10(4): Metcalfe, J.S. (1998), Evolutionary Economics and Creative Destruction, The Graz Schumpeter lectures, London: Routledge. Murmann, J.P. (2003), Knowledge and Competitive Advantage. The Co-evolution of Firms, Technology, and National Institutions, Cambridge: Cambridge University Press. Rohlfs, J. (1974), A theory of interdependent demand for a communications service, Bell Journal of Economics and Management Science, 5(1):

216 200 Network analysis Uzzi, B. (1996), The sources and consequences of embeddedness for the economic performance of organizations: the network effect, American Sociological Review, 61(4): Valletti, T.M. and M. Cave (1998), Competition in UK mobile communications, Telecommunications Policy, 22(2):

217 PART IV Spatial systems

218

219 10. Diversity, stability and regional growth in the United States, Jürgen Essletzbichler* 1. INTRODUCTION Recent years have witnessed important changes in economic governance systems represented as a scalar shift of economic and political power from national states to supra-national entities and subnational entities such as cities and regions (Jessop, 1990, 1994; Brenner, 1998, 2004; Scott, 1998). Cities, regions and city-regions are increasingly forced into direct competition with each other that prompts regional policy makers to actively design and shape regional economic development (Harvey, 1989; Leitner and Sheppard, 1998). Baden-Württemberg, the Third Italy and Silicon Valley exemplify the paradigmatic model of economic development that other regions attempt to emulate (Bartik, 1996). Regional policies are designed to attract clusters of functionally related industries with high growth potential although the value of cluster-based policies is not uncontested (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 even to be based on common misconceptions. The value of industrial specialization for regional economic development is uncertain as theoretical and empirical work on specialization and diversity of cities suggests (Baldwin et 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 stability of regional economies that is largely ignored by policy makers (Baldwin and Brown, 2004). This chapter examines the relationship among diversity, growth and stability of regional production systems. Empirical work by regional scientists and new geographical economists yields ambiguous results. While Kort (1981) and Baldwin and Brown (2004) find strong evidence for a positive 203

220 204 Spatial systems relationship between stability and diversity and a negative relationship between employment growth and diversity, Attaran (1986) and Smith (1990) contest these findings. Overall, the review by Dissart (2003) suggests that more diversity leads to more stability and less growth in unemployment. The regional science literature is strongly focused on the identification of empirical relationships, while the theoretical links are not fully developed (Conroy, 1974, 1975; Siegel et al., 1995; Chandra, 2003). New geographical economists examine the theoretical and empirical relationship between diversity and economic growth (Krugman, 1991; Glaeser et al., 1992; Henderson, 1997; Brakman, et al., 2001). Some of these researchers are influenced by the ideas of Alfred Marshall (1920) and Jane Jacobs (1969) linking variety (technological and/or industrial) to external economies, efficiency of regional production systems and economic growth (Henderson, 2003). Conclusions from both areas of research have potentially important policy implications. Should regional policy makers stimulate or curtail the production of diversity to foster regional economic growth? Should policy makers focus on generating conditions for high rates of economic growth or should they focus on minimizing growth rate fluctuations? The existing work has developed important theoretical arguments to understand the relationship between diversity and growth (Henderson, 1988; Glaeser et al., 1992, Quigley, 1998), but the relationship between diversity and stability has been undertheorized. In part, this might be explained by the influence of neoclassical economics on regional science and new geographical economics focusing on market competition as the only 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 competition is based on very different principles from regional competition. While firms have to maximize profits to stay in business, regions cannot go bankrupt (Krugman, 1994). Furthermore, regional policy makers are responsible to different interests in the region including the provision of technical and social infrastructures and social services. Only if it is assumed that increased regional efficiency translates into welfare gains for everybody can the exclusive focus on economic growth be justified. Instead of using economic growth as a vehicle to achieve other goals such as equity or sustainability, higher rates of economic growth become the policy target. This exclusive focus on growth might be problematic if it leads to a reduction in technological, industrial, social and institutional diversity in the region. Because growth rates are maximized if less-efficient routines, technologies, skills and industries are eliminated, this is likely to be the case. A lack of diversity might reduce the adaptive potential of the region to future change.

221 Diversity, stability and regional growth in the United States 205 This chapter addresses explicitly the trade-off between regional employment growth and regional economic stability drawing on insights from evolutionary theory and ecological economics. Evolutionary theories highlight the importance of diversity as fuel for the selection process (Nelson, 1995). Selection winnows on existing variation and, given a stable selection environment and no introduction of new diversity, will ensure that only the most efficient entities survive. In reality, new diversity is added through innovation and firm entry and coupled with a continually changing environment, efficiency and optimality criteria are perpetually redefined. Perfect adaptation towards a global optimum is therefore impossible (Hodgson, 1993, 1997). Applications in evolutionary economics focus primarily on the impact of firm diversity in populations of competing firms on population (for example, industry) averages. In this work, Fisher s principle is employed, stating that the rate of change is proportional to the variance in efficiency characteristics (for example, profit rates, unit costs or productivity levels) (Metcalfe, 1994, 1998). Recent work suggests that intra-population dynamics has to be linked to interpopulation dynamics and include selection processes at various analytical scales such as the firm, industry, region and nation (Gowdy, 1992; Andersen, 2004). Moving the focal level to the regional scale complicates the analysis considerably. Ecological economists and evolutionary biologists have long argued that a trade-off between adaptive efficiency and the adaptability (the ability to adapt to environmental changes) of ecosystems exists (Gould and Lewontin, 1979; Levins and Lewontin, 1985; Vrba and Gould, 1986). Like ecosystems, regions might be confronted with an explicit trade-off between adaptation and flexible adaptivity/resilience. Adaptation refers to the optimal adjustment to current environmental circumstances. Adaptation is achieved through enhanced efficiency of individual agents (for example, through innovation and imitation) and the elimination of redundant features such as undesired skills, inefficient technologies, industries, organizations and institutions. While boosting current efficiency levels (and rates of economic growth), lower levels of diversity decrease the likelihood of pre-adaptive features and the potential to react to changing environmental conditions ushered in by technological paradigm shifts, exogenous shocks or changes in the institutional environment (Holling, 1973, 2001). However, there are limits on the extent of diversity. Without commonalities between different entities, no synergies arise, and certain efficiency thresholds necessary for the economic survival of regions might never be reached. In this chapter, the theoretical arguments from evolutionary theory and ecological economics are summarized to put the trade-off between regional economic diversity and regional economic growth on stronger theoretical

222 206 Spatial systems foundations. For this purpose, Section 2 reviews work by evolutionary theorists and ecological economists and their emphasis on the relationship among diversity, growth and stability. Section 3 presents a simple empirical model that links regional economic diversity to stability and growth and Section 4 concludes this chapter. 2. DIVERSITY, STABILITY AND GROWTH The case for diverse regional production systems hinges on the premise that diversity reduces volatility 1 (or enhances stability). Stability is seen as a positive property of regional production systems for two reasons. First, high levels of volatility are often coupled with higher rates of unemployment, because contracting economies destroy jobs and release workers and because it takes time to match workers to new jobs. Second, high volatility complicates planning decisions to provide adequate investment in technical and social infrastructures (Schoening and Sweeney, 1992; Baldwin and Brown, 2004). The maintenance of diversity is therefore useful from a policy point of view. However, diversity affects not only stability but also regional efficiency by stimulating or constraining innovation, technology spillovers and supplier customer interaction (Jacobs, 1969). Whether or not diversity will generate external economies through spillovers is likely to depend on the exact mix of industries, firms, workers, organizations and institutional practices in a region. Too much diversity might stifle the formation of spillovers through a lack of synergies. Too little results in increasing specialization and makes the region vulnerable to changes in technological paradigms, demand and supply shocks. Current market-driven policies largely based on David Ricardo s theory of comparative advantages drive the formation of trade areas and the globalization process. The erosion of national boundaries is likely to result in increasing regional specialization. This might increase overall efficiency (at the supra-national or global level) but at the expense of increased vulnerability at the subnational or regional level. Unfortunately a theoretical discussion of these intertemporal and interspatial trade-offs is largely absent from economic geography. A substantial body of literature theorizing these trade-offs, however, is emerging in evolutionary theory, ecological economics 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 summarized. Although there are limitations to the transferability of knowledge and concepts from the physical and biological sciences to the social realm, some of the conclusions from this literature pertain to all complex, adaptive

223 Diversity, stability and regional growth in the United States 207 systems (Giampietro and Mayumi, 1997). The review of this literature is not expected to generate a series of testable hypotheses but a series of general principles on the relationship among diversity, stability and growth of regional production systems that can be explored through empirical work. Evolutionary Theory, Diversity and Stability Diversity and selection Evolution is driven by the creation and destruction of diversity. Diversity is expressed as variation at the genetic level, as biodiversity at the level of ecosystems, as technological diversity at the level of industries, as industrial and institutional diversity at the level of regions and countries. In biological systems, diversity is created by random mutation. In socio-economic systems, diversity is generated primarily by the processes of innovation and plant 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. Selection rewards those species, firms, regions or countries that are best adapted to narrow conditions at the moment. In this sense, selection operates as a short-term adaptive force (Rammel and Staudinger, 2002). Adaptation is interpreted as temporary feature providing a benefit over its alternatives under specific environmental conditions (Rammel and van den Bergh, 2003: 123, emphasis added). This view on adaptation has a number of important implications relating to questions of optimality, efficiency, equilibrium and causal relationships. Without going into details about debates on adaptationism in evolutionary biology (see, for instance, Gould and Lewontin, 1979; Depew and Weber, 1995), a few clarifications need to be made in the context of this contribution. Spencer interpreted selection as a process that guaranteed the survival of the fittest. In this view, selection is regarded as a (global) optimization process. This reading of selection entails a closed universe, one that can be described by a unique and optimal equilibrium configuration towards which the system gravitates. Natural selection is the mechanism that ensures that this state will be reached eventually. In equilibrium, only those traits, species and populations survive that are perfectly adapted to environmental conditions describing this equilibrium. In equilibrium a global optimum is reached. This view of evolution was adopted by neoclassical economists who interpreted competitive markets as selection environments that ensured that only firms with optimal technologies and organizational routines survived. Because inefficiency was considered to be

224 208 Spatial systems only a transitory phenomena, diversity in firm behavior could be ignored and firms be treated as if they were profit maximizers (Friedman, 1953). During the movement towards global optimum, diversity becomes eliminated (see Vromen, 1995). In the absence of changing environmental conditions and creation of new diversity, selection would indeed reduce variation until only the profit maximizers would survive (Alchian, 1950; Jovanovic, 1982; Iwai 1984a, b; Metcalfe and Gibbons, 1986; Metcalfe, 1994, 1998). In reality, firms are confronted with moving targets in the form of shifting fitness landscapes, continuous introduction of new diversity in the form of innovation and technological change, random shocks and non-linear feedback mechanisms and complex patterns of interactions whose outcomes cannot be predicted ex ante. In this environment of uncertainty and unpredictability, optimization must be understood as local and myopic (Nelson, 1995). In that sense it might be better to talk about survival of the fitter or sufficiently fit (Rammel and van den Bergh, 2003) or survival of the fitting (Boulding, 1981). According to this view, selection does not entirely eliminate diversity. Although the persistence of diversity might be undesirable from a neoclassical point of view, the rejection of the existence of a global optimum makes diversity in the form of redundant, suboptimal and inefficient technologies, skills, firms and industries not only acceptable but a necessary condition for long-term survival of firms and regions. Diversity, optimality and stability As in regional science, the exact relationship between diversity and stability is still debated in ecological theory (Holling, 2001; Rammel and Staudinger, 2002). In ecology, diversity is negatively related to stability if species diversity reflects a diversity in functional entities in an ecosystem with minimum redundancy (Rammel and Staudinger, 2002: 305). Translated to economic geography, this case would describe a region with a diversity of sectors whose technological inputs and demand are highly correlated, that is, whose input output structures are almost identical. In this case, the existing industrial diversity would not protect the region from demand shocks and/or shifts in technological paradigms (Wagner and Deller, 1998; Frenken et al., 2005). Independent of the specific expression of diversity, the rejection of the assumption of a global optimum complicates definitions of efficiency 2 suggesting that current regional policies based on some notion of economic and social efficiency are driven by the ideology of efficiency (Bromley, 1990) and not derived from solid theoretical foundations. In the presence of shifting adaptive landscapes and moving equilibria, the focus on efficiency

225 Diversity, stability and regional growth in the United States 209 (in the maximum power sense) entails a prioritization of short-term adaptation/optimization that comes, potentially, at the expense of long-term stability. If optimality exists it will be temporary, because through evolution, selection, and innovation it is easily transformed into maladaptive traits. Under such conditions, diversity is a key element of long term stability 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 of alternative options, which increases the probability that pre-adaptations to altered conditions exist. This is referred to as evolutionary potential (ibid.). While selection operates as a short-term adaptive force that reduces diversity to narrow and temporally adapted features, selection does not guarantee survival in the long run (Matutinovic, 2001). Diversity persists because of imperfect adaptation and the counter acting influence of other sorting mechanisms. Selection rewards those individuals or firms that are relatively more efficient (generally characterized by lower input output ratios) but a firm s competitive position is also improved by exaptation and exogenous shocks (Gowdy, 1992). For instance, exaptation could refer to an increase in efficiency of suppliers of a firm A that translates into lower costs of inputs and in turn a lower input output ratio of firm A. This improvement of efficiency is achieved without any actual technological or organizational changes by firm A. Exogenous shocks, such as a rise in energy prices, can influence firm A s efficiency through a shift in relative prices of input factors. Firms that use relatively small amounts of energy will improve their efficiency relative to firms that use larger amounts 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, because these sorting mechanisms might reward relatively inefficient firms. Diversity might therefore be as much an evolutionary outcome as specialization. Within bounds, regions should therefore embrace rather than eliminate redundancy. In ecology, redundancy of agents (and pathways) stands for ecosystem overhead. Ulanowitz in Matutinovic (1992) argues that ecosystem overhead evolves: (1) as a response to the opportunity for the complete use of available resources (efficiency in the second law sense); (2) to prevent system brittleness; (3) to preserve its adaptive response and creativity; and (4) to preserve its reliability (Matutinovic, 2002: 434). Similarly, there are good reasons for economic systems to embrace overhead or redundancy. From an evolutionary point of view, firms, institutions, regions and countries are likely to be forced into a trade-off between realizing short-term

226 210 Spatial systems profits (adaptation to current conditions to achieve a local optima) and long-term flexibility to enhance the adaptive potential and the ability to react to technological paradigm shifts, exogenous shocks and industrial shifts (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 of view this entails a strengthening of existing well-performing sectors (probably clusters) but without completely eliminating those firms and sectors that appear less efficient and redundant at present. This view also resonates with arguments made by innovation system researchers (Lundvall and Johnson, 1994; Edquist, 1997). Contrary to biological systems, socio-economic systems actively produce diversity. This means that the diversity-selection feedback works much faster in social systems and hence, any reduction in diversity might be translated more rapidly into adaptability problems. Furthermore, economic systems are often characterized by increasing returns based on internal and external economies, cumulative technological change, learning and network externalities and complementary production factors that can result in path-dependent evolution and lock-in. Diversity helps to break lock-in and path dependence (Grabher, 1993; Arthur, 1994; Grabher and Stark, 1997). Diversity at the level of the region refers to diversity in labor (skills), firms, industrial sectors, organizations and institutional environments but also the network connections between local and non-local agents (Granovetter, 1973; Grabher and Stark, 1997; Matutinovic, 2002). Diversity can thus be seen as a risk-minimizing strategy similar to portfolios in business economics (Chandra, 2003). The theoretical arguments on the relationships among diversity, stability and resilience are rather general and biology, ecology and complex adaptive systems theory have yet to solve the exact linkages between them. Despite these shortcomings, the theoretical arguments put forward demonstrate that a narrow policy focus on regional efficiency is problematic. Strategies to maximize efficiency in the short term might pose problems for economic prosperity over longer time horizons and hence, prioritize implicitly the needs of current generations at the expense of future generations. To complicate matters further, the impact of economic policies will lead to conflicting outcomes not only at various temporal but also at various spatial scales. Competition between regions might yield positive economic returns for some regions, but also result in the unnecessary duplication of infrastructure, services and organizations that appear wasteful from the perspective of the national state (Harvey, 1989; Hubbard and Hall, 1998). On the other hand, regional specialization might result in

227 Diversity, stability and regional growth in the United States 211 positive region-specific externalities that maximize wealth at the national level at the expense of intra-regional diversity (Krugman, 1991; Fujita et al., 1999, Neary, 2001). Nations might manage risks by maintaining a portfolio of specialized regions similar to assets of companies. Decline in some regions will be compensated by growth in other regions. In this case, the national scale receives priority over the regional scale where the average well-being of national citizens will increase at the expense of declining welfare in declining regions. If these intertemporal and interspatial tradeoffs do indeed exist, the trade-offs at various temporal and spatial scales have to be made explicit in regional policy templates rather than hidden behind the assumption that free markets will lead to a (global) welfare optimum. While evolutionary theory provides us with interesting insights into the trade-off between diversity, stability and growth, the exact relationship between technological and industrial diversity and economic growth and stability have been insufficiently developed so far. The relationship between diversity and economic growth has been addressed extensively by new geographical economists (this literature cannot be discussed in this chapter but for overviews on the new geographical economics, see Martin (1999), Sheppard (2000a, b), Neary (2001), Duranton and Puga (2004), Frenken et al. (2005) and Robert-Nicoud (2005) and for an empirical attempt to disentangle the effects of urbanization and localization economies on metropolitan labor productivity, see Rigby and Essletzbichler (2002)) while regional scientists applied portfolio theory to examine the empirical relationship between diversity and stability. Industrial Diversity and Portfolio Theory In business economics and industrial organization, the concept of portfolio refers to the valuation of the collection of a company s assets to examine the impact of product diversity on corporate profitability growth. The basic underlying principle is that diversity of assets reduces risk. Ideally a company diversifies into technologically related industries/products in order to maximize economies of scope, but also industries that are characterized by unrelated demand in order to protect overall sales from demand shocks in individual product markets. This reasoning has a striking similarity to the arguments by Giampietro and Mayumi (1997) on the behavior of complex adaptive systems. Although regions cannot go bankrupt in the same way as corporations do, regions expand and contract over the business cycle. Regional contraction manifests itself through plant closures, low entry rates and a shrinking employment base. Once a negative cumulative cycle is set in motion it is often hard to switch to a new path of regional

228 212 Spatial systems economic growth, to attract businesses and jobs. In severe cases of economic decline, regions are confronted with very fast rates of employment decline. This is particularly the case if the economic base is dependent on a few companies and/or industrial sectors. If individual plants are closed down because of structural problems occurring in this sector, related industries follow rapidly and whole areas can be transformed into ghost towns in 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 unfolding processes (Oswalt, 2004). Although the region does not go bankrupt in the same sense as firms do, capital has to be scrapped, workers laid off and, in the case of prolonged crisis, have to move to other regions. In most circumstances, not all sectors of an economy decline at the same time or at equal rates. Borrowing from portfolio theory, it is therefore possible to think of regional diversification as a strategy to reduce the risk of economic decline. Developing a portfolio of industries whose demand is largely uncorrelated might be a useful strategy of regions to avoid big fluctuations in rates of economic growth and to shield them in part from economic decline during recessions (Baldwin and Brown, 2004). Clearly these arguments are rather abstract and require refinement. The same levels of regional diversity might result from very different industry mixes, and some of them might be more favorable than others. Even if the levels of regional diversity remain constant, the underlying industry mix of regions might change over time. And finally, what are the appropriate temporal and spatial scales to examine the evolution of regions? Diversity might be useful for the long run at the expense of short-run economic growth. Regional economic specialization might yield high levels of efficiency that benefit actors at the regional and national scales and that maintain industrial diversity at the national scale. Frenken et al. (2005) describe new geographical economics and portfolio approaches as static because variety at a single point in time relates to regional growth. Boschma and Lambooy (1999) argue that urbanization economies and Jacobs externalities are more important during the emergence of new industries and technological paradigms when industries have not yet generated their specific skills, or supplier and institutional requirements, but that localization economies might become more important once these factors are created (see also Boschma and Frenken, 2003). This literature is important as it brings a dynamic perspective to the literature and demonstrates how industrial and technological diversity influence regional growth at various stages of industry life cycles, but regions are somehow 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

229 Diversity, stability and regional growth in the United States 213 distribution of characteristics within regions and the changes in regional aggregates such as growth, productivity, profitability and stability. This avenue is pursued in the following empirical analysis: regions are conceptualized as an assembly of industries and the distribution of employment among these industries (indicating regional industrial diversity/concentration) is expected to exert an influence on changes in regional economic growth and stability. 3. EMPIRICAL ANALYSIS The empirical part of this chapter is based on employment data from the US county business patterns ( ). The goal of the analysis is the establishment of a negative statistical relationship between economic growth and stability and a positive relationship between industrial diversity and economic stability at the level of the economic areas of the Bureau of Economic Analysis (BEA) through the application of spatial econometric techniques. In order to measure the impact of industrial diversity on regional volatility, it is necessary to control for the influence of additional explanatory variables. The choice of variables is informed by the theoretical discussion and empirical work in regional science (for example, Baldwin and Brown, 2004). Regional stability/volatility is measured as the variance of annual regional employment growth rates. According to Baldwin and Brown, the variance will be influenced by the diversity of a region s industrial structure, the variance of its industries growth rates and the covariance between those growth rates. The correlates of volatility are chosen from a set of structural characteristics of regions. The names, definitions and expected signs of these variables are summarized in Table The simplest and most widely used measure of diversity is probably the Herfindahl index (Duranton and Puga, 2000; Chandra, 2003; Baldwin and 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)

230 214 Spatial systems Brown, 2004). Experimentation with entropy measures of diversity did not change the conclusions of the chapter. The Herfindahl index H of a BEA region r is measured as H r i s 2 ir (10.1) where s ir E ir i E ir and E ir refers to employment in sector i in BEA region r. The index varies between 1 (all employment is concentrated in one sector) and 1/n (employment is distributed equally among all sectors). A higher value indicates greater concentration of employment in fewer sectors (lower diversity), while a lower value indicates a more even distribution of employment across sectors (higher diversity). The index has been constructed for the base year (1975) using the 1972 SIC 3 3-digit system. Based on the theoretical discussion on diversity and volatility, a positive relationship between the level of concentration (a high Herfindahl index) and volatility (high variance of annual growth rates) is expected. Employment levels and growth rates are also expected to relate to volatility. Although it is expected that total regional employment is correlated with diversity and hence, the effect of employment size subsumed by the effect of diversity on volatility, Malizia and Ke (1993) argue that larger regions are more stable than smaller regions and that size (measured as total employment) might have a positive effect on regional stability independent of the effect of diversity. The impact of size on volatility will also depend on the geographic concentration of markets for products. If firms in larger regions sell a larger share of their product in local markets, then the growth rates of a region s industries are more likely to be correlated because they will be dependent on the same market and subject to the same economic shocks. In this case, larger regions will be more volatile than smaller regions even if they are characterized by similar levels of industrial diversity. The relationship between size and volatility is therefore ambiguous. Malizia and Ke found a U-shaped relationship between growth and volatility, suggesting that regions that have concentrations in fast-growing industries have higher growth rates, while regions with concentrations in fast-declining industries have lower growth rates. On the other hand, more diverse regions are characterized by more stability and average growth rates. The U-shaped relationship between growth and volatility has been confirmed by Baldwin and Brown (2004) for Canadian census regions and manufacturing industries. Contrary to the U-shaped relationship detected by Malizia and Ke and Baldwin and Brown, a linear (positive) relationship between growth and volatility for BEA regions was discovered (see Figure

231 Diversity, stability and regional growth in the United States , below). The squared growth rate was thus omitted from the set of independent variables. The average plant size in a region is expected to exert a positive influence on regional economic stability because smaller plants are generally newer and 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 of commodities and are thus less vulnerable to market fluctuations affecting a particular product. Product variety tends to be lower in smaller firms which will decrease their ability to adapt to market fluctuations affecting a specific product. Hence, we would expect a negative relationship between average plant size and volatility. The negative correlation coefficient between average plant size and stability confirms this relationship (see Table 10.4, below). Baldwin and Brown (2004) add export shares and the share of employment in different types of industries (for example, resource based) as explanatory variables. Unfortunately, export data are not available for BEA regions but the share of regional employment in resourced-based industries was included. It can be expected that regions with a high share of resourcebased industries such as mining, agriculture and forestry, logging, lumber and 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 of resource-based industries and volatility is expected. The data used in this analysis are based on county business patterns from 1975 to The data have been aggregated to the level of BEA regions because 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 and For many counties, actual figures have been suppressed and replaced by employment size classes. Because of the large amount of undisclosed information (in particular for smaller counties) and in order to reduce measurement error, county employment at the SIC 4-digit level has been aggregated to SIC 3-digit employment and the diversity measures have been calculated at the SIC 3-digit level. If information for individual counties was not reported at the SIC 3-digit level, the average value of the employment size class (for example, 10 for employment size class 0 19) was used to impute the missing information. Because counties were then aggregated to the new BEA regions, measurement error will be relatively small and is unlikely to influence the results. Changes in county and BEA definitions have been considered in order to keep the geography constant over the whole period. Overall, the dataset spans 27 years and includes the 177 BEA areas of the continental United States.

232 216 Spatial systems (a) (b) (c) Key (1 = lowest; 5 = highest) Source: County Business Patterns, Figure 10.1 Volatility (a), diversity (b), growth (c) Figure 10.1 maps the key variables volatility, diversity and growth for the 177 BEA areas. Growth and volatility are measured over the whole period from 1975 to 2002 while diversity is measured for the base year, Table 10.2 lists the top 10 and bottom 10 cities with respect to stability, diversity and growth. Figure 10.1 and Table 10.2 reveal the following geographic pattern. The regions characterized by the highest levels of stability tend to

233 Table 10.2 BEA area rankings by stability, diversity and growth Rank BEA area Stability BEA area Diversity BEA area Growth 1 Philadelphia, PA-NJ Little Rock, AR Las Vegas, NV 6.87 DE-MD 2 Lincoln, NE Harrisburg, PA Austin, TX New York, NY-NJ-CT Jackson, MS Sarasota, FL 5.42 PA 4 Albany, NY New York, NY-NJ-CT Phoenix, AZ 5.16 PA 5 Dover, DE Memphis, TN-MS-AR Fayetteville, AR-MO Richmond, VA Atlanta, GA-AL Bend, OR Harrisburg, PA Philadelphia, PA-NJ Flagstaff, AZ 4.94 DE-MD 8 Scranton, PA Dallas-Fort Worth, TX Sacramento, CA-NV Omaha, NE-IA Chicago, IL-IN-WI Colorado Springs, CO Rochester, NY Columbia, SC Orlando, FL San Angelo, TX Morgantown, WV State College, PA Pendleton, OR Bend, OR Waterloo, IA Eugene, OR Port Arthur, TX Springfield, IL Sarasota, FL Pueblo, CO Erie, PA Farmington, NM Reno, NV Buffalo, NY Reno, NV Odessa, TX Odessa, TX Odessa, TX Flagstaff, AZ Davenport, IA-IL Lafayette, LA Waterloo, IA Cleveland, OH Casper, WY Gulfport, MS Pittsburgh, PA Bend, OR Las Vegas, NV Port Arthur, TX 0.76 Source: County Business Patterns,

234 218 Spatial systems be concentrated in the Mid-Atlantic region including areas such as Philadelphia, New York and Rochester, while among the most volatile are many of the resource-based economies of Oregon, Texas and Wyoming. The most diverse regions are those surrounding large urban areas such as New York, Philadelphia, Dallas, Chicago, Atlanta or Memphis, while the more specialized areas tend to be either resource-based economies or those focusing on tourism such as Las Vegas or Reno. The fastest-growing regions are located in Florida and the southwest of the country and include retirement areas and high-tech centers. The slowest-growing regions are found in the old manufacturing heartland and include regions such as Buffalo, Cleveland and Pittsburgh. Although some regions score high/low on several variables, others do not follow a clear pattern. Figure 10.2 plots the relationship between volatility on the vertical axis and growth/diversity on the horizontal axes. Both relationships are positive and significant at the level. Descriptive statistics of all dependent and independent variables as well as the logarithmic values are presented in Table 10.3, while the raw correlation coefficients are presented in Table Table 10.3 highlights considerable variation in diversity, growth and stability, although the variation is considerably smaller than for Canadian Census regions (Baldwin and Brown, 2004). The correlation coefficients reveal that most of the independent variables are correlated with stability. Table 10.4 indicates that specialized regions and those characterized by higher rates of economic growth, smaller employment size and smaller average plant size tend to be more stable (that is, have lower variances of growth rates). Also of interest is the positive relationship between growth and diversity: specialized regions appear to grow more rapidly but are also characterized by higher volatility. Table 10.3 suggests the presence of extreme outliers with respect Volatility Corr = 0.31 (p = ) Growth Volatility Corr = 0.44 (p = ) Diversity Source: County Business Patterns, Figure 10.2 Relationship between volatility and growth/diversity

235 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 HERF MGROWTH EMP PLSIZE R LOGVARGROWTH LOGHERF LOGMGROWTH LOGEMP LOGPLSIZE LOGR Source: County Business Patterns, to the dependent and independent variables that might drive overall results. Hence, the natural logarithms of the variables are taken to remove the effect of those outliers. Table 10.4 suggests that the linear relationships between the logged variables tend to become stronger. The basic regression model estimated may be written as: y X, (10.2) where y is an N 1 vector of observations on the dependent variable, X is an N K matrix of observations on K independent variables, is a K 1 vector of regression coefficients, and is an N 1 vector of errors assumed to be normally and independently distributed. As discussed above, the dependent variable is the variance of regional growth rates (VARGROWTH) and the independent variables are HERF75, GROWTH, EMP75, PLSIZE75 and R75. Figure 10.1 suggests considerable spatial autocorrelation in the dependent variable. Employing spatial contiguity weights, a Moran s I value of (significant at the level) suggests the presence of strong spatial autocorrelation of the independent variable. In the presence of spatial autocorrelation, ordinary least squares (OLS) estimates may be inconsistent (Anselin, 1988; Anselin and Rey, 1991). Spatial dependence is of two basic forms, error dependence and lag dependence. In the spatial error model the errors can no longer be assumed independent and identically distributed and the regression model takes the following form: y X W, (10.3)

236 220 Spatial systems Table 10.4 Correlation coefficients between dependent and independent variables (p-values in parentheses) VARGROWTH HERF75 MGROWTH EMP75 PLSIZE75 R75 VARGROWTH (0.0001) (0.0001) (0.0042) (0.0001) (0.0001) HERF ( (0.246) (0.003) (0.0796) (0.0001) MGROWTH (0.0001) (0.246) (0.0514) (0.0001) (0.7212) EMP (0.0042) (0.003) (0.0514) (0.0001) (0.0018) PLSIZE (0.0001) (0.0796) (0.0001) (0.0001) (0.0001) R (0.0001) (0.0001) (0.7212) (0.0018) (0.0001) LOG LOG LOG LOG LOG LOG VARGROWTH HERF75 MGROWTH EMP75 PLSIZE75 R75 LOG VARGROWTH (0.0001) (0.0001) (0.0001) (0.0001) (0.0001) LOGHERF (0.0001) (0.6416) (0.0001) (0.0001) LOG MGROWTH (0.0001) (0.6416) (0.0343) (0.0001) (0.3718) LOGEMP (0.0001) (0.0001) (0.0343) (0.0001) (0.0001) LOGPLSIZE (0.0001) (0.0007) (0.0001) (0.0001) (0.0001) LOGR (0.0001) (0.0001) (0.3718) (0.0001) (0.0001) where is the spatial autoregression coefficient, W is an N N matrix of spatial weights representing the geography of the observational units (BEA s), and is an N 1 vector of errors assumed to possess the usual properties. In this form, spatial dependence influences the error term only and it has been shown to influence the power of tests for heteroscedasticity and the structural stability of regression coefficients. In the spatial lag model, the standard regression equation may be rewritten as: y Wy X, (10.4) where is the spatial autoregression coefficient. In this form, the value of the dependent variable at a particular location is jointly determined by its

237 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 error Form Linear Log-linear Log-linear Log-linear Dependent LN LN LN variable VARGROWTH VARGROWTH VARGROWTH VARGROWTH Constant 4.87* 5.27** 3.87** 5.14** (2.05) (14.03) (8.41) (12.76) (LN)HERF ** 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)EMP e (0.81) (1.25) (0.44) (1.01) (LN)PLSIZE * ( 1.72) ( 2.06) ( 1.19) ( 1.89) (LN)R ** 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.) p p Log-likelihood AIC SC Diagnostics for heteroscedasticity Breusch Pagan 51.22** Koenker Basset 14.79* Diagnostics for spatial dependence LM-ERROR 5.85* 21.76** LM-LAG 5.51* 27.38** Notes: OLS ordinary least squares; ML maximum likelihood; **, *, : significant at the 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 ML estimation. AIC Akaike information criterion; SC Schwartz criterion. 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 Specialized and faster-growing regions and those with higher shares of resource-based industries are characterized by more volatility in growth rates. All relationships are significant at the 0.01 level. The average plant size

238 222 Spatial systems is negatively related to volatility, supporting the theoretical arguments discussed above. The relationship is significant at the 0.1 level only. The size of the region is positively related to volatility, lending some support to Fujita et al. s (1999) argument that the demand for products in large regions is more likely to be correlated, exacerbating demand shocks. However, Table 10.4 revealed high correlations between regional size and most other independent variables, suggesting that the size effects might have been picked up by other variables (for example, HERF75). Furthermore, Table 10.5 shows that the positive relationship between EMP75 and VARGROWTH is not statistically significant. In addition to the parameter estimates and t-values, a set of diagnostic statistics have been added. An adjusted R-square value of indicates a relatively good fit of the original model. However, tests on heteroscedasticity and spatial dependence reveal that both are present in the model. Heteroscedasticity could have been the result of correlated spatial errors. However, even after correcting for spatial lags/errors, heteroscedasticity posed a problem. The scatterplots depicted in Figure 10.2 seemed to suggest that the variance of volatility increases with higher rates of both growth and diversity, and that no single variable could be easily identified to cause heteroscedasticity. Gujarati (2003) suggests that the log transformation of variables often helps to eliminate heteroscedasticity. Model versions 2 4 present the results for the log-linear models. The fact that parameter estimates can be interpreted as elasticities is an added advantage of the log-linear model. Model 2 presents the results for the OLS estimates without correction for spatial dependence. The signs of the parameter estimates for the log-linear version do not change, although the parameter estimate for average plant size is now significant at the 0.05 level and the adjusted R-square value indicates a moderately worse model fit. On the other hand, the Breusch Pagan and Koenker Bassett tests reveal no heteroscedasticity, while spatial dependence is still present in the models. Lagrange multiplier tests suggest the presence of both spatial lag and spatial error. Model 3 provides the results for the spatial lag model. The model results are based on maximum likelihood estimation and the values in parentheses are z- rather than t-values. The signs of the parameter estimates are consistent with those of Models 1 and 2. The negative estimate for the size of plants is no longer significant, but all the goodness-of-fit measures indicate a clear improvement from model version 2. Furthermore, the parameter estimate for the spatially lagged dependent variable is positive and significant at the 0.01 level. The results suggest that volatility of growth in a region is also influenced by volatility of growth in the neighboring regions. The results for the spatial error model (Model 4) are similar to the results of the spatial lag model. The estimate for average plant size is significant again at the 0.1 level and the goodness-of-fit statistics suggest an

239 Diversity, stability and regional growth in the United States 223 almost identical performance when compared to the spatial lag model. The spatial lag parameter,, is also positive and significant at the 0.01 level. Anselin (1988, 1992) advises using performance indicators such as the log-likelihood, Akaike information or Schwartz criteria to inform model selection. Because the performance indicators for both models are almost identical, no obvious choice for the best model emerges. Because the parameter estimates are very similar and since the purpose of the model is not predication but the establishment of statistical relationships between variables, this does not pose a major predicament. Independent of the exact model specification, a comparison of the elasticities suggests that a change in diversity has the highest impact on change in regional economic stability, followed by a change in average growth rate and the change in share of resource-based industries (keeping in mind that the impact of average plant size is barely significant). The results confirm the work by other researchers and highlight the importance of industrial diversity for regional economic stability. 4. CONCLUSION This chapter employed evolutionary theory to develop arguments on the trade-off between short-term adaptation and long-term adaptability. At present, little thought is given to the potentially negative impacts of clusterbased regional policies predicated on the spatial concentration of functionally integrated sectors. The concentration of economic activity in a few economic areas is likely to boost short-term productivity growth and profit rates through the exploitation of externalities based on the local skill base, knowledge spillovers, and traded and untraded interdependencies. The negative side of specialization is a decline in adaptive flexibility, the ability to react to continually changing economic environments. The chapter examined the relationship among stability, growth and diversity for 177 BEA areas over the , period using employment data from county business patterns. The analysis revealed a strong positive relationship between diversity and stability on the one hand, and growth and instability, on the other. While these results confirm work in other countries and provide some credibility to evolutionary theories of regional economic change, it is important not to overstate the results. Although regional economic stability is desirable because it facilitates planning for technical and social infrastructure and avoids the pitfalls of fast growth (congestion, rising house prices, environmental degradation, overinvestment in infrastructure), stability (small variances in growth rates) coupled with economic decline is problematic if the decline is rapid and investments have to be written off

240 224 Spatial systems rapidly. In other words, from a policy point of view, it still has to be worked out what kind of regional stability is desirable, bearing in mind the trade-off between growth and stability. Furthermore, the Herfindahl index provides a general measure of industrial diversity, but does not capture the degree of functional relation between sectors. Regions can contain a large number of different but functionally integrated economic sectors that react in a similar fashion to demand shocks. The impact of diversity on stability will be influenced strongly by the degree of functional integration of sectors and hence, future work will have to pay more attention to this aspect of diversity (Frenken et al., 2005). However, based on insights from evolutionary theory and the empirical results, industrial, institutional, skill, technological and social diversity should be elevated to a general principle of regional economic development even at the cost of short-term welfare losses. This is imperative 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 earlier draft of this chapter. The usual disclaimer applies. 1. Volatility is interpreted as the opposite of stability and will be measured as the variance in annual rates of employment change. 2. Ecology offers three different notions of efficiency: (1) first law efficiency, or simply the fraction of energy input that appears as output; (2) efficiency in the second law sense where resources are being used more thoroughly by a diverse set of agents, having different single-use efficiencies (the most efficient agent is the one that effects the most complete use of the available resource, regardless of the rate of use); (3) efficiency in the maximum power sense, where an agent uses a resource to provide either the quickest return or the greatest rate of output (Matutinovic, 2002: 433). Economics prioritizes definition (3). From a maximum power efficiency perspective, less-efficient firms are considered redundant. Second law efficiency is generally absent from economic policy discourse although it might be desirable from an equity point of view. 3. Standard Industrial Classification. REFERENCES Alchian, A. (1950), Uncertainty, evolution and economic theory, Journal of Political Economy, 58: Andersen, E.S. (2004), Population thinking and evolutionary economic analysis: exploring Marshall s Fable of Trees, Danish Research Unit for Industrial Dynamics (DRUID) Working Paper, Anselin, L. (1988), Spatial Econometrics, Dordrecht: Kluwer Academic. Anselin, L. (1992), SpaceStat tutorial: a workbook for using SpaceStat in the analysis of spatial data, Santa Barbara: NCGIA National Center for Geographic Information and Analysis.

241 Diversity, stability and regional growth in the United States 225 Anselin, L. and S. Rey (1991), The performance of tests for spatial dependence in linear regression models, Geographical Analysis, 23: Arthur, B. (1994), Increasing Returns and Path Dependence in the Economy, Ann Arbor, MI: University of Michigan Press. Attaran, M. (1986), Industry diversity and economic performance in U.S. areas, Annals of Regional Science, 20: Baldwin, J., T. Dunne and J. Haltiwanger (1998), A comparison of job creation and job destruction in Canada and the United States, The Review of Economics and Statistics, 80: Baldwin, J.R. and W.M. Brown (2004), Regional manufacturing employment volatility in Canada: the effects of specialization and trade, Papers in Regional Science, 83: Baldwin, R.E., R. Forslid, P. Martin, G.I.P. Ottaviano and F. Robert-Nicoud (2003), Economic Geography and Public Policy, Princeton, NJ: Princeton University 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 ICMA University, pp Begg, I. (ed.) (2002), Urban Competitiveness, Bristol: Policy Press. Berman, E. and L. Bui (2001), Environmental regulation and productivity: evidence from oil refineries, Review of Economics and Statistics, 83: Black, D. and V. Henderson (2003), Urban evolution in the US, Journal of Economic Geography, 3: Boschma, R.A. (2004), Competitiveness of regions from an evolutionary perspective, Regional Studies, 38: Boschma, R.A. and K. Frenken (2003), Evolutionary economics and industry location, Review of Regional Research, 23: Boschma, R.A. and J.G. Lambooy (1999), Evolutionary economics and economic geography, Journal of Evolutionary Economics, 9: 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 territorial restructuring in contemporary Europe, Review of International Political Economy, 5: 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: Chandra, S. (2003), Regional economy size and the growth-instability frontier: evidence from Europe, Journal of Regional Science, 43: Conroy, M.E. (1974), Alternative strategies of regional industrial diversification, Journal of Regional Science, 14: Conroy, M.D. (1975), The concept and measurement of regional industrial diversification, Southern Economic Journal, 41: 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:

242 226 Spatial systems Dosi, G. and R. Nelson (1994), An introduction to evolutionary theories in economics, Journal of Evolutionary Economics, 4: Duranton, G. and D. Puga (2000), Diversity and specialization in cities: why, where and when does it matter?, Urban Studies, 37: Duranton, G. and D. Puga (2001), Nursery cities: urban diversity, process innovation and the life-cycle of products, American Economic Review, 91: Duranton, G. and D. Puga (2004), Micro-foundations of urban agglomeration economies, in V. Henderson and J.-F. Thisse (eds), Handbook of Regional and Urban Economics, vol. 4. Amsterdam: North-Holland, pp Edquist, C. (1997), Systems of Innovation: Technologies, Institutions and Organizations, London and Washington, DC: Pinter. Essletzbichler, J. and D.L. Rigby (2005a), Technological evolution as creative destruction of process heterogeneity: evidence from US plant level data, Economic Systems Research, 17: Essletzbichler, J. and D.L. Rigby (2005b), Competition, variety and the geography of technology evolution, Tijdschrift voor Economische en Sociale Geografie, 95: Feldman, M. and D. Audretsch (1999), Innovation in cities: science-based diversity, specialization and localized competition, European Economic Review, 43: Frenken, K., F.G. van Oort, T. Verburg and R. Boschma (2005), Variety and regional economic growth in the Netherlands, Papers in Evolutionary Economic Geography no Utrecht University, Urban and Regional research center Utrecht. Friedman, M. (1953), Essays in Positive Economics, Chicago: University of Chicago Press. 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 socioeconomics systems based on hierarchy theory and its application to sustainability, Structural Change and Economic Dynamics, 8: Glaeser, E., H. Kallal, J. Schenkman and A. Shleifer (1992), Growth in cities, Journal of Political Economy, 100: Gould, S.J. and R.C. Lewontin (1979), The spandrels of San Marco and the Panglossian paradigm: a critique of the adaptionist programme, Proceedings of the Royal Society of London series, 205(1161): Gowdy, J.M. (1992), Higher selection processes in evolutionary economic change, Journal of Evolutionary Economics, 2: Grabher, G. (1993), The weakness of strong ties: the lock-in of regional development in the Ruhr area, In G. Grabher (ed.), The Embedded Firm,New York and London: Routledge, pp Grabher, G. and D. Stark (1997), Organizing diversity: evolutionary theory, network analysis and postsocialism, Regional Studies, 31: Granovetter, M. (1973), The strengths of weak ties, American Journal of Sociology, 78: 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): Henderson, J.V. (1988), Urban Development: Theory, Fact and Illusion, Oxford: Oxford University Press.

243 Diversity, stability and regional growth in the United States 227 Henderson, J.V. (1997), Externalities and industrial development, Journal of Urban Economics, 42: Henderson, J.V. (2003), Marshall s scale economies, Journal of Urban Economics, 53: Henderson, J.V., A. Kuncoro and M. Turner (1995), Industrial development in cities, Journal of Political Economy, 103: 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: Holling, C.S. (2001), Understanding the complexity of economic, ecological, and social systems, Ecosystems, 4: Holling, C.S. (2004), From complex regions to complex worlds, Ecology and Society, 9(11), available at 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 Hudson, R. (1999), The learning economy, the learning firm and the learning region: a sympathetic critique to learning, European Urban and Regional Studies, 6(1): Iwai, K. (1984a), Schumpeterian dynamics II: technological progress, firm growth and economic selection Journal of Economic Behavior and Organization, 5: Iwai, K. (1984b), Schumpeterian dynamics: an evolutionary model of innovation and imitation, Journal of Economic Behavior and Organization, 5: 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 Johnson, K.P. and J.R. Kort (2004), 2003 redefinition of the BEA economic areas, Survey of Current Business, November: Jovanovic, B. (1982), Selection and the evolution of industry, Econometrica, 50(3): Kitson, M., R. Martin, and P. Tyler (2004), Regional competitiveness: an elusive yet key concept?, Regional Studies, 38: Kort, J.R. (1981), Regional economic instability and industrial diversification in the U.S., Land Economics, 57: Krantz, R. (2002), Employment in business services: a year of unprecedented decline, Monthly Labor Review, April: Krugman, P. (1991), Increasing returns and economic geography, Journal of Political Economy, 99: Krugman, P. (1994), Competitiveness: a dangerous obsession, Foreign Affairs, March/April: Leitner, H. and E. Sheppard (1998), Economic uncertainty, inter-urban competition 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,

244 228 Spatial systems 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 Regional Research, 23: Lundvall, B.-A. and B. Johnson (1994), The learning economy, Journal of Industry Studies, 1: Malizia, E.E. and S. Ke (1993), The influence of economic diversity on unemployment and stability, Journal of Regional Science, 33: 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, Martin, R. and P. Sunley (2003), Deconstructing clusters: chaotic concept or policy panacea?, Journal of Economic Geography, 3: Matutinovic, I. (2001), The aspects and the role of diversity in socioeconomic systems: an evolutionary perspective, Ecological Economics, 39: Matutinovic, I. (2002), Organizational patterns of economies: an ecological perspective, Ecological Economics, 40: Mayumi, K. and M. Giampietro (2001), The epistemological challenge for modelling sustainability: exploring the difference in meaning of risk, uncertainty and ignorance, Paper prepared for the EC High Level Scientific Conference ESEE Frontiers 1: Fundamental Issues of Ecological Economics, Cambridge, UK, 4 7 July. Metcalfe, S. (1994), Competition, Fisher s principle and increasing returns in the selection process, Journal of Evolutionary Economics, 4: Metcalfe, S. (1998), Evolutionary Economics and Creative Destruction, London and New York: Routledge. Metcalfe, S. and M. Gibbons (1986), Technological variety and the process of competition, Economie Appliquée, 39: Neary, J.P. (2001), On hypes and hyperbolas: introducing the new economic geography, Journal of Economic Literature, 39: Nelson, R.R. (1995), Recent evolutionary theorizing about economic change, Journal of Economic Literature, 33: 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 von Postfordismus 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: Quigley, J.M. (1998), Urban diversity and economic growth, Journal of Economic Perspectives, 12: Rammel, C. and M. Staudinger (2002), Evolution, variability and sustainable development, International Journal of Sustainable Development and Global Ecology, 9: 305. Rammel, C. and J.C.J.M. van den Bergh (2003), Evolutionary policies for sustainable development: adaptive flexibility and risk minimizing, Ecological Economics, 47: Rigby, D.L. and J. Essletzbichler (1997), Evolution, process variety and regional trajectories of technological change in US manufacturing, Economic Geography, 73:

245 Diversity, stability and regional growth in the United States 229 Rigby, D.L. and J. Essletzbichler (2002), Agglomeration economies and productivity differences in US cities, Journal of Economic Geography, 2: Rigby, D.L. and J. Essletzbichler (2006), Technological variety, technological change and the geography of production techniques, Journal of Economic Geography, 6: Forthcoming. Robert-Nicoud, F. (2005), The structure of simple New economic Geography models (or, on identical twins), Journal of Economic Geography, 5: Saviotti, P. (1991), The role of variety in economic and technological development, in P. Saviotti and S. Metcalfe (eds), Evolutionary Theories of Economic and Technological Change, Chur: Harwood Academic Publishers, pp 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 evolutionary economics, in P.P. Saviotti and J.S. Metcalfe (eds), Evolutionary Theories of Economic and Technological Change, Chur: Harwood Academic Publishers, pp Schoening, N.C. and L.E. Sweeney (1992), Proactive industrial development strategies and portfolio analysis, Review of Regional Studies, 22: Schütz, J. (1999), The value of systemic reasoning, Ecological Economics, 31: 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, interdependence 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 Sheppard, E. (2000b), Competition in space and between places, in E. Sheppard and T. Barnes (eds), A Companion to Economic Geography, Oxford: Blackwell, pp Siegel, P.B., T.G. Johnson and J. Allwang (1995), A structural decomposition of regional economic instability: a conceptual framework, Journal of Regional Science, 35: Smith, E.D. (1990), Economic stability and economic growth in rural communities: dimensions relevant to local employment creation strategy, Growth and Change, 21: Ulanowicz, R.E. (1997), Ecology, the Ascendent Perspective, New York: Columbia University Press. Van den Bergh, J.C.J.M. (2004), Evolutionary thinking in environmental economics; retrospect and prospect, in J. Foster and W. Hölzl (eds), Applied Evolutionary Economics and Complex Systems, Cheltenham, UK and Northampton, MA, USA: Edward Elgar, pp Vrba, E.S. and S.J. Gould (1986), The hierarchical expansion of sorting and selection: sorting and selection cannot be equated, Paleobiology, 10: Vromen, J.J. (1995), Economic Evolution: An Enquiry into the Foundations of New Institutional Economics, London and New York: Routledge. Wagner, J.E. and S.C. Deller (1998), Measuring the effects of economic diversity on growth and stability, Land Economics, 74:

246 11. Inter-regional knowledge flows in Europe: an econometric analysis Mario A. Maggioni and T. Erika Uberti* 1. INTRODUCTION The aim of this chapter is to analyse the impact of knowledge on regional economic development and, consequently, on regional disparities across five major European countries: France, Germany, Italy, Spain and the United Kingdom. In particular our focus is on the nature of knowledge, not only as a fixed cost in the production process (leading to scale economies), an investment good (influenced by accumulation and depreciation dynamics) and an experience good (whose quality attributes can be detected only upon using, or consuming, the good), but also as a relational good displaying network externalities. In this chapter we analyse the manifold nature of knowledge through the analysis of four distinct but complementary phenomena (Internet hyperlinks, European research networks, European Patent Office (EPO) co-patent applications, Erasmus student mobility) which characterise knowledge as an intrinsic relational structure (directly) connecting people, institutions and (indirectly) regions across five European countries. Two main research questions are addressed: the first deals with the notion of regional disparities; the second refers to the different concepts of distance, namely geographical, functional and sectoral. Regional disparities can no longer be defined only in terms of statistical differences in the values of standard macroeconomic indicators. Knowledge matters more and more in defining both the level and the growth rate of a given region GDP (Sapir et al., 2004). For this reason, new relational indicators have to be built and compared in order to develop a new kind of (relational) analysis able to complement the usual attributional one. Traditionally, regional economic disparities have been ascribed to peripherality measured by the distance from the main centres of population and economic activity and/or to a high level of dependence on declining sectors (mainly mature industries ). The scale of regional and 230

247 Inter-regional knowledge flows in Europe 231 other disparities, as well as the political approach and the specific policy instruments used at the European level to deal with this problem, have changed very much over the years. Europe is lagging behind the USA in terms of growth and investments in knowledge infrastructures, but this general statement, while true, hides a huge variance across European regions and nations (DG Enterprise, 2003). In the last 15 years, income differences among European member states have been strongly narrowing while the process has been matched with a widening of the inter-regional variance within single countries (Martin, 1998). All this casts a shadow on the whole range of European regional policies, explicitly designed to reduce geographical imbalances and strengthen regional cohesion, and raises questions about the consequences of the future Europe enlargement, as the gap is expected to widen. A very odd and worrying aspect of the European context is that the productive capacity agglomeration process as a consequence of market forces may become 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 and dynamic knowledge-based economy in the world, capable of sustainable economic growth with more and better jobs and greater social cohesion (European Commission, 2000) and the Council requested the Commission to report annually on the structural indicators of progress in member states towards the EU s strategic goal. These calls for robust evidence and rigorous monitoring of outcomes led to the development of a set of comprehensive structural indicators to underpin further analyses. In particular, it is interesting to study the different effects of geographical, functional and sectoral distance on the relational activity of different territories. This is the object of the present analysis which, within a gravitational framework, looks at four different relational variables (Internet hyperlinks, European research networks, EPO co-patenting applications and Erasmus student flows) between 110 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 on geodesic path or road distance) between two areas to capture a series of distance-related phenomena which are difficult to measure (such as: transport costs, time elapsed during shipment, synchronisation costs, communication costs, transaction costs and cultural distance). Here, we use geographical distance, calculated as the shortest road distance existing between 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

248 232 Spatial systems the Regional Summary Innovation Index (RSII) contained in the European Innovation Scoreboard (EIS) and the sectoral distance (based on the sectoral distribution of the patenting activity). The chapter is organised as follows: Sections 2 and 3 describe the variables used in the different analyses; Section 4 presents different types of correlations (Pearson, Spearman and quadratic assignment procedure); Section 5 illustrates the use of social network analysis to detect structural properties of different knowledge exchange flows; Section 6 is devoted to the econometric analysis of two gravitational models; and Section 7 concludes. 2. FOUR TYPES OF KNOWLEDGE FLOWS Krugman, in his Geography and Trade, stated that knowledge flows... are invisible; they leave no paper trail by which they may be measured and tracked (Krugman, 1991, p. 53). Jaffe et al. (1993, p. 578) reacted to the previous statement by suggesting that knowledge flows do sometimes leave a paper trail, in the form of citations in patents. Because patents contain detailed geographical information about their inventors, we can examine where these trails actually lead. We attempt to move the approach a little further by focusing on four knowledge-based relational phenomena: digital information exchange (transmitted through Internet hyperlinks), participation in the same research networks (funded by the EU Fifth Framework Programme), EPO co-patent applications and Erasmus student exchange flows. Through these variables we attempt to measure the intrinsic relational structure of knowledge flows which directly connects people and institutions and, indirectly, regions, across five European countries. These four variables capture different types of knowledge (spanning from pure tacit to pure codified knowledge) and different stages of the knowledge creation process. Although information and communication technologies (ICTs) reduce communication and transmission costs, the nature of knowledge and its creation process are very complex and require social processes involving different modalities of interactions. Even in the Internet era, face-to-face relations remain crucial (Feldman, 2002). It is worth noting that the relational variables considered in the analysis span the entire spectrum of relational aspects of knowledge creation, suggesting alternative ways to detect the knowledge trail: from a new and non-material way of information exchange (that is, Internet hyperlinks), to physical and virtual institution-based interactions developed to improve knowledge creation (that is, research networks) by exchanging mostly

249 Inter-regional knowledge flows in Europe 233 codified knowledge, to physical and virtual individual-based relationships aimed at developing marketable innovations (that is, co-patent applications) by exchanging mostly tacit knowledge and know-how, to the physical movement of people leaving their own region in order to acquire a part of their university education in a foreign institution (Erasmus student exchange). Digital Information Exchange (through Internet Hyperlinks) The recent diffusion of ICTs, and in particular of the Internet, stimulated several analyses to measure the diffusion of such a phenomenon across countries, regions, cities, ethnic groups and social classes in order to map the current state and to detect the presence of a digital divide. Different indicators may be used to detect the diffusion of ICTs. The simplest way is to measure the endowment of the ICT equipment (that is, the number of Internet hosts, personal computers and broadband connections) and, more generally, all telecom infrastructures that allow efficient connections. A second way concerns the measurement of the access conditions to ICT services, in terms of the market structures (and prevailing pricing strategies) of the relevant markets (telecoms, Internet service providers (ISPs) and so on). Another way is related to the use of ICT, which may be detected 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 infrastructure of the Internet (comprising cables, routers, satellite and radio connections), and of the www (world wide web), the Internet virtual interface and service platform that allows us to visualise and exchange the information. Since the www is a network of web pages linked through Internet hyperlinks, it can be used to map the inner structure of communication channels and to detect the producers and consumers of digital information. When an Internet hyperlink button is clicked, the content of the target web page is transferred to the clicking computer. One may thus think that the web page containing the hyperlink button acts as an importer of digital information and the target web page represents the information exporter or, more precisely, think of an Internet hyperlink as an index of revealed comparative advantages in the production of specific types of digital information. Internet hyperlinks therefore may be used as an indicator of potential use, 1 since the existence of a hyperlink from one web page to another signals the willingness of the owner of one page to import digital information from another and increases the probability that the targeted web page is actually accessed. 2 One may argue that the number of Internet

250 234 Spatial systems hyperlinks in a web page is uninformative since the inclusion of a new hyperlink button is not constrained by a monetary budget as it does not cost anything. However, the presence of buttons within a page is subject to a harder graphical budget constraint. Web design handbooks show that while the number of hyperlinks is a key element in determining the attractiveness of a web page, such attractiveness is a non-monotonic function of the 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 and development laws. Albert and Barabasi (2002) argue that the www has a scale-free topology in which a small number of central web pages are very popular (are targeted by a huge number of hyperlinks), while the rest of the web is composed of peripheral pages which are almost unconnected and virtually unknown. Other studies show that different typologies of web pages have different organisational structures. Some types of web pages (that is, those of universities and newspapers) display a random network structure in which there are a few extremely central and extremely peripheral nodes, while most of the nodes are targeted by a number of hyperlinks around the average (Maggioni and Uberti, 2005). Uberti and Maggioni (2004) analysed the connectivity of web pages of different institutions (universities, local authorities and chambers of commerce) at the regional level and showed that universities are the most active traders of digital information. In this chapter we therefore include the number of Internet hyperlinks between 308 university web pages located in German, Spanish, French, Italian and UK regions at the NUTS2 level. 4 We chose these 308 universities, members of the European University Association (EUA), located in Germany, Spain, France, Italy and the United Kingdom. The selected sample accounted for 51 universities in France, 61 in Germany, 53 in Italy, 45 in Spain and 95 in the UK. However, its representation of the total population of European universities largely differs from country 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 knowledge flows, we transposed the matrix of the Internet hyperlinks (that is, the presence of an Internet hyperlink from region j to country i, is analysed as the 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 consortia (among firms, universities, research centres and public agencies) in

251 order to increase the competitiveness of the European industry and to foster intra-european cohesion through the exchange and diffusion of scientific and technological knowledge. In both the Single European Act and the Maastricht Treaty (Article 130G), the European institutions have been given competence in the area of science and technology and have developed several actions in order to promote research and development (R&D), create European networks, coordinate R&D and stimulate the European mobility of researchers (Breschi and Cusmano, 2004). Framework programmes always constituted the main planning instrument and funding source for R&D policies in the EU, but as time passed, priorities changed: the latest programmes have shifted the emphasis from supply-side factors, central in the design of the first policies, to diffusionoriented projects and the increase of central skills and knowledge among Europeans (ibid., p. 752). In 1998, the European Council and the European Parliament approved the Fifth Framework Programme (5FP), a programme with a different structure from the previous ones, valid for five years (from 1998 to 2002), and financed with about 15 million. This 5FP is divided into 10 thematic and horizontal programmes, and provides for 12 different types of contracts. 5 In our analysis we focus on two contracts explicitly dedicated to the establishment and use of scientific networks, namely thematic network contracts and research network contracts whose coordinator is a university located in one the 110 regions of the sample, and we included all participants located in these regions, irrespective of their typology (that is, universities, research centres or business organisations). Co-patents Inter-regional knowledge flows in Europe 235 Patents (and patent applications) are one of the most established output indicators of innovative activities. Since the seminal contribution of Scherer (1965), patents have been used in the economic literature (Griliches, 1981, 1990), in order to measure knowledge spillovers and other spatial externality 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 to use a common dataset to analyse the innovative performance of different European countries and regions. In particular, Paci and Usai (2000) and Breschi and Lissoni (2004) have developed systematic analyses of patenting activity throughout Europe at different NUTS levels, showing the existence of significant clustering phenomena (whose agglomeration indices are even higher than those registered by high-tech manufacturing) within a core periphery geographical pattern.

252 236 Spatial systems Later studies analyse patent data as relational variables. Unlike Breschi and Lissoni s (2004) analysis on patent citations, Maggioni and Usai (2005) look at patents as a relation between inventors and applicants at NUTS2 level and study the distributions of these relationships within different European countries. They seek industry-specific patterns and test the hypotheses of a diffused brain-drain dynamics by which peripheral regions host inventors, but do not exploit the economic outcomes of their scientific and technological creativity since applicants (mostly firms) are located in the core regions. In this chapter we consider another relational aspect of patents: the coinvention process. Co-invention (and thus co-patenting) is a process involving both tacit and codified knowledge exchanges. For this reason it implies a series of both face-to-face and over-a-distance relationships between inventors. This is why it is interesting to analyse the relative importance of geographic versus functional distance as forces shaping the interregional (international) structure of this knowledge flow network. Out of a total of more than patent applications belonging to every International Patent Classification (IPC) section (coming from inventors located in the above-mentioned five countries in the period) extracted by the CRENoS files which were based on the original EPO database we selected only those patents whose applications were recorded by more than one inventor. Next we split each patent into equal shares attributed to each inventor. We then added these data for each NUTS2 region in order to built a matrix in which a generic cell ij represents the share of patents 6 recorded jointly by inventors located in regions i and j (where regions i and j could belong to different nations). A total of nearly co-patents was detected. Erasmus Student Exchange The Erasmus student exchange represents another relevant part of the spectrum of relational activities involving knowledge flows among European regions: the mobility of tertiary education students, who represent the basic channel for international training and education. 7 The Erasmus programme, introduced in 1987 and, since 1995, part of the Socrates programme 8 is a European programme devoted to fostering higher education and to creating a European dimension of education. Its popularity, in terms of student participation, is constantly increasing, and has undoubtedly widened after the Lisbon Council emphasised the enforcement of education and training, and student mobility as important goals to be achieved. The Erasmus student exchange reflects several important features, equally contributing to [strengthening] the whole fabric of relations existing

253 Inter-regional knowledge flows in Europe 237 between the peoples of Europe (European Commission, 2005): the institutional integration among European countries; the openness of national tertiary systems; and the relative attractiveness of a country, either in terms of its culture or in terms of the reputation of its tertiary education system. As in the case of digital information flowing through hyperlinks, we are interested in the flow of knowledge; hence we consider the region in which the hosting university is localised as the emitting region (region i) of the knowledge flows embedded in the learned students returning to their receiving region (region j) after their period of study abroad EXPLANATORY (ATTRIBUTIONAL AND RELATIONAL) VARIABLES In this chapter we use some attributional variables to detect differences in the knowledge-based characteristics of 110 European regions. The attributional variables include: GDP, R&D intensity (the ratio between total R&D expenditure and GDP), and three measures of distance (or dissimilarity): the geographical distance (based on road distance between capitals ), the functional distance (based on the RSII contained in the EIS), and the sectoral distance (based on the 2-digit sectoral composition of regional patent application 10 ). What follows is a brief description of variables, their transformations and 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 countries. GDP i and GDP j :Gross domestic product of regions i and j expressed in purchasing power standards (pps). GDP data, expressed in millions of euros refer to year RD i and RD j :Research and development intensity of regions i and j. It is calculated as the ratio between the regional levels of gross expenditure on research and development (GERD) and GDP and refers to various years. GDist ij : Geographical distances among 110 European regions are calculated according to the shortest road distance (in kilometres) between regional capitals. The notion of regional capital implied the use of a certain degree of arbitrariness since NUTS2 levels are administrative meaningful entities in Italy, Germany, Spain and France, but not in the UK. In this last case we used population as the selecting criteria to

254 238 Spatial systems identify the most relevant city (which we called capital ), irrespective of the presence of an administrative capital. FDist ij : Functional distance is measured as the difference (taken in absolute value) of the values registered by the two regions on the RSII contained in the EIS. The RSII measures the European technological leadership and ranks the absolute innovative performance of European regions. It is calculated by re-scaling the regional values of the 13 available indicators 11 according to the following formula, and then taking the unweighted average of the re-scaled values per each region: RSII jj m f 1 X fjj min(x fjj ) max(x fjj ) min(x fjj ), (11.1) where X fjj is the value of an indicator f for region j in country J, and m is the number of available indicators for the j region (DG Enterprise, 2003). This composite index is based in data recorded on different years but officially refers to (DG Enterprise, 2003). SDIST ij : Sectoral distance is measured as the inverse of the technological nearness index (Moreno et al., 2005) calculated as a correlation coefficient between the sectoral composition of patent application registered by region i and by region j at the EPO in the period. CONTIG ij : Contiguity, or adjacency, is a dummy variable which takes the value 1 for contiguous regions (that is, which share a border), and 0 elsewhere. 12 COUNTRY IJ :A dummy variable which is used to control for fixed national effects both on the emitting and the receiving regions. 4. FOUR TYPES OF KNOWLEDGE FLOWS: A CORRELATION ANALYSIS This section shows some results on the correlation existing between the four dependent variables we used to describe knowledge flows (Internet hyperlinks, research networks, co-patents and Erasmus student flows) used in the econometric exercise. In particular we shall focus our attention on simple

255 Inter-regional knowledge flows in Europe 239 Table 11.1 Pearson and Spearman correlations between knowledge flow variables Pearson s correlation Spearman s rank correlation Diginfo ij RN ij Pat ij Eras ij Diginfo ij RN ij Pat ij Eras ij Diginfo ij Diginfo ij RN ij RN ij Pat ij Pat ij Eras ij Eras ij correlation (Pearson), rank correlation (Spearman) and quadratic assignment procedure (QAP) correlation coefficients. Table 11.1 presents both Pearson s simple correlation coefficients and Spearman s rank correlation coefficients for our relational dependent variables. All coefficients are positive and significant, showing that the four variables selected for the analysis measure different elements of the same phenomenon: information and knowledge flows. Digital information and research networks, on one side, and digital information and Erasmus exchange, on the other side, show the highest correlation coefficients. 13 This may be interpreted as a sign of complementarity between virtual and physical interactions among European universities (and regions). One may also note that the high correlation coefficient between research networks and the Erasmus exchange programme shows the existence of hysteresis in the university inter-regional (and international) relationships. Once a relationship is established, both professors and 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 innovations which seem to be partially successful: in fact Pearson s correlation between research networks and co-patenting is quite high (0.264). 14 We further analysed the relationships among these knowledge flows by using the QAP correlation, a bootstrap method that computes correlation indices between entries of two square matrices and assesses the frequency of random measures as large as actually observed. The QAP algorithm proceeds in two steps. In the first step it computes Pearson s correlation coefficients between corresponding cells of the two data matrices. In the second step, it randomly (synchronously) permutes rows and columns of one matrix and re-computes the correlation to the other matrix. The second step is carried out hundreds of times (in our case: 5000 times) in order to compute the proportion of times that a random measure is larger than or equal to the observed measure calculated in step 1. A low proportion

256 240 Spatial systems Table 11.2 QAP correlation between knowledge flow variables Pearson s correlation Diginfo ij RN ij Pat ij Eras ij Diginfo ij RN ij 0.302** (0.031) Pat ij 0.220** (0.043) (0.040) Eras ij 0.270** 0.245** (0.038) (0.036) (0.048) Note: ** Significant at 5%; standard error in parenthesis. (smaller than 0.05) suggests a strong relationship between the two matrices that is unlikely to have occurred by chance (Borgatti et al., 2002). Table 11.2 shows the results of such a procedure. 15 The highest correlation 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 complementarities between these flows of knowledge. QAP procedure shows that the correlation between co-patenting and research networks and co-patenting and Erasmus exchange flows (which registered low Spearman and Pearson correlation coefficients) is not significant, indicating the persistence of frictions between different worlds (that is, the business and academic environments). Although the sample of research networks included in the analysis has been selected, 16 hence suffers from some biases, these results may also show that EU programmes seem to fail in connecting different actors, hence these actions need to be redefined to be really effective across different institutions (and in particular between profit and non-profit organisations). 5. NETWORK ANALYSIS OF KNOWLEDGE FLOWS Network analysis (NA) uses quantitative techniques, derived from graph theory, to study and describe the structure of interactions (edges) between given entities (nodes) (Wasserman and Faust, 1994). Initially used by sociologists and ethnologists to study complex personal interactions, NA has recently been used in economic analyses (Snyder and Kick, 1979; Maggioni, 1993 and 2000; Leoncini et al., 1996; Uberti, 2002; Breschi and

257 Inter-regional knowledge flows in Europe 241 Cusmano, 2004; Breschi and Lissoni, 2004) to analyse institutional, technological and commercial relationships among agents, industries, regions and countries. Therefore, in this chapter, 110 NUTS2 European regions are treated as nodes, while their different knowledge flows are treated as edges. Orthodox approaches describe the innovation process through an atomistic principle that assumes the existence of individual utility maximisation procedures and does not take into account the wider social, economic and institutional framework. By contrast, NA highlights some relevant structural features. The behaviour of a node (in terms of strategy and performance) has to be interpreted in terms of both structural limits and internal features. Internodal relationships must be examined from two complementary perspectives: the single node s and the whole system s perspective. Neither a single node nor a pair of nodes can be meaningfully analysed, when isolated from the system framework (holistic principle). Systems display a surprisingly intrinsic fractal nature: both the macro level (whole system) and the micro level (nodes) are composed by a plurality of structurally interrelated elements. The interdependence of observations does not hinder NA techniques, allowing a wider use of this methodology even when more traditional statistical and econometric techniques based on pure attributional variables suffer. (Bramanti and Maggioni, 1997, p. 327) In the following analysis we shall use the density and clustering coefficients as the index of systemic connection. Density is defined as the ratio between the actual number of edges e and the maximum number of directed edges in a network composed of n nodes 17 or: D e n(n 1). (11.2) The clustering coefficient of node i characterises the extent to which nodes adjacent to it are adjacent to each other (Watts, 1999): C i i, (11.3) where v i is the number of nodes connected to i and is the total number of possible edges in i s neighbourhood. The clustering coefficient for the whole network is obtained by averaging the clustering coefficient of all nodes in the network. The networks analysed in the chapter describe different knowledge flows: co-patenting and research networks are symmetric, and Internet hyperlinks and Erasmus exchange flows are a-symmetric. In the latter case, for each

258 242 Spatial systems 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 dichotomized version 18 of the original innovation flow matrices. The customary procedure implies the choice of an appropriate (often ad hoc) threshold; however, it must be considered that the choice of a given threshold is strategic because different values produce different dichotomised networks. In this analysis we choose the network average as the threshold value. In order to detect the most central actor within the system and the definition of a scale of hierarchy (inequality), centrality and centralisation indices have been designed. 19 Formally, the degree centralisation of a network (system) of n nodes (regions) can be defined as follows: C i b (C* b C i b ) i (n 1)(n 2), (11.4) where C* b is the centrality value of the most central region in the system and the denominator reflects the maximum level of centrality obtainable in a system of n regions. The centralisation indices (which lies between 0 and 1) measure the difference in centrality levels between the most central region and the others. A high centralisation index identifies a very hierarchic system where differences in positions are maximised, and a pivotal node exists. A low centralisation index identifies a structure where most of the positions are similar and interchangeable. Table 11.3 shows the network indices for different knowledge flows. Density indices of dichotomised networks show that the digital information and the co-patenting networks are the least dense, while the research network is the most dense. The data show that knowledge flows do not spread evenly between European regions, which suggests that traditional face-to-face interactions remain one of the most active phenomena of knowledge creation, although virtual ones are cheaper. Table 11.3 Network analysis indices of knowledge flow structures Density Clustering Isolated nodes Centralisation Outdegree Indegree Diginfo ij RN ij Pat ij Eras ij

259 Inter-regional knowledge flows in Europe 243 The ranking based on clustering coefficient is almost the opposite of the density-based one. This can be explained by referring to the number of isolated nodes. The most clustered networks (digital information and copatenting) have many isolated nodes (private club structure): meaning that if one region is connected to another one, then it is very likely that the same region is also connected to the original node neighbours. Centralisation indices show that, in general, the Erasmus student network is the most centralised, while co-patenting is the least centralised network. However since Internet hyperlinks and Erasmus are asymmetric, while research networks and co-patenting are symmetric it is more useful to consider each couplet in isolation. As far as symmetric relationships are concerned, research networks exhibit a more hierarchical regional structure than co-patenting ones, suggesting that educational institutions are tied for better or worse to the region s performance, while individual inventors are more evenly diffused and their interactions follow a more uniform pattern. The co-patenting network has a rather non-hierarchic structure due to the presence of some very central regions (Oberbayern, Darmstadt, Düsseldorf and Île de France) and to a series of other regions that are connected not exclusively to the most central ones but also with their national neighbourhood. In fact, by removing the most central nodes from the network, highly connected national islands emerge (see Figures 11.1 and 11.2). As far as a-symmetric relationships are concerned, Erasmus student flows display a more hierarchical structure (some European regions are highly engaged in the Erasmus programme either as a source or a destination of student flows, while others are almost not involved) than the digital information one (differences in the number of hyperlinks are not so relevant). The difference in centralisation values (referred to as outdegree and indegree) of the Erasmus programme may be interpreted as the existence of a larger difference in participation in the Erasmus programme of different European regions as recipients of students than as senders (a greater number of regions send their students abroad, but their destination is concentrated in a smaller number of regions). A similar (although smaller) difference is shown by the centralisation indices of the digital information networks. European regions show a greater difference in their information exports than in their information imports. In other words while universities (and regions) are more similar in the number of hyperlink buttons inserted in their web pages, few universities (and regions) record a larger share of total Internet hyperlink destinations.

260 Yorkshire and the Humber Figure 11.1 Co-patent network, , including Oberbayern, Darmstadt, Düsseldorf and Île de France 244

261 Yorkshire and the Humber Figure 11.2 Co-patent network, , excluding Oberbayern, Darmstadt, Düsseldorf and Île de France 245

262 246 Spatial systems 6. GRAVITY EQUATIONS Looking for the source of regional disparities we use gravity equations in order to assess whether geographic distance was responsible for such a phenomenon (that is, peripherality exogenously causes poor performances of regions, therefore determining the polarisation of a rich core and a depressed periphery) or whether functional distance (that is, difference in the scientific and technological levels) endogenously plays a major role in determining the existence of a much more dense core (the network of more advanced regions) and a residual sparse set of relations within the periphery. Finally, we tested the influence of the similarity/dissimilarity of the productive structure of different regions by detecting the effects of sectoral distance (measured through patent activity) on knowledge flows. The gravity equation model is a powerful tool of empirical analysis to explain social interactions (for example, international trade, foreign direct investment, migration and tourism) according to the existence of attracting and impeding forces. This range of models is derived from the law of universal gravitation proposed by Isaac Newton in 1687, which stated that gravitational force between masses decreases with the distance between them, according to an inverse-square law.... [T]he theory notes that the greater an object s mass, the greater its gravitational force on another mass (Wikipedia, 2005). In the economic literature these models are commonly used to explain international trade: bilateral trade between two countries is proportional to their economic mass (that is, GDP or population) and inversely related to their geographical distance. These models have been a successful tool for empirical analysis since the 1960s: the signs of parameters of importing and exporting countries GDPs are positive, roughly equal to unity and significant, and the sign of geographical distance is negative and significant (Tinbergen, 1962; Poyhonen, 1963). Recently this empirical success has been 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 particular type of knowledge flows between two generic regions i and j as a function of a series of relational and attributional variables. All variables are taken in logs in order to interpret the estimated coefficients as elasticities. The generic dependent variable, KF ij, stands for four different typologies of knowledge flows: digital information (KF ij Diginfo ij ), research networks (KF ij RN ij ), co-patenting (KF ij Pat ij ), or Erasmus student exchange (KF ij Eras ij ); the independent variables are as defined in Section 3. Table 11.4 shows the results of eight ordinary least squares (OLS)

263 Table 11.4 Gravity equation for knowledge flows Digital information flows Research networks RN ij Co-patents Pat ij Erasmus students Eras ij Diginfo ij (Ia) (Ib) (IIa) (IIb) (IIIa) (IIIb) (IVa) (IVb) Functional Sectoral Functional Sectoral Functional Sectoral Functional Sectoral distance distance distance distance distance distance distance distance Independent variable GDP i 0.761*** 0.758*** 0.076*** 0.075*** 0.748*** 0.733*** 0.621*** 0.604*** (0.021) (0.021) (0.004) (0.004) (0.025) (0.025) (0.018) (0.018) GDP j 0.600*** 0.594*** 0.076*** 0.075*** 0.747*** 0.733*** 0.461*** 0.441*** (0.022) (0.021) (0.004) (0.004) (0.026) (0.025) (0.018) (0.018) RD i 0.240*** 0.236*** 0.045*** 0.041*** 0.391*** 0.394*** 0.065* 0.056** (0.026) (0.026) (0.005) (0.005) (0.030) (0.030) (0.025) (0.024) RD j *** 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) GDist ij 0.449*** 0.445*** 0.020** 0.024*** 0.525*** 0.580*** (0.038) (0.038) (0.006) (0.007) (0.043) (0.039) (0.038) (0.037) Contig ij 0.168** *** 0.975*** 1.220*** 1.169*** (0.100) (0.098) (0.017) (0.017) (0.074) (0.072) (0.158) (0.154) FDist ij 0.051** 0.020*** ** (0.016) (0.003) (0.018) (0.014) SDist ij 0.403** 0.052** 0.520*** 0.677*** (0.118) (0.017) (0.135) (0.108) Constant 7.678*** 7.400*** 1.454*** 1.346*** *** 9.539*** 8.960*** 9.525*** (0.411) (0.397) (0.067) (0.066) (0.477) (0.465) (0.446) (0.441) Number of observations R-squared F-test Note: *** significant at 1%; ** significant at 5%; * significant at 10%; robust standard error in parenthesis; country dummies are included in all regressions but not reported. 247

264 248 Spatial systems regressions, 20 where we considered alternatively functional and sectoral distance in the regressors. Regression (a): ln(kf ij ) 0 1 ln(gdp i ) 2 ln(gpd j ) 3 ln(rd i ) 4 ln(rd j ) Regression (b): 5 ln(gdist ij ) 6 Contig ij 7 ln(fdist ij ) I ij ln(kf ij ) 0 1 ln(gdp i ) 2 ln(gpd j ) 3 ln(rd i ) 4 (RD j ) 5 ln(gdist ij ) 6 Contig ij 7 ln(sdist ij ) I ij where I indicates country dummy variables and, ij and ij are standard error terms. Table 11.4 presents the results of the econometric analysis. Regression I describes the structure of information flows running through Internet hyperlinks established between European universities. These flows are positively influenced by both regions GDP, confirming the existence of a positive relation between the economic size 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 than that of the receiving one. A positive and significant coefficient is registered by the R&D intensity of the emitting region, while the coefficient of the receiving region is not significant. This suggests that the level of intensity of innovation inputs of a region determines the visibility of the local university s website (perhaps via a relationship between public funding of R&D and university relevance 21 ), while the positioning of hyperlink buttons follows a different logic. Geographical distance is negative and significant suggesting that, at least for our sample of university websites, the advent of the Internet did not cause the death of distance. 22 Digital relationships are considered in academia as complementary to physical ones and faceto-face contacts are still crucial. It is, however, worth noting that the coefficient of the contiguity variable is also negative and significant. Such a result may be explained in terms of a limited use of Internet-based information flows between neighbouring universities and some hidden forms of spatial competition on the local pool of prospective students. Functional distance bears a significant and neg-

265 Inter-regional knowledge flows in Europe 249 ative coefficient, thus signalling that university networks of relations as measured by Internet hyperlinks tend to develop between similar regions. A similar result is shown by the measure of sectoral distance, perhaps suggesting the existence of a deep relation between a region s industrial structure and the characteristics of its universities. Regression II analyses the joint participation of research institutions belonging to different regions in two different types of research network under the fifth EU research framework programme. These flows are positively influenced by both regions GDP, confirming the existence of a positive relation between the economic size of a region, its research potential and its scientific networking activity. The R&D intensity coefficients of the emitting and receiving regions are positive and very similar, suggesting that the propensity to be involved in the network is positively correlated with the scientific and technological level of both regions. The coefficient of geographical distance is negative and significant; however, its size is very small, suggesting a limited influence of spatial effects in this activity whose aim is explicitly to link research units from different places all over Europe. This is confirmed by the insignificance of the coefficient on contiguity. The coefficients of both functional and sectoral distance are negative and significant: meaning that both the scientific and technological levels and the sectoral specialisation of a region play a positive role in determining the probability of joining the same research network. In other words, research networks have a club structure in which agents match with others that are similar. If this is the case, then research networks cannot be used as policy tools to support cohesion and inclusiveness since their structure is a segmented one in which stronger regions cooperate with other stronger ones, and weaker with weaker. It is also worth noting that the coefficient on the sectoral distance has a higher value than the geographic distance. Regression III describes the structure of scientific relationships, which derives from the exchange of knowledge and know-how between European inventors. Co-patenting relationships are strongly and positively influenced by both regions GDP and R&D intensity; the coefficients of both variables are very similar. This confirms that both size (that is, larger and richer regions have a greater number of patentable inventions) and technological level play an important role in determining the amount of knowledge exchange needed to develop a patentable innovation. Geographic distance has a significant and negative coefficient. This could be explained in terms of the need for face-to-face contacts in the R&D activity (based on tacit knowledge exchange) leading to a patent application. Since the

266 250 Spatial systems coefficient of functional distance is not significant, we focus the attention on model b in order to test whether the sectoral distance plays a more relevant role. This is exactly the case: the negative and significant coefficient shows that a common sectoral specialisation of the technological activity of the two regions is important to determine the level of scientific collaboration between inventors. The positive and significant coefficient on contiguity registered in both specifications (a and b) confirms Jaffe et al. s (1993) results and shows that the innovation process is deeply rooted in a given territory and that knowledge spillovers easily overcome regional borders. The coefficient of the geographic distance is larger than not only the coefficient of the functional distance (not significantly different from zero) but also the coefficient of 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 knowledge flows, we consider the region in which the hosting university is localised as the emitting region of the knowledge flows embedded in the learned student returning to his/her original and receiving region after their studies. Regional GDP and R&D intensity coefficients are significant and positive. Larger, richer and technologically advanced regions are more aware of the advantages of an international education process and more involved in this programme. The coefficients on both functional and sectoral distance are negative and significant, while the coefficient of geographical distance is insignificant. Taken together, this may be interpreted as showing that the Erasmus programme does foster the geographical mobility of European students but not as much as cohesion and convergence of the scientific level of European regions. Geographical distance does not influence the flow of Erasmus students; however, students from top regions (in terms of their respective RSII) tend to study abroad in better foreign regions than their counterparts coming from the bottom regions. In every model shown in Table 11.4 country dummies included in the estimation to take into account the institutional factors of the emitting and receiving regions which may be determined by national characteristics record significant coefficients. The regression constant in the gravitational models literature refers to a regional fixed effect which is sometimes interpreted as an indirect measure of remoteness (that is, the distance of one region to every other region). If one accepts this reasoning, our results support the common wisdom that peripheral (in a geographical sense) regions are also peripheral in a functional sense and that knowledge and

267 Inter-regional knowledge flows in Europe 251 information flows have a hierarchically segmented structure with limited evidence of a filtering-down process. 7. CONCLUSION The reduction of regional disparities has been one of the main targets of EU policies since its very beginning. However, the digital revolution has given new meanings to this concept. Per capita GDP and unemployment rates are still relevant economic indicators, but so are knowledge and ICT indicators (in terms of endowments, access and use). This chapter which focuses on the structure of knowledge flows as measured by four distinct but complementary variables (Internet hyperlinks, research networks, EPO co-patent applications and Erasmus student mobility) has underlined the intrinsic relational nature of knowledge. We showed that there is a positive correlation between knowledge exchange flows and that these flows are influenced by different types of distance: geographical, functional and sectoral. NA techniques showed that Erasmus student flows and Internet hyperlinks have a more hierarchical structure in their outdegree than in their indegree. These results confirm the existence of a polarised centre periphery hierarchy of European regions which is reflected in the structure of knowledge flows. 23 The NA perspective showed that although the co-patent network displays some international relations connecting European regions, co-patenting still remains a mainly intra-national activity, mostly connecting regions in the same nation. By using a gravitational model we demonstrated that, far from the claim of the death of distance, geographical distance is still relevant for determining the structure of inter-regional knowledge flows. Functional and sectoral distance also play a crucial role, suggesting that knowledge flows easily between similar (according to their scientific, technological and sectoral characteristics) regions. Convergence between lessdeveloped regions towards income levels of richer regions in the EU is thus hampered by the observed network dynamics. If the EU intends to build a truly European research area in which the networking 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. Turati and S. Usai for fruitful discussions, and M. Gioè for research assistance. The following

268 252 Spatial systems 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 (Istituto di studi su popolazione e territorio, Università Cattolica del Sacro Cuore, Milan); S. Usai and B. Dettori (Centre for North South Economic Research (CRENoS), Università degli Studi di Cagliari) and M. Thelwall (School of Computing and Information Technologies, University of Wolverhampton). We acknowledge the financial support of the Ministry of Education, University and Research (MIUR) under the Research Project Dinamica strutturale: tecnologie, reti, istituzioni ( _001). 1. While it is possible for a single website to count and map all access, it is extremely difficult to collect this information on a wider scale since it would involve the cooperation of the web masters of all websites. 2. A further confirmation of the informational content of Internet hyperlinks relates to the fact that several search engines (and in particular the popular Google) use hyperlink counting as ranking criteria of web pages since they consider it a good proxy of the quality and relevance of the web page. 3. A new discipline (webmetrics) devoted to the analysis of the www using bibliometric procedures has been created, as well as a scientific association (the International Society for Scientometrics and Informetrics, ISSI) (Maggioni and Uberti, 2005). 4. The retrieval of Internet hyperlinks following Thelwall and Smith (2002) was run in 2003 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 measures, research grants (individual fellowship), research network contracts, study contracts, assessment contracts and thematic network contracts. Some of these contracts are assigned to single applicants (that is, research grants), while others require the creation of research networks among the participants (that is, research and thematic network contracts). 6. A patent registered by three inventors located in three distinct regions i, j and z, would be split into n*(n 1) cells and, respectively, into i with j and z, j with i and z, and z with i and j. Hence an invention co-patented by three individuals in three different regions is registered with a value of 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, hence in this analysis we focus only on the Erasmus student flows. 8. The Socrates programme is the European programme for education, and includes eight actions; it was developed to promote the European dimension and to improve the quality 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 Agency responsible 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 hightech services; public R&D; business R&D; EPO high-tech patent applications; EPO patent applications; share of innovative enterprises in the manufacturing and service sectors; innovation expenditures in manufacturing and in services; and sales of new to the firm but not new to the market products. 12. Sometimes a regional border in our sample may also be a national border, and borders are 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 the contiguity 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 typology of the research networks considered in this study, which excludes those coordinators that are not universities.

269 Inter-regional knowledge flows in Europe These correlations were calculated using binary matrices dichotomised according to the average of raw matrices: the cell ij value above the mean would be registered 1 and 0 otherwise. 16. See Research networks in Section For symmetric networks with undirected edges, the density is calculated as follows: D 2e [n(n 1)]. 18. A value equal to 1 is substituted for the actual value of the edge when it is greater than 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, networks 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 outward measures (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 further specification identify the outward measure of the indices. 20. Since we are mainly interested in the significance and signs of the coefficient, simple OLS estimation provides valid results. Alternative estimation procedure (either count data models or OLS with Box Cox transformation) would allow detailed analysis of the coefficient 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 sectoral distance. 23. A larger number of regions send their students abroad, but their destination is concentrated in a small number of regions. Also, it is easier to be the origin of a hyperlink than to be the target. REFERENCES Albert, R. and Barabasi, A.L. (2002), Statistical mechanics of complex networks, Reviews of Modern Physics, 74, Anderson, J.E. (1979), A theoretical foundation for the gravity equations, American Economic Review, 69, Bergstrand, J.H. (1985), The gravity equation in international trade: some microeconomic foundations and empirical evidence, Review of Economics and Statistics, 71, Borgatti, S.P., Everett, M.G. and Freeman, L.C. (2002), UCINET 6 for Windows: Software for Social Network Analyis, Cambridge, MA: Harvard, Analytic Technologies. Bramanti, A. and Maggioni, M.A. (1997), The dynamics of milieux: the network analysis approach, in Ratti, R., Bramanti, A. and Gordon, R. (eds), The Dynamics of Innovative Regions, Aldershot: Ashgate pp Breschi, S. and Cusmano, L. (2004), Unveiling the texture of a European research area: emergence of oligarchic networks under EU framework programmes, International Journal of Technology Management, 27, Breschi, S. and Lissoni, F. (2004), Knowledge networks from patent data: methodological issues and research targets, CESPRI Working Papers 150, Centre of Research on Innovation and Internationalization (CESPRI), Milan.

270 254 Spatial systems Dalgin, M., Mitra, D. and Trindade, V. (2004), Inequality, nonhomothetic preferences, 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 a neoclassical world?, in Frankel, J.A. (ed.), The Regionalization of the World Economy, Chicago: University of Chicago Press, pp DG Enterprise (2003), 2003 European Innovation Scoreboard, Technical Paper No. 3, Regional innovation performances, trendchart.cordis.lu/, accessed January European Commission (2000), The Lisbon strategy for economic, social and environmental renewal, available at index_en.html, accessed February European Commission (2005), accessed: February Feenstra, R.C. (2002), Border effects and the gravity equation: consistent methods for estimation, Scottish Journal of Political Economy, 49, Feldman, M.P. (2002), The Internet revolution and the geography of innovation, International Social Sciences Journal, 54, Griliches, Z. (1981), Market value, R&D and patents, Economic Letters, 7, Griliches, Z. (1990), Patent statistics as economic indicators: a survey, Journal of Economic Literature, 28, 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 of knowledge spillovers as evidenced by patent citations, Quarterly Journal of Economics, 108, 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 Italy and Germany, Research Policy, 25, Maggioni, M.A. (1993), Network analysis of regional industrial dynamics and local economic policies, paper presented at the 5th Society for the Advancement of Socio-Economics (SASE) conference, New York, March. Maggioni, M.A. (2000), Intersectoral innovation flows within and between nations and regions: network analysis and systems of innovation, in Punzo, L., Farina, F. and Fabel, O. (eds), European Economies in Transition, London: Macmillan, pp Maggioni, M.A. and Uberti, T.E. (2005), Webmetrics, in Pagani, M. (ed.), Encyclopedia of Multimedia Technology and Networking, London: Idea Group Inc., pp Maggioni, M.A. and Usai, S. (2005), Patents as relations: the organisation and the evolution of the innovative activity in two European countries, paper presented at 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, Moreno, R., Paci, R. and Usai, S. (2005), Spatial spillovers and innovation activity in European regions, Environment and Planning A, 37:

271 Inter-regional knowledge flows in Europe 255 Paci, R. and Usai, S. (2000), Technological enclaves and industrial districts. An analysis of the regional distribution of innovative activity in Europe, Regional Studies, 34, Poyhonen, P. (1963), A tentative model for the volume of trade between countries, Weltwirtschaftliches Archiv, 90, 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 of patenting inventions, American Economic Review, 55, Snyder, D. and Kick, E.L. (1979), Structural position in the world system and economic growth, : a multiple network analysis of transnational interactions, American Journal of Sociology, 84, Thelwall, M. and Smith, A. (2002), Interlinking between Asia-Pacific university web sites, Scientometrics, 55, Tinbergen, J. (1962), Shaping the World Economy: Suggestions for an International Economic Policy, New York: 20th Century Fund. Uberti, T.E. (2002), Flussi commerciali, tecnologici ed informativi: elementi per un 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 potenziale. Un esercizio di hyperlinks counting a livello sub-nazionale, Quaderno Diseis 0402, Milano: Università Cattolica del Sacro Cuore. Wasserman, S. and Faust, K. (1994), Social Network Analysis: Methods and Applications, Cambridge: Cambridge University Press. Watts, D. (1999), Small Worlds: The Dynamics of Networks between Order and Randomness, Princeton, NJ: Princeton University Press. Wikipedia (2005), Low of universal gravitation, Law_of_universal_gravitation, accessed April 2005.

272 12. Explaining the territorial adoption of new technologies: a spatial econometric approach Andrea Bonaccorsi, Lucia Piscitello and Cristina Rossi* 1. INTRODUCTION The idea that information and communication technology (ICT) will reduce the economic importance of geographic distance has been expounded energetically in post-internet literature (Cairncross, 2001). According to this view, the New Economy works in a space rather than a place, transport costs will be drastically reduced, distance will be less important, and peripheral regions will benefit from opportunities that are not available in an economy based on manufacturing industry (Negroponte, 1995; Kelly, 1998; Compaine, 2001). Since it is mostly based on non-material and human capital investment, ICT offers new areas of potential growth to regions or areas that have historically suffered from isolation, high transport costs or a lack of private and public physical infrastructure. Consequently, according to this view, the concentration of income opportunities and wealth should decrease over time. Although other predictions have also been made in 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 in the intensity with which ICT is adopted across countries, but big differences still persist within industrialised countries. Indeed, differences in economic development still shape the rate of the adoption of these technologies, at firm, regional and national levels. The reasons for these stylised facts have been investigated at length in recent years. This chapter contributes to current literature in several ways. First, it focuses on intra-national or regional differences, which is a much less explored dimension of the digital divide. Second, it uses a new metric for ICT adoption, namely the number of second-level Internet domain names registered under the country code Top Level Domain (cctld).it. Finally, 256

273 Explaining the territorial adoption of new technologies 257 it explicitly combines the analysis of determinants with a spatial econometric approach. The chapter is organised as follows. Section 2 surveys the literature on the digital divide and the relation between local development and adoption of ICT. Section 3 describes data and methodology. Section 4 contains a description of the model and the empirical results. Section 5 summarises the main conclusions of the chapter. 2. THE LOCAL DIGITAL DIVIDE: THE RELATION BETWEEN DEVELOPMENT AND ICT ADOPTION The conceptual link between economic development and ICT adoption is a widely researched issue in economics literature. It can be claimed that the nature of ICT makes it possible to overcome territorial peripherality. Unlike traditional heavy and light manufacturing investment, ICT may increase regional attractiveness as a strategic location factor, thus enhancing territorial competitiveness (Gillespie et al., 1989; Kraemer and Dedrick, 1996; Steinmuller, 2001; Camagni and Capello, 2005). The successful experiences of Ireland and India as emerging regions in the provision of software services, due to the availability of efficient communication infrastructures, are often quoted. Contrary to most expectations, however, the overall empirical reality is one of big geographic differences in the rate of ICT adoption, which seems to reinforce rather than reduce disparities and inequalities. 1 Most studies have revealed astonishing differences in Internet and computer penetration between North America and Europe on the one hand and African and Asian countries on the other (see Chinn and Fairlie, 2004, for a comprehensive survey of this literature). These disparities have mainly been explained in relation to differences in income, but also in human capital, telecommunication infrastructures (Dasgupta et al., 2001; Oyelaran-Oyeyinka and Lal, 2003; Pohjola, 2003; Wallsten, 2003), demographic variables and regulatory regimes (Wallsten, 2003). 2 Although these explanations are fairly convincing, it is puzzling as to why there is still scant evidence of a process of convergence on the part of less-developed countries in the adoption of these technologies. Less investigation has been devoted to the local dimension of the phenomenon; indeed digital inequalities do not divide only developed from developing countries but also regions within the same country ( local digital divide ; see, for instance, Gareis and Osimo, 2004; Ramsay, 2004). Both developed and developing countries suffer from severe regional disparities in ICT adoption. Evidence has been gathered with reference to the United States (NTIA, 2002; Mills and Whitacre, 2003), Canada (Dryburgh, 2001),

274 258 Spatial systems Portugal (Nunes, 2004), Spain (Billon Curras and Lera Lopez, 2004), Italy (Bonaccorsi et al., 2002; Assinform, 2004) and China (Qingxuan and Mingzhi, 2002; Wensheng, 2002). One clear-cut stylised fact that emerges from this literature is that regional disparities are at least as important as cross-country differences, at least within industrialised nations. In Italy, for instance, Bonaccorsi et al. (2002) found that the geographic concentration of Internet adoption is much greater than that of population or income. This would seem to suggest that, far from reducing regional disparities, ICT actually reinforces them. Empirical studies have shown that the determinants of local inequalities are associated with economic, social and demographic disparities. In particular, differences in the spatial diffusion of ICT have been explained in terms 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 spatial factors. Investigating the geography of second-level domain names in Portugal (.pt), Nunes (2004) recently proposed that the Internet might contribute to reinforcing the tendency to territorial disintegration, promoting geographic disparities in a more pronounced way than is the case in the real economy space. He found that the role of ICT in overcoming spatial inequalities in Portugal is less important than expected, since these technologies, far from changing the existing spatial structure, are deeply influenced by it. The importance of a spatial perspective for the analysis of Internet adoption has been addressed by Aztema and Weltevreden (2004) and by Weltevreden et al. (2005), who combine a conventional innovationadoption approach with a detailed geographical analysis to study the adoption of business-to-consumer e-commerce by Dutch retailers. In line with the most recent studies, mainly framed within models of technology diffusion (Geroski, 2000), several groups of factors that potentially influence the territorial adoption of ICT were distinguished (for an excellent recent survey, see OECD, 2004). One category of factors that are positively related to ICT adoption concerns the local technological endowment and the relevant absorptive capacity. The latter refers both to the ability of firms to assess technological opportunities (which depends on their endowment of human and knowledge capital, Cohen and Levinthal, 1989), and to learning effects. The former may arise from earlier use of ICT or a predecessor of a specific ICT element which already embodies constituent elements of later applied, more advanced vintages (McWilliams and Zilberman, 1996). In addition, according to Hollenstein (2004: 41).

275 Explaining the territorial adoption of new technologies 259 [T]hese aspects of absorptive capacity refer to the standard epidemic model of technology diffusion and to the relevant information spillovers from users to non users of the technology. This model basically states that a firm s propensity to adopt a technology at a certain point in time is positively influenced by the present (or lagged) degree of its diffusion in the economy as a whole or in the industry to which the firm is affiliated. A second category of variables concerns market characteristics. It has been amply demonstrated that the sectoral specialisation of a region impacts significantly upon the adoption of ICT (Pohjola, 2003). Likewise, the characteristics of firms have traditionally been employed as explanatory variables in most studies on adoption. In particular, firm size captures the Schumpeterian hypothesis about the positive relation between innovativeness and dimensional scale. The same holds for firm age, although the theoretical arguments are not conclusive (positive experience effects versus negative adjustment cost effects in the case of older firms, see Lal, 2001; Hollenstein, 2004). The adoption of ICT may also be affected by the market conditions in which firms are operating, particularly the competitive pressure to which they are exposed. In markets where competition is stronger, firms are expected to be more inclined to innovative activities or rapid technology adoption (Porter, 1990; Majumdar and Venkataraman, 1993; Feldman and Audretsch, 1999; Hollenstein, 2004). Finally, we explicitly take into account the role that spatial externalities play 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 to the Canadian Statistics Bureau, it includes desktop and laptop computers, software, peripherals and connections to the Internet that are intended to fulfil information processing and communications functions. 3 Such a variety poses severe methodological problems when it comes to measuring the level of territorial adoption of these assets. According to Pohjola (2003), two kinds of metrics reveal disparities in ICT adoption across countries: data on ICT equipment and its use, and indicators of ICT spending. Most of the studies that have analysed geographical inequalities at the international level have identified ICT with the Internet, focusing on the number of Internet hosts (OECD, 2001; Kiiski and Pohjola, 2002) and users (Norris, 2002; NTIA, 2002), 4 although restricting the issue of

276 260 Spatial systems differences in ICT adoption simply to Internet access is misleading (Oden and Rock, 2004). Indeed, while data regarding Internet hosts are readily available and highly reliable (Press, 1997; Wolcott et al., 2001), 5 this metric suffers from two major shortcomings: the data are gathered only at a national level and they do not provide any information about the adopters. Regional-level analysis benefits from the availability of larger sets of indicators, ranging from the share of sales of electronic goods to mobile phones; survey data are also available. 6 Recently, the use of domain names as a proxy of Internet diffusion has been proposed (Zook, 2000; Zook et al., 2004). Domains may be a valid proxy for ICT adoption, mainly because they operationalise the intention to actively supply contents through the Net. Those who register a domain name use the Internet in a more conscious manner, with a view not only to demanding but also to adding content to it. 7 In general, the registration of a domain name by a firm is the first step towards setting up a website through which to present their products or even to undertake electronic commerce activities. Domains therefore provide an underestimation of ICT adoption 8 as: (i) ICT adoption does not necessarily involve registering a domain; and (ii) Internet service providers (ISPs) often offer their users room (on their servers) for adding new content. Thus, domains constitute a lower bound as any registrant is unquestionably an ICT adopter. Additionally, every domain name is uniquely associated with a registrant whose geographical location and nature are unambiguously recorded in the databases of the organisations that manage the cctld (Mueller, 1998; Grubesic, 2002). The availability of information at the subnational level makes domains a valid metric to explore the territorial dimension of ICT adoption, while data on the nature of the registrants make it possible to take into account different adoption determinants for different populations of potential adopters. In the present study, domain name registrations by Italian firms are used as a proxy for ICT adoption at the NUTS3 (Nomenclature des Unités Territoriales Statistiques) level (103 provinces). In the period, the Institute of Informatics and Telematics (IIT) of the National Research Council (CNR), the Sant Anna School of Advanced Studies and the University of Pisa built up a database of domain name registrations, organised on a subregional basis into different categories of actors (individuals, business firms, universities and research centres, third sector associations and local government bodies). Data were extracted from the databases of the registrations under the cctld.it that are managed by the Italian Registration Authority (RA) hosted by IIT. A total of domain names were inspected for classification; multiple names registered by the same registrant were carefully checked and eliminated.

277 Explaining the territorial adoption of new technologies 261 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 ICT adoption, the penetration rate in each province was calculated as the percentage of firms in the province that had at least one domain name registered in the RA databases as of July Table 12.1 summarises the descriptive statistics of the variable. As expected, the data show the existence of a North South divide. The regions of Italy are highly differentiated from social, cultural, demographic and economic points of view, and ICT adoption is no exception. The figures in Table 12.2 reveal that the level of ICT adoption in Italy is quite low, with an average penetration rate of less than 4 per cent, and that differences between the macro areas are highly significant (the Kruskall Wallis test is significant at p 0.01). No southern province ranks in the top 50. The best-performing province in the South ranks 55th, with just eight northern provinces below that position. Conversely, the 20 worst-performing provinces are all located in the South (see Table 12.3 for the top 10 and bottom 10 positions). The geographical disparities in ICT adoption mirror the inequalities in economic development (measured by value added per employee) both among and within geographical macro areas (Figure 12.1). As the analysis of the map in Figure 12.1 suggests a possible spatial autocorrelation with regard to the provinces located in the North Centre, the overall pattern in the local spatial pattern was disaggregated by means of the local indicator of spatial association (LISA). In fact, Figure 12.2 shows the emergence of 3 H-H clusters, all located in the North Centre, and 2 L-L Table 12.1 ICT adoption: descriptive statistics ICT No. Min Max Mean Std Skewness Kurtosis adoption dev Table 12.2 ICT adoption in macro areas Area No. Mean Std dev. Kruskal Wallis test p-value North Centre South Total

278 262 Spatial systems Table 12.3 Ranking of Italian provinces by ICT adoption and per capita income (PCI) in 2001 top 10 and bottom 10 provinces Ranking Ranking Province Region Macro region ICT ICT PCI (NUTS3) (NUTS2) (NUTS1) adoption adoption 1 1 Milano Lombardia North Firenze Toscana Centre Bolzano Trentino Alto North 6.7 Adige 4 26 Lecco Lombardia North Bologna Emilia Romagna North Torino Piemonte North Varese Lombardia North Udine Friuli Venezia North 6.2 Giulia 9 28 Como Lombardia North Roma Lazio Centre Benevento Campania South Potenza Basilicata South Agrigento Sicilia South Brindisi Puglia South Matera Basilicata South Caltanissetta Sicilia South Vibo Valentia Calabria South Crotone Calabria South Enna Sicilia South Campobasso Molise South 1.2 clusters in the South. This is in line with Fabiani et al. (2003), who found very big differences between firms in the South of Italy and in the North and 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 concerned. In addition, Figure 12.2 also evidences that ICT adoption is more locally concentrated than added value per employee. 9 Following the literature on the spatial distribution of innovation (Audretsch and Feldman, 1996; Audretsch, 2003), spatial dependence was expected to exist between the observations. As Le Sage notes, spatial dependence in a collection of sample data observations refers to the fact that one observation associated with a location which we might label i depends on other observations at locations j i (Le Sage, 1998: 3). Table 12.4 reports the results of tests normally used for detecting spatial dependence. 10 ICT adoption is the percentage of firms in each province that

279 Explaining the territorial adoption of new technologies 263 North North Centre Centre South South Std Deviation: ICT_END (19) (37) (26) (19) > 7.08 (2) Std Deviation: VAA (25) (19) (44) (13) > (2) Figure 12.1 Distribution of ICT adoption and added value per inhabitant across Italian provinces: the Three Italies had at least one domain name registered as of All three tests confirm the existence of a spatial dependence, and we can conclude that the adoption of ICT by each province i is related to the adoption of other provinces j i. This shows that geography matters.