Innovation and Regional Growth in the European Union

Advances in Spatial Science Editorial Board Manfred M. Fischer Geoffrey J.D. Hewings Anna Nagurney Peter Nijkamp Folke Snickars (Coordinating Editor)

For further volumes: http://www.springer.com/series/3302

.

Riccardo Crescenzi

l

Andre´s Rodrı´guez-Pose

Innovation and Regional Growth in the European Union

Riccardo Crescenzi London School of Economics Geography and Environment Houghton Street WC2A 2AE London, United Kingdom [email protected]

Andre´s Rodrı´guez-Pose London School of Economics Geography and Environment Houghton Street WC2A 2AE London, United Kingdom [email protected] IMDEA Social Sciences MADRID, Spain

Advances in Spatial Science ISSN 1430-9602 ISBN 978-3-642-17760-6 e-ISBN 978-3-642-17761-3 DOI 10.1007/978-3-642-17761-3 Springer Heidelberg Dordrecht London New York Library of Congress Control Number: 2011930832 # Springer-Verlag Berlin Heidelberg 2011 This work is subject to copyright. All rights are reserved, whether the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations, recitation, broadcasting, reproduction on microfilm or in any other way, and storage in data banks. Duplication of this publication or parts thereof is permitted only under the provisions of the German Copyright Law of September 9, 1965, in its current version, and permission for use must always be obtained from Springer. Violations are liable to prosecution under the German Copyright Law. The use of general descriptive names, registered names, trademarks, etc. in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use. Cover design: eStudio Calamar S.L. Printed on acid-free paper Springer is part of Springer Science+Business Media (www.springer.com)

Acknowledgments

The authors are grateful to Roberta Capello, Fabrizio De Filippis, Guido Fabiani, Carlo Pietrobelli, Rikard Stankiewicz and Michael Storper who have greatly contributed to shaping many of the arguments included in this book. Riccardo Crescenzi is grateful for the support received from the Italian Ministry of Education, University and Research (Scientific Research Programs of National Relevance 2007 on European Union policies, economic and trade integration processes and WTO negotiations). He is also grateful to the Robert Schuman Centre for Advanced Studies, European University Institute and to the Department of Economics, University of Roma Tre for support during the writing of this book. Andre´s Rodrı´guez-Pose is grateful for the support of the European Research Council under the European Union’s Seventh Framework Programme (FP7/20072013)/ERC grant agreement nº 269868, of a Leverhulme Trust Major Research Fellowship and of the Prociudad-CM programme. Both authors are also members at the LSE Spatial Economics Research Centre (SERC). Some of the chapters are partially based on previously published material by the authors in different scholarly journals. The articles from which material has been included are: Crescenzi R. (2005) “Innovation and regional growth in the enlarged Europe: the role of local innovative capabilities, peripherality and education.” Growth and Change 36(4), 471-507. Rodrı´guez-Pose A. and Crescenzi R. (2008) “R&D, spillovers, innovation systems and the genesis of regional growth in Europe” Regional Studies, 42(1), 51–67. Crescenzi R. (2009) “Undermining the principle of territorial concentration? EU regional policy and the socio-economic disadvantage of European regions” Regional Studies, 43(1), 111-133. Chapter Six is based on a revised version of the paper “The territorial dynamics of innovation: a Europe-United States Comparative Analysis” (The Journal of Economic Geography, 7(6), 673-709, 2009) published by the authors jointly with Michael Storper.

v

vi

Acknowledgments

Finally, Riccardo would like to dedicate this book to his parents, Roberto e Maria Teresa, his brother Enrico and his wife Amalia. Their constant support and unconditional love made this endeavour possible. Andre´s would like to dedicate this book to his sister Pepa, for remaining so positive despite all the odds.

Contents

1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

2

Theoretical Framework: A Spatial Perspective On Innovation and the Genesis of Regional Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.2 Innovation and Regional Growth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.3 A Broader View On the Process of Innovation: The Innovation Systems Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.4 The Regional Perspective: Regional Systems of Innovation and the “Social Filter” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.5 R&D, Innovation Systems and Knowledge Spillovers . . . . . . . . . . . . . . . . 2.6 The Diffusion of Knowledge Spillovers: How Institutional Factors and Global Networks Shape the Spatiality of Knowledge Flows . . . . . 2.7 The Success of the “Core”: Where Proximities, Relational Networks and Institutions Generate Local “Buzz” . . . . . . . . . . . . . . . . . . . . 2.8 Explaining Core and Periphery Patterns: The Foundations of an Integrated Empirical Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

3

Geographical Accessibility and Human Capital Accumulation . . . . . . . 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2 The Key Relationships: Accessibility and Human Capital Accumulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.2.1 How Accessibility Influences the Process of Innovation . . . . . . 3.2.2 Education: A “Preliminary” Proxy for Innovation “Prone” and Innovation “Averse” Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3 The Empirical Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.1 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.3.2 The Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.1 Estimation Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3.4.2 Regional Innovation Activities and Economic Performance in the EU-25 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

9 9 11 13 15 17 19 20 22 31 31 32 32 34 35 35 36 38 38 39 vii

viii

Contents

3.4.3 The Translation of Innovation into Growth: The Interaction of Innovative Effort with Accessibility and Education . . . . . . . . 46 3.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49 4

5

6

The Role of Underlying Socio-Economic Conditions . . . . . . . . . . . . . . . . . . . 4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2 The Socio-Economic Conditions of the EU Regions and Their Innovative Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.1 An Empirical Definition for the Regional System of Innovation: The Social Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.2.2 The Geography of the Social Filter Conditions in the EU . . . . . 4.2.3 The Social Filter Index and Its Association with R&D Expenditure: Four Typologies of Regions and Their Differential Growth Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3 An Empirical Model for the EU Regional Growth Performance . . . . . 4.3.1 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.2 Estimation Issues . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.3.3 Innovation, Spillovers and Social Filter: Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Knowledge Flows and Their Spatial Extent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.2 Contextualising the Analysis of Knowledge Spillovers into an “Integrated Framework” . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3 Measuring the Spatial Extent of Knowledge Spillovers . . . . . . . . . . . . . . 5.3.1 Model Specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.3.2 The Spatial Extent of Innovative Spillovers: Empirical Evidence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Innovation In an Integrated Framework: A Europe-United States Comparative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.2 The Innovation Gap Between Europe and the United States and Its Structural Determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.3 Geography and Innovative Performance in the EU and the US . . . . . . 6.4 The Model: A Modified Knowledge Production Function in an “Integrated” Territorial Framework . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5 Results of the Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.1 Specific Estimation Issues and Units of Analysis for the Comparative Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6.5.2 Comparing Territorial Dynamics in Europe and the US . . . . . . .

51 51 52 52 59

62 64 64 67 68 71 75 75 76 78 78 79 81

83 83 84 87 91 98 98 99

Contents

ix

6.5.3 Agglomeration, Specialisation and Absolute Size of Clusters . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108 6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113 7

8

9

What Can We Learn From the “Integrated Approach” To Regional Development? The Impact of EU Infrastructure Investment . . . . . . . 7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.2 Infrastructure and Regional Economic Development . . . . . . . . . . . . . . . . 7.2.1 The Rationale for Public Infrastructure Development . . . . . . . . 7.2.2 A Broader Theoretical Framework for the Assessment of the Impact of Transport Infrastructure Development . . . . . . 7.3 The Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4 Results of the Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.1 Estimation Issues, Data Availability, and Units of Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.2 Stylised Facts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7.4.3 Transport Infrastructure and Regional Growth in the EU . . . . 7.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . The EU Regional Policy and the Socio-economic Disadvantage of European Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2 Regional Policy and Structural Disadvantage . . . . . . . . . . . . . . . . . . . . . . . . 8.2.1 The EU Regional Policy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.2 Territorial Concentration and Correlation with Structural Disadvantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.2.3 Where Are the Funds Most Needed? The Empirical Evidence Produced So Far . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3 Where Do the Funds Actually Go? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.1 A Measure for Socio-economic Conditions . . . . . . . . . . . . . . . . . . . 8.3.2 The Empirical Model for the Allocation of Funds Across Regions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.3.3 The Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4 Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.1 Spatial Concentration: Structural Funds Versus Socio-economic Disadvantage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.2 The Drivers of the Regional Allocation of Structural Funds . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8.4.3 Socio-economic Disadvantage and Regional Convergence . . . 8.5 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

115 115 116 116 120 121 127 127 129 133 145

147 147 148 148 150 151 152 153 154 155 156 156 158 162 167

Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 169 9.1 What We have Learnt on Regional Innovation and Growth Dynamics: Synopsis of Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . 169

x

Contents

9.2 What Are the Implications for Local and Regional Economic Development Policies? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.1 The Territorial “Integrated” Framework as a Common Platform for Top-Down and Bottom-Up Policies . . . . . . . . . . . . . 9.2.2 Translating the Territorial “Integrated” Approach into a Diagnostic/Policy Tool . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.2.3 A Diagnostic Policy Tool for Locally-Suited Economic Development Policies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9.3 Concluding Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

172 172 174 176 179

.................................................................

183

....

195

..................

199

Appendix C: Technicalities of the Principal Component Analysis and Results for the EU and the US ......................................

201

............

203

.................................

211

Appendix F: Spatial Autocorrelation Test for the Regression Residuals (Chap. 7) .......................................................

213

References

Appendix A: Data Availability and Description of the Variables Appendix B: The Weight Matrix and the Moran’s I

Appendix D: List of the Regions Included in the Analysis Appendix E: Unit Root Tests (Chap. 7)

Chapter 1

Introduction

Over the past three decades technological change has not only been the single most important force behind the process of economic growth but it has also enabled the “widening, deepening and speeding up of worldwide interconnectedness in all aspects of contemporary social life, from the cultural to the criminal, the financial to the spiritual” (Held et al. 1999: 2) that may be referred to as globalization. The progressive liberalisation of the movements of capital and labour, the sharp reduction in the cost of international and intercontinental travel, as well as the purportedly progressive convergence towards “global” cultural models, and, above all, the frictionless availability of information seem to suggest an ever-decreasing influence of both physical distance and the underlying contextual conditions upon economic interactions. Faster and cheaper access to information and technology has also led to a restructuring of how we conduct business all over the world and contributed to dismantle the barriers that anchored economic activity to specific locations. At a first superficial look the consequence of all these changes seems to be a world where neither the distance between the economic actors, nor the contextual condition in which their interactions take place would matter any longer; a world where information “once available only to the few would be available to the many, instantly and (in terms of distribution costs) inexpensively” (Cairncross 1997: 4); a world where every economy has a similar chance of exploiting and maximizing the opportunities of global interaction, regardless of its geographical location and its indigenous conditions. In brief, a world where more and more people are empowered by this access to information and become more conscious of the need to engage and compete as individuals in an integrated world. However, not all people and territories can benefit equally from the changes that globalization brings about. The process of globalization is shaping a world where there are winners and losers; where the winners are precisely those that can maximize the opportunities for innovation, economic activity, and growth that real time access to information offers. The information revolution has opened new windows of opportunity so that new actors may emerge in the global arena while others have been closed, provoking the relative decline of some previously leading regions. In addition, some economies have remained marginal in the world economic panorama. The new technological regime is producing a thoroughgoing reorganisation of the world economy, rather than a global trend towards R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_1, # Springer-Verlag Berlin Heidelberg 2011

1

2

1 Introduction

similar development levels made possible by ubiquitous economically productive knowledge. Technology has, over the past three decades, produced an enormous change in the world economic panorama. The pre-existing knowledge base of many territories and regions that flourished under the previous technological regime has been made obsolete thus bringing about a profound reorganisation of the world economy. In the US, for example, high-tech clusters have emerged not only in the southwest and the west, far away from the manufacturing belt of Cleveland, Detroit and Dayton (Acs 2002). In Europe new areas, such as Southern England (rather than the English Midlands) in the UK, Midi-Pyre´ne´es (rather than Nord-Pas de Calais) in France or Baden-W€urttemberg and Bavaria (rather than the Rhine-Ruhr) in Germany have gained economic leadership. Technological change (and the associated socioeconomic restructuring) has thus produced a specific pattern of winners and losers which, in the European Union (EU) context, has resulted in the cross-country “tendency of metropolitan areas and the ‘new industrial foci’ built around research centres in intermediate regions to grow above average, whereas traditional industrial zones and peripheral regions endure lower growth rates” (Rodrı´guez-Pose 1998a; p. 22). Furthermore the change in the technological paradigm has led to (under the pressure of a complex set of other non-technological factors1) the decentralisation of production from traditional industrial zones towards new growth spaces – such as in the north-eastern regions in Italy – where employment in research-intensive firms and innovative output has progressively increased over the past decades (Cicciotti and Rizzi 2003; D’Antonio and Scarlato 2004). This book’s aim is to illustrate the nature of the relationship between technological change and territorial development, shedding some new light on the primary causes of the observed pattern with a special reference to the case of the EU regions (also in a comparative perspective with the US). In this context the general scope of this book is to analyse what features (related to both physical space and socioeconomic realm) have allowed some areas to gain from the information revolution while preventing others from doing the same. In addition this book will examine how the current development in Information and Communication Technology (ICT) will further affect this pattern, either facilitating new development opportunities or accentuating existing disparities. The understanding of such trajectories is of fundamental importance for the future of many lagging areas in the EU 15 and for the large majority of the regions of the new member states of the Union. If R&D investments are per se sufficient to deliver economically productive innovation (thus exploiting the benefits of the so-called ‘new economy’), if knowledge, thanks to the developments in ICT is “universally available” and easily accessed, then 1

As an example, D’Antonio and Scarlato (2004), when analysing the growth dynamics of Italian regions over the past 3 decades, highlight the “decentralisation of production from large to small firms and from old industrial areas towards north-eastern and Southern Italy. Such a process has been caused, on the one hand, by the search for higher labour flexibility in order to reduce the impact of the extra-costs associated to congested metropolitan areas and, on the other hand, by policies promoting the development of the Italian Mezzogiorno“(p. 283).

1 Introduction

3

peripheral and lagging regions should be able to fully benefit from the knowledgebased economy. In this perspective the future of currently lagging regions is a longrun convergence towards the wellbeing of “core” regions and development policies should only provide incentives for increasing local innovative efforts, which are supposed to be equally conducive to economic growth both in the core and in the periphery, in “advanced” and in socio-economically disadvantaged regions. This book will question this line of reasoning and cast doubt on its fit with the empirical evidence on the regional growth dynamics of the EU 25. It will consider, on the contrary, that some intrinsic features of each territory are factors which influence its capability to translate innovative efforts and external knowledge flows into economic growth. In particular the analysis will focus upon the role of geographical peripherality (location disadvantage) and innovation averse indigenous socio-economic conditions (adverse internal conditions) by showing that their combination may persistently hamper the capability of some regions to grow and develop in the context of the knowledge-based economy. In the perspective advanced by this study, peripheral and/or socio-economically disadvantaged regions suffer from a persistent structural competitive disadvantage, which prevents them from catching up with the leading regions. Consequently the mere increase in their innovative efforts (R&D investments) may not be a sufficient condition to promote their endogenous economic development. Successful economic development policies may thus need to address the factors of socio-economic disadvantage simultaneously with the lack of innovative efforts. This work will investigate the theoretical foundations and look for empirical evidence of the explanation proposed for the differential impact observed of technological change and innovation on the development pattern of the EU regions. In particular the book aims at answering the following research questions: l

l

l

l

How can differential innovative efforts (in the form of R&D expenditure) help explain the differential regional growth performance? What geographical factors influence this relationship? Does the process of innovation in peripheral areas exhibit specific features that differentiate it from the process of innovation in other areas? What is the role of indigenous socio-economic conditions in the process of absorbing externally produced knowledge and the translation of endogenous innovative efforts into economic growth? Are the EU’s regional development policies – one of the most important largescale regional policy experiences – likely to produce an improvement in the capability of peripheral and disadvantaged areas such as to satisfy their individual needs?

Whether we can answer these questions depends on the development of a coherent theoretical framework able to handle the complexity of real world processes. There is consensus in the literature on the idea that technological change, as the single most important force behind the process of secular growth (Abramovitz 1956; Solow 1957), can be considered the key engine of the transformation taking place in the world economy at every territorial scale. There is also widespread

4

1 Introduction

agreement on the idea that innovation, (i.e., the search, development and adoption of new processes and products) is a persistent source of competitive advantage whose understanding is crucial for the assessment of the longer term economic impact of the information revolution and the analysis of its trajectory. However, where technological progress is seen as a true public good (i.e., non rivalrous and unexcludable2) its role is confined to the “residual” of the growth process, after the traditional factors of production have been accounted for (Solow 1957; Borts and Stein 1964). Exogenous and freely available technological knowledge immediately spills over into a full set of intended and unintended beneficiaries allowing (together with decreasing returns to capital) for a long-run convergence process between the individual economies. By contrast, when the accumulation of knowledge (together with the efforts necessary for its creation and accumulation) is endogeneized, its production is related to the deliberate efforts of the economic agents (by means of existing knowledge and human capital), its nature as a quasipublic good is acknowledged (due to its partial and – at least temporary – excludability) and the different absorptive capacities of the various actors3 are taken into account, it becomes a localised source of competitive advantage able to determine persistently different economic development patterns. New growth theories, while more realistically modelling the relationship between innovation and growth, still rely however upon a linear view of the process of innovation as they are unable to account for the complex set of interactions and feedbacks between the innovative agents and the external socio-economic and institutional environment. Such interactions play a crucially important role in a theory of innovation consistent with the observed differentiated capabilities shown by the various territories in translating innovative efforts into economic growth. When this kind of systemic interaction is introduced into the analysis, i.e., when the linear model of innovation is definitely abandoned, the geographical and territorial dimension of the process emerges. Since knowledge is embodied in human and institutional forms that are organized by the underlying geography, distance and space in fact exert an influence on the creation and diffusion of innovation in the form of localised knowledge spillovers that determine the creation of place-specific concentrations of knowledge. The systemic-relational and the spatial dimensions of the process are strictly interrelated: while proximity has been shown to strengthen the exchange of knowledge through the network of socio-institutional human interactions, an appropriate local socio-institutional setting needs to be in place in order to allow an effective use of exogenous knowledge flows. 2

Non-rivalrous and non-excludable goods are said to be public goods. Non-rivalrous refers to the consumption of the good by one individual that does not reduce the amount of goods available for consumption by others. Non-excludable indicates the impossibility of preventing individuals from consuming the good if they want to (Stiglitz 1986). 3 “Prior related knowledge confers an ability to recognize the value of new information, assimilate it, and apply it to commercial ends. These abilities collectively constitute what we call a firm’s “absorptive capacity.”” (Cohen and Levinthal 1990, p. 128).

1 Introduction

5

While necessary for the consistency of any analysis of the process of innovation (and its economic impacts), both the systemic interaction of the kind mentioned above and the spatial dimension of the process of innovation have been rarely combined with the rigorous approach specific to endogenous growth models and thus with the formal analysis of the relationship between innovative efforts and economic growth. From a theoretical point of view this work aims at filling this gap in the literature by combining in one empirical model innovative efforts, geographically-bounded knowledge flows (spillovers) and the underlying socio-economic conditions (as proxies for the local system of innovation). Given the limitations of both the “formal” approach of endogenous growth theories and the “qualitative” view adopted by the innovation systems approach, the need for a theoretical framework emerges, which, while preserving an acceptable degree of generality, can account for factors shown to be relevant not only by the systemic approach but also by other theoretical contributions of the literature on economic geography and regional and local development. Our contribution aims at fitting the various factors that the existing literature has demonstrated to be relevant for the process of innovation within a common “integrated” framework, thus attempting a cross-fertilization of formal growth models with other theoretical approaches to innovation. The objective of integrating theoretical contributions from different backgrounds into the same framework requires the adoption of a model able to highlight their interrelations. As underlined by the modern theories of regional development and economic geography, interactions between the socio-economic fabric, institutions and the creation and diffusion of knowledge take place at the local and regional level. Consequently, the local and regional level may, in practice, be considered the most suitable level of analysis. The core of the analysis is no longer the individual firm, the sector, the “cluster”, or “mileu”, the network of institutions but the territory where all these factors interact and whose geography influences its growth trajectory. An “integrated” and territorially oriented approach to innovation is, as this research aims to demonstrate, the most effective for the purpose of understanding the structural preconditions for the successful exploitation of the knowledge economy. In this context geographical peripherality (i.e., the reduced accessibility of an area) emerges per se as a factor of disadvantage, in addition to other conditions that may be in place in the specific territory. The low degree of accessibility of a territory not only reduces its exposition to external knowledge flows but also shapes its socio-economic structure (and its network dimension, in particular) making its innovative efforts less conducive to economic growth through a variety of mechanisms that need to be carefully investigated. It is not, however, only the relative position in the geographical space that matters for the competitive success of a territory. The endogenous socio-economic conditions that, as discussed above, may help or prevent the development of an adequate pattern of systemic interaction between local actors also need to be assessed. The capacity to single out these specific factors of disadvantage is particularly necessary in the context of the debate about the evolution of structural and

6

1 Introduction

development policies in the EU. Since the Lisbon strategy4 the EU has accepted the knowledge-based economy as its economic development model. As a consequence, the EU development policies have to be progressively more targeted toward making the EU’s lagging regions more innovative thus improving their economic performance. However, this policy approach may overlook the need to adequately address the structural sources of competitive disadvantage, which, instead, in the framework previously outlined are of crucial importance for reaping the benefits of an innovation-based economic paradigm. The reduced impact of current EU development policies, lamented by many empirical analyses, may thus be the result of the reduced capacity of past policies to properly address the contextual factors singled out in our analysis. The book will analyse the theoretical issues briefly outlined in this introduction and will subsequently test, by means of quantitative statistical methods, their empirical relevance for the EU context (and for the US, in a comparative perspective). The original theoretical framework proposed early in the book is used as the foundation for the subsequent empirical analysis. This analysis has been implemented using “standard” econometric techniques, as has been customary in a number of similar works in the field of Innovation Studies, Economic Geography, Regional Studies and Regional Economics. Comparability of results is of paramount importance in order to test our conceptual framework and draw the policy indications that we discus in the conclusions. The argumentation will progress through the various chapters as follows. Chapter 2 will present the theoretical framework of the analysis and set the foundations for the subsequent empirical investigation. The chapter combines three different approaches to the process of innovation in a homogenous theoretical framework (a) the analysis of the link between investment in R&D, patents, and economic growth; (b) the study of the existence and efficiency of regional innovation systems; and (c) the examination of the geographical diffusion of regional knowledge spillovers. The theoretical analysis will attempt to overcome the important operational and methodological barriers that, up to now, have thwarted any potential cross-fertilization between these fields of research. Using this theoretical framework, the subsequent three chapters will analyse, from a theoretical and an empirical perspective, different processes through which geography and space shape the capability of each region to produce innovation and translate it into economic growth. Chapter 3 will analyse the role of physical accessibility, showing

4 The European Council, which met in Lisbon in 2000, set the goal of making the EU “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” (Presidency Conclusions, par. 5). The regional dimension of social cohesion is, together with full employment, explicitly mentioned as the ultimate expected outcome of the strategy. Crucially, the Lisbon strategy relies on the capability of knowledge to be translated into growth in order to deliver economic development. Furthermore, by focusing policy efforts on the creation and diffusion of knowledge, growth is not only supposed to be increased but also qualitatively improved in terms of sustainability, quality of employment, and (social and regional) cohesion.

1 Introduction

7

how peripheral regions may suffer from a “locational disadvantage”, curbing their growth potential. The analysis will test to what extent regional innovative activities (for which a specific measure is developed) can play a significant role in determining differential regional growth patterns. Furthermore, the analysis will shed light on how geographical accessibility and human capital accumulation, by shaping the regional system of innovation, interact with local innovative activities, thus determining how such activities become more (or less) effectively translated into economic growth. The chapter will empirically assess whether an increase in innovative effort is likely to produce the same effect in all EU-25 regions or not. On the basis of the general relationships tested in Chap. 3, Chap. 4 will focus more specifically upon the identification of the set of socioeconomic conditions that make some regions more prone to innovation than others. After a careful quantitative analysis of the regional systems of innovation of the EU-25 regions, a multiple regression analysis is conducted including measures for R&D investment and proxies for regional innovation systems in order to uncover the importance of both endogenous innovative efforts and socioeconomic conditions for the creation of economically productive innovation and regional growth. In addition the model will include, for each region, proxies for the innovative efforts and socio-economic conditions of neighbouring areas assessing the potential location advantage determined by both a higher exposition to knowledge spillovers and by an “innovation prone” interconnected neighbourhood. Chapter 5 will further develop the empirical analysis of the differential effects of “indigenous” factors and “external” knowledge flows in the process of innovation and growth of EU regions. The analysis will not only test whether knowledge flowing from neighbouring regions improves regional growth performance but also if such knowledge spillovers are geographically bound and decay with distance. In addition the model will aim at measuring the spatial extent of knowledge spillovers. By empirically assessing the existence of knowledge spillovers and their spatial reach, the analysis will shed additional new light on the role of geography in the process of innovation. In particular, the tension between two forces: the increasingly homogeneous availability of standard “codified” knowledge and the spatial constraints of “tacit” knowledge and contextual factors will be explained. The analysis of the impact of the interaction between global (“codified” knowledge) and local (“tacit” knowledge and socio-economic conditions) factors upon the spatial organisation of economic activities will allow us to make a contribution – from an informational and cognitive perspective – to the interpretation of the EU regional growth dynamics in the light of the more general relationship between “local” economic development and “global” processes and markets (Becattini 2000; Dicken 1994; Guerrieri et al. 2001). In Chap. 6 the ‘integrated’ framework progressively developed in previous sections will be further enriched in order to explicitly take into account the impact of agglomeration and specialisation patterns and will be applied to the comparative analysis of the innovation dynamics in Europe and the United States. The quantitative comparative analysis – based on two distinct regional-level datasets – will shed

8

1 Introduction

new light on the geography of innovation in the two continents and on its capability to explain the existing innovation gap between Europe and the US. Having completed the development of a fully-functional analytical framework for the study of regional innovation and growth dynamics, the final two chapters of the book will move into the analysis of existing regional development policies, shedding light on the benefits of an ‘integrated’ approach for the detection of the broader possible set of intended and unintended effects of territorial policies including potential trade-offs between various policy targets. In this perspective Chap. 7 will look at the role of infrastructure endowment and investment in the genesis of regional growth in the European Union, assessing the economic effect of the existence and improvement of transport networks, in light of their interactions with the process of innovation and local socio-institutional conditions. In Chap. 8, instead, the empirical results of the previous chapters will be applied to the analysis of the structure of the regional policy of the European Union. The spatial distribution of the various socio-economic factors shown to influence innovation and growth (in the foregoing sections) will be compared with the spatial distribution of the EU funds devoted to compensating disadvantaged regions. This analysis is aimed at uncovering the coherence of the EU regional policies with regard to the structural disadvantage of EU regions, identifying in the process whether there is a potential inconsistency between policy objectives (favouring disadvantaged areas) and the beneficiaries of the funds. A potential mismatch between structural disadvantage and EU funds would provide a potential alternative explanation for the limited impact of the EU regional policy. Chapter 9 will present a synopsis of the conclusions of the individual chapters and discuss some more general implications from the previous analysis. In particular this chapter will argue that the “integrated” approach to the analysis of regional growth dynamics – theoretically developed and empirically tested in this book – can simultaneously work as a meso-level foundation for top-down policies and as a unifying coordination device for bottom-up local economic development policies.

Chapter 2

Theoretical Framework: A Spatial Perspective On Innovation and the Genesis of Regional Growth

2.1

Introduction

Technological change seems to be making innovation not only more “globalised” but also more “territorially-specific”. Innovation relies on “global” knowledge flows of formal codified knowledge, but as these flows become progressively easier to access and exchange, the territorial aspect of innovation and learning has become a key resource in competitive advantage. In order to understand this process, however, it is necessary to reconsider the linear model of innovation. As we will discuss in this chapter, innovation is a collective learning and socially embedded process that is crucially dependent on tacit knowledge and “untraded interdependencies”. Consequently a dialectical linkage has been established between innovation and space. While territories, with their social, cultural and institutional realm, are crucial for successful innovation, innovation is in turn a key source of competitive advantage for territories and regions. However, different streams of literature have shed light upon specific factors and “conditions” involved in the process without bringing them together in an analytical model. The capacity to innovate and to assimilate innovation have regularly been considered as two of the key factors behind the economic dynamism of any territory (Feldman and Florida 1994; Audretsch and Feldman 1996; Cantwell and Iammarino 1998; Furman et al. 2002). Yet, despite this agreement, researchers have tried to untangle the link between research, innovation, and economic growth in very different ways. Three different approaches to this relationship predominate. The first is the so-called “linear model” (Bush 1945; Maclaurin 1953), whereby basic research leads to applied research and to inventions, that are then transformed into innovations, which, in turn, result in greater growth. Empirically, this type of analysis focuses fundamentally on the link between R&D and patents, in the first instance, followed by that between patents and growth. These types of analysis are fundamentally conducted by “mainstream economists” and, despite criticisms (e.g., Rosenberg 1994), the approach remains popular with academics and policy makers. A second group can be classified under the appellations of “systems of innovation” (Lundvall 1992) or “learning region” (Morgan 1997) approaches. These approaches, associated with evolutionary economics (Dosi et al. 1988;

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_2, # Springer-Verlag Berlin Heidelberg 2011

9

10

2 Theoretical Framework: A Spatial Perspective On Innovation

Freeman 1994), concentrate on the study of territorially-embedded institutional networks that favour or deter the generation of innovation. The capacity of these networks to act as catalysts for innovation depends, in turn, on the combination of social and structural conditions in every territory, the so-called “social filter” (Rodrı´guez-Pose 1999). These approaches tend to be fundamentally qualitative and mainly conducted by geographers, evolutionary economists, and some economic sociologists. Finally, there is a large group of scholars who has mainly concentrated on the diffusion and assimilation of innovation (Jaffe 1986; Audretsch and Feldman 1996a; Cantwell and Iammarino 2003; Sonn and Storper 2008). This knowledge spillovers approach has been generally adopted by economists and geographers, using both quantitative and qualitative methods. Although such a wide variety of approaches contributes significantly to improve our understanding of the process of innovation and of the linkages between innovation and economic development, the theoretical mechanisms developed by these different, but nevertheless, complementary strands of literature have rarely been combined. There has been little cross-fertilisation. Major operational and methodological barriers have hitherto kept any potential interaction to a bare minimum. The main reasons for this lack of interaction are related to the different disciplinary backgrounds of the researchers working on innovation, to the different methods used by different approaches, and to the difficulties in operationalising some of the concepts used by different strands of the literature on the topic. This chapter represents an attempt to try to bridge this gap in the literature by combining in one model linear, innovation systems, and spillover approaches in order to develop an “integrated framework” for the understanding of regional growth dynamics, setting out the foundations for the analyses to be pursued in the subsequent chapters. In order to achieve this aim, we ground our approach on a series of fundamental theoretical mechanisms which make knowledge and its transmission an important explanation for differential growth performance. First, as highlighted by the linear model of innovation, local innovative activities are crucial for the “production” of new knowledge and the economic exploitation of existing knowledge, given the presence of a minimum threshold of local innovation capabilities (as put forward by evolutionary economics and neo-Schumpeterian strands). Such activities are not geographically evenly distributed and thus become a localised source of competitive advantage for some areas rather than others. Second, information is not automatically equivalent to economically-useful knowledge (Sonn and Storper 2008). A successful process of innovation depends on “localised structural and institutional factors that shape the innovative capacity of specific geographical contexts” (Iammarino 2005, p. 499), as indicated by the systems of innovation (Lundvall 2001), regional systems of innovation (Cooke et al. 1997) and learning regions (Morgan 2004; Gregersen and Johnson 1996) approaches. And third, technological improvements in “communication infrastructures” have not affected all kinds of information in the same way. While “codified information” can be transmitted over increasingly large distances, “tacit” knowledge is geographically bound thus determining the increasing concentration of innovation and the

2.2 Innovation and Regional Growth

11

geographical boundedness of knowledge spillovers (Audretsch and Feldman 2004; Cantwell and Iammarino 2003; Sonn and Storper 2008). The integration of these streams of literature will allow us not only to assess how differences in innovative activities can explain differential regional economic performance, but also to show how exposure to localised knowledge spillovers together with indigenous socio-economic conditions, may significantly influence a region’s capacity to translate such efforts into economic growth. This chapter is organised into seven further sections. In the first six sections the theoretical relationships are analysed in order to explain the mechanisms driving and conditioning the translation of innovative efforts into growth. Building on such analysis, the final part outlines an empirical model which extends the neoclassical linear growth model and in which each variable is justified and grounded in the previously developed theoretical framework.

2.2

Innovation and Regional Growth

The possibility of explaining regional growth differentials on the basis of differentiated innovative activities relies on the understanding of the mechanisms through which knowledge is created and translated into growth. Where technological progress is seen, as in the traditional neoclassical perspective, as independent of capital accumulation, economies of scale, human capital and, above all, as a truly public good, its creation is unrelated to the rest of the economic system and the understanding of its effect on growth rates is regarded as a “residual” matter, after the traditional factor of production have been accounted for (Solow 1957; Borts and Stein 1964; Richardson 1973 when regions are considered). In such a perspective exogenous and freely available technological knowledge together with the assumption of decreasing returns to capital produce, in the long run, automatic convergence in regional growth rates without any role being assigned to indigenous innovative activities. When, in contrast to neoclassical assumptions, technology and human capital accumulation are fully recognised as the result of explicit decisions of economic agents, economic growth becomes “an endogenous outcome of an economic system, (and) not the result of forces that impinge from outside” (Romer 1994, p. 3). Technology, technological progress, and human resources – considered as the main forces “behind perpetually rising standards of living” (Grossman and Helpman 1991, p. 24) – become endogenous, and change differently in different territories according to the quality of human resources and to the amount of human and physical capital devoted to research and development (Romer 1986; Lucas 1988; Rebelo 1991). Innovation takes place where the adequate endowments of human and physical capital are located and, vice versa, innovation generates economic dynamism which attracts more human resources and more capital. Hence, under an endogenous growth framework, innovation and human capital will tend to co-locate in relatively compact geographical areas.

12

2 Theoretical Framework: A Spatial Perspective On Innovation

When “the underlying source of sustained growth in per-capita income, namely the accumulation of knowledge” is endogeneized “through formal education, on the job training, basic scientific research, learning by doing, process innovations, and product innovations” (Aghion and Howitt 1992, p. 1) the relevance of indigenous innovative activities changes completely. When the “endogenous growth model” fully incorporates the view of innovation as a result of deliberate efforts, a wholly new light is shed on the contribution of innovation towards the understanding of growth dynamics. New knowledge is not a pure public good but is “produced” by existing knowledge and human capital through investment in R&D, remunerated by the temporary extra-rent provided by the (partial and at least temporary) appropriability of the results of innovation (Romer 1990, Grossman and Helpman 1991 in a monopolistic competition framework). However, as the benefits of innovative efforts are not fully appropriated by the innovative firm/nation/region because knowledge “spills over”, the existing pool of knowledge is also increased thereby also benefiting other operators in their additional innovative activities. Thus the inclusion of innovation efforts into the determinants of growth allows this theoretical perspective to account for permanent regional disparities in the rates of growth. Therefore, regions that are well endowed in terms of knowledge and human capital, thanks to their accumulated pool of knowledge, will have a “continuous advantage over less well-endowed regions which rely on exogenously embodied technology (in the shape of new capital equipment purchased from other regions) because they are not capable of producing their own” (Armstrong and Taylor 2000, p. 87). In this context the ways in which knowledge is transmitted from its “producers” to the whole set of potential (intended or unintended) beneficiaries becomes a crucial issue. Technological knowledge is not exhausted after use. It is cumulative, being based on an existing pool of knowledge, and rarely completely appropriable by whoever holds its property rights. It produces spillovers whose effects offer an additional important explanation of differentiated development patterns. Industry-specific spillovers suggest that the different specialisation pattern of each region may result in different growth performances, while geographically concentrated spillovers (as those analysed in Audretsch and Feldman 1996a) produce a local accumulation of knowledge, which, through a cumulative causation mechanism, facilitates the clustering of economic activity (Maurset and Verspagen 1999). By bringing innovation to the fore, it is often assumed that greater investment in basic R&D will lead to greater applied research and to an increase in the number of inventions, that, when introduced in the production chain become, growth-enhancing innovations. This linear perception of the innovation process has localised R&D investment has the key factor behind technological progress and, eventually, economic growth. In essence, the implications of this approach are that the higher investment in R&D, the higher innovative capacity, and the higher the economic growth. Despite being much derided (e.g., Fageberg 1988; Verspagen 1991; Rosenberg 1994; Morgan 1997), the linear model remains popular with academics and policy makers because of its simplicity and powerful explanatory capacity: nations and regions that invest more in R&D, generally tend to innovate more, and

2.3 A Broader View On the Process of Innovation

13

often grow faster. But by focusing on local R&D, the linear model completely disregards key factors about how innovation is actually generated. These factors are related to the context in which innovation takes place and to the potential for territories to assimilate innovation being produced elsewhere. The potential for the concentration of economic activity and for divergence becomes more evident when issues such as the minimum thresholds of R&D and of appropriability of technology – highlighted by the neo-Schumpeterian strand of the endogenous growth approach – are considered. For R&D investment to be effective a minimum threshold of investment is necessary, making the relationship between investment in R&D and economic growth not linear. Furthermore there are strong threshold effects and external economies associated with R&D investment and returns from R&D rely heavily on the quality of the workforce conducting research, on the concentration of R&D centres in limited spaces, on the quality of the local human capital (Audretsch and Feldman 1996a; De Bondt 1996; Engelbrecht 1997), and, above all, on the amount of investment (Scherer 1983; Dosi et al. 1988). Hence, limited and/or dispersed investment in R&D in lagging areas may not yield the expected returns, as most R&D projects may lack the adequate dimension to conduct competitive research and local scientists and researchers are likely to be more isolated than in advanced technological centres. In addition, as will be discussed in further detail below, the local economic fabric may lack the capacity to successfully achieve the passage from technological progress to innovation and to economic growth (Rodrı´guez-Pose 1999).

2.3

A Broader View On the Process of Innovation: The Innovation Systems Approach

As underlined in the previous paragraphs, knowledge and innovation are important sources of regional growth. However, there are other factors which have to be accounted for because, ceteris paribus, regions show differential capabilities to absorb and translate available knowledge into (endogenous) economic growth. Many of the agglomeration effects of the endogenous growth theories are reinforced by the predictions of numerous institutional theories that underline the role of institutions and institutional factors on economic activity. These theories, despite their different origins, coincide on the role played by institutions in fostering economic concentration. Many studies have unearthed a close link between “good” institutional conditions or the presence of strong communities and the clustering of economic activities. Qualitative work on clusters and industrial districts (e.g., Piore and Sabel 1984; Kristensen 1992; Semlinger 1993; Burroni 2001), “learning regions” (Gertler et al. 2000; Henry and Pinch 2000; Bathelt 2001), and regional systems of innovation (Cooke and Morgan 1998) stresses how complex institutional and governance arrangements create the conditions for economic activity to thrive

14

2 Theoretical Framework: A Spatial Perspective On Innovation

and ultimately – as good institutional conditions are hard to replicate – to agglomerate. Factors such as the close interaction among local political actors, the presence of a functioning civil society, regional administrations, employers organizations and trade unions – in what Trigilia (1992) calls an “institutionalized market” – favour economic development and agglomeration. Well developed traditions, strong trade unions co-operating with employers, and nation-wide institutions work in a similar direction. Conversely, the absence of poles of collective action often leads to the formation of vicious circles of low growth. The lack or relatively little importance in social life of collective organizations, the presence of clientelistic practices, or the governing of social activity by simple social structures (often characteristic of relatively remote and backward spaces) facilitate migration and discourage economic activity. Many quantitative analyses reach similar results. Putnam’s (1993) work on Italian social capital shows how differences in levels of community institutions between Northern and Southern Italy are at the base of their sizeable income inequalities. Other research has found that different institutional proxies of community, such as group participation, help explain higher economic performance (Knack and Keefer 1997; Zak and Knack 1998; Beugelsdijk et al. 2004; Guiso et al. 2004), or that, conversely, excessive divisions within societies limit their growth potential (Easterly and Levine 1997; Rodrı´guez-Pose and Storper 2006). Taken to its limits, some analysts indicate how having a high density of closelyknit institutional networks in close physical proximity – called “institutional thickness” by Amin and Thrift (1995) and “institutional capital” by Healey (1998) – is a key condition for economic development. Combinations of “intellectual capital” (i.e., knowledge resources), “social capital” (trust, reciprocity, cooperative spirit and other social relations), and “political capital” (capacity of collective action) within these institutional networks determine the potential for development. The greater the density of complex institutional networks within a given territory, the greater the potential for higher growth and development (Amin and Thomas 1996; Morgan 1997; Cooke and Morgan 1998). How can this heterogeneous body of literature be reconciled with the theories of innovation and growth discussed in the previous paragraph? The empirical evidence shows that “the ability to adapt new technologies depends on the institutional infrastructure, education, geography and resources devoted to R&D” (Maurseth and Verspagen 1999, p.152). These factors can be included in the analysis of the process of innovation following different, but reconcilable (as we will see), theoretical approaches to such a process. According to the formal neoclassical perspective of the “technology gap theory” (Abramovitz 1985; Fagerberg 1988; Verspagen 1991) differentiated “absorptive” capabilities – in their turn the result of different economic and institutional infrastructures – determine the clustering and geographical concentration of economic activity. Instead, moving towards an “institutionalist” perspective, where innovation is not directly the outcome of a linear production function, the institutional environment acts directly as the generator of creative synergies and externalities (Dosi et al. 1988; Freeman and Soete 1997 among the others). The “evolutionary

2.4 The Regional Perspective: Regional Systems of Innovation the “Social Filter”

15

approach” (e.g., Nelson and Winter 1982), on the other hand, focuses upon the process of transformation of new ideas (genotypes) into technological advances through a series of mutations whose “evolutionary success” is determined by a natural market selection. All these heterogeneous (but only up to a certain extent1) views fit into the common conceptual framework of “systems of innovation”: “the network of institutions in the public and private sector whose activities and interactions initiate, import, modify and diffuse new technologies” (Freeman 1987, p. 1). Lundvall (1992) broadens this systemic view to include all the parts of the economy and its institutional infrastructure that affect the learning process and suggests that historical and theoretical aspects be jointly considered in order to discriminate what (and what not) to include in the analysis of the individual case under scrutiny. Thus, in this perspective, the definition of innovation not only includes the creation of the Schumpeterian “New combinations”, but also their “diffusion” (as in Nelson) and “new forms of organisation” and “institutional innovation”. The operational translation of the concept of systems into a concrete spatial (national or regional) or sectoral dimension2 gives rise to the concepts of sectoral, national and regional systems of innovation. In what follows we will focus upon the geographical approach, even if the strong complementarities between the sectoral and the geographical approach (Edquist 1997), will allow them to be considered in a common framework.

2.4

The Regional Perspective: Regional Systems of Innovation and the “Social Filter”

Within each national system of innovation a marked degree of unevenness in the geographical pattern of innovative activities has to be registered. This phenomenon does not only depend on the degree of centralization of the individual national territories as “even countries with fairly uniform rates of innovation geographically can hide quite marked disparities on a local or regional level (in terms of systems of innovation)” (Howells 1999, p. 69). The analysis of such disparities leads to the hypothesis of the existence of “self-consistent” regional systems within the

1

Evolutionary and interactive learning approaches can be easily reconciled when, following Edquist (1997), interactive learning is considered as a selection mechanism in an evolutionary process. 2 In this area there is no agreement in the literature about the most convincing approach. Some authors (Nelson and Rosemberg, 1993 and Lundvall 1993) even question the rationale for the study of national systems of innovation preferring the idea of the emergence of a truly global system.

16

2 Theoretical Framework: A Spatial Perspective On Innovation

framework of national ones. In a “top–down” perspective, as pointed out by Howells (1999), the regional system fundamentally originates from “discontinuities” and regional unevenness in the national components of the system (governance structure and institutional arrangements, regional industry specialisation, educational policies etc.) or by the different ways in which both national institutions and regulatory frameworks actually work at a local level. On the contrary, when, in a bottom-up perspective, the role of face-to-face contacts and tacit knowledge is considered local and regional systems are no longer proto-systems at the lower level of a hierarchical scale (from global to local) but key building-blocks and the engine of the innovative process. In this perspective the process of innovation is not separate from the entire functioning of the socio-economic sphere but embedded in the (various) territorialized processes responsible for the economic performance of each economic space. Innovation thus needs to be linked to the cluster structure of the economy and the regional innovation system should be understood in terms of the relationships and flows between the various actors and parts of the innovation system itself (Cooke 1997). In this sense the regional innovation systems approach shows important points of contact with the most advanced theories of regional development. The concepts of socio-cultural milieu, heterarchies and networks (as opposed to hierarchical relationships), tacit knowledge and embedded interfirm organizations are part and parcel of a common language (Cooke 1998) which can be referred to as the “network” or “associational” paradigm (Morgan 1997). These perspectives are brought into a systematic theoretical view in the Storper’s (1997) “holy trinity of the reflexive turn” of regional development (technologies, organizations and territories). “The regional development problem associated with building different systems of innovations thus turns essentially on building the capacities for reflexive, collective action and the forms of coordination consistent with the kind of action required in each world (of production)3” (Storper 1997, p. 126). Each “basic kind of product” is thus associated with a different system of innovation thereby accounting for the “fundamental diversity” of the world economy. This theoretical framework, which would be over-ambitious to try to analyse here, not only harmonically “embeds” innovation into its socio-economic context but also effectively integrates sectoral and territorial systems of innovation. Some of the most relevant findings related to these approaches are the relevance of proximity, local synergy, and interaction (Camagni 1995a; Camagni and Capello 2003) and the importance of “inter-organization networks, financial and legal institutions, technical agencies and research infrastructures, education and training systems, governance structures, innovation policies” (Iammarino 2005, p. 499) in shaping innovation. The explanatory capacity of such approaches is, however, somewhat constrained by the problems of operationalising in a relatively homogenous way across territories the territorially-embedded networks, social economic structures, and institutions that are at the heart of these approaches. By nature the systemic

“Each (. . .) set of conventions describes a framework of action, different for each basic kind of product, which we label a world of production.” (Storper 1997, p. 112). 3

2.5 R&D, Innovation Systems and Knowledge Spillovers

17

interactions between (local) actors are intrinsically unique and thus hard to measure and compare across different systems. A potential solution to this problem is the “evolutionary integrated view of the regional systems of innovation” (Iammarino 2005). By comparing national (macro-level) and regional systems of innovation (micro-level i.e., that of firms and localised institutional networks), a meso-level emerges characterised by “local structural regularities from past knowledge accumulation and learning” (Iammarino 2005, p. 503). This implies the existence of a series of “external conditions in which externalised learning and innovation occur” (Cooke 1997, p. 485) which can be identified across innovation systems and on which innovation strategies can be based. These conditions act as “conditions that render some courses of action easier than others” (Morgan 2004) or “social filters” or, in other words, the unique combination “of innovative and conservative components, that is, elements that favour or deter the development of successful regional innovation systems” (Rodrı´guez-Pose 1999, p. 82) in every space.

2.5

R&D, Innovation Systems and Knowledge Spillovers

Territories can rely not just on their internal capacity to produce innovation either through direct inputs in the research process or through the creation of innovation prone systems in the local environment, but also on their capacity to attract and assimilate innovation produced elsewhere. Within the framework outlined in the previous sections innovation is not the result of a linear process. On the contrary, it has been shown to be the result of a complex set of interactions between innovative units (R&D departments within firms, universities, research centres etc.) and their external environment through the “network” structure of the regional economy. Such interactions produce the transmission of knowledge in the form of “knowledge spillovers” (Jaffe 1986; Acs, Audretsch and Feldman 1992) that are reaped by local actors. The origin of knowledge spillovers can be local, but they can also be generated outside the borders of the locality or region object of the analysis, as “there is no reason that knowledge should stop spilling over just because of borders, such as a city limit, state line or national boundary” (Audretsch and Feldman 2003, p. 6). If there are internal and external sources of spillovers important questions arise. The first relate to the balance between internally generated innovation and externally transmitted knowledge and the extent to which a territory can rely on externallygenerated knowledge for innovation. The second group of questions concern the local and external conditions that will maximise the diffusion of knowledge. While the final group deals with the capacity of knowledge spillovers to travel and the potential for distance decay effects. In order to address these questions we have to resort to the theoretical distinction between different communication technology of codifiable information and tacit knowledge. “Codifiable information is cheap to transfer because its underlying symbol systems can be widely disseminated through information infrastructure” (Leamer and Storper 2001, p. 650). However, information is not completely codifiable due to some specific features which, in some cases, make

18

2 Theoretical Framework: A Spatial Perspective On Innovation

codification impossible or too expensive. “If the information is not codifiable, merely acquiring the symbol system or having the physical infrastructure is not enough for the successful transmission of a message” (Storper and Venables 2004, p. 354). In this latter case we can include “face-to-face” contacts that act as communication technology. Face-to-face contacts exhibit at least two relevant features for the processes under analysis: they are an intrinsically “spatial” communication technology and they are “socially” shaped. The space-sensitivity of such contacts is the mechanism through which geography and distance exert their influence over the process of innovation. The combination of physical geography, communication and transport infrastructures and urbanisation patterns determines how easy (difficult) and dense (sparse) such contacts will be, thus emphasizing (hampering) their “potential” as communication technology. However, while face-to-face contacts as a communication technology make the transmission of innovation possible they also pursue other functions that make communication more effective. Among these functions Storper and Venables (2004) cite the following: trust and incentives in relationships, screening and socialising and motivation. These functions are clearly influenced by the socio-institutional environment in which they take place. As a consequence not only is the “production” of innovation itself shaped by the “social filter” in place in the local economy but this latter factor also influences the extent and the effectiveness of the diffusion of innovation and knowledge. For this reason the role of interconnection between territories and the corresponding knowledge flows cannot be fully understood unless associated with the underling socio-economic conditions. It is also reasonable to expect that face-to-face contacts are maximised within the region due to the effect of closer proximity and common socio-institutional infrastructure and networks. However, part of the “uncodifiable” knowledge produced in a region, could overcome the limits of the “institutionally defined” region thus “flowing” into neighbouring interconnected territories. However, and in contrast with codifiable information, the process of transmission of tacit knowledge is costly and would suffer from strong distance decay effects. Face-to-face contacts are maximised within relatively small territories, due to a combination of proximity and the presence of common socio-institutional infrastructure and networks. The potential to reap knowledge spillovers will thus be maximised within the region. Some of this knowledge will nevertheless spill over beyond the borders of the region or locality flowing into neighbouring areas, as a consequence of the existence of different forms of inter-regional contacts. Flows of interregional knowledge are thus important as agents of innovation, but their influence is likely to wane with distance (Anselin et al. 1997; Adams and Jaffe 2002; Adams 2002), as the potential for face to face and other forms of interaction decay. Hence not only the local innovative efforts have an influence on the innovative output and on economic performance, but also on the capability to access external sources of “uncodified” knowledge. Such “accessibility to extra-regional innovation” in its turn must be confronted with the internal social-filter conditions which make communication more effective and determine to what extent such knowledge is translated into economic growth. The evidence of the spatial boundedness of knowledge spillovers not only contradicts the idea of ubiquitous knowledge evenly

2.6 The Diffusion of Knowledge Spillovers

19

available everywhere, but also helps explain how peripherality can persistently hamper regional innovative capacity after controlling for indigenous innovative efforts: the smaller the spatial extent of knowledge spillovers, the lower the exposition of peripheral areas to externally produced knowledge. While highly-accessible core regions can benefit from innovative activities pursued in their proximity, the spatial boundedness of spillovers prevents them from reaching peripheral remote regions. As a consequence, the stronger the spatial decay of the spillovers the more accentuated their tendency to develop localised pools of knowledge in central locations.

2.6

The Diffusion of Knowledge Spillovers: How Institutional Factors and Global Networks Shape the Spatiality of Knowledge Flows

As O’Brien (1992), Cairncross (1997), and Friedman (2005) posit, there is little doubt that, in theory, progress in telecommunications and in the capacity to store and diffuse massive amounts of information online has greatly reduced the role of physical proximity for the development of economic activity. However, physical or geographical proximity (the focus of previous paragraphs) is only one dimension of proximity. Boschma (2005, p. 62) identifies four other dimensions: cognitive, organizational, social, and institutional. Cognitive proximity is related to the fact that “knowledge and innovations are often cumulative and localised outcomes of search processes within firms with a high degree of tacit knowledge” (Boschma 2005, p. 63). Organizational proximity refers to the organizational practices and interdependencies that facilitate interactive learning, while social proximity highlights the fact that economic activity is embedded in a social context (Granovetter 1985; Grabher 1993). Finally, institutional proximity refers to the presence of similar institutions, such as “a common language, shared habits, a law system securing ownership and intellectual property rights, etc” (Boschma 2005, p. 68) that provide the support for economic co-ordination. While Boschma (2005) is careful to state that these different types of proximity do not necessarily relate to geographical proximity, we will argue that the reason behind the emergence of core-periphery patterns is precisely the strong interdependence of all the different types of proximity and how these different proximities coalesce in large metropolitan areas and clusters (and hence in relatively reduced geographical scales from a world perspective). Our tenet is that large urban agglomerations (high density of face-to-face contacts) in “central locations” (i.e., where exposure to external knowledge flows is maximised thanks to high accessibility) provide the setting where economic and social actors benefit from proximity to other economic and social actors with whom they can relate from a cognitive, organizational, social, and institutional dimension, creating the adequate environment for exchanges of ideas, Jacobs’ type externalities, innovation, and ultimately, economic activity and growth (Duranton and Puga 2001). Large (urban and industrial) agglomerations provide the anchor for the flows generated by

20

2 Theoretical Framework: A Spatial Perspective On Innovation

the information and knowledge society to take hold: it is true that advanced economic activity can now happen in more areas of the world than before, but, even in these places, it will tend to increasingly concentrate in a series of agglomerated relational nodes. In other words, physical proximity acts as an enabling force that makes other proximities more likely to be developed: physical proximity (at least on a temporary basis) is a necessary (though not sufficient) condition for the exchange of existing knowledge and the generation of innovative ideas. In this perspective, the innovative performance of countries and regions is not only the result of their innovative efforts but it is also influenced by the spatial configuration of their innovative agents. Such a spatial configuration not only influences the exposure to internal and external knowledge flows but also impacts upon the development of other crucial proximities by interacting with indigenous socio-institutional conditions. The set of proximities needed to generate a virtuous circle of innovation – by allowing the emergence of complex innovative network relationships, operating between and across different scales (from local to transnational) – further contributes to the emergence of core-periphery patterns in the world economic geography. From this perspective “innovation systems are a combination of intra-local, extra local and transnational network connections” which “are not just intra or inter-corporate in nature (as highlighted in Faulconbridge 2006), but may also encompass other forms of social networks” (Coe and Bunnell, 2003: 454). These networks generate a multifaceted geography of relations in the world economy which may systematically favour some actors (those enjoying the best balance of the various proximities with the most innovative actors), while further marginalising those at the geographical, cognitive, organisational, social, and/or institutional periphery.

2.7

The Success of the “Core”: Where Proximities, Relational Networks and Institutions Generate Local “Buzz”

The mechanisms discussed above provide us with a clear framework for the understanding of the economic success of leading cities and industrial agglomerations. The synergic co-existence of cognitive, organizational, social, and institutional proximities brought together in a reduced geographical environment and framed into a favourable socio-institutional environment gives rise to local “buzz”. As discussed before, the concentration of local innovative activities improves local economic performance of “core” areas but also produce localised knowledge spillovers whose beneficial effects not only depend on proximity relationships, but also on the presence of local institutions (or social filters) enabling their absorption and translation into further economic growth. In addition, by enabling face-to-face contacts and the transmission of uncodified/tacit (or uncodifiable) knowledge – often in conjunction with their role of major nodes of (material and immaterial) global network relations – “buzz” areas in the economic “core” benefit from an

2.7 The Success of the “Core”

21

enduring competitive advantage over other territories which reinforces other agglomeration forces in a process of cumulative causation. The success of “buzz” areas also depends on other localised factors such as a favourable balance between specialisation and diversification. While increasing specialisation is likely to foster MAR (Marshall-Arrow-Romer) externalities within the same industry, the diversity of economic activities pursued locally allows local actors to benefit from knowledge base complementarities and across-industry exchange of ideas (Jacobian externalities). The empirical literature suggests that both MAR (Glaeser et al. 1992; Henderson 1999) and Jacobian externalities (Andersson et al. 2005; Carlino et al. 2001; Feldman and Audretsch 1999) may play an important role in fostering innovation either in different industrial contexts4 or at different phases of a product life cycle.5 A crucial issue for the prosperity and success of cities stems from the capability to efficiently exploit MAR and Jacobian externalities. When other forces (historical, institutional, political) prevent the evolution of the cluster from reaching its most efficient equilibrium at any moment in time between both types of external economies, overall economic performance could be hampered. Diversified cities tend to be larger while specialised cities are generally smaller in size. Whereas both diversified and specialised cities can in principle perform equally well, the potential risks for specialised cities are greater. These risks are related to their lower innovative capacity and their greater exposure to rise and fall patterns of specific sectors of specialised cities (Duranton and Puga 2000). In the long-run, intervention in the form of policies that encourage labour mobility (mainly to larger diversified cities) in order to address the decline of specialised cities may be needed. Hence it is fundamentally the unique mix of social, institutional, cognitive, and organizational proximities found in large metropolitan areas that once again allows for the adequate linkages to be developed and for the right mix of specialisation versus diversification to emerge. Once this process is activated it has an enormous cumulative potential: the productivity of local innovative activities is significantly enhanced when the conditions mentioned above are met, generating the economic incentive for further investment. New investments in innovation, in their turn, not only produce localised spillovers but also directly and indirectly increase local absorptive capabilities and stimulate the continuous updating of the local socio-institutional environment. A favourable socio-institutional environment is, in its turn, prone to the development of outward connections, extra-regional interdependencies, and global network relations. The most important buzz cities (e.g., London, New York, L.A.) are nodes of international business, financial and cultural networks, locations of the headquarters of many multinational corporations; they are at the very centre of “global” traveland-meeting activities. However, “the highest levels of international business require 4

Henderson et al. (1995) find that Jacobs-type externalities prevail in high tech and MAR in capital goods industries. 5 Duranton and Puga (2001) suggest that firms develop new products in diversified creative urban contexts, subsequently, relocating to specialised cities in the mass production phase in order to exploit cost advantage.

22

2 Theoretical Framework: A Spatial Perspective On Innovation

insertion into locally-grounded government and political networks in order to function efficiently” and although “the precise mix of activities involving faceto-face contacts and collocation will change, they (. . .) will continue to generate agglomeration of highly skilled individuals, firms and bureaucracy in high-cost urban centres” (Storper and Venables 2004, pp. 366 and 368). This is reflected in Bathelt et al.’s (2004) “local buzz, global pipeline” model, which explicitly brings extraregional dynamics to light: extra-agglomeration knowledge flows complement local buzz by means of investments in channels of communication (pipelines). If learning is “increasingly inserted into various forms of networks and innovation systems (at regional, national and international levels)” (Asheim and Coen, 2006, p. 171), cities are likely to become the centres of the knowledge based economy thanks to their capacity to act both as buzz environments and major nodes of immaterial/a-spatial/ temporary networks. This process is not only about a few major world centres, but has produced a complex roster of cities where “buzz” cities are functionally interconnected by an uneven world city system (Beaverstock, Taylor, and Smith, 1999). Furthermore, the increasing importance of cities is likely to be complemented by the emergence and reinforcement of a number of highly specialised high-tech centres of excellence where the importance of global interconnections may complement and even exceed that of local buzz (Moodysson et al. 2005). This set of economic forces is continuously shaping and re-shaping the world economic geography. However, the whole system is highly dynamic and big radical shifts in the technological frontier may allow new windows of opportunity to be opened (and others to be closed) thus allowing new cities and agglomerations to emerge in the global landscape but, at the same time, condemning other areas to economic decline.

2.8

Explaining Core and Periphery Patterns: The Foundations of an Integrated Empirical Framework

So far we have analysed how different theoretical approaches shed light upon a variety of mechanisms that link the process of innovation to regional economic performance. The analysis of these mechanisms has uncovered the complexity of the role of innovation in the genesis of regional growth and has made explicit the insufficient explanatory power of the linear neo-classical view of this process. However, in this same perspective, also the growth theories which attempted to incorporate the Schumpeterian legacy (Fagerberg 1988) – while providing a more realistic view of the process of innovation – still fail to account for the dynamics that emerge when the spatial/territorial dimension of the process is dealt with. The analysis of the complex factors shaping the relationship between innovative efforts and economic performance at the territorial level calls for an appropriate analytical framework. Such a framework should possess a number of important features that would make it a suitable foundation for empirical analysis and policy guidance: not only a tool for the detection of the factors of success in leading regions (e.g., as in the local buzz

2.8 Explaining Core and Periphery Patterns

23

theory or in the “global pipelines” perspective) but a full-embracing conceptualisation of the determinants of regional growth in both core and peripheral areas. Let us briefly discuss the key features of such a framework. In a first instance the empirical framework will take the linear relationship between local innovative efforts and economic growth as a point of departure. This will on the one hand provide us with solid foundations for quantitative analysis – linking directly our model with the endogenous growth perspective (Romer 1990; Cheshire and Magrini 2000) and the Knowledge Production Function approach (Griliches 1986; Audretsch and Feldman 1996; Audretsch 2003) – and, on the other, will ensure comparability with other existing studies on regional growth and convergence. In addition, by adopting the relationship between innovative efforts and the generation of new ideas/ knowledge as its milestone, our conceptual framework will be a suitable foundation for both quantitative (as in the “mainstream” economics literature) and qualitative (as in large part of the literature on technological development and systems of innovation) analysis. The linear relationship between R&D, innovation and economic growth (typical of existing quantitative analyses) will progressively be developed in order to incorporate proxies for the relevant more “qualitative” conditioning factors singled out by other streams of literature (those on the role of institutions, geographical factors and distance in relation to the process of innovation). In this perspective the model will combine insights from different streams of literature in an eclectic fashion. This book, throughout its chapters, will progressively develop the basic linear relationship between innovative input and economic performance and incorporate the complexity introduced by other streams of literature taking into account the role of space and socio-institutional conditions in a progressively more sophisticated way. While cross-fertilising linear and non-linear, quantitative and qualitative approaches to the generation of regional economic dynamism, this conceptual framework will also reconcile the often contradictory views resulting from the adoption of either a top–down or a bottom–up perspective. The relative importance of the various determinants of regional growth varies significantly where assessed from opposed perspectives. Top-down analyses are designed to capture macrodevelopmental dynamics: as an example they can assess the aggregate average impact of innovative efforts on the economic performance of a cross-section of regions, highlighting common and “general” trends. In so doing these models will treat as a “residual” all specific idiosyncratic factors that differentiate one region from the other. Conversely, bottom-up analyses will focus their attention on the specificities of a set of regions capturing their unique internal dynamics but overlooking inter-regional trade offs and general trends. While these limitations are logical consequences of the perspective adopted, they significantly hamper the capability of these models to provide an accurate picture of the real world and act as foundations for development policies. In order to address this weakness of the existing literature, the empirical framework developed in this book aims to bridge top-down and bottom-up views on the process of regional and local economic development by adopting a meso-level perspective based on the inclusion into a macro-quantitative framework of proxies for factors generally treated as

24

2 Theoretical Framework: A Spatial Perspective On Innovation

idiosyncratic (and confined into the residual) by mainstream analysis. By explicitly looking at quantitative variables grounded into a largely qualitative literature (e.g., systems of innovation) we will be able to test their generality as drivers for innovation and growth. Key qualitative concepts – traditionally addressed by means of case studies analyses – will be developed in order to be treated in a quantitative framework: this will imply further theoretical elaboration of the key concepts in order to emphasize their “non-strictly-idiosyncratic” component and indentify the most appropriate proxies. As an example let us consider how regional economic performance is profoundly influenced by extra-regional trans-local factors that have to do with the different ways in which regions interact with each other across distance. External, extra-regional factors impact upon local innovative performance by means of spillovers and externalities that can only be captured in a macro perspective by covering simultaneously a plurality of regions by means of the appropriate spatial weights (rather than by focusing upon one single regional case in isolation). However, the impact of such flows crucially depends on a set of localised (largely idiosyncratic but still partially generalizable) socio-institutional conditions that the meso-level integrated approach outlined above can effectively capture, providing us with a more realistic view on the functioning of the regional economy. The realism provided by an eclectic integrated approach comes, however, at the price of a lower degree of internal coherence when compared to quantitative empirical models directly derived from macro-regional theories. But it will also mean that the following chapters of this book will show the empirical relevance of more qualitative factors brought into the picture by an integrated approach. In addition, the final conclusive chapter will show how this approach can serve as a needed meso-level foundation for local and regional development policies. Following this line of reasoning, in this section, we show how the various theoretical perspectives discussed so far, can be brought together in a joint framework able to drive the specification of the empirical models presented in the following chapters. In other words we will attempt to mould different theoretical perspectives into an eclectic approach to the territorial genesis of innovation. In so doing we aim at singling out the “forces at work” in regional growth dynamics rather than testing the explanatory power of one theory of innovation against the other. We assume Fagerberg’s (1998) approach as the starting point of our analysis. This approach explicitly aims at overcoming the limitation of previous research on the explanations for differential cross-country growth rates: the “catch-up” analysis, the “growth accounting” and the “production-function” studies. The “catch-up” literature (e.g., Abramovitz 1986) seems unable to explain the forces behind the opening up of the observed technological gaps which, once established, become the engine for the diffusion of innovation through imitation mechanisms. The mainly descriptive contributions in this stream of literature have placed too much emphasis on diffusion processes while neglecting the importance of innovation for developments in leading countries and the persistent disadvantage of non-converging countries. The “growth accounting approach”, for its part, fails to distinguish between “active” and “passive” factors of growth i.e., between the true determinants and supportive processes as

2.8 Explaining Core and Periphery Patterns

25

well as failing to include innovation among such determinants. Finally, the production function approach, relying upon equilibrium assumptions, is not only methodologically inconsistent with the introduction of “disequilibrium” factors (i.e., unemployment) but necessarily fails to account for “differences in technological levels and innovative performance across countries, which we believe to be one of the most fundamental disequilibrium mechanisms in the world economy” (Fagerberg 1988, p. 438). Building on the limitations of previous contributions and assuming disequilibrium conditions “right from the start”, Fagerberg (1988) proposes a “technology-gap theory” of economic growth as an application of Schumpeter’s dynamic theory of capitalist development which explicitly assumes the interaction of two “conflicting” forces [innovation (which generates the technology gap) and imitation or diffusion (which tend to reduce it)] as an “engine of growth”. The capacity to single out these two fundamental forces behind the process of economic growth makes Fagerberg’s model particularly interesting for our approach. The production function takes the form: Q ¼ ZDa N b Ct

(2.1)

where Q is level of production of a country, Z is a constant, D is the level of knowledge transferred from abroad, N the level of knowledge created in the country or “national technological activity”, C is the country’s capacity for exploiting the benefits of knowledge (both from D and N). Following Fagerberg, by differentiating and dividing through with Q, denoting growth rates with small-case letters, we obtain: q ¼ ad þ bn þ tc

(2.2)

As a further step Fagerberg (1988) assumes, “as customary in the diffusion literature, (. . .) that the contribution of the diffusion of internationally available knowledge to economic growth (d) is an increasing function of the distance (T/Tf) between the level of knowledge appropriated in the country (T) and that of the country on the technological frontier (Tf)” (p. 439). Thus substituting the resulting equation for d in (2.2), we obtain the final specification of Fagerberg’s model: q ¼ aðT=Tf Þ þ bn þ tc

(2.3)

Equation (2.3) is the point of departure for the specification of our empirical model which only partially follows Fagerberg’s (and Fagerberg et al. 1997 for an application to European regions) specification. The main point of contact between our model and those of the technology gap literature – and that of Fagerberg et al. (1997), in particular – is the assumption of GDP per capita as a proxy for the potential for imitation (T/Tf). As the total level of knowledge appropriated in the country (T), i.e., the total set of techniques in use, cannot be measured directly, a measure for the output of the process embedding such techniques has to be considered. This measure is the level of productivity (Q/L), usually proxied by

26

2 Theoretical Framework: A Spatial Perspective On Innovation

GDP per capita. Thus we suppose that the lower the productivity (GDP per capita) of a region the farther the technological frontier and, consequently, the higher the potential for imitation. However, as pointed out by Fagergerg in first instance, the process by which a country could fill such gap is not automatic (as assumed, through different mechanisms, by equilibrium-based models), but depends upon national innovative performance (n). This emphasis on innovative efforts is the main point of contact between the “technology-gap” approach a` la Fagerberg and some (later) contributions in the “endogenous growth” literature which have more explicitly assumed the Schumpeterian idea of innovation as an endogenous factor explaining productivity growth in the economy (Romer 1990; Grossman and Helpman 1991; Aghion and Howitt 1992). As a consequence the idea that “the scope for imitation combined with investment or education or both works well to explain differences in growth across countries” (Fagerberg, Verspagen, von T€ unzelmann 1994, p. 10) fairly represents the change in the perspective of the “formal analytical” literature that in different ways attempted at incorporating the Schumpeterian legacy (“technology gap” and “endogenous growth”) from the traditional Cobb-Douglas production function approach (Fig. 2.1). This same idea is incorporated in our empirical model which, as discussed above, aims at overcoming its limitations by including into the analysis the processes singled out by the “systems of innovation approach”, on the one hand, and by the literature on localised knowledge spillovers and the spatial dimension of the process of innovation, on the other. Consequently, the proposed model is substantially different from that of Fagerberg, as we do not consider capital investment as “an indicator of growth in Innovation as technological progress

International and regional cross-sections “Formal” quantitative analysis

“Technology-gap” “Endogenous growth” models

Schumpeterian approach: “new combination” “Qualitative” analysis Regional and local “case studies”

Systems of innovation

Innovation as a collective learning process

Fig. 2.1 Conceptual map of existing literature on innovation and growth Source: Authors’ elaboration

2.8 Explaining Core and Periphery Patterns

27

the capacity for economic exploitation of innovation and diffusion” (Fagerberg 1988, p. 447) as capital investment should be seen as an effect rather than a cause of growth (as later acknowledged in Fagerberg et al. 1997). As an alternative, we introduce human capital accumulation as a better proxy for the diffusive mechanism of innovation. We also propose a simple model which tries to combine the key factors from the three strands of literature presented in the previous sections (Fig. 2.2): endogenous innovative efforts, socially and territorially embedded factors, and spatially-bound knowledge spillovers. The model is aimed at understanding – and, to a certain extent, discriminating among – the role of the different innovation factors proposed by different strands in order to generate economic dynamism. The systems of innovation approach provides fundamental insights into such dynamics, but the lack of “theoretical borders” for this approach6 means that it is not a theory of innovation but a “focusing device” for factors relevant for the process of innovation (Edquist 1997). The concept, similarly to that of the “innovative milieu”, should thus be interpreted not as a full explanatory model but rather as a “meta-model” able to show the relevance of some specific factors for a broader phenomenon7 (Camagni 1995).

Link between investment in R&D, patents, and economic growth. (Fagerberg 1988, 1994 and 1997;Grossman and Helpman 1991;Maurseth and Verspagen 1999)

Geographical diffusion of regional knowledge spillovers; (Anselin et al. 1997, Adams and Jaffe 2002; Audretsch and Feldman 2003, Leamer and Storper 2001, Storper and Venables 2004, Sonn and Storper 2005)

Existence and efficiency of regional innovation systems. (Camagni 1995, Becattini 1987, Morgan 1997 and 2004, Cooke et al. 1997,

Fig. 2.2 Streams of literature combined in the model of empirical analysis Source: Author’s elaboration

“. . . at the present state of the art, defining the limits of a system of innovation in this way (as all determinants) is a “catch 22” problem. (. . .) We will, for the time being, specify systems as including all important determinants of innovation. (. . .) In this way the approaches serve as ‘focusing devices’ (. . .): interesting conjectures that need to be specified and then verified or disproved through further research.”(Edquist 1997, p. 15). 7 Cantwell and Iammarino (2003) make this specificity explicit when stating that “proper regional systems of innovation are found only in a few well-defined areas: in most regions systemic interactions and knowledge flows between relevant actors are simply too sparse and too weak to reveal the presence of systems of innovation at work” (p. 5). 6

28

2 Theoretical Framework: A Spatial Perspective On Innovation

However, these specificities of the innovation systems approach do not prevent us from including the “lesson” of such a model in a broader theoretical framework able to deal with more general empirical analysis (Fig. 2.1). The innovation systems approach has challenged the linear relationship between innovative efforts, productivity and growth, developed in the endogenous growth framework, stimulating and enriching the theoretical perspective on innovation. However the inclusion of this broader set of explanatory factors in the explanation of growth performance, as we underlined above, is not new in the field of quantitative analysis in economic geography and regional development (e.g., Rodrı´guezPose 1994, 1998a and 1998b; Cheshire and Carbonaro 1996, Cheshire and Magrini 2002). Building on such contributions we support the idea that the variety of factors brought into “focus” by the innovation systems approach, while resistant to any generalisation when used as linear predictors for innovative performance, may instead play a (statistically significant) role as “supportive” variable whose explanatory power lies in their interaction with innovative efforts. The same line of reasoning applies to the literature on the spatial diffusion of knowledge flows. In many cases this literature has remained either theoretical (Storper and Venables 2004), disconnected from regional growth analysis (Sonn and Storper 2005) and/or limited to case studies (literature on local “buzz”). Even where in-depth analyses have been implemented in order to capture the empirical relevance of spatially-mediated knowledge flows (Adams and Jaffe 2002), their interaction with the underlying absorptive conditions has remained unexplored. The lack of a common theoretical perspective and the different languages of separated streams of literature have hampered the possibility to look at spatial knowledge flows jointly with the factors conditioning their impact on the regional economy. Our empirical framework, by building upon the commonalities of these different streams of literature and incorporating them into a formal innovation-led regional growth model, will allow us to directly capture the role of innovation systems as potential catalyser for the absorption of localised knowledge flows and their translation into economic activity. The operational translation of all these concepts and further theoretical justification for their introduction into our growth model will be discussed in greater detail in the chapters specifically devoted to the analysis of the insights offered by each of these factors for the analysis of regional growth and innovation dynamics. In general terms (2.3) is extended by combining, as discussed above, proxies for the regional system of innovation [c in (2.4)] and for the spatial dimension of the process of innovation [the variable (g) in (2.4)]: q ¼ aðT=Tf Þ þ bn þ tc þ gg

(2.4)

The basic model to be estimated in following chapters is:   1 Yi;t ln ¼ a þ b1 lnðyi;tJ Þ þ b2 innovationit þ b3 innovation systemsit J Yi;tJ þ b4 geographyit þ b5 D þ e (2.5)

2.8 Explaining Core and Periphery Patterns

29

where: 1 J

 ln

Yi;t Yi;tJ



a lnðyi;tJ Þ Innovation Innovation system Geography

D e

Is the usual logarithmic transformation of the ratio of regional per capita GDP at the two extremes of the period of analysis (tJ,t); Is a constant; Is the log of the GDP per capita at the beginning of the period of analysis (tJ) Is a measure for innovative efforts; Is a proxy for the local system of innovation including the level of human capital accumulation; Is a proxy for the effect of space and geographical distance (e.g., aggregate measure of accessibility, innovative efforts and socio-economic conditions in neighbouring regions); Is a set of national dummies; Is the error term.

The various specifications of this model presented in the following chapters combine inputs in the innovation process (R&D expenditure) with the socioeconomic local factors that make the presence of favourable regional systems of innovation more likely, while controlling for the wealth of European regions. These factors are considered locally, i.e., the R&D and the local conditions in the region under study, and externally, i.e., the conditions in neighbouring regions in order to assess the importance of the geography of these factors in the form of spillovers. Finally we control for the influence of national factors, such as the presence of national systems of innovations. However, the need for a feasible specification of the innovative process, which inevitably implies some simplistic assumptions, must not hide the complexity of the real world as represented by the systems approach. We should thus resist the temptation of a “linear interpretation”8 of the results of our empirical analysis, but rather take full account of the complexity of the underlying mechanisms behind the results. The models of empirical investigation employed in the following chapters of this book are based on different specifications of this general base model. In particular, in Chap. 3 the role of peripherality in explaining differential capabilities to translate innovation into economic growth will be assessed by means of an aggregate index of accessibility. In Chaps. 4 and 5 the analysis will be broadened by including a more in-depth quantitative analysis of regional systems of innovation and of the spatial extent of knowledge flows respectively. In Chap. 6 this approach will be further expanded by accounting for specialisation and agglomeration patterns and applied to the comparative analysis of the innovation dynamics in Europe and the United States and, finally, in Chap. 7 to the analysis of the impact of the transport infrastructure policy of the EU on regional growth performance.

8 “(. . .) the mere introduction of sets of social and political variables into linear models of growth does not by itself solve the problem of what type of relationship there is between the socio-political setting and economic growth” (Rodrı´guez-Pose 1998, p. 46).

.

Chapter 3

Geographical Accessibility and Human Capital Accumulation

3.1

Introduction

This chapter performs a preliminary empirical exploration of the key conceptual links developed in Chap. 2. It looks at the relationship between innovation and regional growth by means of an empirical model that explicitly accounts for the impact of spatially mediated processes (geography) and place-specific socioinstitutional conditions. The cross-sectional analysis, covering the EU-25 regions, sheds light on how regional innovative activities influence differential regional growth patterns. The thrust of the analysis will be to examine how geographical accessibility and human capital accumulation, by shaping the regional system of innovation, interact with local innovative activities, thereby enhancing (or impeding) economic growth. The analysis will support the idea that an increase in innovative effort is not, per se, necessarily or likely to produce the same effect in all EU-25 regions (as the linear model would predict), whereas geography, together with human capital accumulation, does, on the contrary, influence this relationship and shall be shown to do so. This will constitute the first empirical confirmation for the theoretical hypotheses developed in the previous chapter, and which will be further analysed in Chaps. 4 and 5 with a more in depth treatment of respectively systems of innovation conditions and geography. This chapter is organised into four sections. In the first section the theoretical relationships are analysed in order to explain how accessibility and human capital accumulation condition the translation of innovative efforts into growth. In the second section the model is outlined and the key proxies for the various conceptual dimensions are presented, in section three the empirical results are discussed. In the final section some preliminary conclusions are drawn.

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_3, # Springer-Verlag Berlin Heidelberg 2011

31

32

3.2

3.2.1

3 Geographical Accessibility and Human Capital Accumulation

The Key Relationships: Accessibility and Human Capital Accumulation How Accessibility Influences the Process of Innovation

A number of empirical studies has focused on the dynamics and attempted to measure the extent of “knowledge flows” (see, Jaffe 1989, Anselin et al. 1997, Fischer and Varga 2003), providing support for many important contributions in the field of economic geography – from the breakthrough studies by Storper (1995) to the literature on industrial districts and innovative milieus as in Camagni (1995a) – based on such concept. In this framework “localised technological spillovers” flow mainly through “local” interactions such as face-to-face relationships and local networks, often relying on a common technical language and tacit knowledge. However “knowledge spillovers do not (. . .) transmit costlessly with respect to geographical distance” (Audretsch and Feldman 1996b, p. 256). Consequently, in our model, we introduce the idea that the concept of “accessibility”, defined as “relative opportunity of interaction and contact between households, firms or industries across geographical space afforded by location in a particular town or region” (Keeble et al. 1988, p. 2), may be relevant to measure the probability not only of local interactions but also of those taking place on a wider spatial scale (e.g., interregional or even international in the EU context). Obviously when such a measure is introduced for knowledge flows it becomes more difficult to sort out the various “transmission mechanisms” of such flows (Breschi and Lissoni 2001). On a wider scale the incidence of free and spontaneous flows probably becomes less relevant while more “market mediated” flows become predominant. Having acknowledged this – thus also leaving to qualitative research the aim of distinguishing between these “mechanisms” – the concept of accessibility emerges as a good proxy for the effect of distance on innovation, as it accounts not only for localised spillovers represented by the “density” of local (within the region) interactions but also for the cost/opportunity barrier of distant (non local, between regions) interactions. This effect can be explained by a variety of mechanisms grounded in different approaches to the process of innovation which attribute different roles to knowledge flows: 1. From a neo-classical perspective, it is intrinsically difficult (as discussed in Chap. 2) to assess the role of distance between innovative agents in the process of generating and spreading innovation. However, even in the restrictive framework of the linear approach it is possible to single out a (somehow indirect) impact of accessibility on innovative performance by taking into account the reduction of risks associated with the introduction of new products as a consequence of improvements in market potential. In other words, accessibility would affect innovation through risk-minimization, as the possibility for the innovative

3.2 The Key Relationships: Accessibility and Human Capital Accumulation

33

firm to reach a greater number of (distant) potential buyers increases the chance of success for innovative products; 2. From an endogenous growth perspective, we can assume that better accessibility may increase the probability for firms and individuals to come into touch with different people and thus with new ideas. Romer (1990) links this probability to population growth; 3. Knowledge-intensive regional networks have been considered a crucial issue for regional development. However these networks may also be detrimental for the innovative capacity of the firms involved, due to a possible “lock-in effect” as “networks that are geographically closed may, in the long run, hamper rather than stimulate innovation” (Lundvall 2001, p. 281). In this sense there is the need for “outward networking” which is afforded by accessibility to other innovative firms outside the network and the immediate proximity; 4. From an evolutionary economics perspective, better accessibility simply allows firms to access a wider diversity of knowledge sources which are necessary for (evolutionary) successful innovation (i.e., learning by interacting) (Cooke 1997). When linking together these different mechanisms, we may assume that the effect of accessibility on innovation is related to what may be called “exposition to different types of knowledge”, the possibility/probability/frequency to be in touch with ideas, organisations, products different from those locally available. Better “access to diversity” may either directly affect growth by providing an additional source of knowledge or produce an indirect effect, when interacting with indigenous innovative efforts making them more growth enhancing. Our empirical model will allow us to test both these effects. When accessibility is introduced into the analysis of economic growth, it could be conducive to a certain degree of endogeneity, as better accessibility may be viewed both as a cause and an effect for improved economic performance. However, we believe this problem does not affect the robustness of our analysis, once clear assumptions on the mechanics of this relationship have been made explicit. At first, we have to consider that accessibility inevitably depends on physical spatial factors which influence, but are not influenced, by the economy. Some regions thus benefit from better accessibility simply because of a location advantage. Obviously “second nature geography” matters too, compensating for or reinforcing natural advantage. Better economic performance by a region may influence this second “source of accessibility” due to improvements in infrastructure and incentives for contacts. However we have to consider that: (a) There is a consistent time-lag between the emergence of the need for better accessibility (due to improved economic performance) and its effective achievement (e.g., in the case of infrastructure which requires significant lead times in terms of decision taking, planning and implementation), significantly reducing room for endogeneity (b) Dramatic changes in accessibility, at least in the European experience, seem to depend more on political factors – such as the reduction in the obstacles in cross-border contacts produced by the EU enlargement – or on economic policy

34

3 Geographical Accessibility and Human Capital Accumulation

objectives – such as that of the Trans-European Transport Network projects – than on observed growth rates (c) Our empirical findings do not support the hypothesis of a linear relationship between accessibility and growth performance (in line with the existing literature) whatever the direction of the causal relationship Once accessibility is assumed to be an exogenous factor, empirical results have to be interpreted coherently, preventing us from considering the role of accessibility in a causal way and, consequently, as a feasible policy-tool. Our results reject the hypothesis of accessibility as a direct predictor of regional economic performance, and prefer to shed light on its indirect contribution when interacting with innovative activities.

3.2.2

Education: A “Preliminary” Proxy for Innovation “Prone” and Innovation “Averse” Regions

In line with the theoretical perspective developed by Rodrı´guez-Pose (1999) for the regional process of innovation, the differentiated capability to translate indigenous innovative activity into innovation and economic growth depends on there being different “social filters”: the interaction of a complex set of economic, social, political and institutional features that makes some regions “prone” and others “averse” to innovation. The empirical definition of the features that make a region “prone” to innovation, is complex as a consequence of the intricacies of the dynamics involved and the scarcity of data. Consequently, the next chapter will be entirely focused upon the identification and theoretical justification of a set of regional structural conditions able to systematically enhance the local capability to produce and absorb innovation and translate it into economic growth. At this intermediate stage of the analysis, where we gradually depart from the endogenous growth perspective on the process of innovation in order to include the role of space in the form of accessibility into the analysis, analytical attention is focused on a specific feature of the local socio-economic environment: the level of human capital accumulation. Many “observable” (e.g., youth migration) and “unobservable” (local quality of educative institutions, social “reward” for education, regional policy, etc.) human capital factors may indirectly affect innovation insofar as they can condition this variable. The introduction of the level of education of the population as a first proxy for the capability of a region to exploit and spread innovation is particularly appropriate for a variety of reasons. When compared to capital investment, the risk of reverse causality associated with this variable seems to be much lower, as the creation of skills in the population is a long-term process and thus is less likely to be influenced by the current growth rate. In addition, our approach transfers the legacy of the so-called “human capital school” (Lucas 1988; Mankiw et al. 1990), which highlights the capacity of human capital to “embody” formal (and tacit) knowledge into a growth model with endogenous regional innovative activities.

3.3 The Empirical Model

35

The mechanism whereby education interacts with innovation is, in our view, convincingly made explicit by the innovation systems approach, which emphasizes its effects on the “learning capability” of local society. Furthermore, as pointed out by Lundvall (1992) the level of education of the population matters for the creation of innovative forms of organisation (i.e., learning organizations). Consequently, our model, allows us to test its capacity to operate as a statistically significant explanation for differentiated social filters and, consequently, for non-homogeneous innovative performance.

3.3

The Empirical Model

3.3.1

Model Specification

Within the theoretical framework developed in Chap. 2, the full estimated model here is:   1 Yi;t ln ¼ a þ b1 lnðyi;t
J Þ þ b2 innovation þ b3 education Yi;t
J J þ b4 accessibility þ e

(3.1)

where: 1 J

 ln

Yi;t Yi;t
J



a lnðyi;t
J Þ innovation education accessibility e

Is the usual logarithmic transformation of the ratio of regional per capita GDP at the two extremes of the period of analysis (t
J, t); Is a constant; Is the log of the GDP per capita at the beginning of the period of analysis (t
J); Is the index of innovation as defined below; Is a proxy for human capital accumulation; Is the index of accessibility; Is the error term.

Departing from this first specification of the empirical model where the direct effect of each variable is assessed, we introduce an interaction term between innovation efforts (innovation) and both accessibility and education in order to explicitly model their interaction, rather than their direct effect, with the generation of knowledge in determining growth performance. Thus: (1) When the interaction term between innovation and accessibility is introduced:   1 Yi;t ln ¼ a þ b1 lnðyi;t
J Þ þ b2 innovation þ b3 education þ b4 accessibility J Yi;t
J þ b5 innov  accessibility þ e

36

3 Geographical Accessibility and Human Capital Accumulation

from which, where considering the partial effect of innovation efforts on growth (holding all other variables fixed) we obtain:    Yi;t D 1J ln Yi;t
J Dinnovation

¼ b2 þ b5 accessibility

(3.2)

Thus implying, with b5 > 0, that, ceteris paribus, for more accessible regions there is a greater probability that additional innovative efforts may translate into economic growth. (2) When the interaction term between innovation and education is introduced:   1 Yi;t ln ¼ a þ b1 lnðyi;t
J Þ þ b2 innovation þ b3 education þ b4 accessibility Yi;t
J J þ b5 innov  education þ e where the partial effect is: D



1 J

 ln

Yi;t Yi;t
J



Dinnovation

¼ b2 þ b5 education

(3.3)

Which implies, with b5 > 0, that, other things being equal, innovation may efforts produce an increase in growth proportional to the level of human capital accumulation. In the next sub section we analyse in further detail the mechanisms through which each variable is supposed to influence economic growth.

3.3.2

The Variables

3.3.2.1

Innovation: An Index for EU Regions

Many attempts have been made in order to develop indexes able to capture the various components of the process of innovation. One of the most important contributions in this direction is the “ArCo Indicator of Technological Capabilities” (Archibugi and Coco 2004) which is based on three “main dimensions” of technological capability: the creation of technology, technological infrastructures and the development of human skills. As we specifically look for a measure of “innovative effort” and the other dimensions of the process of innovation are represented individually in our model, we base our index of innovation on the sub-index elaborated by Archibugi and Coco for “technology creation”. This index is based on the “number of patents” and the “number of scientific articles” but “data on R&D would have nicely complemented the measure of technological creation”

3.3 The Empirical Model

37

(Archibugi and Coco 2004, p. 13). Given that we face different kinds of data constraints when dealing with regions, compared to a cross-country analysis, we can include both R&D expenditure and personnel in our index, but we have to exclude the number of journal articles as this information is not available at the regional level and probably would lose most of its force if it were detached from the national context. Furthermore this structure of the index is conceptually coherent with the “knowledge creation” component of the European Innovation Scoreboard, a similar exercise promoted by the European Commission (2001). Our innovation index thus includes: “R&D expenditure as% of GDP”, “R&D personnel as% of total labour force”, “number of high-tech patents per million in the labour force”. Standardised1 indicators are obtained for each variable on the basis of the following formula: Observed value
Minimum observed value Maximum observed value
Minimum observed value The unweighted average of the indicators will produce the index whose value will range between 0 and 1. In the economic literature on the “measurement” of innovation (see e.g., Archibugi and Coco 2004), as in similar exercises developed by the European Commission (e.g., the European Regional Innovation Scoreboard), the adoption of equal weights for the different components of the indices is a rule of thumb, as there are no theoretically grounded a priori alternatives. However, it is crucial to bear in mind that the various sub-indexes measure different processes and for this reason we provide separate estimations for each component, as also for the overall index. Our index, in addition to giving a succinct measure of the capability of “producing” technology for ranking purposes, is also able to minimize the bias associated with the differential “innovative productivity” linked to the different sectoral composition of each region. The index by including both financial input (R&D expenditure), human resources input (R&D personnel) and output (patents) avoids the distortion introduced for example by “capital intensive” kinds of R&D activities which may artificially increase R&D expenditure, biasing a measure of creation of innovation based on this variable alone or, on the contrary, the effect of the differential propensity to patent of the different sectors.2 Overall the index, when included in the regression analysis, has showed better performance, both in terms of goodness-of-fit and significance, than the individual variables used to compute it.

1

The standardisation procedure used is the common procedure whereby we subtract from the observed value of each variable the minimum value taken over the whole set of regions under analysis and divide the result by the range of variation (the maximum minus the minimum value taken) of the same variable. The final result is an indicator which ranges from 0 to 1 for each region. 2 We consider high-tech patents only in order to further minimize this bias introduced by differential propensity to patent of different sectors.

38

3 Geographical Accessibility and Human Capital Accumulation

3.3.2.2

Measuring Accessibility: The Peripherality Indicator

Accessibility is introduced in our empirical model by means of an “accessibility indicator”. In particular we use one of the indicators developed by the IRPUD (2000) for the DG Regional Policy of the European Commission. Such indicator is based on the traditional “potential” structure where “activities” or “opportunities” to be reached are “weighted” by “the effort, time, distance or cost needed to reach them” (Sch€urmann and Talaat 2000, p. 2). The indicator we have selected is a “standardised accessibility indicator” (called Peripherality Indicator 2), it expresses regional accessibility as a percentage of average accessibility of all regions and the higher the value of the index the higher the relative “accessibility” of the region. In particular, we adopted the indicator based on accessibility of population by car. This measure (rather than “market potential” accessibility based on GDP and trucks as alternative mode of transport) seems the most appropriate as a measure of “opportunities” required by our model as it emphasizes the probability of contacts between people (see par. 3.2.1).

3.4 3.4.1

Empirical Results Estimation Issues

In this section we test the model outlined above by means of HeteroskedasticityConsistent OLS (Ordinary Least Square). The estimation of the parameters, however, suffers from some econometric problems whose effects we attempted to minimize. The most serious problem is spatial autocorrelation: the lack of independency among the error terms of neighbouring observations (Armstrong 1995; Quah 1996). In order to minimize this problem and its distorting effect on the estimation of the parameters, we followed a procedure similar to that adopted in other analyses of regional processes (e.g., Rodrı´guez-Pose 1999; Rodrı´guez-Pose and Fratesi 2004), and “nationally weighted” the dependent and independent variables by measuring them relative to the respective country mean. The Moran’s I test on the regression residuals (Cliff and Ord 1972) has been computed for each specification of the empirical model in order to detect the possible presence of spatial autocorrelation. The test statistic confirmed the absence of spatial autocorrelation. Another major problem concerns the problem of endogeneity. We dealt with this by including (where possible3) in the model the value of the explanatory variables as a mean over the period (t
J
5) – (t
J), while the average growth rates have 3 In the case of the New Member States data availability has prevented us from calculating the mean of the explanatory variables over the five year period (t
T
5) forcing us to use a shorter time span. For some EU 15 countries slightly differential time spans have been used according to data availability for each variable. The specification of each individual case would take up too much space.

3.4 Empirical Results

39

been calculated over the period from t
J to t. In addition, in order to resolve the problem of different units of account, explanatory variables are expressed, for each region, as a percentage of the respective GDP or population. In order to answer our research questions using the available data we followed a two-step procedure. First, we tested a “reduced” version of the model by regressing the growth rates onto innovative efforts and (log) initial GDP per capita only. This reduction in the number of variables allowed us to consider a time span (1990–2003) able to minimize the effects of the economic cycle on the variables and capture the structural component of the relationships under analysis while keeping a reasonable number of regions in the sample (the 1990–2003 time span allowed us to include into the analysis regions from eight EU countries: Austria, Belgium, France, Germany, Greece, Italy, Spain and Portugal). Secondly, where the full empirical model was estimated, we reduced the time span considered (1995–2003) and, consequently, obtained the advantage of being able to include all the EU members of the “enlarged” Europe for which regional data are available (see Appendix A for data availability and description of the variables).

3.4.2

Regional Innovation Activities and Economic Performance in the EU-25

The enlarged European Union exhibits significant differences in terms of innovative effort. Where the innovation index (together with its individual components) is calculated for the 25 members of the Union, it is possible to appreciate such disparities (Table 3.1). The rankings – similar to that of Archibugi and Coco (2004), but based on a larger set of indicators – allow to distinguish between leading (innovation index between 0.92 and 0.40), intermediate (0.39–0.20) and latecoming (less than 0.20) countries. The first group includes almost all the countries in the “core” of EU-15, while comparatively less innovative existing members, plus Slovenia4 as a new member of the Union, belong to the second group. The rest of the new members of the EU fall into the third group: the latecomers. However, these data do not provide an exhaustive picture of the geography of innovation of the EU. When regions are considered and the innovation index is calculated on nationally weighted regional data, significant disparities are still observable. In Fig. 3.1 the distribution of the innovation index is plotted, clearly showing a differentiated pattern in regional innovative effort: some regions “produce” more innovation than others. Furthermore, the relationship between such differentiated innovative efforts and regional growth pattern is apparent and this is probably one of the reasons that have led policy-makers to underestimate the importance of the other factors that interact 4

Slovenia ranks 26th out of the 162 countries considered in Archibugi and Coco 2004 when applying their index.

Table 3.1 EU-25 Countries ranked by Innovation Index 2001 Index for high High tech patents (per tech patents million in the labour force) FI Finland 320.33 1 SE Sweden 211.07 0.657985 DK Denmark 86.06 0.266669 DE Germany 107.15 0.332686 NL Netherlands 142.27 0.442622 FR France 72.41 0.22394 BE Belgium 59.32 0.182965 AT Austria 43.02 0.131941 UK United 75.31 0.233018 Kingdom SI Slovenia 15.79 0.046704 IE Ireland 75.56 0.233801 IT Italy 17.7 0.052683 HU Hungary 13.47 0.039442 GR Greece 5.23 0.013648 ES Spain 9.23 0.026169 CZ Czech 1.53 0.002066 Republic EE Estonia 4.94 0.01274 LT Lithuania 0.96 0.000282 SK Slovak 2.44 0.004915 Republic PT Portugal 1.31 0.001377 PL Poland 0.87 0 LV Latvia 1.13 0.000814 a 1998 data b Higher Education sector not included Source: own calculations on Eurostat data 0.785 1 0.5325 0.56 0.405 0.49 0.475 0.4075 0.405 0.325 0.225 0.21 0.17 0.0925 0.17 0.2375 0.1275 0.105 0.0925 0.145 0.0925 0.0425

3.41 4.27 2.4 2.51 1.89 2.23 2.17 1.9 1.89 1.57 1.17 1.11 0.95 0.64 0.95 1.22 0.78 0.69 0.64 0.85 0.64 0.44

2001 R&D expenditure Index for R&D (% GDP) expenditure

0.71 0.74 0.55

1 0.89 0.89

1.28 0.95 0.92 1.03 1.28 1.04 0.91

R&D personnel (% total labour force) 2.53 2.45 1.89 1.59 1.54 1.51a 1.52 1.38a 0.74b

0.272 0.284 0.208

0.388 0.344 0.344

0.5 0.368 0.356 0.4 0.5 0.404 0.352

1 0.968 0.744 0.624 0.604 0.592 0.596 0.54 0.284

1999 Index for R&D personnel

0.139459 0.1255 0.083771

0.17608 0.149761 0.147138

0.290568 0.2756 0.206228 0.203147 0.202049 0.200056 0.197189

0.928333 0.875328 0.51439 0.505562 0.483874 0.435313 0.417988 0.359814 0.307339

Innovation Index

40 3 Geographical Accessibility and Human Capital Accumulation

3.4 Empirical Results

41

25

20

15

Mean Median Maximum Minimum Std. Dev. Skewness Kurtosis

0.192647 0.156695 0.733604 0.007349 0.131206 1.458452 5.360039

Jarque-Bera Probability

118.4908 0.000000

10

5

0 0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Fig. 3.1 The distribution of the innovation index for the EU25 regions Source: Own elaboration on Eurostat data

in this process. The scatterplots of the logarithm of the nationally weighted average annual growth rate (1995–2003) of the EU-25 regions versus the individual components of the Innovation Index (Fig. 3.2) show similar patterns for this relationship. This seems to support the idea of a similar positive relationship between different measures of innovative activities and regional growth performance. Additionally, in order to understand how similar the patterns described in the scatterplots are (i.e., whether the same regions occupy the same position in each scatterplot), we calculated the Spearman’s Rho for rank correlation (Kendall 1990) between the individual components of the innovation index. As expected (see par. 3.2.1), the sub-components of the innovation index are reciprocally correlated as they measure different aspects of the same process. However, this correlation is neither complete, nor of the same intensity for each component. While the sub-indexes for R&D Expenditure (In_Exp) and R&D personnel (In_per) – both measures for R&D inputs – show a high degree of rank correlation, their correlation with the output measure (High Tech Patents) is less accentuated (though still statistically significant). Thus, while the same regions in the scatterplots of growth rate against both R&D Expenditure and personnel tend to occupy the same positions, this is true to a lesser extent when growth rate is scattered against the sub-index for High Tech Patents. This provides, in our view, a first empirical confirmation of the need for a multifaceted measure (the Innovation Index) as we showed in our theoretical argumentation and will be confirmed by the increase in the goodness-of-fit provided by the Innovation Index in the regressions below when compared to its individual components. Figure 3.3 plots growth rates against the Innovation Index confirming the positive relationship between the two phenomena.

42

3 Geographical Accessibility and Human Capital Accumulation Nationally Weighted Growth Rate vs.R&D Personnel Nationally Weighted Growth Rate (1995-2003)

Nationally Weighted growth rate (1995-2003)

Nationally weighted Growth Rate vs. R&D expenditure 3

2

1

0

–1 0.0

0.2

0.4

0.6

0.8

1.0

Innovation Index - R&D expenditure

3

2

1

0

–1 0.0

0.2

0.4

0.6

0.8

1.0

Innovation Index - R&D Personnel

Nationally Weighted Growth Rate (1995-2003)

Nationally Weighted Growth Rate vs. High T ech Patents 2.0 1.5 1.0 0.5 0.0

– 0.5 0.0

0.2

0.4

0.6

0.8

1.0

Innovation Index - High T ech Patents

Fig. 3.2 Scatter diagram of nationally weighted growth rate (1995–2003) versus the individual components of the innovation index: R&D expenditure, R&D personnel, High-Tech patents Source: own elaboration on Eurostat data Cambridge Econometrics data

However, in Fig. 3.3, it is equally apparent that many regions are above or below the regression line, showing that the same level of innovative efforts may be translated into very different economic performance. This is visible in Table 3.2, where the nationally weighted average annual growth rate of each region has been categorized by the value of the innovation index. Although, the higher the extremes of the innovation index category, the higher the mean growth rate, the standard deviation within each group is also high, indicating high variability in terms of average growth rates within similarly innovative regions (in line with previous research e.g., Fagerberg et al. 1997). The estimation of the “reduced” and the “full” versions of our model, by Heteroskedasticity-Consistent OLS, provides with a better understanding of the relationship between innovative efforts and regional growth patterns as also of the differentiated economic performance associated with similar innovative effort. In Table 3.3 we present the estimation of our “reduced” model, which allows taking into account the period 1990–2003 thus minimizing the effect of cyclical components but forcing us to reduce the number of variables and regions considered. The logarithm for the initial level of GDP per capita is significant and shows a negative sign, suggesting that regions whose productive structures are farther

3.4 Empirical Results

Table 3.2 Nationallyweighted average rate of growth (1995–2003) categorized by the value of innovation index

Nationally Weighted Growth Rate vs.Innovation Index

Nationally Weighted Growth Rate (1995-2003)

Fig. 3.3 Scatter diagram of nationally weighted average growth rate (1995–2003) versus the innovation index Source: own elaboration on Eurostat and Cambridge Econometrics data

43

3

2

1

0

–1 0.0

Innovation Index

0.2 0.4 0.6 Innovation Index - Global Indicator

Mean growth rate

[0, 0.2) 0.900344 [0.2, 0.4) 1.020586 [0.4, 0.6) 1.186509 [0.6, 0.8) 1.285959 All 0.962215 Source: own calculations on Eurostat data

0.8

Standard deviation of growth rate 0.376735 0.302381 0.443582 0.270402 0.368868

Table 3.3 Heteroskedasticity-Consistent OLS estimation of the “reduced” model in (3.1): the relationship between innovation and regional growth performance (1990–2003) 1 2 3 4 Dependent variable: Nationally weighted regional growth rate Constant logðy90 Þ In_EXP In_PER In_HPAT In_INNOV

0.929683*** (17.99810)
0.510784*** (
3.066982) 0.280987* (1.762797)

0.882594*** (12.93278)
0.550720*** (
2.752301)

0.971988*** (15.36617)
0.331009 (
1.433080)

0.925032*** (17.92436)
0.533848*** (
3.116949)

0.333822*** (2.079908) 0.149163 (0.709207)

0.266353* (1.836826) Adj. R-squared 0.10 0.09 0.013 0.1 F-statistic 10.13*** 6.40*** 1.30 6.52*** Moran’s I
0.014
0.018
0.018
0.019 t-Statistics in parentheses – *10% significance; **5% significance; ***1% significance – n ¼ 101 Source: Own calculations on Eurostat and CamEcon data

44

3 Geographical Accessibility and Human Capital Accumulation

removed from the technological frontier can exploit this potential to grow faster than others. However the scope for imitation is only a part of the story, as the local innovative effort also plays a significant role in explaining differential regional economic performance. Two out of three measures of innovative effort (represented by the individual components of the innovation index – columns 1–3 in the table) show positive and significant coefficients: R&D expenditure and personnel, but not high tech patents. However, when a broader measure of innovative effort is introduced (column 4), by mean of the innovation index, the regression analysis confirms the hypothesis that regional economic growth is positively associated with indigenous innovative effort. The regression coefficient of the innovation index is positive and significant, the F-statistic confirms the robustness of the model (as in the other cases) and the Adj. R-squared confirms a sufficient strength of the relationship between the variables. However, when considering this last point we have to remember that the analysis of economic performance at the regional level implies a different magnitude of the phenomena in question when compared to international cross-section analyses. The role of innovation in explaining differential economic performance between countries can explain a larger share of total variance (R-squared generally higher but significantly smaller samples, as e.g., in Fageberg 1988), making it easier to include innovative efforts into the “primary drivers”. When this approach is translated into a regional scale a careful reconsideration of the whole set of process is necessary, as we underlined in Chap. 2. Thus, in this context, our purpose is not to explain regional disparities per se, as we are aware of the complexity of factors which the economic geography literature has shown to be at work, but rather to show that the impact of innovative efforts (on which the Lisbon and Europe 2020 strategies are based) is influenced by a variety of features of the local socio-economic environment thus contributing differentially to regional growth performance. The value of the R-squared has to be assessed in such a perspective, in particular when dealing with a very large cross section of the EU regions analysed through a relatively small set of explanatory variables, as in our case. Additionally we tested for the presence of multicollinearity and the V.I.F (Variance Inflation Factor) test allowed us to exclude this possibility. The next step was to test if these results are confirmed after the fully-specified model is estimated. Relying on 1995–2003 data for almost all the EU-25 regions, Table 3.4 shows the results of the estimation of the full model. Once again, the log of the initial average income per capita is highly significant and negatively correlated with the average growth rate, confirming the relevance of the potential for imitation in favouring less advanced regions. To the degree that the (more backward) regions of New Members of the EU are now included in the analysis, the confirmation of this first result is an important indication of their growth potentialities. Also the innovation index and its various components show the expected (positive) sign and are all significant, confirming the role of local effort for the creation and exploitation of innovation. Innovative activities are able to foster economic performance even in the framework of the “enlarged” Europe, but (as the full specification of the model allowed us to investigate) the structure of the

1.342179*** (4.912562) 0.060314 (0.552789)

1.206573*** (4.416165)


0.053361 (
0.677751) 0.12 7.53***
0.020

5 0.726294*** (5.219857)
0.487662*** (
2.736782)

4 0.753032*** (13.74617)
0.44008*** (
2.899668)

Adj. R-squared 0.042 0.11 0.07 0.11 F-statistic 5.15*** 13.22*** 6.71*** 13.37*** Moran’s I
0.021
0.020
0.020
0.022 t-Statistics in parentheses – n ¼ 181 – **5% significance; ***1% significance Source: Own calculations on Eurostat and CamEcon data

Accessibility

Educ_POP

Educ_LF

Educ_LLL

Table 3.4 H–C OLS estimation of (3.1): the “full”model (1995–2003) 1 2 3 Constant 0.885287*** 0.740202*** 0.887561*** (20.09472) (12.74658) (24.64971) logðy95 Þ
0.161545
0.484752***
0.434886*** (
1.128836) (
0.095730) (
2.807201) In_EXP 0.668243*** (2.465387) In_PER 0.928619*** (4.400992) In_HPAT 0.612835*** (3.389211) In_INNOV


0.090153 (
1.185439) 0.15 9.87***
0.019

0.379349*** (2.717120)

1.011485*** (3.445833)

6 0.516331*** (3.572495)
0.489344*** (
2.955945)

0.380516*** (3.088652)
0.153029** (
1.975598) 0.16 10.01***
0.018

0.875639*** (2.988347)

7 0.606814*** (5.219058)
0.469018*** (
3.262094)

3.4 Empirical Results 45

46

3 Geographical Accessibility and Human Capital Accumulation

regional system of innovation exerts a crucial differential impact in making such effort more (or less) conducive to growth (this point will be discussed in further detail in the subsequent chapter). The percentage of the labour force with tertiary educational qualifications, our proxy for human capital accumulation, shows a positive and significant correlation with the average growth rate. This result provides support for the role of regional “learning capabilities” in determining economic performance through their contribution to the productivity of innovative effort for the generation of innovation and its diffusion (as discussed in the previous section). In contrast, the index of accessibility, at this stage, is not statistically significant casting doubt on the existence of “direct” effect of peripherality on the process of translation of innovation into growth. This relationship will be further discussed in the next section. The probability of the F-statistics lets us reject the null hypothesis that all of the regression coefficients are zero. The r-squared confirms the overall goodness-of-fit of the model (with same caveats discussed for the estimation results of the “reduced” model). V.I.F tests have been conducted for this specification of the model, excluding the presence of multicollinearity. We also test alternative measures of the level of human capital by substituting the “% of the labour force with tertiary education” with the “% of population with tertiary education” and “% of adult in Life Long Learning”. While the former was significant, the latter was not. The significant coefficient for the educational achievements of the population has, in our view, to be interpreted as an indication of the importance for the process of innovation of the whole educational environment, making the pool of human resources of the region highly relevant. The lack of significance, instead, of the second variable might be related to the fact that adult education programmes (the so called Life Long Learning Programmes) are often pursued as a social policy measure to tackle long-term unemployment rather than as a real instrument for updating skills and providing knowledge relevant for the innovation process.

3.4.3

The Translation of Innovation into Growth: The Interaction of Innovative Effort with Accessibility and Education

In Sect. 3.3.1 we reported the steps which lead to the specification of (3.2) and (3.3) through the introduction of an interaction term in the model under (3.1). This change in the specification of the model calls for particular care in interpreting the parameters and in testing their significance. The purpose of the interaction coefficient is to identify when the partial effect of the dependent variable (average growth rate) with respect to an explanatory variable (innovative effort) depends on the magnitude of another explanatory variable (accessibility or level of education). As we can see from (3.2) and (3.3) the relevant coefficient is b5 which, when positive, implies that an increase in innovative efforts yields a higher increase in the average growth rate in regions that benefit from better accessibility (3.2) or more educated

3.4 Empirical Results

47

labour (3.3). Table 3.5 shows the results for the estimation of (3.2) and (3.3). As we can immediately see the parameter for the interaction coefficient is positive in both cases. The F-statistic lets us, once again, reject the null hypothesis that all of the regression coefficients are zero and the R-squared confirms the goodness-of-fit of the model. Instead, additional care is called for in the interpretation of the tests on the significance of the coefficients, as we do not have to test separately the significance of b2 and b5 but rather the joint hypothesis Ho:b2 ¼ 0, b5 ¼ 0 (Woolgridge 2003). To test this joint hypothesis we implemented the Wald Coefficient Test whose results, shown in Tables 3.6a, b, let us reject Ho at the 1% level, thus allowing us to conclude that the estimated coefficients are significant in both cases. Additionally, the Wald Test is not affected by the intrinsically lower Table 3.5 H–C OLS estimation of (3.2) and (3.3): the interaction terms Equation (3.2) Constant a 0.736677* (2.990571) logðy95 Þ b1
0.560223* (
3.210771) In_INNOV b2 0.053030** Educ_LF b3 0.209716 (1.079190)
0.190810 Accessibility b4 (
1.350885) In_INNOV* Accessibility b5 1.056172** In_INNOV* Educ_LF Adj. R-squared 0.118793 F-statistic 6.122663* Moran’s I 0.023 t-Statistics in parentheses *1% significance **1% significance in the Wald Test reported below Source: own calculations on Eurostat and CamEcon data

Equation (3.3) 0.622034* (2.554417)
0.548873* (
3.041183) 0.821887** 0.195521 (0.700528)
0.048429 (
0.602703) 0.246766** 0.111211 5.754828* 0.019

Table 3.6a Wald test for the significance of the coefficients b2 and b5 in (3.2): interaction term between innovation (In_innov) and accessibility (Accessibility) Wald test: Null Hypothesis: b(2) ¼ 0 b(5) ¼ 0 F-statistic 6.883089 Probability 0.001308 Chi-square 13.76618 Probability 0.001025 Table 3.6b Wald test for the significance of the coefficients b2 and b5 in (3.3): interaction term between innovation (In_innov) and human capital accumulation (Educ_LF) Wald Test: Null Hypothesis: b(2) ¼ 0 b(5) ¼ 0 F-statistic 6.171001 Probability 0.002544 Chi-square 12.34200 Probability 0.002089

48

3 Geographical Accessibility and Human Capital Accumulation

orthogonality of the two variables (correctly detected by the VIF test) which, instead, tends to decrease the individual t-scores making them unreliable. The empirical analysis thus confirms our hypotheses about the role of accessibility in the process of translation of innovation into economic growth. Our empirical findings are in line with other analyses as accessibility does not enter the growth equation directly (Eq. 3.1, Table 3.4) but we cannot conclude that the role of accessibility is completely mediated by other factors typical of a “peripheral location” and, consequently, not autonomously measurable (e.g., in Richardson 1973). However, thanks to the interaction coefficient, we have been able to directly measure the role of accessibility and its significance for the process of innovation. The role of accessibility is significant and directly measurable when assessed in its interaction with indigenous innovative effort (as we pointed out in the previous section). Better accessibility, through the various mechanisms we outlined in our theoretical analysis (i.e., risk-minimization, higher probability of getting in touch with new ideas, outward networking, access to wider diversity), makes innovative efforts more productive in terms of economic growth. As we can see from (3.2), other things being equal, an increase in innovative efforts produces   ðDinnovationÞ

 an increase in the average growth rate D 1=J ln Yi;t Yi;t
J which is proportional to the accessibility of the region (ðb2 þ b5 accesibilityÞ). In our view this finding opens up some interesting considerations. In terms of overall efficiency it would seem more productive to concentrate innovative efforts in regions benefiting from better accessibility, as this seems to lead to a bigger increase in the average growth rate. However, more accessible regions are also the most advanced and, consequently, those benefiting less from imitating other regions. By contrast the “technology gap”, representing the scope for imitation, would make innovative activities more conducive to growth in backward regions. Thus there seem to be two forces at work in the translation of innovative efforts into growth: one related to the degree of accessibility, which makes indigenous innovative efforts less productive in “peripheral” regions and the other related to the exploitation of the technology gap potential which, on the contrary, should boost growth in less-advanced regions (see also Rodrı´guez-Pose 2001). However, the number of forces in fact at work are far higher than two and low accessibility is only one dimension of an “innovation averse” social filter which, by reducing the productivity of R&D efforts is able to prevent lagging regions from exploiting their “scope for imitation”. By acknowledging once again the multifaceted nature of “real world” processes, our empirical findings may provide small clues to the functioning of at least one of the components of the social filter. The estimation of (3.3), shows that education may interact with innovative effort in a way similar to that of accessibility. The equation shows that the higher the level of human capital accumulation, the more productive, in terms of growth, an increase in innovative efforts may be. Thus, education not only contributes to economic growth directly (as shown by the estimation of (3.1) in Table 3.4) but also by making indigenous innovative efforts more productive, showing the potentiality for compensating the negative effects of poor accessibility.

3.5 Conclusions

3.5

49

Conclusions

This chapter has empirically tested the relevance of innovative efforts for the regional growth performance of the European regions. However, in accordance with the theoretical analysis, our results have shown that different factors influence the capability of each region to translate its innovative efforts into economic growth. The empirical results suggest that geographical distance exerts an influence on local capability to translate innovative efforts into economic growth. Consequently, for innovative efforts in peripheral regions to be as productive as in core areas, they need to be complemented by adequate investments in human capital as a way to compensate for their lower accessibility and to make their social filter more prone to innovation. When this line of reasoning is applied to the new member states of the EU we can understand how the focus on human capital may be crucial for the exploitation of their growth potential as such members’ scope for imitation is still significant and simultaneously some of them may benefit from relatively good (and improving) accessibility. In the following chapter we will try to disentangle the socio-economic preconditions for a successful innovation process by developing more effective proxies for the regional system of innovation conditions. Subsequently, in Chap. 5 the association between peripherality and innovative capabilities will be investigated in further detail by analysing knowledge spillovers and their spatial extent.

.

Chapter 4

The Role of Underlying Socio-Economic Conditions

4.1

Introduction

The exploratory analysis pursued in Chap. 3 has shown that the economic success of EU regions in an innovation-based model of sustainable economic growth depends on their capability to produce and access innovation. The uneven geographical distribution of R&D activities has been shown to be a localised source of competitive advantage for some areas rather than others. Coherently with such empirical evidence regional, national and supra-national policies – not only in Europe but also the United States, as will be discussed in Chap. 6 – devoted to the promotion of the so called “knowledge-based economy” have been mainly focused on various forms of support for R&D activities needed not only for the “production” of new knowledge but also for the economic exploitation of existing knowledge (Cohen and Levinthal 1990). However this emphasis on R&D overlooks at least one other important factor which influences the translation of innovation into regional growth. A successful process of innovation depends on “localised structural and institutional factors that shape the innovative capacity of specific geographical contexts” (Iammarino 2005, p. 499) as highlighted by the systems of innovation (Lundvall 2001), regional systems of innovation (Cooke et al. 1997), and learning regions (Morgan 2004; Gregersen and Johnson 1996) approaches. The idea of innovation prone and innovation averse societies (Rodrı´guez-Pose 1999), from the point of view of socioeconomic filters within regions also plays an important part. This chapter focuses on these factors in order to show that while regional growth is fostered by local innovative activities and benefits from external knowledge flows (in turn improved by the increased interconnection between territories) underlying socio-economic conditions play a crucial role as well. Not only is the “production” of innovation itself shaped by the system of innovation in place in the local economy but this latter factor also influences the extent and the effectiveness of the diffusion of (un-codified) knowledge. In this sense our analysis, by combining R&D, knowledge flows, and innovation systems approaches into one model, will show that contextual socio-economic factors need to be addressed when designing innovation-based regional development policies. In addition, the

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_4, # Springer-Verlag Berlin Heidelberg 2011

51

52

4 The Role of Underlying Socio-Economic Conditions

theoretical and empirical tools developed for the analysis of the regional system of innovation conditions of EU regions provide interesting insights into the geography of socio-economic disadvantage in Europe. This chapter is organised into three sections. The first section introduces the empirical analysis of the socio-economic conditions of the EU regions. In the second section the specification of the empirical model is outlined and the empirical results are discussed. The final section concludes with some preliminary economic policy implications that will be further developed in Chapters 7, 8 and 9 of this book where the integrated framework for the analysis of regional economic performance will directly be applied to policy analysis.

4.2

4.2.1

The Socio-Economic Conditions of the EU Regions and Their Innovative Performance An Empirical Definition for the Regional System of Innovation: The Social Filter

In Chap. 2, when developing the theoretical framework for the empirical analysis of regional growth dynamics and laying the foundations of an integrated approach to regional development, the need for a broader view of the process of innovation than that provided by the “technology-gap” and the “endogenous growth” literature was put forth. The relevance of the insights offered by the systems of innovation approach was highlighted while underlining its methodological orientation towards case-studies analysis. As a consequence, the eminently qualitative “localised structural and institutional” constituents of the regional system of innovation would need to be operationally translated into a set of “measurable” features of the regional realm. However, by definition, the “system of innovation” is a complex and multifaceted concept which not only refers to institutional density but also, in a dynamic sense, to the nature and intensity of the interactions among the innovative agents which resist generalisation. Furthermore, in those cases where an empirical analysis aims at covering simultaneously a large number of regions in a quantitative fashion (necessary if we are to reach relevant conclusions e.g., about EU-level regional policy-making), the amount of innovation-system related suitable statistical information is very limited. In consequence, and in order to allow the systems of innovation literature to be integrated into a cross-sectional growth analysis, it is necessary to single out some structural regularities in the socio-economic structures of innovation-prone regions by focusing our attention on the “external conditions in which externalised learning and innovation occur” (Cooke 1997, p. 485) which can be identified across innovation systems and on which innovation strategies can be based. In other words, rather than trying to “represent” the system of innovation itself, we insert in a quantitative framework a set of proxies which represent the

4.2 The Socio-Economic Conditions of the EU Regions

53

“conditions that render some courses of action easier than others” (Morgan 2004). We define these conditions as “social filters” or, in other words, the unique combination “of innovative and conservative components, that is, elements that favour or deter the development of successful regional innovation systems” (Rodrı´guez-Pose, 1999, p. 82) in every space. In particular, the variables which seem to be more relevant for shaping the system of innovation of a regional space – taking into account the data availability constraints for a sufficiently large cross-section of regions1 – are those related to three main domains: educational achievements (Lundvall, 1992; Malecki 1997), productive employment of human resources and demographic structure (Fagerberg et al. 1997; Fischer et al. 2009c; Rodrı´guez-Pose 1999). For the first domain, tertiary educational attainment (of both the population and the labour force) and participation in Lifelong Learning programmes2 are considered as proxies for the accumulation of skills at the local level. For the second area the percentage of labour force employed in agriculture and the long-term component of unemployment are included in the analysis in order to capture the amount of human resources excluded from productive employment. On the one hand, long term unemployment represents the incidence of people whose possibilities of being productively involved in the labour market is persistently hampered by inadequate skills (Gordon 2001). On the other hand, in particular in the new members of the European Union, agricultural employment is virtually synonymous of “hidden unemployment” (Fabiani 1991; Crescenzi 2004 for an application to Central Eastern European Countries). For the third area, the percentage of population aged between 15 and 24 is used as a proxy for the flow of new resources entering the labour force, thus potentially “renewing” the existing stock of knowledge and skills.3 Table 4.1 shows some descriptive statistics of the proxies included in the “social filter” for the regions of the EU-25. The general inspection of the table suggests that the EU regions are quite differentiated in terms of their socioeconomic conditions.

1

See Appendix A on data availability. “The lifelong learning approach is an essential policy strategy for the development of citizenship, social cohesion, employment and for individual fulfilment” (European Commission 2002, p. 4). In the framework of the Lisbon Agenda’s objectives by means of lifelong learning the European Commission aims – among the other things – to ensure that people’s knowledge and skills match the changing demands of jobs and occupations, workplace organisation and working methods (2002, p. 5). 3 The European Commission made explicit the challenges created by ageing population when countries/regions have to rely upon the benefits of the knowledge based society, and presented the investment in human capital as a tool for counterbalancing the potential negative impact of ageing upon innovative performance. Where policies targeted towards human capital accumulation are implemented, “the future cohorts of older workers will benefit from higher levels of training, reducing the risk of a slower spread of new technologies that could be associated with ageing”. (European Commission 2006; p. 6). 2

54

4 The Role of Underlying Socio-Economic Conditions

Table 4.1 Descriptive statistics of the social filter variables Variable Mean StDev CoefVar Education population 0.125 0.054 43.270 Education labour force 0.187 0.073 39.040 Life-long learning 0.055 0.048 87.020 Agricultural labour force 0.098 0.094 95.230 Long term component of 0.464 0.121 26.160 unemployment Young people 0.133 0.027 20.330 Source: EUROSTAT data – 166 Regions

Min 0.042 0.061 0.002 0.001 0.030

Median 0.122 0.187 0.040 0.072 0.462

0.087 0.127

Max 0.259 0.433 0.207 0.456 0.737

Range 0.217 0.372 0.205 0.456 0.707

0.214 0.127

Some interesting preliminary insights into the spatial organisation of these variables are provided by Figs. 4.1a–f. The boxplots (or box-and-whisker plots) in the figures show the distribution of the socio-economic variables presented above across member states at the beginning of the period of analysis (1995). As far as the educational variables (of the population and the labour force) are concerned (Figs. 4.1a–b), the graphs highlight the differences between three groups of member states: the majority of the EU-15 countries, a group of EU-15 laggards (Italy, Portugal and Greece) and the new members of the EU. The first group shows, on average, a higher level of human capital accumulation together with a more marked degree of internal regional differentiation. In the other two groups, which show lower than average levels of education achievements, the degree of internal variability is less accentuated (with the exception of the Slovak Republic). The cross-country variability in terms of the participation of adults in education and training is more limited (Fig. 4c). Only Finland, the Netherlands and the UK show significant above-average values due to the implementation of extensive adult education programmes in these countries, which are also an important pre-condition for tackling unemployment by keeping the knowledge base of the labour force up-to-date. However, the European Commission (2002) has emphasized the problems encountered in encouraging the weakest segments of the labour market to participate in these programmes and highlighted how the effectiveness of such programmes is crucially dependent on a previous experience of formal education. For these reasons the contribution of this variable in shaping of the regional social filter conditions and its distribution needs to be addressed together with the (formal) educational variables discussed above. Fig. 4.1d shows the distribution by country of the weight of agriculture on total regional employment. There are no significant differences among national averages, with the exception of Greece, for EU-15 member-states, and Poland, for the EU-25. What matters more is the variability of this indicator around its average, which highlights the magnitude of the rural-urban economic divide in the various countries and the degree of dependence of some areas on agricultural income and employment. If, as discussed earlier in this section, the concentration of regional employment in agriculture may be considered as a symptom for the under-utilisation of human resources, the long term component of regional unemployment (Fig. 4.1e) provides complementary information on the stratification and persistence of unemployment which is reinforced by the lack of adequate skills.

4.2 The Socio-Economic Conditions of the EU Regions

55

The latter indicator shows very similar average values in the EU countries (with the exception of Austria and Finland) and the Central-Eastern European countries seem to perform relatively well for this indicator while Belgium, Greece, Italy and Spain display the highest long-term share of total unemployment.

a Educational achievements of the population in EU-25 regions, 1995 FR1

% of people with tertiary education

0.25

ES21 ES3 ES22

0.20

CZ01

GR3

0.15

HU01

0.10

I T60

0.05 AT

BE

CZ

DE

ES

FI

FR GR HU Countries

IT

NL

PL

PT

SK

UK

b Educational achievements of employed people in EU-25 regions,1995 % of employed people with tertiary education

0.45 0.40 0.35

FR1

0.30 0.25

GR3

CZ01

HU01

AT13

0.20 0.15 0.10 0.05 AT

BE

CZ

DE

ES

FI

FR

GR HU

Countries

Fig. 4.1 (continued)

IT

NL

PL

PT

SK

UK

56

4 The Role of Underlying Socio-Economic Conditions

c

% of adults involvedin education and training

Rate of involvement in Life-long learning in EU-25 regions,1995 0.20

0.15 UKN

CZ01

0.10 DE3 DE6

0.05

0.00 AT

BE

CZ

DE

ES

FI

FR

GR HU

IT

NL

PL

PT

SK

UK

PT

SK

UK

Countries

d Agricultural employment as % of total employment

Agricultural labour force in EU-25 regions, 1995 0.5

0.4

0.3

0.2

0.1

0.0 AT

Fig. 4.1 (continued)

BE

CZ

DE

ES

FI

FR GR HU Countries

IT

NL

PL

4.2 The Socio-Economic Conditions of the EU Regions

57

e Long-term unemployment in EU-25 regions, 1995 0.8

%Long-term of total unemployed

0.7 UKN

0.6 0.5 0.4 NL21 FR42

0.3

PL08

0.2 0.1 0.0 AT

BE

CZ

DE

ES

FI

FR GR HU Countries

IT

NL

PL

PT

SK

UK

f

People aged 15-24 as% of total population

Young people in EU-25 regions, 1995 0.225

0.200

0.175 CZ02

UKN

0.150

GR42 HU01

0.125 CZ01

0.100

AT

BE CZ

DE

ES

FI

FR GR HU Country

IT

NL

PL

PT

SK

UK

Fig. 4.1 Descriptive statistics: education population, (b) Descriptive statistics: education labour force, (c) Descriptive statistics: life-long learning, (d) Descriptive statistics: agricultural labour force, (e) Descriptive statistics: long term unemployment, (f) Descriptive statistics: young people

58

4 The Role of Underlying Socio-Economic Conditions

Finally, Fig. 4.1f compares the distribution of the percentage of young people in the EU member states which reflects the different demographic trends in place in these countries. On average Austria, the Netherlands and Spain for the EU15 and the Czech Republic and Hungary for the new member states benefit from relatively more favourable demographic structure while all other countries show a lower (and similar) % of young population. In the case of this indicator regional differences within each country are also particularly evident. Overall, from the general inspection of the distribution of all these “social filter” indicators it is apparent that, together with different “national” characteristics (variability in national averages), relevant regional patterns emerge (within country variability). As a consequence regions from countries with very different national averages may show similarly high or low values of a certain indicator. Each of these variables is thus assumed to be representative of one “side” of the intrinsically multifaceted concept of social filter. While keeping in mind such a composite nature of the social filter concept and the autonomous role played by each domain, Principal Component Analysis (PCA)4 is applied to the set of variables discussed above in order to achieve two distinct objectives. First, Principal Component Analysis allows us to represent a large part of the global information provided by the original set of variables in two-dimensional space. Such a representation yields a first “classification” of the EU regions in terms of homogenous categories based on their social-filter conditions. Second, PCA enables the development of a “joint” measure for the social filter conditions of each regional space by “reducing” them into an individual variable able to preserve as much of the initial information (variability) as possible. Such a measure also enables us to deal with the problem of multicollinearity, which would prevent the simultaneous inclusion of the individual (highly correlated) social filter variables in a regression model. Tables 4.2a, b show the output of the Principal Component Analysis. Table 4.2a Principal component analysis: eigenanalysis of the correlation matrix PC1 PC2 PC3 PC4 PC5 Eigenvalue 2.5886 1.2723 0.9083 0.6418 0.5661 Proportion 0.431 0.212 0.151 0.107 0.094 Cumulative 0.431 0.643 0.795 0.902 0.996

Table 4.2b Principal component analysis: principal component’s coefficients

4

Variable Education population Education labour force Life-long learning Agricultural labour force Long term unemployment Young people

For the technicalities of PCA see Appendix C.

PC1 0.576 0.554 0.395
0.43
0.14 0.019

PC6 0.0229 0.004 1

PC2
0.224
0.313 0.26
0.285
0.459 0.701

4.2 The Socio-Economic Conditions of the EU Regions

59

The Eigenanalysis of the Correlation Matrix (Table 4.2a) shows that the first principal component (PC) alone accounts for around 43% of the total variance with an Eigenvalue significantly larger than 1, the second PC accounts for an additional 21% of the total variability with an Eigenvalue still larger than 1. The first two principal components therefore explain a significant part of total variability (64%). The coefficients of the first PC (Table 4.2b) emphasize the educational dimension of the social filter by assigning a large weight to the educational achievements of the population (0.576) and of the labour force (0.554) and to the participation in Life Long Learning Programmes (0.395). A negative weight is, as expected, assigned to the agricultural labour force (
0.430) and, with a smaller coefficient, long-term unemployment (
0.140). The weight of the young population (0.019) is much smaller, in the first principal component. The second PC, instead, emphasizes precisely this demographic dimension by assigning a relatively small and negative value to educational (except life long learning) and “resources employment” variables in contrast with the large and positive coefficient of the “young population” variable. The first principal component will provide us with the “joint measure” for each region’s social filter and will be referred to as the “social filter index”. Consequently, the first principal component’s scores are computed from the standardised5 value of the original variables by using the coefficients listed under PC1 in Table 4.2b. The second principal component will be scattered against the first in order to represent the social filter variables in a two-dimensional space. As discussed above, this second PC (PC2) contrasts the incidence of young people in the total population with the other features of the regional social filter and, for explanatory purposes, can be referred to as the “demographic dynamism index”.

4.2.2

The Geography of the Social Filter Conditions in the EU

A first representation of the socio-economic “infrastructure” of the EU regions is provided by the PCA, which yielded the scatter plot showed in Fig. 4.2. For each region this graph plots the score of the first principal component on the X-axis and that of the second PC on the Y-axis. As discussed above the first two principal components account for a high percentage of the global “information” of the original variables. Consequently, the scatterplot in Fig. 4.2 accurately represents the relative position of the EU regions in terms of their “social filter” conditions. The more favourable the socio-economic conditions (societies more prone to innovation) the higher the scores of the first PC (Social Filter Index). This means that the regions are ordered on the X-axis by the value of their social filter index so that innovation prone regions appear furthest to the right of the graph. The value of the second principal component (Demographic Dynamism Index) is higher for the regions where the percentage of the young population is higher (the only variable 5

Standardised in order to range from zero to one.

60

4 The Role of Underlying Socio-Economic Conditions

Social Filter:first vs second PC Q1 (0.053) Demographic Dynamism Index (PC2)

0.50

Q3 (0.6)

A T33 A T34

NL21

A T11

A T32 NL34 NL12 PT11 I T32AAT31 T21 NL31 NL11 NL23 ES42 ES61 HU02 ES62 CZ02 AI T33 T22 CZ05 NL41 ES52 NL13 GR42 FI 15 HU03 CZ04 CZ03 ES53 PT14 HU07 CZ06 ES43 I T4 NL22 NL33 CZ07 AI T12 HU06 T2 NL32 CZ08 HU04 I TB NL42 UKE PL04 I T11 HU05 SK02I T91 ES51 UKCUKG UKD UKH UKF UKL UKK UKM FI 13 SK04 SK03 PL0C PT13 I T53 PL08 I T12 PT12 I T92 UKJ DE2 I T51 PL0F FI 14 PL02 UKN CZ01 I ITA FR3 T8 PL01 FI 2 FR42 FR22 FR43 PT15 FR21 HU01 FR41 DEB I T52 SK01 PL0B A T13 PL0GPL06 I T93 DEF DE1 FR23 PL0E FR83 FR51 DE9 ES24 I T71 I T60 GR22 DE7 ES22 FR63 GR3DE6 DEA FR25 FR26 PL03 I T72 I T13 PL07 PL05 FR24 ES23 GR12 ES41 DE8 ES3 GR41PL09 GR11 FR81 FR82 DE5 ES13 FR52 FR53DEC DEEDEGDED FR72 FR61 PL0A GR24 PL0D FR62 ES12 ES21 GR43 GR13 ES11 FR71 BE2 DE3 DE4 GR23 BE3 FR1 GR14GR21 GR25

0.25 0.00 –0.25 –0.50

A

–0.75

BE1

C UKI

Ave(-0.21)

B

–1.00 -0.5

0.0

Ave (0.336) 0.5

1.0

1.5

Social Filter Index (PC1)

Fig. 4.2 Social filter: first versus second principal component

associated to a large and positive coefficient) when compared to other social filter components. Thus the Y-axis contrasts with the X-axis by differentiating the regions essentially on the basis of their demographic dynamics. E.g. two regions with a similar value for the Social Filter Index (thus being plotted in a very close position on the X-axis) may have differential demographic dynamics and thus are plotted far away from one another on the Y-axis. Scrutiny of Fig. 4.2 suggests that variability in terms of the Demographic Dynamism Index tends to be lower as concerns the intermediate values of the Social Filter Index (thus highlighting a more homogenous demographic structure for these regions) while increasing with low and high values of the Social Filter Index. Low values (below the first quartile – Q1) and high values (above the third quartile – Q3) of the Social Filter Index (PC1) the Demographic Dynamism Index (PC2) result in three distinct clusters of regions, circled in the graph. Conversely, when intermediate values of the Social Filter Index, no particular pattern is apparent. Some of the lowest social filter index regions, which are grouped together on the left-hand side of the graph, show particularly weak demographic dynamics thus suggesting that an ageing population is a prominent characteristic of their generally unfavourable social filter conditions (group A in the graph). At the opposite extreme of the distribution of the Social Filter Index, the Demographic Dynamism Index (PC2) allows us to distinguish two further groups (a) the regions where generally favourable social filter conditions are associated with a relatively older population (group B) and (b) those, on the contrary, where the demographic dynamics are relatively more favourable (group C). While group C regions benefit from a more

4.2 The Socio-Economic Conditions of the EU Regions

61

favourable social filter and from favourable demographic dynamics that keep the percentage of the young relatively high, group B regions compensate for a less favourable demographic structure with better performance in other social-filter indicators. Conversely, in group A regions the process of ageing of the population is not compensated for by other more favourable aspects. As the European Commission (2006) has highlighted, the most effective tool to compensate for the negative impact of ageing population on productivity and employment is the investment in human capital which, however, in this last group of regions tends to be low as well. In general, when interpreting the meaning of the Demographic Dynamism Index, it must be borne in mind that a large part of the demographic dynamism of the European Union is generated by migration flows which far exceed the natural increase in the European population (European Commission 2006). After 1990, net inward migration has become the major component of population growth in the EU-25: “since 2000 more than three-quarters of the total population growth in the EU-25 is due to net migration. However, while the former EU-15 countries fully account for the population growth by international migration, net migration is (still) negligible in the new Member states” (Eurostat 2006; p. 46). As a consequence the Demographic Dynamism Index (PC 2) mainly highlights the differential migration trends: some regions have been able to benefit from inward migration flows which have accentuated the incidence of young people in the total population (those in the higher part of the graph), while in other regions low birth rates and the natural ageing of the population prevail (those in the lower part). This “classification” of the EU regions according to their social filter conditions reflects the underlying diversity of the socio-economic geography of the Union, which is more directly analysed in Fig. 4.3. Figure 4.3 plots the social filter index of each region against the corresponding accessibility to innovation prone extra-regional areas i.e., the distance weighted average of the neighbouring regions’ social filter index.6 In other words the graph compares the local social filter conditions with those of neighbouring regions in order to highlight the presence of a spatial pattern in the distribution of regional competitive disadvantage. This spatial pattern, which is clearly highlighted in Fig. 4.2, shows that less well-endowed regions (left-hand side part of the graph) tend to have disadvantaged neighbours as opposed to more innovation-prone areas (right-hand side) whose neighbours tend to be equally favourable to the development and absorption of innovation. In other words, the regions with more favourable socio-economic conditions also benefit from a better “accessibility” to innovation prone extra-regional areas while the more disadvantaged ones, due to the spatial concentration of the factor of disadvantage, tend to have a less direct access to more advantaged territories. The empirical model presented in Sect. 4.4 assesses whether this particular geography of these socio-economic factors has an influence on local economic performance or not.

6

The calculation of this last variable is discussed in further detail in the section justifying its inclusion in the regression model.

4 The Role of Underlying Socio-Economic Conditions

Accessibility to Innovation Prone Extra-Regional Areas

62

Endogenous vs External Socio-Economic Conditions 0.6

FR3

0.5

UKH NL23 NL41 UKG NL33 NL32 BE2 NL22UKF UKE NL34 NL12 NL21 UKD UKK UKL NL13 NL42 UKCBE3 DEA NL11

NL31 UKJ

UKI

BE1

FR22

DE5 UKM DE9 DEF DE6 ES23 UKN FR23DEC DEB DE7 DEE FR25 FR24 FR41 FR1 FR51 DE1 DE8DEG DE4 ES22 FR53 FR42 FR63 FR26 FR52FR61 DE3 ES24 ES21 FR43 ES13 FR72 DED FR62 FR81 ES41 ES42 FR71 ES51 ES12 CZ02 ES52 AT34 CZ04 ES62 PL0G ES3 FI 2 ES53 DE2 FR82 ES11ES43 FIFI 1314 FI 15 AT33 I T12 CZ03 ES61 PT11 PT12 PT15 AT31 AT32 PL04 PT14 CZ05 I TB FR83 I T11 I T13 AT12 AT21 AT22I T2 CZ06PT13 PL0EPL0F IAT11 sk02 I T33 CZ01 T32 HU03 PL0B I TACZ07 HU04 HU07 HU02 I T51 AT13 I T4 PL01 PL02 PL08 PL0A I T93 IIT53 T52 HU06 CZ08 HU05 I T71 GR41 GR43 sk03 HU01 sk01 I T92 IGR42 T91 I T6 I T72 sk04 PL05 I T8 GR11 PL03 PL0D PL07 PL0C GR22 PL09 PL06 FR21

0.4

0.3

GR25 GR23 GR24

GR14

GR13 GR21

0.2

GR12

GR3

– 0.5

0.0

0.5 Social Filter Index

1.0

1.5

Fig. 4.3 Endogenous versus external socio-economic conditions

4.2.3

The Social Filter Index and Its Association with R&D Expenditure: Four Typologies of Regions and Their Differential Growth Performance

Socio-economic conditions, as discussed in Chap. 2, exert a special influence on regional economic performance when interacting with local innovative activities. Fig. 4.4 presents a visual representation of this relationship by analysing all the possible combinations between local innovative efforts (as proxied by R&D expenditure) and social filter conditions (proxied by the Social Filter index). The theoretical analysis suggests that the “optimal conditions” for economic success are met where a considerable amount of innovative effort is pursued in an innovation prone socio-economic environment (top right quadrant of the graph). Conversely, economic performance seems to be significantly hampered by the coexistence of “inadequate social filter conditions and innovative efforts”. In this latter case indegenous efforts are reduced and their effects on the local economy are negatively affected by the inadequate environment. In addition to these two opposite cases a variety of intermediate situations are possible. In particular, a region with a favourable “social filter” may pursue “sub-optimal research efforts” or relevant investment in R&D may be undertaken in an unfavourable

4.2 The Socio-Economic Conditions of the EU Regions +

Social Filter Conditions

Fig. 4.4 Local innovative efforts and social filter conditions

63

Sub-optimal research efforts

Optimal conditions

Inadequate conditions and innovative efforts

Detachment between R&D and local conditions

-

R&D expenditure

+

socio-economic environment i.e., where an “inconsistency between R&D and local conditions” is recorded. In the former case the theory suggests the possibility of benefiting from externally produced innovation which may partially compensate for the inadequacy of local efforts. In the latter case, instead, R&D investments run the risk to produce the “cathedrals in the desert” effect. Coherently with this analytical framework, Fig. 4.5 plots the observed R&D expenditure in the EU regions against the corresponding value of the social filter index. As a result we notice that the large majority of the EU lagging regions are clustered in the lower left corner of the graph: a combination of underinvestment in R&D and adverse social filter conditions might be contributing towards their economic disadvantage. Conversely, the innovation leaders are clustered in the top right-hand side area: in these regions the innovative efforts find the most appropriate socio-economic environment, making their leadership possible. This group includes: Berlin, Baden-W€ urttemberg, Bavaria and (to a lesser extent) Bremen in Germany, the entire South East England, and the ˆIle de France and MidiPyre´ne´es regions in France. Between these two “poles” it is possible to observe a variety of intermediate possible combinations of local efforts and social conditions associated with very different economic outcomes. This kind of evidence suggests that the large majority of the EU regions are simultaneously affected (even if with place-specific combinations of each factor) by inadequate social filter conditions and insufficient innovative efforts. The inadequacy of these regions with respect to both indicators can be assessed both with respect to the EU average and, in the case of R&D expenditure (for which it is possible), with respect to the US average (both highlighted in the graph). Such heterogeneity of situations and their association

64

4 The Role of Underlying Socio-Economic Conditions

R&D expenditure and "Social Filter" conditions EU25 Ave US Ave 1.5 UKI

BE1 NL31 UKM

1.0

ES22

ES21

Social Filter Index

DE4 CZ01

NL32 ES3

0.0

UKJ

NL33

UKD DEG UKF UKL BE2 UKG NL22 BE3 DEE FI15 UKC DE7 UKN SK01 DE6 FR71 NL42 NL21 FI2 FI14 ES51FI13DEF HU01 ES24 FR82 GR3 NL12 ES12 ES41 NL13 DEB ES13 ES52 FR42 FR61 FR81 DEADE9 FR41 NL34 ES61 FR52 FR43 ES23 ES62 AT32 DEC FR72 FR24 FR3 ES53 FR51 FR23 FR25 FR26 ES42 FR53 PT13 FR22 AT34 AT33 FR21 FR63 IT60 ES43 CZ06 HU02 AT21 PL07 HU05 AT12 GR12 PL01 PL0B IT2 IT4AT31 HU03 PL0C ES11 PL06 IT33 AT22 IT52 IT53 AT11 IT32 HU06 CZ08IT13 CZ03 CZ04 CZ05 IT51 HU04 IT12 CZ07IT11 FR83 PL0G PL04 IT8 IT71 ITA GR42 PL0F HU07 IT91 ITB GR13 IT72 IT93 SK03 PL0E PL02PL08 PL09GR21 PL05 SK04 SK02 PT15 PT11IT92 PL03 PL0A PT14 GR41 GR22 GR14 PL0D PT12 GR43 GR11 GR24 GR23 UKE DE8

0.5

DE3

FR1 UKK DED

UKH NL11 NL41 NL23 DE2 AT13

DE1 FR62 DE5

Ave (0.336) CZ02

GR25

–0.5 0.00

0.01

0.02

0.03

0.04

R&D expenditure (% Regional GDP)

Fig. 4.5 R&D expenditure and “Social Filter” conditions

with regional growth performance deserves a more in-depth investigation by means of regression analysis.

4.3

4.3.1

An Empirical Model for the EU Regional Growth Performance Model Specification

In previous chapters we pointed out that an important part of the process of innovation relies upon three crucial factors: internal innovative efforts, socially embedded factors and spatially bound knowledge spillovers. In this section such factors are integrated into a formal empirical model in order to study how local innovative efforts are translated into regional growth and assess the role of systems of innovation and knowledge spillovers in this process. As succinctly presented in Table 4.3 the model combines the mechanisms discussed in the Chap. 2 (and preliminarily explored in Chap. 3) in a common “integrated” analytical framework in order to assess their empirical relevance and their reciprocal interactions. In line with this, the standard specification of an endogenous growth model for the EU regions (where innovation is considered as the key determinant for regional growth performance) is extended in

4.3 An Empirical Model for the EU Regional Growth Performance

65

Table 4.3 Structure of the empirical model based on the “integrated” approach to regional growth dynamics Internal Spillovers R&D Investment in R&D in the Investment in R&D in neighbouring region regions Regional systems Regional system of innovation Regional system of innovation in of innovation neighbouring regions GDP per capita As a proxy for initial Initial conditions in neighbouring conditions and potential regions National effect Controlled for by a set of National Dummies

order to take into account both the “indigenous” characteristics of the regional system of innovation and the “spillover effects” due not only to neighbouring regions’ innovative activities, but also to their structural socio-economic features. The full estimated model, directly derived from the “prototype” model in (2.5), is:
 1 Yi;t ln ¼ a þ b1 lnðyi;t
J Þ þ b2 RDi;t
j þ b3 SocFilteri;t
J þ b4 Spillovi;t
j Yi;t
J J þ b5 ExtSocFilteri;t
J þ b6 ExtGDPcapi;t
J þ b7 D þ e

ð4:1Þ

where:
 1 Yi;t ln Yi;t
J J a lnðyi;t
J Þ RDt
j SocFilteri;t
J Spillovi;t
j ExtSocFilteri;t
J ExtGDPcapi;t
J D e

Is the usual logarithmic transformation of the ratio of regional per capita GDP in region i at the two extremes of the period of analysis (t-J,t) Is a constant Is the log of the GDP per capita of region i at the beginning of the period of analysis (t-J) Is expenditure in R&D as a % of GDP in region i at time (t-J) Is a proxy for the socio-economic conditions of region i representing its “social filter” Is a measure of accessibility to extra-regional sources of innovation Is a measure of the “social filter” of neighbouring regions Is a measure of the GDP per capita in neighbouring regions Is a set of national dummy variables Is the error term

Initial level of GDP per capita – As customary in the literature on the relationship between innovation and growth, the initial level of the GDP per capita is introduced in the model in order to account for the region’s stock of existing knowledge. R&D expenditure – The percentage of regional GDP devoted to R&D is a measure of the economic input in order to generate innovation in each region. Social Filter – As a proxy for the socio-economic conditions we introduce in the model the “social filter index” discussed in Sect. 4.2.1 and calculated by means of the Principal Components Analysis.

66

4 The Role of Underlying Socio-Economic Conditions

Spillovers – This variable proxies the capability of exogenously produced innovation to influence regional economic performance in addition to “endogenous” innovation. For this purpose a measure for the accessibility of “extra-regional” innovative activities has been developed and introduced in the analysis by means of a standardised “accessibility of innovation index”. The index is a potential measure of the “innovative activities” (in terms of nationally weighted millions of Euros invested in R&D activities) that can be “reached” from each region at a “cost” which increases with distance. Our index is based on the customary formula for accessibility indices: Ai ¼

X

gðrj Þ f ðcij Þ

(4.2)

j

where Ai is the accessibility of region i, rj is the activity R to be reached in region j, cij is the generalised cost of reaching region j from region i and g(·) and f(·) are “activity” function (i.e., the activities/resources to be reached) and “impedance” function (i.e., the effort, cost/opportunity to reach the specific activity) respectively. In our index the “activity” to be reached is R&D expenditure, thus: gðrj Þ ¼ ðR&D expenditureÞj and the “impedance” is the bilateral trip-time distance between region i and region j: 8 wii ¼ 0 > > > > > 1 < f ðcij Þ ¼ dij > wij ¼ P > > 1 > > : d ij j

if i ¼ j if i 6¼ j

(4.3)

where dij is the average trip-length (in minutes) between region i and j and w the corresponding inverse-distance weight. We base our analysis on the travel time calculated by the IRPUD (2000) for the computation of the Peripherality Indicators and made available by the European Commission.7 We chose road distance, rather than straight line distance, as (in particular on a smaller scale) it gives a more realistic representation of the real “cost” of the contacts across distance. In addition the use of trip-length rather than kilometres allows us to take account of “different road types, national speed limits, speed constraints in urban and mountainous areas, sea journeys, border delays (. . .) as also congestion in urban areas” (IRPUD 2000, p. 22) which significantly affect real-world interactions.

7

As the time distance-matrix is calculated either at the NUTS1 or at the NUTS2 level, in order to make it coherent with our data which combine different Nuts levels we rely on the NUTS distance matrix using the NUTS 2 regions with the highest population density in order to represent the corresponding NUTS1 level for Belgium, Germany and the UK.

4.3 An Empirical Model for the EU Regional Growth Performance

67

Thus, the amount of knowledge flowing from outside the region is proxied by the average magnitude of all other regions’ R&D expenditure weighted by the inverse of the bilateral time-distance. The resulting variable is then standardised by making it range from zero to one, in order to be comparable with the social filter index. In this chapter, knowledge spillovers are assumed to decay with distance in linear fashion: R&D activities in all EU regions are weighted with the inverse of their travel-time distance. The objective of this part of the analysis it to empirically test the existence of localised spillovers and assess their interaction with Social Filter conditions. Chapter 5 is, instead, specifically devoted to the analysis of the distance decay process of knowledge flows and to the “measurement” of its spatial extent by means of different cut-off points in the inverse distance weighting matrix. As consequence Chap. 5, where the spatial boundeness of knowledge spillover is analysed in detail, includes several different specifications for this variable. Extra regional social filter – Following a similar procedure we calculate, for each region, the inverse-distance-weighed average of the social filter index of all other regions thus aiming at assessing the existence of spillovers from innovation prone regions towards neighbourhood as far as their social filter is concerned.  their  As aconsequence f c remains the same as in equation (4.3), while: ij  g rj becomes the Social Filter Index GDP in neighbouring regions – Again the same weighing procedure is pursued in order to introduce the initial economic conditions (GDP per capita) of neighbouring regions. This variable accounts for the advantage of proximity to relatively “advanced” regions which could potentially increase the scope for imitation. In this case:  g rj denotes GDP per capita in (4.3).

4.3.2

Estimation Issues

In this section we estimate the model outlined above by means of Heteroskedasticity-Consistent OLS (Ordinary Least Square). Spatial dependence (i.e the lack of independency among the error terms of neighbouring observations) violates the assumptions of the traditional regression analysis and calls for appropriate “remedies” in the specification of the empirical model. Several alternative strategies have been proposed in the literature with no consensus on the “best” solution. The introduction in the empirical model of an additional regressor in the form of a spatially lagged dependent variable is customary in the empirical literature. However, in this case, the inclusion of the spatially lagged depended variable would substantially change the structure of our empirical model forcing us to rely on a very specific set of assumptions on the diffusion of knowledge spillovers that would not be fully justifiable in light of the theoretical analysis and, in any case would preclude any ceteris paribus interpretation of the estimated coefficients (Anselin 2003, p. 158). As a consequence, in order to minimize the effect of spatial autocorrelation we include in the analysis a set of national dummy variables

68

4 The Role of Underlying Socio-Economic Conditions

accounting for the “national fixed effect” which, in turn, accounts for a consistent part of the similarities between neighbouring regions. Furthermore, by including spatially lagged variables in our analysis, we aim to model explicitly the interactions between neighbouring regions, minimizing their effect on the residuals. Given that the Moran’s I test (computed for each specification of the model) does not detect spatial autocorrelation in the residuals, we believe that the combination of “national” variables and spatially lagged explanatory variables are able to capture a significant part of the total spatial variability of the data. Another major problem concerns endogeneity, which we partially address by including8 in the model the value of the explanatory variables as a mean over the period (t-J-5) – (t-J), while the average growth rate was calculated over the period from t-J to t. In addition, in order to resolve the problem of different accounting units, explanatory variables are expressed, for each region, as a percentage of the respective GDP or population. The empirical model is estimated for the period 1995–2003, including all the EU-25 members for which regional data are available (see Appendix A for further details on the dataset).

4.3.3

Innovation, Spillovers and Social Filter: Empirical Results

The estimation results for the empirical model outlined in the previous section are presented in Table 4.4. In regressions 1–3 the variables for “social filter” and “accessibility to external sources of innovation” are progressively introduced. In regressions 4–9 the various components of the social filter are autonomously introduced in the model. In equations 10–12 the effect of neighbouring regions endowment in terms of social filter and economic wealth is assessed. The r-squared confirms the overall goodness-of-fit of all the regressions presented and in all cases the probability of the F-statistics lets us reject the null hypothesis that all of the regression coefficients are zero. The V.I.F test has been pursued for the variables included in all the specifications of the model excluding the presence of multicollinearity. Additionally, we have tested for the presence of spatial autocorrelation in the residuals and the Moran’s I statistic allow us to exclude this possibility for all the regressions presented in the table. From a general inspection of Table 4.4 we notice that the initial level of the GDP per capita is significant in a few cases only, suggesting that for the period under analysis, neither regional convergence, nor divergence can be recorded. Only when social conditions are explicitly controlled for (regressions 3, 10, 11 and 12) there is evidence of a weak degree of regional convergence.

8 In the case of the New Member States data availability has prevented us from calculating the mean of the explanatory variables over the five year period (t-T-5) forcing us to use a shorter time span. For some EU 15 countries slightly differential time spans have been used according to data availability for each variable. The specification of each individual case would take up too much space.

x

0.665 0.626 17.27
0.0185667

0.659 0.62 16.84
0.0193012

0.013236 (0.008148) x

0.09406*** (0.02572)
0.003098 (0.003255) 0.2682** (0.1174)

2

0.672 0.631 16.7
0.0189041

0.12182*** (0.02796)
0.00663* (0.003543) 0.1791 (0.1218) 0.010787** (0.004598) 0.01387* (0.008031) x

3

0.681 0.642 17.45
0.0194612

0.017003*** (0.005341)

0.013157* (0.007908) x

0.1126*** (0.02563)
0.00574* (0.003267) 0.1366 (0.1212)

4

0.676 0.636 17.03
0.0198153

0.019224*** (0.006986)

0.013733* (0.007975) x

0.10707*** (0.02561)
0.005112 (0.003268) 0.166 (0.1208)

5

0.66 0.618 15.82
0.0193265

0.00385 (0.01076)

0.012717* (0.0083) x

0.09655*** (0.02671)
0.003359 (0.003346) 0.2556** (0.1229)

6

0.66 0.618 15.85
0.0198503

0.003802 (0.006528)

0.012262 (0.008336) x

0.08491*** (0.03019)
0.00196 (0.003803) 0.2664** (0.1177)

7

0.659 0.618 15.81
0.0195195

0.001892 (0.006205)

0.013353 (0.008182) x

0.08989*** (0.0292)
0.002733 (0.003478) 0.2653** (0.1182)

8

0.665 0.624 16.19
0.0199182


0.009089 (0.005882)

0.013807* (0.008119) x

0.10777*** (0.02709)
0.004345 (0.003339) 0.2548** (0.1172)

9

0.67 0.63 16.61
0.0188243

0.012617*** (0.005656)

0.014184* (0.008052) x

0.12054*** (0.02802)
0.006577* (0.003571) 0.1883 (0.1213)

10

*, **and ***denote significance at a 10%, 5% and 1% level respectively. SE in parentheses; n. of observations ¼ 166 in all regressions

Extra-Regional Social Filter Total accessibility to innovation prone space Accessibility to Innovation Prone Extra-Regional areas Accessibility to wealth neighbouring regions R-Sq R-Sq (adj) F Moran’s I

Young People

Long Term Unemployment

Agricultural Labour Force

Life-Long Learning

Education Labour Force

1

0.12284*** (0.02814)
0.005756 (0.00353) 0.1424 (0.1207) 0.01052** (0.004626)

Social Filter Individual Components: Education Population

Accessibility to ExtraRegional Innovation National Dummies

Social Filter Index

R&D expenditure

Log GDP 95

Constant

Dependent variable: Regional Growth Rate (1995–2003)

Table 4.4 H-C OLS estimation of the empirical model. R&D, social filter and knowledge spillovers

0.12059*** (0.02809)
0.007705* (0.003929) 0.1909 (0.1234) 0.011422** (0.004713) 0.014229* (0.008067) x

12

8.8E-07 (0.00000138) 0.672 0.672 0.629 0.63 15.72 15.77
0.0188376
0.0189403


0.00808 (0.0261)

0.12187*** (0.02805)
0.006349* (0.003668) 0.177 (0.1223) 0.010538** (0.004682) 0.013936* (0.008059) x

11

4.3 An Empirical Model for the EU Regional Growth Performance 69

70

4 The Role of Underlying Socio-Economic Conditions

Local R&D expenditure generally shows a positive and significant relationship with economic growth in all regressions, in line with similar research (Bilbao-Osorio and Rodrı´guez-Pose 2004; Cheshire and Magrini 2000; Crescenzi 2005; Fagerberg et al. 1997; Fischer 2010; Rodrı´guez-Pose 1999, 2001). Investing in R&D seems to be a more effective means to achieving economic growth than relying upon knowledge spillovers from neighbouring regions. When the two factors are considered together (regression 1) the coefficient of local R&D expenditure is positive and significant, while access to innovation generated outside the region is insignificant. However, local R&D investment is not per se a guarantee for achieving greater growth: the relationship is not always robust after controlling for social conditions (“social filter” variable). Regression 2 highlights that local socio-economic conditions are a better predictor of regional growth performance than local R&D investment. The social filter variable is always positively associated with economic growth and statistically significant. This suggests that the regional system of innovation conditions are the crucial asset in terms of the local capability to innovate. Local “innovativeness” is sometimes better proxied by an innovation prone social filter that impacts upon the “quality” of the local innovative effort than by the amount of resources devoted to R&D investment. Furthermore the relevance of the “social filter” is enhanced when R&D investment and access to knowledge spillovers are considered in conjunction with local conditions (Regression 3). Since knowledge may flow also from outside the region (both in the form of codified knowledge and spillovers), local socio-economic conditions may prove the true differential competitive factor by enabling the translation of all source of knowledge into successful innovation and economic growth. The analysis of the individual sub-components of the social filter uncovers the specific importance of the educational endowment of both the population and the labour force (regressions 4 and 5). By contrast, other social filter variables are not statistically significant when autonomously introduced in the model (regressions 6–9). Accessibility to extra-regional innovation, our proxy for knowledge spillovers, is related in a positive and statistically significant way to regional growth performance, in particular where associated to an appropriate measure for socio-economic conditions. This confirms that knowledge spillovers, by increasing the “amount of knowledge” available in the region, reinforce the effect of local innovative activities, and, up to a certain extent, they may even compensate for a weak contribution of the innovative activities pursued locally. Thus, other things being equal, a region with an innovative neighbourhood is more advantaged than one located close to less innovative areas. On the contrary both the socio-economic endowment (regr.11) and the level of economic wealth (regr.12) of neighbouring regions seem to have no significant effect on local economic performance. The extra-regional social filter is significant only when considered jointly with “endogenous” features, as in regression 10 where the total accessibility to innovation prone space is considered by including in a single variable both the region’s features and that of its neighbourhood. On the basis of these results the potential of a region in terms of economic performance is enhanced by local R&D investment but only maximized when an appropriate social filter (a better predictor of economic growth) is in place. The

4.4 Conclusions

71

reception of R&D spillovers from neighbouring regions is an important additional source of advantage which, in any case, requires an appropriate social infrastructure in order to be productively translated into new innovation and economic growth.

4.4

Conclusions

In this chapter – by following the combination of the theoretical insights of various strands of literature developed in Chap. 2 i.e., by focusing upon different, though complementary, aspects of the process of innovation – we have been able on the one hand, to explain the role of a broad set of socio-economic contextual factors in the process of innovation (not limiting ourselves to human capital accumulation as in the previous Chapter) and, on the other hand, to overcome the case-study approach of many existing empirical analyses that aim at addressing these factors. In this perspective it has been possible to analyse the regularities of the innovative dynamics of the EU regions and their “social-filter” preconditions, and draw a quantitative map of the EU-wide sources of competitive disadvantage. Furthermore the empirical analysis has confirmed, in a regression framework, the importance not only of internal innovative efforts but, even more so, of local socio-economic conditions for the genesis and assimilation of innovation and its transformation into economic growth across European regions. Our results show that, while innovative output spills over into interconnected areas, the regional system of innovation only benefits the local economy. As a consequence, internal socioeconomic conditions are crucial for the successful translation of innovation into economic growth, as the lack of an adequate system of innovation cannot be compensated for by external factors. In other words, our results depart from a neo-Schumpeterian approach and support the idea that the capacity of the local population to assimilate whatever research is being generated locally or in neighbouring regions and transform it into innovation and economic activity matters more than the threshold of expenditure. This piece of evidence has important regional policy implications. When innovation is recognized as the key source of sustained economic growth, the mechanics of its contribution to economic performance becomes crucial for an effective policy targeting. Policies based on innovation may deliver, at the regional level, very different results, according to the possibility of simultaneously benefiting from endogenously and externally produced knowledge and favourable underlying socioeconomic conditions. Referring to Fig. 4.6 we can imagine two alternatives strategies for the development of lagging, less innovative areas. A first strategy, ideally represented by the white arrow, might pursue the objective of increasing the innovative competitiveness mainly relying upon an increase in R&D activities, expecting contextual conditions to adjust autonomously. The empirical evidence discussed in this chapter suggests that a more effective innovation-based development strategy should, instead, follow the pattern stylized by the black arrow in the graph: address social-filter conditions and stimulate

72

+

Social Filter Conditions

Fig. 4.6 Alternative innovation-based economic development strategies

4 The Role of Underlying Socio-Economic Conditions

Sub-optimal research efforts

Optimal conditions

In adequate conditions and innovative efforts

Detachment between R&D and local conditions

-

R&D expenditure

+

innovative activities jointly. In less innovative regions specific policies are necessary in order to address the underlying social filter conditions. The local endowment needs to be reinforced in order to improve the local capability to produce and assimilate innovation thus also guaranteeing the greatest return from policies based on R&D. Even if the direct relationship between investment in R&D and regional growth has not always proven to be robust after controlling for social conditions, investment in R&D itself remains an important determinant of local capacity to “absorb” external knowledge flows. This is even more important in a “difficult learning environment (such as that of poor social filter regions, ideally close to the origins of the axes in Fig. 4.6) which increases the marginal effect of R&D on absorptive capacity” (Cohen and Levinthal 1990, p. 140). Our results also suggest that both local capability to produce growth-enhancing knowledge and “absorptive” capacity are crucially influenced by the contextual socio-economic conditions. Following this line of reasoning, innovation-based policies should be targeted towards the improvement of the regional system of innovation conditions and should stimulate R&D activities as part and parcel of a balanced regional development strategy. Such development strategy will also need to address the dynamic nature of the regional system of innovation – in terms of its institutional and relational dimension – of which our empirical analysis has been able to analyse only the socio-economic pre-conditions by means of the “social filter” concept. While the next chapter will analyze in further detail the spatial scope of knowledge spillovers in order to fully develop the third component of our integrated approach (i.e., by looking at the role of geography), Chap. 6 will show the empirical relevance of these concepts for the understanding of regional growth not only in Europe but also in the United States. The policy implications of

4.4 Conclusions

73

this approach will be developed in particular in Chap. 8 which will directly investigate to what extent the EU regional development policies are able to address the sources of socio-economic disadvantage as suggested by the empirical evidences discussed so far.

.

Chapter 5

Knowledge Flows and Their Spatial Extent

5.1

Introduction

The empirical evidence produced in Chap. 4 has suggested that not only innovative activities pursued locally affect regional economic performance but also those pursued in neighbouring regions play a positive and significant effect. On the basis of this evidence we analysed how different indigenous socio-economic conditions may result in a differential capability to translate exogenously produced knowledge into regional growth. In this chapter, by building upon the analysis pursued so far, we focus our attention on the spatial extent of such knowledge spillovers. In other words we aim at assessing if knowledge spillovers tend to decay with distance and to what extent. The objective is to shed new light on the role of geographical distance in the process of innovation by supporting the idea of there being a “continuing tension between two opposing forces” (Storper and Venables 2004 p. 367): the increasingly homogeneous availability of standard “codified” knowledge and the spatial boundedness of “tacit” knowledge. In this sense we test the hypothesis that technological improvements in “communication infrastructures” have not affected all kinds of information in the same way. While “codified information” can be transmitted over increasingly large distances, “tacit” knowledge is geographically bound thus determining the increasing concentration of innovation and the geographical boundedness of knowledge spillovers (Audretsch and Feldman 2004; Cantwell and Iammarino 2003; Fischer et al. 2009a, b; Guerrieri et al. 2001; Sonn and Storper 2008). The evidence of the spatial boundedness of knowledge spillovers not only contradicts the idea of a ubiquitous knowledge evenly available everywhere but also helps explaining how peripherality can persistently hamper regional innovative capacity after controlling for indigenous innovative efforts: the smaller the spatial extent of knowledge spillovers, the lower the exposition of peripheral areas to externally produced knowledge. While highly-accessible core regions can benefit from innovative activities pursued in their proximity the spatial boundedness of spillovers prevents them from reaching peripheral remote regions. As a

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_5, # Springer-Verlag Berlin Heidelberg 2011

75

76

5 Knowledge Flows and Their Spatial Extent

consequence the stronger the spatial decay of the spillovers, the greater their tendency to develop localised pools of knowledge in central locations.1 This chapter is organized into three further sections. In the first section the theoretical framework of the analysis and the evidence produced so far are briefly re-examined. The second part discusses new empirical results. The final section concludes.

5.2

Contextualising the Analysis of Knowledge Spillovers into an “Integrated Framework”

As extensively discussed in previous chapters, the “integrated framework” for the analysis of regional growth put forth in this book relies upon the combination of three complementary factors: internal innovative efforts, socially and territorially embedded factors, and spatially-bound knowledge spillovers. In line with this approach we proposed a simple empirical model which combines the key factors from these three approaches in the study of the process of innovation and of how innovation influences economic growth. The model has aimed at understanding – and, to a certain extent, discriminating between – the role of the different innovation factors in order to generate economic dynamism in the regions of the EU-25. As discussed in Chap. 2, the model combines inputs in the innovation process (R&D expenditure) with the socio-economic local factors that make the presence of favourable regional systems of innovation more likely and controls for the wealth of European regions. These factors are considered locally, i.e., the R&D and the local conditions in the region, and externally, i.e. innovative efforts pursued in neighbouring regions. Several implications can be extracted from the results of the empirical analysis pursued so far. For the EU regions investing in R&D seems to be an important source for economic development (Chap. 3). However, relying exclusively on local R&D inputs is not a guarantee for achieving greater growth, as such relationship proves to be not always robust when the social filter variable is introduced (Chap. 4). The local socio-economic conditions – the “social filter” – are a better predictor of economic growth than investment in R&D. These results highlight that while investing in R&D locally enhances economic growth, relying of knowledge spillovers is a viable alternative for regions with adequate socio-economic structures that would guarantee the reception and assimilation of those spillovers (see Table 4.4, regression 3). This does not mean that local innovative efforts are unimportant for regional 1 The indicator of accessibility to innovation used in this book is purely geographical. While we acknowledge that geographical distance may neither be a sufficient, nor a necessary condition for the assimilation of spillovers, and cognitive, organizational, social, and institutional proximity play an important role in the diffusion of knowledge (Boschma 2004; see also queryIammarino and McCann 2006), the quantitative nature of the analysis prevents us from focusing on these other forms of proximity. Hence we measure the geographical distance between different socioeconomic structures in regions, but not the social distance between these same structures.

5.2 Contextualising the Analysis of Knowledge Spillovers

77

economic performance thanks to their impact upon local absorptive capacity (Cohen and Levinthal 1990). However, as far as knowledge may flow also from outside the region (both in the form of codified knowledge and spillovers), local socio-economic conditions may prove the true differential competitive factor by enabling the translation of all source of knowledge into successful innovation and economic growth. This set of results is particularly important when assessing the specific role of knowledge spillovers and their spatial boundedness: accessibility to extra-regional innovation, our proxy for knowledge spillovers, is related in a positive and statistically significant way to regional growth performance, in particular when associated to an appropriate measure for socio-economic conditions (Table 4.4, regg. 3–6 and 10–12). This confirms that knowledge spillovers, by increasing the “amount of knowledge” available in the region, reinforce the effect of local innovative activities, and, to a certain extent, may even compensate for a weak contribution of the innovative activities pursued locally. Thus, other things being equal, a region within an innovative neighbourhood is more advantaged than one located close to less innovative areas. On the basis of these results, the potential of a region in terms of economic performance is maximized when an appropriate social filter is combined with local investment in R&D. The reception of R&D spillovers from neighbouring regions is an important additional source of advantage which, in any case, requires an appropriate social infrastructure in order to be productively translated into new innovation and economic growth. In this framework the analysis of the spatial scope of such spillovers becomes particularly important for the understanding of the role of geography in a knowledge based economy. The understanding of the spatial scope of knowledge spillovers is extremely relevant from both a theoretical and a political point of view. Even if, as discussed in Chap. 2, a variety of contributions provided significant evidence in support of the role of proximity as a relevant condition for the transmission of knowledge, in a recent “state of the art” of the research on geographical knowledge spillovers, D€ oring and Schnellenbach (2006) highlight that “no consensus is reached about the spatial range that can be attributed to knowledge spillovers, and in fact the majority of studies refuse to quantify the range at all” (p. 384). Since the seminal work by Anselin et al. (1997) on the influence of the location of universities and private R&D facilities on local innovative productivity, the spatial extent of knowledge flows in the US has been extensively analysed. Acs (2002 Chap. 3) compares the results of a number of earlier studies based on different estimation techniques and concludes that university research spills over a range of 50 miles from the innovative MSA, while the spillovers from private R&D tend to be contained within the MSA itself. Even if such results adjust downward the 75 mile radius previously measured by Varga (2000), for the US case, the range 50–75 miles seems to provide a “consolidated” measure for the geographical extent of knowledge spillovers. In contrast, at the EU level, the scarcity (and heterogeneity) of research efforts in this direction have a priori prevented the formation of any consensus. Greunz (2003) finds a positive and significant effect on local patenting activity of innovative efforts pursued in the first and the second order neighbouring regions (190 miles or 306 km on average). The magnitude of this effect sharply decreases at the third order neighbourhood

78

5 Knowledge Flows and Their Spatial Extent

(274 miles or 441 km on average) and is no longer significant thereafter. Bottazzi and Peri (2003) find evidence of a spillover effects, with a positive impact of neighbouring regions’ R&D efforts on local productivity, only within a 200–300 km limit. In the same analytical vein, Moreno et al. (2005a, b) estimate a similar spatial scope of regional spillovers: “innovative activity in a region is positively related to the level of innovative activity in regions located within 250 km of distance, but no further” (p. 7). Our analysis helps filling the existing gap in the empirical literature on the measure of the spatial extent of regional spillovers in the EU by including the regions of the entire EU25. In addition, our empirical analysis, while delivering comparable results, differs from previous studies in that: (a) It is not based on a Knowledge Production Function but on a regional growth model thus capturing the effects of neighbouring regions’ innovative efforts on the overall productivity of the regional economy rather than on the production of innovative output only (b) Distance is introduced into the model by means of a (time-based) trip-length measure which captures more accurately the differential quality of connections between regions (c) The model explicitly accounts for the underlying socio-economic conditions

5.3

Measuring the Spatial Extent of Knowledge Spillovers

5.3.1

Model Specification

In this section we estimate a reduced form of the model outlined in the previous chapter in order to provide a measure of the spatial extent of knowledge spillovers. In order to reach this objective we employ different spatial weights in the calculation of the spillovers variable. The full estimated model is:   1 Yi;t ¼ a þ b1 lnðyi;t
J Þ þ b2 RDi;t
j þ b3 SocFilteri;t
J þ b4 Spillovi;t
j ln J Yi;t
J þ b5 D þ e where: 1 J

 ln

Yi;t Yi;t
J



a lnðyi;t
J Þ RDt
j SocFilteri;t
J

Is the usual logarithmic transformation of the ratio of regional per capita GDP in region i at the two extremes of the period of analysis (t
J,t) Is a constant Is the log of the GDP per capita of region i at the beginning of the period of analysis (t
J) Is expenditure in R&D as a % of GDP in region i at time (t
J) Is a proxy for the socio-economic conditions of region i representing its “social filter” (continued)

5.3 Measuring the Spatial Extent of Knowledge Spillovers Spillovi;t
j D e

79

Is a measure of accessibility to extra-regional sources of innovation Is a set of national dummy variables Is the error term

Initial level of GDP per capita – As customary in the literature on the relationship between innovation and growth, the initial level of the GDP per capita is introduced in the model in order to account for the region’s stock of existing knowledge. R&D expenditure – The percentage of regional GDP devoted to R&D is a measure of the economic input in order to generate innovation in each region. Social Filter – As a proxy for the socio-economic conditions we introduce in the model the “social filter index” discussed in Sect. 4.2.1 and calculated by means of the Principal Components Analysis. Spillovers – This variable proxies the capability of exogenously produced innovation to influence regional economic performance in addition to “endogenous” innovation. For this purpose a measure for the accessibility of “extra-regional” innovative activities has been developed and introduced in the analysis by mean of a standardised “accessibility of innovation index”. The index is a potential measure of the “innovative activities” (in terms of nationally weighted millions of Euros invested in R&D activities) that can be “reached” from each region at a “cost” which increases with distance as discussed in further detail in par.4.3.1. However, in this Chapter the “accessibility to innovation index” will be calculated for different cutoff points in the spatial weighting matrix thus allowing to measure the spatial extent of the spillovers effect. The estimation procedures and the dataset remain the same as in Chap. 4.

5.3.2

The Spatial Extent of Innovative Spillovers: Empirical Evidence

In what follows, we focus in greater detail on the relevant “spatial scale” for the transmission of growth-enhancing knowledge spillovers, by attempting to quantify the concept of “proximity” for the regions of the EU-25. In Table 5.1 we present various estimations of our empirical model in which regional spillovers’ proxies are calculated by means of different “spatial weights”. As in the case of the regressions presented in Chap. 4 all usual diagnostic statistics confirm the robustness of our results. In particular the Moran’s I test (computed for each specification of the model) does not detect spatial autocorrelation in the residuals, confirming – as in Chap. 4 – that the combination of “national” dummy variables and spatially lagged explanatory variables are able to capture a significant part of the total spatial variability of the data. Regression 1, which we use as a benchmark, shows our estimation results when regional spillovers are proxied by the index of accessibility to extra-regional innovation as in all regressions in the previous table. The regression not only

0.12182*** (0.02796)
0.00663 (0.003543) 0.1791 (0.1218) 0.010787** (0.004598)

1

0.134*** (0.02838)
0.007635** (0.003612) 0.1486 (0.1194) 0.01074** (0.004579)

2

0.002556 (0.004712)

x 0.672 0.631 16.7
0.0189041

x 0.674 0.634 16.89
0.0196286

x 0.666 0.625 16.25
0.0186123

x 0.666 0.625 16.28
0.019055


0.005154 (0.007263)

0.12551*** (0.02844)
0.005813 (0.003537) 0.1475 (0.1211) 0.010379** (0.004638)

4

0.006191 (0.004619)

0.010656** (0.004538)

0.12176*** (0.02799)
0.005661 (0.003506)

6

0.006103 (0.004628)

0.010685** (0.004538)

0.1216*** (0.02799)
0.005642 (0.003505)

7

0.010782** (0.00455)

0.12116*** (0.028)
0.005572 (0.003506)

8

0.005447 (0.004506) x x x x 0.665 0.666 0.666 0.665 0.626 0.627 0.627 0.627 17.27 17.34 17.33 17.28
0.0189909
0.0192397
0.0191901
0.0189931

0.005349 (0.004505)

0.01081** (0.00455)

0.12107*** (0.028)
0.005554 (0.003506)

5

0.009091** (0.004518)

0.09202*** (0.02533)
0.001913 (0.003168)

10


0.000643 (0.004707)

0.08063*** (0.02512)
0.000093 (0.003078)

11

0.09103*** (0.02533)
0.001779 (0.003168)

12

0.00836* (0.004402) x x x x 0.652 0.653 0.644 0.652 0.615 0.616 0.606 0.615 17.46 17.55 16.84 17.47
0.0188665
0.0191502
0.0165446
0.0188604

0.008264* (0.004401)

0.09082*** (0.02532)
0.001745 (0.003166)

9

*, ** and *** denote significance at a 10%, 5% and 1% level respectively. SE in parentheses. 166 observations in all regressions

National Dummies R-Sq R-Sq (adj) F Moran’s I

600 min cut-off

300 min cut-off

180 min cut-off

3

0.12317*** (0.02822)
0.006016* (0.003571) 0.1458 (0.1211) 0.01101** (0.004724)

Total (Extra þ Intra regional) accessibility to Innovation Continuous Space

600 min cut-off

Accessibility to ExtraRegional Innovation Continuous Space 0.01387* (0.008031) 180 min cut-off 0.00983** (0.00481) 300 min cut-off

Social Filter Index

R&D expenditure

Log GDP 95

Constant

Dependent Variable: Regional Growth Rate (1995–2003)

Table 5.1 H-C OLS estimation of the empirical model: accessibility to innovation

80 5 Knowledge Flows and Their Spatial Extent

5.4 Conclusions

81

confirms that knowledge flowing from neighbouring regions improves regional growth performance, as was underlined before, but also shows that spillovers are geographically bounded and that they decay with distance. The weighing mechanism on which the variable is based makes the importance of other regions’ innovative activities decrease with distance thus emphasizing the effect of innovative activities pursued in neighbouring regions. More precisely, regions can rely on the research strength of regions within a three hour drive (ca 200 kms) as shown by the increase in significance of the spillover variable once a 180 min cut off is introduced in the weighing matrix (Regression 2). When more remote regions are taken into consideration, by fixing the cut off trip length at 300 and 600 min (Regressions 3 and 4 respectively), the variable is no longer significant thus showing that beyond a 180 min trip-time the returns to extra-regional innovative activities are inexistent. Such measure for the spatial extent of regional spillovers is in line with the empirical evidence produced so far (as discussed above). However, trip-length distance has allowed a more accurate measure of distance as a barrier to human interactions across geographical space. These results are confirmed also where total accessibility to innovative activities is considered by introducing a variable capturing both internal and distance-weighed R&D expenditure (Regressions 5–12). In this second case the “institutional” borders of the region are overcome by focusing upon a “continuous” space which results from the aggregation, in an individual variable, of the total R&D expenditure that can be reached from a certain location regardless of regional borders. In doing this, we aim to measure the total impact of R&D agglomeration on economic performance. Our results show once again that only the variables combining the strength of internal efforts with those pursued in more proximate (within the 180 min limit) areas produce a positive and significant effect on regional growth performance. The 180 min limit for interregional knowledge flows is in line with the idea of a “human-embodied” transmission technology since it allows the maximization of face-to- face contacts between agents. Agents within driving distance one from another can exchange their information face-to-face (potentially) on a daily basis, at much lower marginal cost in comparison to those where an overnight stay is necessary (Sonn and Storper 2008).

5.4

Conclusions

The objective of this chapter has been to “measure” the importance of proximity for the transmission of economically productive knowledge. The results highlight that not only knowledge flowing from neighbouring regions improves regional growth performance, but also that spillovers are geographically bound and that there is a strong distance decay effect, which in the European case expands to more or less a 200 km radius. These outcomes shed additional light on the role of geography in the process of innovation, and confirm the existence of a tension between two forces:

82

5 Knowledge Flows and Their Spatial Extent

the increasingly homogeneous availability of standard “codified” knowledge and the spatial boundedness of “tacit” knowledge and contextual factors. The analysis also has important regional policy implications. When innovation is recognized as the key source of sustained economic growth, the mechanics of its contribution to economic performance becomes crucial for an effective policy targeting. Policies based on innovation may deliver, at a regional level in Europe, very different results, according to the possibility of every region of benefiting from knowledge spillovers (location advantage) and favourable underlying socioeconomic conditions (internal conditions). Thus R&D investments in core regions, which benefits from both a location and social filter advantage, are overall more conducive to economic growth due to their impact on both local and neighbouring regions’ performance. Conversely, in peripheral regions investment in R&D may not yield the expected returns. Their limited R&D investment capacity, their inadequate social filters, and their lower exposure to R&D spillovers, because of their location, are likely to undermine the R&D efforts conducted within the borders of these regions. Does this mean that it is not worth investing in innovation in these regions? The results presented in this chapter indicate that very different policies from those of the core may be needed in order to render peripheral societies in Europe more innovative. These policies will need to rely less on R&D investment and much more on removing the local social and economic barriers that prevent the reception and assimilation of external innovation. Any incentive for local innovative activities would have to be complemented by reinforcing the local endowment in terms of education and skills in order to guarantee the greatest returns from innovation policies. The emphasis on skills is also likely to set the foundations for a future transformation of these regions into innovation prone societies, in which the returns of any investment in R&D will yield substantially higher results than at present. These kinds of policy objectives are line with some of the policy recommendations produced by Dosi et al. (2006) in their – though very general and “aggregate” – comparison between the innovative performance of the United States and the European Union. When assessing the critical factors to be addressed by EU policies in order to effectively contribute to filling (or at least reducing) the innovation gap with the US, they suggest a reducing policy emphasis on physical infrastructures and “networking” and, conversely, devoting more attention to encouraging “high quality basic research”, on the education system as also on the diffusion of “open research results”. Overall investments in all these directions would contribute to improve on the one hand the general social filter conditions of the European Union and the diffusion of knowledge flows on the other.

Chapter 6

Innovation In an Integrated Framework: A Europe-United States Comparative Analysis

6.1

Introduction

In the first part of this book we have shown how different streams of literature – the linear model, the systems of innovation approach and the geographical analysis of the diffusion of knowledge spillovers – can be effectively combined into an “integrated” analytical framework, providing us with a more complex and perhaps realistic view on the territorial determinants of innovation and economic growth. This chapter is aimed, on the one hand, at further developing the “integrated framework” discussed so far by explicitly including into the picture specialisation and agglomeration processes and, on the other hand, at using this framework as a “common ground” to compare the drivers of innovation (and their geography) in Europe and in the United States. Section 2.7 has extensively discussed how different specialisation patterns can give rise to differentiated combinations of MAR and Jacobian externalities that, in turn, are supportive of different innovative trajectories. As an example, different stages of products’ life-cycles may require differentiated combinations of such externalities, making certain regions more “attractive” than others for the location of innovative activities. In addition, shifts in the technological frontier may affect the sustainability of existing specialisation/diversification patterns. This important dimension of the capability of regions and territories to adapt to technological development will be fully incorporated in the analytical framework developed in previous chapters and put in relation with agglomeration processes and factor mobility. Specialisation and agglomeration patterns play a particularly important role for the understanding of the different determinants of innovation in different contexts such as Europe and the United States. In addition, a number of “macro” structural factors contribute to the overall innovative performance of the two continents. Existing analyses have mainly addressed the differences between the two continents in terms of the major “inputs” to innovation, such as R&D investments, the level of human capital accumulation, the structure of the educational system and the capacity to generate and retain top-level scientists. Other analyses emphasise the ways that organisational and institutional settings shape the use of such innovative

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_6, # Springer-Verlag Berlin Heidelberg 2011

83

84

6 Innovation In an Integrated Framework

inputs. Even though these factors account for a large part of differential innovative performance, it stands to reason that their spatial organisation may also play a role, as extensively discussed in previous chapters for the case of Europe. The spatial organisation of the sources of innovation determines levels of localised economies of scale and localised knowledge externalities, and may in this way affect the level of innovative output (see Sect. 2.5). Studies of the geography of innovation have indeed emphasised agglomeration of innovation inputs, proximity effects, and knowledge spillovers (see Sect. 2.6). Much less attention has been devoted to the dynamic process through which agglomeration and specialization are constructed and reconstructed over time, through the spatial mobility and matching of innovative factors. This process, or flow, gives rise to specific levels and types of proximity and spillovers (see Sect. 2.7). This chapter aims at focusing on the “territorial dynamics of innovation” in the EU and the USA. These two areas of the world are characterised by a different historical geography of innovation systems. Behind such histories, we argue, there are different contemporary institutions, rules, and incentives governing the creation and geographical mobility and combination of such inputs to innovation. In order to analyse the geographical processes underlying innovation, this chapter will further develop the model of quantitative analysis presented in previous chapters in order to fully incorporate these territorial dynamics in conjunction with the factors evaluated in standard models. The analysis shows that the higher mobility of capital, population, and knowledge in the economically- and culturally-integrated US market enables combinations of factors that respond rapidly to shifts of the technological frontier, and allow full exploitation of local innovative activities and (informational) synergies. In the European Union, by contrast, imperfect market integration, and institutional and cultural barriers across the continent produce a spatial configuration both less dynamic and less coherent at the local level.

6.2

The Innovation Gap Between Europe and the United States and Its Structural Determinants

In 2000, the Conclusions of the presidency of the Lisbon European Council established the goal of making the European Union the “most competitive and dynamic knowledge-based economy in the world”. In so doing, they explicitly acknowledged the gap separating the EU from the present world leader – the USA – and announced their intention to catch up within a ten-year period. A year earlier, the President of the United States had set the goal of “maintaining world leadership in science, mathematics, and engineering” (NTSC 1999). Nowadays, a decade later, US leadership is still undisputed and – as highlighted by the International R&D scoreboard published by the UK Department of Trade and Industry in October 2006 – the transatlantic technology gap has widened. Though there are limitations of the available proxies for innovative output, all standard indicators reveal a significant disadvantage in the innovative capacity of

6.2 The Innovation Gap Between Europe and the United States

85

the European Union. When considering scientific activity, such as the number of scientific publications and citations weighted by population (as reported in Dosi et al. 2006 using OECD data), the output gap between the EU-15 and the US is immediately apparent, with 4.64 publications per thousand inhabitants in the US versus 3.6 in the EU-15 in the 1997–2001 period. This gap is even wider when the impact of such scientific production is assessed in terms of article citations (39.75 per thousand inhabitants in the US against 23.03 in the EU-15) or shares in the top 1% most cited publications (0.09 top 1% publications per thousand inhabitants in the US against 0.04 in the EU-15). When considering technological output, the United States shows the best innovative performance, as measured by its share of the total triadic patent families1 (36.4% in 2003 against the 30.3% of the EU). When triadic patent families are weighted by population, US patent intensity is 47% higher than for the EU-15 and almost double that of the EU-25 (our calculations based on OECD 2006). In spite of the magnitude of this innovation gap and the political emphasis attached to it on both sides of the Atlantic, very little systematic comparative analysis has been pursued on its causes. The innovation output gap between Europe and the USA is most frequently attributed to differences in inputs to innovation production. The quantity and quality of inputs, as well as the broader “innovative infrastructure” in the two contexts – by reflecting their cultural, institutional, and economic diversity – are indeed substantially greater in the United States. First, the total amount of the resources devoted to innovative activities varies significantly. In 2004, 1.9% of GDP was spent on Research and Development in the EU-25 (1.95% in the EU-15) (Eurostat 2006a) compared to 2.6% of GDP (NSF 2006) for the US. Furthermore, the nature of such expenditure differs considerably. On the one hand, as Dosi et al. (2006) point out, “the usual claim concerning the higher share of publicly funded R&D in the EU as compared to the US is simply groundless”(p. 1456): government financed R&D expenditure as a percentage of GDP was 0.66% in the EU-25 against 0.70% in the US (in 2003 and 2004 respectively). However, a large percentage of this US public expenditure is for R&D carried out by private firms, about double the corresponding figure in the EU. American private firms not only benefit from a larger share of public funds than their EU counterparts, they also devote a higher proportion of their internal resources to R&D. Industry-financed R&D expenditure is about 1.9% of GDP in the US (NSF 2006), but only around 1% of GDP in the EU (Eurostat 2006a). Second, the gap in human resources devoted to R&D is large and significant: “in 2003, the number of researchers (in full-time equivalents) per thousand of the labour force amounted to only 5.4 in the EU against 10.1 in Japan and 9.0 in the US. This EU deficit is mainly located in the business sector” (European 1

“A patent is a member of the triadic patent families if and only if it has been applied for and filed at the European Patent Office (EPO), at the Japanese Patent Office (JPO) and if it has been granted by the US Patent and Trademark Office (USPTO)” (Eurostat 2006: 6). Patent families are supposed to improve international comparability by suppressing the home advantage.

86

6 Innovation In an Integrated Framework

Commission 2005a, p. 6). Furthermore the advantage of the US in this area is not only “quantitative” but also “qualitative”, as the US attracts and retains a large proportion of high impact researchers: of the top 1,222 most cited individuals in 14 scientific fields, 66% live and work in the US, while only 20% are from the sum of the EU countries (Batty 2003). This situation in the research sector reflects a more general trend in humancapital accumulation, which, in turn, is the result of different structures and levels of investment in the educational system. In 2004, only 34.1% of all 20 year olds were enrolled in higher education in the EU-25 (and 33.4% in the Euro zone), in comparison to 46.2% in the US (European Commission 2005b). Furthermore, public and private investment in education is significantly higher in the US than in the EU: in 2003 the expenditure per student in Higher Education – adding public and private expenditure – was just 39.3% that of the US in the EU-25 and 41.1% in the Euro zone2 (Eurostat data). The combination of higher US investment in higher education with existing gaps in R&D investment and in the capacity to generate, attract, and retain top scientists is reflected in a growing gap in the standing and influence between American and European universities. According to the ranking produced by Shanghai Jiao Tong University, among the top 20 universities in the world 17 are in the US and only two in the EU (both in the UK)3 (Institute for Higher Education, Shanghai Jiao Tong University 2006). The two continents also show marked differences in the institutions and policies governing the invention, development, and adoption of new technologies. “The foundations of the US ‘national system of innovation’ were largely put in place during 1945–1950 (when) demobilization for peace was replaced by Cold War rearmament” (Mowery 1998, p. 640). In contrast, despite the recent and rapid formal institutional-building efforts at EU level, there is yet no analogous Europe-wide system in place, i.e., which could complement and integrate existing national systems in the way the US institutions do (Gregersen and Johnson 1996; Borra´s 2004; Stein 2004). While the US’s integrated (though decentralized) system was forged by the implementation of consistent innovation policies with large-scale federally-funded projects largely benefiting private firms and basic research, the European innovation system still suffers from fragmented, small-scale projects and highly bureaucratic government policies. As a result, the US national innovation system seems oriented towards technological “shifting” rather than “deepening”: radical innovations are more easily achieved in the US, because it is capable of rapidly reallocating resources in line with the requirements of new technological paradigms (Ergas 1987). The “shifting” capacity of the US system is supported by the way its research universities develop complex interactions with the world of € 8,049.5 in the EU-25 and € 8,422.6 in the Euro Zone vs. € 20,487 in the US, measured in PPS, based on full-time equivalents. 3 When the ranking is extended to the top 100 universities we find that 57 are in the USA and 35 in the EU (of which 11 in the UK). The ranking of the top 500 universities in the world is based upon a variety of performance indicators (see http://ed.sjtu.edu.cn/rank/2005/ARWU%202005.pdf for further details). 2

6.3 Geography and Innovative Performance in the EU and the US

87

business. Furthermore the antitrust and intellectual property regulatory frameworks seem to offer a fertile environment for the marketing of new technologies4 (Hart 2001). This generally more favourable business environment allows a more effective adjustment of the sectoral structure of the economy with the prompt startup of new firms in response to new market opportunities in emerging sectors. The higher degree of specialisation in R&D intensive sectors and the stronger presence of small and medium-sized R&D intensive firms in the national market (Smith 2007) are both causes and consequences of the US “shifting” capacity, in a circular process supported by the intrinsically different system of innovation conditions. Conversely, EU firms, on average, have weaker entrepreneurial culture and greater resistance to organizational change (Delmas 2002). Major constraints also arise from barriers to the access to venture capital (a major source of funds for US innovation) and European labour market rules, which frequently lead to mismatches between staffing patterns and the true demand for skills, since they slow down recomposition of staff in response to technology and market shifts. All these factors directly influence innovative performance. But in addition to this, they also lead to different patterns of spatial organisation of innovation in each continent. As we will argue in the next section, these territorial dynamics further differentiate the rate and direction of innovation in the US and the EU.

6.3

Geography and Innovative Performance in the EU and the US

The spatial distribution of innovative output in both Europe and in the United States, as proxied by patents, exhibits a strong tendency towards disproportionate concentration in a few locations: “during the 1990s, 92% of all patents were granted to residents of metropolitan areas, although these areas account for only about three-quarters of the US population, and for about 20% of land area of the continental United States” (Carlino et al. 2001, p. 1). In the EU, patenting is “highly concentrated” as well (Eurostat 2006b). The cumulative percentage of total patents recorded by the 100 most innovative EU-15 regions and US MSAs (Fig. 6.1) is similar in the two continents and in both contexts the twenty most innovative regions account for around 70% of total patents. The literature on the geographical determinants of innovation in the two continents has focused on the role of agglomeration and the density of economic interactions as the key catalysers of innovation. Agglomeration and density are indeed relevant forces behind a variety of economic processes. As illustrated by Ciccone and Hall (1996), average labour productivity is significantly greater where employment density is higher. In line with this approach, different studies have found that agglomeration increases innovative output even after controlling for differences 4

Though the effects of the 1982 patenting system reform are debated (see Jaffe and Lerner 2004).

88

6 Innovation In an Integrated Framework

Cumulative percentage (total patents)

20 Variable Cumulative distribution EU Cumulative distribution US

100 80 60 40

33.6 (US) 30.8 (EU)

20 0

100 90 80 70 60 50 40 30 20 10 1 Regions' Ranking (less innovative to most innovative)

Fig. 6.1 Distribution of total patents across regions and MSAs 100 most innovative EU regions and US MSAs

in human capital, high-tech industry structure and R&D university infrastructure, both in the US (Carlino et al. 2004; Sedgley and Elmslie 2004) and in some EU countries (e.g., Andersson et al. 2005, for the case of Sweden). As Ciccone (2002) points out, the “agglomeration effects in European countries (France, Germany, Italy, Spain, and the UK) are only slightly lower than in the US and do not vary significantly across countries” (p. 214). Agglomeration influences economic outputs and innovative performance through a mix of different sources of Marshallian agglomeration economies (labour market interactions, linkages between intermediate and final good suppliers, knowledge spillovers) that is present in each place.5 Following Duranton and Puga (2003), the forces behind agglomeration economies can be broken down into “sharing” (e.g., sharing of indivisible facilities, gains from variety of input suppliers), “matching”, and “learning” mechanisms. The creation, accumulation, and diffusion of knowledge also rely upon different types of coordination enabled by face-to-face contacts (Storper and Venables 2004). Close proximity thus becomes a condition for the dissemination of information, which would otherwise be impossible or too expensive to codify (Charlot and Duranton 2006).

5

Duranton and Puga (2003) use as an example a model “in which agglomeration facilitates the matching between firms and inputs. These inputs may be labelled workers, intermediates, or ideas. Depending on the label chosen, a matching model of urban agglomeration economies could be presented as a formalisation of either one of Marshall’s three basic sources of agglomeration economies even though it only captures a single mechanism” (p. 2).

6.3 Geography and Innovative Performance in the EU and the US

89

From an empirical perspective, it is however difficult to isolate the learning, matching, and sharing components of agglomeration. Empirical analysts have thus had to rely upon indirect (output) measurements of learning and proximity – notably the geography of patenting – as a means to distinguish informational spillovers from the other “Marshallian forces” (Fujita and Thisse 2002). Along these lines, research has established a connection between density and patenting, where density is a proxy for agglomeration and patenting for learning. Yet a simple dense/not dense dichotomy does not adequately capture the potential complexity of the geography of matching and learning processes. Knowledge matching and learning are logical outcomes of the many ways in which agents move, signal, and match to other agents. This means not just “being” in established patterns of proximity to other agents, but also the dynamic process by which such densities and proximities are achieved, and how they adjust over time. Such adjustments refer to the types of agents (who they are and what they bring to the innovation process), in relation to changing technologies, markets, and types of knowledge required to innovate. We call these “flow” dimensions of the problem the “territorial dynamics of innovation” and view its analysis as a complement to what the literature has to say on levels of density, proximity, and innovation. Let us now consider these geographical processes and their socio-institutional underpinnings in greater detail. First of all, rather than thinking of agglomerations one by one, it is helpful to consider their interrelations and connections to other places. Even similarly “dense” economic fabrics may be exposed to external knowledge flows to different degrees, with different levels of knowledge spillovers from neighbouring areas. Even if the use of tacit and highly specialised knowledge is maximised in the “core” of dense agglomerations, some of that knowledge travels ´ cs 2002; Sonn and Storper 2008, more widely (Anselin et al. 1997; Varga 2000; A for the US, and Bottazzi and Peri 2003; Greunz 2003; Crescenzi 2005; Moreno et al. 2005a ; Rodrı´guez-Pose and Crescenzi 2008, for the EU case). This means that core regions may potentially benefit from proximity to other innovative neighbourhoods, such that they form a wider network structure through which knowledge flows are transmitted. Inter-agglomeration knowledge flows are different in the two continents. The higher average population density of the EU, with major metropolitan areas relatively closer together than in the United States (where instead metropolitan areas are farther away from one another), may allow a stronger continentwide circulation of knowledge, and possibly limit the distance decay of useful knowledge. In addition, an agglomeration is not merely a stock of resources; it has a dynamic, consisting of the flow of resources into and out of it (a “churn” or turnover). Migration flows contribute to the creation of new knowledge at the local level, by “increasing” the density of the local skill pool and changing its quality, in terms of the variety of skills and cultures it may contain (De Blasio 2006; Ottaviano and Peri 2006). In the most innovative places, we expect migration to update the matching of knowledge, skills, and competencies in line with the evolution of the technological frontier. On the contrary, where agglomerations are less often re-composed through migration, innovative agents may benefit from proximity relationships but find it

90

6 Innovation In an Integrated Framework

more difficult to dynamically match existing agents to new knowledge producers. Migration trends are crucially influenced by the costs of mobility – in turn affected by issues such as culture, identity, or social and personal links – and by institutional incentives to labour mobility, which are very different in the US and Europe. The degree of (domestic) labour mobility is substantially higher in the United States than in the EU, as extensively documented by Zimmermann (1995 and 2005), Vandamme (2000), and Puhani (2001), for the EU6 and by Peri (2005) in a comparative perspective, and there are considerable differences in foreign in-migration as well. Third, these forces that influence the composition and recomposition of clusters and agglomerations may affect the nature and level of innovation through the types of knowledge they match. If they generate increasing specialisation, then they will likely foster MAR (Marshall-Arrow-Romer) externalities within the same industry; if, on the other hand, they promote diversity, they allow local actors to benefit from knowledge base complementarities and across-industry exchange of ideas (Jacobian externalities). The empirical literature suggests that both MAR (Glaeser et al. 1992; Henderson 1999) and Jacobian externalities (Feldman and Audretsch 1999; Carlino et al. 2001; Andersson et al. 2005) play an important role in fostering innovation, but they do so in different sectors. Henderson (2003) finds that Jacobs-type externalities prevail in high-tech and MAR in capital goods industries. Duranton and Puga (2003) suggest that agglomeration economies play roles in innovation at different phases of the product life cycle: firms develop new products in diversified creative urban contexts, subsequently relocating to specialised cities in the mass production phase in order to exploit cost advantages. Where the broader historical, institutional, and political forces inhibit mobility, they could prevent the cluster from adjusting in such a way that the most efficient combination of the two types of external economies is maintained, thus hampering innovative productivity. This could be the case of Europe, where incomplete economic integration, “national” redundancies, and duplications in economic structures may have lead to a suboptimal pattern of specialisation. Finally, the territorial dynamics of innovation is deeply rooted in a complex institutional process that shapes the capacities and attitudes of the population toward innovation, and distribute these populations in geographical space (Morgan 1997). These capacities and attitudes can be captured empirically as the “social filters” of the local population, i.e., characteristics of people that either favour or deter the development of successful regional innovation systems (Rodrı´guez-Pose 1999, 82).

6

Zimmermann (2005) points out that the EU shows “a split labour market that is characterized by high levels of unemployment for low-skilled people and a simultaneous shortage of skilled workers. This lack of flexible high-skilled workers and the aging process has created the image of an immobile labour force and the eurosclerosis phenomenon (thus preventing) the best allocation of resources and hence economic efficiency” (p. 448).

6.4 The Model: A Modified Knowledge Production Function

6.4

91

The Model: A Modified Knowledge Production Function in an “Integrated” Territorial Framework

The empirical models developed in previous chapters have focused regional growth as their dependent variable and have looked at innovation dynamics as its determinants. In this chapter, the focus of the empirical analysis will be restricted to the process of innovation itself (before its translation into economic growth) in order to shed light on its territorial differentiation in an EU–US comparative perspective. Regional economic performance remains at the “core” of our analysis but the complexity of the comparative analysis suggests to narrow down our focus in order to minimize the “white noise” that inevitably arises when comparing two continents. As a consequence, the empirical analysis pursued in this chapter will depart from the models estimated in previous chapters being based on the Knowledge Production Function (KPF), formalised by Griliches (1979, 1986) and Jaffe (1986)7 where innovative output is the dependent variable of the KPF. However, coherently with previous chapters’ analyses, rather than focusing our attention upon the firm as unit of observation, we adopt a geographical unit (NUTS regions for the EU and MSAs for the US) similar to that of Audretsch (2003), Audretsch and Feldman (1996a), Feldman (1994), Fritsch (2002), and Varga (1998). Though our research questions differ from the existing literature in that we focus upon the “territorial dynamics of innovation” in the two continents, our use of the knowledge production function is very similar to them. All this literature, including the present study, is constrained by the limited availability of comparable data at the sub-national level for the US and the EU. Still, the comparison of the estimated Knowledge Production Function coefficients for the EU and the US allows us to account for the role of technological externalities and other geographical dimensions of innovation in the two areas in a direct and consistent way. The drawbacks of the KPF approach – based on a simplified representation of the generation of innovation and relying upon very rough proxies for both its inputs and output – are compensated here by its capability to deliver clear-cut insights from a comparative perspective. Finally, the “standard” Knowledge Production Function is “modified” in line with the conceptual framework developed in Chap. 2: the linear relationship between innovative output and input (in the form of R&D activities) will be developed to account for innovation systems conditions, spillovers and specialization thus remaining fully coherent with our “integrated” framework. The modified Cobb-Douglass knowledge production function (KPF) takes the form: Ii ¼ AKib RDgi SpillRDdi Czi SpillC
i

(6.1)

where I is the innovative output of region i, A is constant, K is the initial stock of knowledge available in the region i, RD is the knowledge created in the region or

7

For a review of the theoretical and empirical works based on this approach and for a discussion of its limitations see Wieser 2005.

92

6 Innovation In an Integrated Framework

“regional technological activity”, SpillRD is a vector of neighbouring regions’ innovative efforts which may spill over into and contribute to the local production of innovative output, C is a vector of local economic and socio-institutional characteristics, SpillC is a vector of broader socio-institutional characteristics in neighbouring regions. The choice of the proxies for the arguments in function (6.1) is determined according to the following matrix: Internal factors Initial patent applications Investment in R&D in the region

Initial patent intensity R&D Agglomeration economies Specialisation of the local economy Human capital mobility Social filter

National effects

Spillovers Investment in R&D in neighbouring regions

Total regional of state GDP/ Population density Krugman index Migration Similar conditions in neighbouring Structural characteristics that regions would make a region more “innovation prone”, including: 1. Education 2. Life-long learning 3. Sectoral composition 4. Use of resources (unemployment) 5. Demographics National dummies (in the case of Europe) and Geographical dummies (for the US)

By developing this framework, equation (6.1) allows us to specify the following empirical model:
 1 Pai;t ln ¼ a þ b lnðPai;tT Þ þ gRDi;tT þ dSpillRDi;tT Pai;tT T þ z1 SocFilteri;tT þ z2 Migi;tT þ z3 KrugmanIndexi:tT þ z4 Agglomi;tT þ ZSpillSocFilteri;tT þ yDi þ E

(6.2)

where: 

Pai;t 1 T ln Pai;tT



a lnðPai;tT Þ RDtT SocFilteri;tT Migi;tT

Is the logarithmic transformation of the ratio of patent applications in region i at the two extremes of the period of analysis (tT,t) Is a constant Is the log of the initial level of patent applications per million inhabitants at the beginning of the period of analysis (tT) Represents expenditure in R&D as a % of GDP in region i at time (tT) Is a proxy for the socio-economic conditions of region i, representing its “social filter” Denotes the migration balance of region i at time (tT) (continued)

6.4 The Model: A Modified Knowledge Production Function

93

KrugmanIndexi:tt Represents the level of specialisation of local employment of region i at time (tT) Agglomi;tT Is either (the natural log of) population density or the regional percentage of national GDP of region i at time (tT), as proxies for agglomeration economies; Spill Indicates the presence of these factors in neighbouring regions D Denotes a set of national/geographical dummy variables e Is the error term.

Patent growth rate – Patent statistics provide a measure of innovative output (OECD 2006). Their strength is to provide comparable information on inventions across a broad range of technological sectors. However, patent indicators suffer from a number of limitations in their ability to proxy innovation, and hence must be interpreted with care. Such limitations include the heterogeneous value or degree of novelty of patented products or processes; the different propensity to patent across countries and sectors; and the non-patentability of many inventions or the better cost-effectiveness of other protection methods (e.g., secrecy) (OECD 2001; Sedgley and Elmslie 2004). Moreover, for the United States, even ostensibly minor changes in the organisation of patenting institutions, by altering the structure of incentives of all participants in the patenting process (applicants, patent office employees, potential imitators), have produced a substantial inflation in the number of patent applications without any underlying “real” change in inventive performance (Jaffe and Lerner 2004). Despite all these ´ cs and Audretsch (1989) show that regression analyses based on patent caveats, A counts deliver results highly comparable with those based on more direct measures of innovation, thus allowing us to consider the growth rate of patents as an effective proxy for changes in local innovative performance. Furthermore, the use of European Patent Office and U.S.Patent Office patent data for comparative analysis has become common practice in both policy documents (e.g., European Commission 2005a, 2007) and academic work (Dosi et al. 2006), even though Criscuolo and Verspagen (2006) have highlighted the intrinsic differences between the two sources when pursuing more complex analyses, such as patent citation analysis. Initial level of patents per million inhabitants – The initial level of utility8 patents granted/applied for9 in the region is a proxy for the existing technological 8

The majority of patents issued by the USPTO are utility (i.e., invention) patents. Other types of patents and patent documents issued by USPTO, but not included in this report, are plant patents, design patents, statutory invention registration documents, and defensive publications. While in 1999 the number of utility patents granted reached 153,493, just 14,732 design patents, 448 reissue, and 421 plant patents were awarded. Our data do not include these other categories. 9 The USPTO provides data at the sub-state level on utility patents granted from 1990 to 1999 with a first-named inventor who resided in the United States. For the EU, instead, patents are organized by EUROSTAT according to the application years rather than the grant years. However, the US patent data at the national level show that the numbers of patent applications and patents granted are highly correlated over time (0.94 for the period 1989–2002) and across geographical units (0.98 for 1990).

94

6 Innovation In an Integrated Framework

capacity of the area and its distance from the technological frontier. This variable also controls for the differential overall propensity to patent, which reflects different initial sectoral specialisation patterns. R&D expenditure – is the investment in R&D as a percentage of GDP (R&D intensity) as in previous chapters. The value of R&D intensity expresses the relative innovative effort of a region. It is the main input in the knowledge production function. Time lags from R&D expenditure to innovative output in the Knowledge Production Function (KPF) are extremely hard to model in this context. Since the seminal contribution by Griliches an extensive scholarly literature has flourished on the time lag between innovative input and output. The lag changes significantly in different technological areas, sectors and is heavily dependent on the nature of the R&D activities considered (public vs. private, intra vs. extra mural etc) and on the life-cycle stage of specific products. When the KPF is estimated at the firm-level by means of micro-data it is (in principle) possible to model this time lag. However, when aggregate regional-level KPF are estimated (as in our analysis), this is not practically feasible. Given the nature of the available data it is impossible to control for different sectors and different R&D activities. This is customary in the international literature on aggregate regional KPF (Moreno et al. 2005a, b; Oo˜huallachain and Leslie 2007). Spillovers – In the knowledge production function, inter-territorial spillovers contribute to the creation of new local knowledge. For this purpose, in previous Chapters (and in particular in Chap. 5, devoted to the analysis of the knowledge spillovers and their spatial extent) we have developed a measure of the “innovative activities” (in terms of R&D expenditure) that can be “reached” from each region at a “cost” which increases with distance. Consequently, for each region the R&D expenditure recorded in neighbouring regions is weighted by the inverse of their bilateral distances. Coherently with this approach the proxy Spillxi for spillovers from variable xi flowing into region i is calculated as:

Spillxi ¼

X j

P 1 xj dij dij j xj P ¼ P 1 dij j d ij j

8i 6¼ j

(6.3)

where xi is the variable under analysis, dij is the average trip-length (in minutes)/ distance between region i and j. The amount of knowledge flowing from outside the region is thus proxied by the average magnitude of all other regions’ R&D expenditure weighted by the inverse of the bilateral time-distance. The use of distance-weights over the whole EU territory (rather than simple contiguity weights) minimizes the potential bias not only due to the heterogeneity in the number of neighbours of each region but also due to the effect of the EU borders. For the EU the measure of distance is based on the travel time calculated by the IRPUD (2000) for the computation of peripherality indicators and made available

6.4 The Model: A Modified Knowledge Production Function

95

by the European Commission10 and extensively used in previous chapters of this book. We chose road distance,11 rather than straight line distance, as, in particular on a smaller scale, it provides a more realistic representation of the real “cost” of interaction and contacts across space. However for the US, this kind of distance measure is not available and we are forced to rely on straight line distance.12 The aggregate nature of the available data on EU and US regions has not allowed us to distinguish between different categories of knowledge spillovers (universityindustry, inter-industry, intra-industry etc.). In this sense our measure of spillovers captures the combined effect of all these sources/transmission mechanisms (including those discussed by Fischer and Varga 2003 and Varga 2000). Krugman Index – Following Midelfart-Knarvik et al. (2002) we call the index K the Krugman specialisation index, used to measure the specialisation of local employment by calculating: (a) For each region, the share of industry k in that region’s total employment: nki ðtÞ (b) The share of the same industry in the employment of all other regions: nki ðtÞ (c) The absolute values of the difference between these shares, added over all industries: Ki ðtÞ ¼

X

absðnki ðtÞ  nki ðtÞÞ with nki ðtÞ ¼ k

X

xk ðtÞ= j6¼i i

X X k

xk ðtÞ j6¼i i

(6.4)

The index takes the value zero if region i has an industrial structure identical to the rest of the EU/US regions, and takes the maximum value of two if it has no industries in common with the rest of the EU/US. For the US, the Krugman Index has been calculated on the basis of the industry classification system developed for the 1990 census13 which consists of 235 categories for employed persons, classified into 13 major industry groups. For the EU, we rely on the Branch Accounts ESA95 data for employment in NUTS1 and 2 regions, which are based on a 17-branch classification of economic activities (NACE Rev. 1.1 A17) available from 1995 onwards.

10

As the time distance-matrix is calculated either at the NUTS1 or at the NUTS2 level, in order to make it coherent with our data which combine different NUTS levels we relied on the NUTS distance matrix using the NUTS 2 regions with the highest population density in order to represent the corresponding NUTS1 level for Belgium, Germany, and the UK. 11 The distance matrix does not take into account the impact of railway and/or air connections on the average trip length. Only road travel-time is available for the EU regions. 12 Data on distances between MSAs are calculated on the assumption that a 1 degree difference in latitude is constant regardless of the latitude being examined. This assumption is not problematic for smaller countries, but for a large country like the US, it may result in an underestimation of the distance between Southern cities and an overestimation of that between Northern cities. 13 The 1990 census classification was developed from the 1987 Standard Industrial Classification (SIC) Manual published by the Office of Management and Budget Executive Office of the President.

96

6 Innovation In an Integrated Framework

Social Filter – The social filter variable introduced in this model has precisely the same definition and conceptualisation as in previous chapters. The theoretical analyses pursued in Chaps. 2 and 4 (completely devoted to the exploration of this concept) have suggested three principal aspects of the “social filter” of a region: educational achievements (Lundvall 1992; Malecki 1997), productive employment of human resources, and demographic structure (Fagerberg et al. 1997; Rodrı´guezPose 1999). A set of proxies is selected for each domain (given data availability at the regional level and comparability between the EU and the US). Educational attainment is measured by the percentage of population and labour force having completed tertiary education. Participation in lifelong learning programmes is used as a measure for the accumulation of skills at the local level for the EU, while for the US we use the number of people who completed “some college (or associate) level education but no degree” and the number of people with “bachelor’s, graduate or professional degrees”.14 For the second area (structure of productive resources), the percentage of the labour force employed in agriculture was available for both the EU and the US. Long-term unemployment is only available for the EU, thus requiring us to use the US unemployment rate (rather than its long term component). These two variables are used because of the traditionally low productivity of agricultural employment in relationship to that of other sectors, and because agricultural employment, in particular in some peripheral regions of the EU but also in some southern states of the US, is in reality synonymous to “hidden unemployment”.15 The rate of unemployment (and in the case of the EU especially its long-term component) is an indicator of the rigidity of the labour market and of the presence of individuals whose possibilities of being involved in productive work are hampered by inadequate skills (Gordon 2001). Demographic structure is indicated by the percentage of population aged between 15 and 24, with the substantive goal of identifying dynamic demographic trends. Young people contribute towards the renewal of the local society, which should influence the collective attitude towards innovation and social change in general. As in Chaps. 4 and 5, we deal with the problems of multicollinearity, which prevent the simultaneous inclusion of all these variables in our model, by means of principal component analysis (PCA). PCA allows us to merge the variables discussed above into an individual indicator that preserves as much as possible of the

14

The first category includes people whose highest level of schooling is an associate degree (for example: AA, AS) or some college credit, but no degree. The second group includes those whose highest level of schooling is a bachelor’s degree (for example: BA, AB, BS), a master’s degree (for example: MA, MS, MEng, MEd, MSW, MBA); or a professional degree (for example: MD, DDS, DVM, LLB, JD) (US Census Bureau). 15 Unemployment is “hidden” in the fabric of very small farm holdings in many EU peripheral areas and in many Southern states of the US (Demissie 1990; Caselli and Coleman 2001). In both these contexts agricultural workers show low levels of formal education, scarce mobility, and tend to be aged.

6.4 The Model: A Modified Knowledge Production Function

97

variability of the source data as discussed in Appendix C. The output of the PCA is shown in Tables C.1–C.4 in Appendix C for both the US and the EU. The eigenanalysis of the correlation matrix shows that the first principal component alone is able to account for around 40% and 34% of the total variance for the EU and US case, respectively, with an eigenvalue significantly larger than 1 in both cases (Tables C.1 and C.3). The first principal component scores are computed from the standardised value of the original variables by using the coefficients listed under PC1 in Table C.2 and C.4. These coefficients assign a large weight to the educational achievements of the population in both continents; this is a major component of the “social filter” part of our model. In the EU, educational achievement, employed people and, to a lesser extent, the participation in life-long learning programmes complete the index with a positive weight. Both in the EU and the US a positive weight is also assigned to the percentage of young people, but it is significantly more important in the US. A negative weight is assigned, in both contexts, to the rate of unemployment (US) and to its long term component (EU). The percentage of agricultural labour has a negative influence in the EU, while in the US, it has a small but positive influence on the magnitude of the index.16 Agglomeration and economies of scale – The degree of agglomeration of the local economy is proxied by the log of population density, as customary in the literature. In addition, the presence of regional economies of scale is proxied by the relative concentration of economic activities (regional percentage of national GDP). Migration – The degree of internal (EU17 and US) labour mobility is reflected by the regional rate of migration (i.e., the increase or decrease of the population due to migration flows as a percentage of the initial population). A positive rate of migration (i.e., an inflow of people from other regions) is a proxy for the capacity of the region to attract new workers, thus increasing the size of its labour pool and its “diversity” in terms of skills and cultural background.

16

We are aware of the potential endogeneity arising from the introduction of the social filter variable into the Knowledge Production Function. An effective strategy in order to address this issue would imply the use of several time lags of the social filter variables as instruments in an instrumental variables framework. However, due to the constraints on data availability discussed before, we are forced to limit ourselves to considering the value of this indicator at the beginning of the period of analysis, while assessing the patent growth rate over subsequent years. 17 Migration data are provided by Eurostat in the “Migration Statistics” collection. However there are no data for Spain and Greece. Consequently, in order to obtain a consistent measure across the various countries included in the analysis, we calculate this variable from demographic statistics. “Data on net migration can be retrieved as the population change plus deaths minus births. The net migration data retrieved in this way also includes external migration” (Puhani 2001: 9). The net migration was standardised by the average population, obtaining the net migration rate. Consequently, while for the EU it is impossible to distinguish between national, intra-EU, and extra-EU migration flows, for the US domestic in-migration and out-migration data consist of moves where both the origins and destinations are within the United States.

98

6.5 6.5.1

6 Innovation In an Integrated Framework

Results of the Analysis Specific Estimation Issues and Units of Analysis for the Comparative Analysis

We estimate the model by means of heteroskedasticity-consistent OLS (Ordinary Least Square) regressions.18 The effect of spatial autocorrelation (i.e., the lack of independence among the error terms of neighbouring observations) is minimized by including a set of national dummy variables for the EU case and a set of geographical dummies for the US, accounting for the “national fixed effect”. Furthermore, by introducing spatially lagged variables in our analysis, we explicitly consider the interactions between neighbouring regions and thus minimize their effect on the residuals. In particular the Moran’s I test (computed for each specification of the model) does not detect spatial autocorrelation in the residuals, confirming – as in Chaps 4 and 5 – that the combination of “national”/geographical dummy variables and spatially lagged explanatory variables are able to capture a significant part of the total spatial variability of the data. Another concern is endogeneity, which we address by incorporating the value of the explanatory variables into the model as a mean over the period (t-T-5) – (t-T), while the average growth rate of patents was calculated over the period from t-T to t. In addition, in order to resolve the problem of different accounting units, explanatory variables are expressed, for each region, as a percentage of the respective GDP or population. For the USA, the model was estimated for 1990–1999, the period for which patent data are available at the sub-state level from the US Patent Office. The analysis is based upon 266 MSA/CMSAs19 covering all continental US States (and the District of Columbia), while MSAs in Alaska, Hawaii, or in other non mainland territories of the US are excluded from the analysis. The lack of sub-state level data for R&D expenditure was addressed by relying upon Standard & Poor’s Compustat20 North American firm-level data which provide a proxy for private R&D expenditure in 145 MSAs out of the total of 266. The proxy was calculated by

18

Different estimation techniques have been considered in order to minimize potential bias due to omitted variables (panel data) and/or modifiable aerial unit problem (e.g., Hierarchical Linear Models, HLM). However, severe constraints in terms of both the spatial scale and the time-series dimension of the existing regional data prevent us from implementing these alternative methodologies. Further research in this direction remains in our agenda, given the continuous progress in the availability of regional statistics. 19 The MSA/CMSA list is based on Metropolitan Areas and Components, 1993, with FIPS Codes, published by the Office of Management and Budget (1993). 20 Standard & Poor’s Compustat North America is a database of financial, statistical, and market information covering publicly traded companies in the U.S. and Canada. It provides more than 340 annual and 120 quarterly income statements, balance sheets, flows of funds, and supplemental data items on more than 10,000 active and 9,700 inactive companies.

6.5 Results of the Analysis

99

summing up firms’ R&D expenditure in each MSA. Though rough, this is the only measure available and similar proxies have been commonly used in the literature on the MSA innovative activities (e.g., Feldman 1994). All other US variables are based on US-Census data included in the USA Counties 1998 CD-Rom. For the EU, the model is run for 1990–2002. Lack of data led to the exclusion of the new member states of the Union. This is actually fortuitous, because the new, central and eastern European member states have much lower development levels than the EU-15 and are less economically integrated with them. In order to maximise the comparability between the EU and the US “territorial dynamics of innovation” the analysis would ideally be focused on FURs (Functional Urban Regions)21 rather than on NUTS administrative regions. Unfortunately, as extensively discussed in Appendix A, the lack of available data for many of the relevant explanatory variables has prevented any consideration of functional regions. For purposes of our analysis, the most appropriate scale of analysis seems to be the one closest to each regional system of innovation. Hence – and considering data availability – the unit of analysis with the greatest relevance in term of the institutional infrastructures that support innovation was selected for each country as in previous chapters (see Appendix A). The use as unit of analysis of NUTS1 and NUTS2 regions for the EU and MSAs for the US could produce distortions in the comparative analysis. While the former cover the whole EU territory, the latter correspond to the main US urban areas and their functional hinterlands, making them not necessarily spatially contiguous. In part, this difference reflects the lower population densities in the US, where more of the territory is truly outside of the direct influence of metropolitan areas than in Europe. In the denser US states, such as Connecticut, the entire territory is covered by metropolitan areas, making it more like Europe. In any case, the robustness of our results against this potential source of bias has been confirmed when reproducing the contiguity structure of the EU NUTS regions by the estimation of the same model with the US states as geographical units of analysis.

6.5.2

Comparing Territorial Dynamics in Europe and the US

The results for model 6.2 are presented in Tables 6.1 and 6.2 for the US and Table 6.3 for the EU. Table 6.1 includes the R&D expenditure variable which is available for only 145 MSAs. However, the regions for which R&D data are available are not a random sample of the total 266 MSAs: on the contrary, when this variable is introduced, the less economically successful and less innovative

21

The concept of FURs has been defined as a means to minimize the bias introduced by commuting patterns. A FUR includes a core city, where employment is concentrated, and its hinterland, from which people commute to the center. For a detailed analysis of this concept see Cheshire and Hay (1989).

(1)

(2)

145 0.12 3.59***

Observations R-squared F

(4)

(5)

(6)

(7)

145 0.24 10.48***

X

145 0.26 12.17***

X

0.446*** (0.090)

145 0.13 3.74***

X

0.150 (0.107)

145 0.16 4.17***

X

0.316*** (0.104)

145 0.19 5.23***

X

0.010*** (0.004)

145 0.13 3.24***

X

0.003 (0.003)

Robust standard errors in parentheses – *significant at 10%; **significant at 5%; ***significant at 1%

X

Geographical Dummies (North–East, South, Great Lakes and Planes)

Krugman Index

Total regional personal income as a percentage of total US

Rate of NET DOMESTIC migration Population density (Ln)

% of agricultural labour force

Rate of unemployment

% Population aged 15–24

(3)

(8)

(9)

(10)

(11)

(12)

145 0.32 12.63***

X

0.003*** (0.001)

0.017*** (0.003)

145 0.24 11.96***

X

0.002 (0.005)

0.016*** (0.003)

145 0.24 11.73***

X

0.002 (0.003)

0.016*** (0.003)

145 0.24 12.81***

X

0.046 (0.042)

0.016*** (0.003)

145 0.32 11.88***

X

0.003*** (0.001) 0.009 (0.007) 0.001 (0.003) 0.031 (0.046)

0.018*** (0.003)

0.220*** 0.140*** 0.162*** 0.162*** 0.267*** 0.215*** 0.200*** 0.222*** 0.214*** 0.207*** 0.162*** (0.045) (0.041) (0.057) (0.043) (0.046) (0.046) (0.038) (0.043) (0.034) (0.034) (0.051) 0.028*** 0.030*** 0.021*** 0.020** 0.023*** 0.019** 0.032*** 0.028*** 0.028*** 0.029*** 0.033*** (0.008) (0.007) (0.008) (0.008) (0.008) (0.008) (0.007) (0.007) (0.007) (0.007) (0.007) 0.011** 0.008 0.009* 0.010* 0.012** 0.008 0.014*** 0.011** 0.012** 0.012** 0.013** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.006) 0.010 0.002 0.023 0.033 0.007 0.031 0.020 0.012 (0.036) (0.031) (0.035) (0.035) (0.033) (0.036) (0.033) (0.034)

0.016*** (0.003) Spat. Weigh. Average of neighbouring MSAs’ 0.005 (0.022) Social Filter % population with bachelor’s, graduate or professional degrees % population with some college level education (no degree)

0.205*** (0.042) Natural Log of patents per 0.019** million inhab., 1990 (0.008) Private R&D expense (% of 0.009* regional personal income) (0.005) 0.029 Spat. Weigh. average of (0.035) neighbouring regions’ R&D Social Filter

Constant

Table 6.1 H–C OLS estimation of the empirical model. Annual Patent growth rate 1990–1999, US MSAs with R&D (145 Obs) 100 6 Innovation In an Integrated Framework

(1)

(3)

(4)

(5)

(6)

(7)

266 0.17 10.83***

Observations R-squared F

266 0.15 10.82***

X

0.374*** (0.058)

266 0.05 3.39***

X

0.219** (0.094)

266 0.05 2.23**

X

0.194* (0.100)

266 0.14 7.92***

X

0.012*** (0.002)

266 0.04 1.84

X

0.002 (0.002)

266 0.21 11.14***

X

0.002*** (0.001)

Robust standard errors in parentheses – *significant at 10%; **significant at 5%; ***significant at 1%

X

Geographical Dummies (North–East, South, Great Lakes and Planes)

Total regional personal income as a percentage of total US Krugman Index

Population density (Ln)

Rate of NET DOMESTIC migration

% of agricultural labour force

Rate of unemployment

(2)

(8)

(9)

(10)

(11)

(12)

266 0.17 10.27***

X

0.007* (0.004)

266 0.17 10.43***

X

0.003 (0.003)

266 0.16 10.32***

X

0.002 (0.035)

266 0.23 7.91***

X

266 0.2242 7.65***

X

0.002*** 0.002*** (0.001) (0.0005) 0.011** 0.0000 (0.005) (0.0000) 0.003 0.003 (0.003) (0.003) 0.014 0.0045 (0.712) (0.037)

0.154*** 0.073*** 0.046 0.072** 0.230*** 0.117*** 0.162*** 0.128*** 0.155*** 0.155*** 0.120*** 0.166*** (0.025) (0.026) (0.039) (0.029) (0.033) (0.028) (0.025) (0.027) (0.026) (0.027) (0.032) (0.028) 0.019*** 0.017*** 0.010** 0.007 0.017*** 0.009* 0.022*** 0.020*** 0.019*** 0.018*** 0.025*** 0.023*** (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.005) (0.006) (0.005) 0.019*** 0.017*** 0.018*** 0.018*** 0.018*** 0.016*** 0.016*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) (0.002) 0.024 (0.025)

Spat. Weigh. Average of neighbouring MSAs’ Social Filter % population with bachelor’s, graduate or professional degrees % population with some college level education (no degree) % Population aged 15–24

Natural Log of Patents per million inhab. Social Filter

Constant

Table 6.2 H-C OLS estimation of the empirical model. Annual Patent growth rate 1990–1999, US MSAs with R&D (Full sample, 266 Obs.)

6.5 Results of the Analysis 101

X

0.047 (0.101)

0.069** (0.030) 0.023*** (0.008) 1.018 (0.699) 8.433** (3.985)

X

0.009 (0.007)

0.010* (0.006)

97 0.44 5.26***

X 97 0.45 5.62***

97 0.45 6.27***

97 0.43 5.61***

97 0.43 5.26***

Robust standard errors in parentheses – *significant at 10%; **significant at 5%; ***significant at 1%

97 0.45 4.65***

X

97 0.46 5.11***

97 0.46 5.71***

X

0.004 (0.005)

0.012 (0.007)

0.008 (0.006)

97 0.46 5.33***

97 0.47 5.58***

X

0.009 (0.007) 0.005 (0.005) 0.087** (0.040)

0.010 (0.007)

97 0.21 2.11**

X

0.171* (0.099)

0.062 (0.055)

0.388 (0.285) 0.055** (0.021) 4.830** (2.267) 45.968* (23.190)

A 1995–2002

0.086* (0.044) 0.029*** (0.008) 0.702 (0.716) 9.305** (4.198)

(11) 1990–2002

0.067** (0.029) 0.025*** (0.007) 0.359 (0.764) 8.260** (3.904)

(10) 1990–2002

0.108** (0.049) 0.024*** (0.007) 0.766 (0.706) 7.782* (3.949)

(9) 1990–2002

0.078*** (0.026) 0.029*** (0.008) 0.969 (0.704) 9.357** (3.999)

(8) 1990–2002

X

X

0.007 (0.047)

0.064 (0.040) 0.021*** (0.007) 0.994 (0.713) 8.282** (4.002)

(7) 1990–2002

97 0.43 6.11***

X

0.362* (0.204)

0.010 (0.050) 0.021*** (0.006) 1.245* (0.713) 8.830** (4.008)

(6) 1990–2002

X

X

0.346** (0.155)

0.043 (0.029) 0.023*** (0.006) 0.233 (0.708) 8.018** (3.581)

(5) 1990–2002

Observations R-squared F

0.155 (0.112)

0.050* (0.029) 0.022*** (0.006) 0.556 (0.719) 8.218** (3.710)

(4) 1990–2002

National Dummies

0.011* (0.006) 0.014 (0.037)

0.094*** (0.032) 0.025*** (0.007) 0.712 (0.773) 7.066* (3.575)

(3) 1990–2002

0.069* (0.038)

0.060** (0.026) 0.021*** (0.006) 0.960 (0.691) 8.311** (3.884)

(2) 1990–2002

Percentage Regional of National GDP Krugman Index of Specialisation (Employment in 16 NACE sectors)

Population Density (Ln)

Migration rate

% Agricultural Labour Force

Long Term Unemployment

Spatially weighted average of neighbouring regions’ Social Filter % of employed persons with tertiary education % of total population with tertiary education % of Population aged 15–24

Natural Log of Patents per million inhab. R&D Expenditure (% Regional GDP) Spatially weighted average of neighbouring regions’ R&D Social Filter

Constant

(1) 1990–2002

Table 6.3 H-C OLS estimation of the empirical model. Annual Patent growth rate 1990–2002, EU Regions 102 6 Innovation In an Integrated Framework

6.5 Results of the Analysis

103

regions22 (i.e., those where it is less likely to find firms included in S&P database on which we rely for our R&D data) are excluded from the sample. Consequently, in order to account for the sample bias,23 in Table 6.2 we exclude R&D expenditure, and estimate the model for all 266 observations. As will be highlighted when commenting on the specific results, the sample selection bias affects only some of the results reported in Table 6.1. In regressions 1–2 of Table 6.1 (US) and Table 6.3 (EU) the initial level of patents, the measure for local innovative efforts, the proxy for knowledge spillovers, and the social filter variable are successively introduced. In regressions 3–7 the individual components of the social filter are included separately in order to discriminate among them. From regression 8 onwards the variables relative to the territorial organisation of the local economy (migration, agglomeration, and specialisation) are introduced sequentially. Table 6.2 follows the same order but without controlling for R&D expenditure and knowledge spillovers. The Adj-R2 confirms the overall goodness-of-fit of all the regressions presented and in all cases F-statistics probability lets us reject the null hypothesis that all regression coefficients are zero. VIF tests were conducted for the variables included in all the specifications of the model excluding the presence of multicollinearity. There was no spatial autocorrelation in the residuals detected using the Moran’s I test (Cliff and Ord 1972). These results offer a number of insights into the territorial dimension of knowledge production in the EU and the US. One of the key similarities in the geography of regional innovation of the US and Europe is the presence of territorial convergence in the regional distribution of innovative outputs over the last few years. The coefficient on the initial level of patenting activity is in both cases negative and significant (Tables 6.1, 6.2, and 6.3). This suggests that adequate conditions for the generation of patents – e.g., the emergence or re-location of innovative firms, the changing balance between positive and negative externalities arising from agglomeration at different stages of product life cycles, and changes in the competitive advantage of some locations in response to changes in technological paradigms – have spread to formerly peripheral areas, making the regional distribution of innovative output more evenly spread (in line with the results of Moreno et al. 2005b for the EU regions). This generalised trend towards the dispersion of innovative activities seems to be less accentuated in the United States than in the EU: when all the 266 MSAs are included in the analysis (thus also taking into

22

The 145 MSAs for which R&D data are available account for 89,9% of the GDP generated in all 266 MSAs and show an average of 225.19 patents per million inhabitant against 176.83 for the whole sample. 23 Beeson et al. (2001) discuss the “sample selection bias” introduced when choosing cities as unit of analysis rather then county-level data: only places that experienced successful growth in the past are considered in this way. The use of standard metropolitan statistical areas minimises this first bias. However, in order to keep the bias at a minimum, we not only report the results for the most innovative subsample of MSAs, but also for all MSAs in the continental US.

104

6 Innovation In an Integrated Framework

account the less innovative MSAs), the convergence parameter is smaller and less significant in the US than in the EU. In the US the relatively higher stability of the geography of the innovative output is associated with a positive and statistically significant impact (Table 6.1 Regression 1) of local innovative activities on innovative output: the higher the level of local R&D expenditure, the higher the patent growth rate at the local level. The production of knowledge and innovation are more localised in the US than in Europe, as implied also by the lack of evidence of inter-MSA spillovers: the spatially-weighted average of neighbouring MSAs’ R&D expenditure does not exert any statistically significant influence upon patent growth rate. In the EU the opposite territorial dynamic occurs. Local innovation productivity is not directly related to the level of R&D expenditure or, at least, the relationship seems to hold only in the short run. When patent growth over the 1990–2002 time span is regressed on initial R&D expenditure the coefficient is not significant (Table 6.3 Regression 1) but, when the shorter period 1995–2002 is considered, the coefficient is positive and significant (Table 6.3 Regression A). Instead, the positive and significant coefficient of the spatially-weighted average of neighbouring regions R&D expenditure (Table 6.3 Regression 1) shows that the long-run growth rate of innovative activities in the EU is shaped more by exposure to interregional knowledge spillovers. Three potential factors may lie behind the differences between the US and Europe in the impact of innovative inputs on innovative outputs: the distance between innovative centres, the composition of R&D investments, and labour mobility. Large innovative areas tend to be physically closer in Europe than in the US. In addition, empirical analyses of the diffusion of spillovers have highlighted the presence of very strong distance decay effects in the US – knowledge spillovers, in general, do not spread beyond a 80–110 kms radius from the MSA where they are generated (Varga 2000; Acs 2002) – whereas in Europe the geographical diffusion of spillovers is felt in a radius of 200–300 kms from the point of origin as showed in Chap. 5 of this book and confirmed by Bottazzi and Peri (2003) and Moreno et al. (2005a). Greater proximity and lower distance decay suggest a greater potential for European regions to rely on innovative inputs in neighbouring areas as a source of innovation and develop inter-regional relational networks between innovative actors supported by spatial proximity (Maggioni et al. 2006). Greater distance and stronger distance decay effects are, in contrast, likely to lead to the creation of self-contained innovative areas in the US, which necessarily have to rely on their own innovative inputs rather than on spillovers from other MSAs. R&D inputs in the US tend also to be more specialised and better targeted than in Europe. A legacy of efforts by virtually every European nation-state to have a presence in a large number of areas of knowledge has resulted in duplications and redundancies in R&D that European integration has, so far, failed to overcome (Gambardella and Malerba 1999; Mariani 2002). The existence of a much more integrated market in the US has favoured the formation of a more specialised geographical innovation structure.

6.5 Results of the Analysis

105

Last but not least, the literature has extensively documented the difference in labour mobility between the two continents, which may also leave an important imprint on their respective geographies of innovation. In the US, high levels of mobility may allow better ongoing matching of innovative actors in space; consequently, they interact more strongly on a local basis, relying less on spillovers from other MSAs than in Europe. The weaker convergence parameter in the USA is an outcome not only of the stronger impact of local R&D expenditure, but also because of the higher and more variable productivity of such expenditure, thanks to the better spatial matching obtained through higher factor mobility. In contrast, weaker intra-regional synergies force European innovators to rely upon neighbouring regions’ innovative efforts (extra-regional spillovers). The more stable EU population distribution – with a lower degree of internal mobility – produces a more interregionally integrated and “redistributive” innovation system, which depends for its functioning on inter-regional communication and “distance” learning. Matching at the local level is severely hampered by low levels of mobility.24 While the spatial diffusion of knowledge exhibits different territorial dynamics in the EU and the US, empirical evidence suggests that, in both contexts, the internal socio-economic factors, which enable the translation of both endogenously produced and exogenous knowledge into innovative output, are quite similar. Both in the US and in Europe (as confirmed in Chap. 4) the social filter variables show positive and significant signs (Table 6.1 Regression 2; Table 6.3 Regression 2). The existence of a set of local socio-economic factors may be a pre-condition for the establishment of a successful regional system of innovation and seems to play an important role in explaining differential innovative performance in the regions of the EU and in US MSAs. In addition, in both contexts, the local economy is not strongly affected by the socio-economic conditions of neighbouring areas: the spatially weighted average of the socio-economic conditions of neighbouring regions is not significant for either the US or the EU (Table 6.1 Regression 2 and Table 6.3 Regression 2, respectively). However this evidence needs to be placed in the context of the different geographical processes in the two continents. The fact that neighbouring regions’ social filter conditions have no impact on local innovative performance is consistent with the more “localistic” and “self-sufficient” nature

24

When assessing this phenomenon it must, however, be borne in mind that the unit of analysis in the case of the EU are NUTS regions i.e., territorial units for the production of regional statistics for the European Union whose definition mainly serves administrative purposes. As a consequence, NUTS regions might not always approximate the functional borders of the regional economy. Conversely, US MSAs are closer to the concept of ‘functional urban regions’ (Cheshire and Hay, 1989) and likely to be more “self-contained” in terms of economic interactions. Consequently, part of the difference between the empirical evidence recorded in the two cases may be due to the different nature of the spatial unit of analysis. However, since we rely on inverse linear distance (and not on contiguity) for the specification of our spillovers variable, the impact of heterogeneous spatial units in Europe and the US should be minimized supporting our results that are, in any case, largely in line with the existing literature. As a robustness check we have reestimated our empirical model for the US, using state-level data (contiguous) with very similar results for the spillover variables. These results are available from the authors on request.

106

6 Innovation In an Integrated Framework

of the US territorial dynamics of innovation. In the European Union the similarly localised impact of the social filter conditions is at odds with the less localistic, inter-regional communications deemed necessary to enable knowledge matching. This suggests a potential inconsistency in the EU system of innovation, where social filter conditions are unable to extend their benefit over distance thus reinforcing the need for long-distance knowledge flows as a way of compensating the consequences of low factor mobility. When the individual components of the social filter are assessed separately, the significant and positive effect of higher education is apparent in both contexts (Regressions 3 and 4 in Tables 6.1 and 6.2). More specifically, from the inspection of Table 6.1, the key resource for the US regions seems to be the population holding bachelors, postgraduate, or professional degrees. The percentage of people holding only some “college level education”, without a full degree, is insignificant. However, this result in the case of the US may be partially due to the sample selection bias discussed above: when the whole sample of MSAs is considered (Table 6.2 Regressions 2 and 3) both education variables are significant. This implies that for the most innovative MSAs, where the majority of the R&D expenditure is concentrated, the true differential competitive factor seems to be a higher level of specialised skills while, in the overall sample of the MSAs, more generic skills may also play a role in improving local innovative performance. In the EU context, where slightly different indicators are available, the educational achievements of the population (Table 6.3 Regression 4) exert a positive and significant influence on innovative output (while those of employed people only are not significant – Table 6.3 Regression). An additional factor that seems to foster innovation in both the EU and the US is the presence of a favourable demographic structure: the regions where the incidence of young adults is higher tend to produce more innovation (Tables 6.1 and 6.3 Regression 5). Only in the US a higher rate of unemployment seems to discourage the production of innovation (Table 6.1 Regression 6 and Table 6.2 Regression 5), while in the EU this effect is not statistically significant. The percentage of the labour force concentrated in agriculture is not significant in either context. From this we conclude that the productive use of human resources is less important than the quality of such resources. The final part of the empirical analysis takes up the territorial organisation of the factors of production: migration flows and the density of human interactions, on the one hand, and the agglomeration and specialisation of economic activities, on the other. In US MSAs the rate of net domestic migration exercises a positive and significant effect on patent growth rates (Table 6.1 Regression 8; Table 6.2 Regression 7). By contrast, the innovative productivity of the EU regions is unable to benefit from this kind of flow due to lower mobility of European workers: as Peri (2005) points out, not only does the US receive a much larger flow of immigrants (in absolute and relative terms) from the rest of the world than the EU, but “the US also complements these large inflows of immigrants with a very high internal mobility of its citizens” (p. 22). Our results can be effectively interpreted in the light of these different features of the labour market,

6.5 Results of the Analysis

107

underscoring that there are territorial mechanisms through which they exert their influence upon the innovative performance of the two continents (echoing the results of Ottaviano and Peri 2006). US MSAs benefit from the inflows of skilled labour that result in higher productivity and innovation, but innovation also acts as a magnet for skilled individuals that would, in turn, contribute to further innovation. A virtuous circle of innovation and migration is thus generated in the US. In Europe, due to the greater cultural and institutional barriers to mobility, this virtuous circle is much weaker. Looking at the role of agglomeration effects, Table 6.1 suggests that in American MSAs neither population density (Regression 9), nor the percentage of total US personal income (Regression 10) have a significant effect. However, when the sample selection bias discussed before is accounted for, by considering the whole 266 MSA sample, “population density” shows a positive and significant sign (Table 6.2 Regression 8) – and its significance increases when assessed together with migration rate25 (Table 6.2 Regression 11) – while “regional % of national GDP” remains insignificant in either case. In the EU context the opposite is true: population density is not significant (Table 6.3 Regression 9) while “regional % of national GDP” is positive and significant both alone and after controlling for population density (Table 6.3 Regression 10 and 11 respectively). It appears that in the US, the density of human contacts is important for innovative productivity since it enables the maximisation of intra-regional spillovers. As previously discussed, these exchanges prevail over inter-regional knowledge flows. Conversely, the agglomeration of economic activities, a proxy which emphasizes the “scale” side of agglomeration economies, is not a differential source of competitive advantage for US MSAs: the optimal scale of production is easily achieved by the US MSAs, which are larger on average than their European counterparts. Consequently the “scale component” of agglomeration economies does not emerge as a differential innovative factor (i.e., in addition to the density of human interactions/localised knowledge spillovers). In the EU, local interaction density does not seem to stimulate innovation productivity. The positive effect of population density on human interaction – predicted by the theory of agglomeration and observed empirically in the US case – seems to be offset, in the EU, by a broader set of territorial forces. Population density, in a context of low labour mobility as in the EU, may encourage the pooling and stratification of inadequate skills while US agglomerations are usually newer and, in absolute terms, less densely populated. Competitive advantage for the EU regions is instead promoted by the relative concentration of wealth which, as seen above, proxies the “scale” side of agglomeration economies.

25

This is consistent with the notion that because mobility is higher in the US, innovation systems have more local matching and learning and hence are more “local” than in Europe, where long distance communication is necessary in order to match relatively immobile agents.

108

6.5.3

6 Innovation In an Integrated Framework

Agglomeration, Specialisation and Absolute Size of Clusters

There is a vigorous debate in the literature over whether agglomeration economies can be fully captured by measures of relative concentration, or whether the absolute size of clusters is more relevant to understanding the effects of geographical concentration on productivity (see Duranton and Puga 2003, for a review). Some recent literature places increasing emphasis on the absolute size of clusters as the basis for calculating the level of specialization, arguing that this level is systematically underestimated for larger metropolitan areas when relative levels of concentration are used (Drennan and Lobo 2007). We address this issue by proxying the absolute economic size of each MSA/region by its total population, total employment, and total GDP. We then calculate an interaction term between the degree of specialisation of the regional economy (Krugman Index) and its absolute size (proxied as specified above). The results are reported in Table 6.4 – for the 145 MSAs for which R&D data are available, Table 6.5 – for the full US sample (266 MSAs), and Table 6.6 – for the EU. All proxies for the absolute size of the MSAs/Regions show a positive and significant effect both in the US (but only when all the 266 MSAs are considered, Table 6.5) and in the EU (Table 6.6). Thus, larger clusters are able to produce more innovation both in Europe and the US. The interaction term between absolute size and specialisation is negative and significant in the US (when the full sample is considered, Table 6.5) while it is not significant for the EU regions (Table 6.6). When the interaction term between absolute size and specialisation is introduced:
 1 Pai;t ¼ a þ b1 lnðPai;tT Þ þ b2 Krugman Index þ b3 RD ln Pai;tT T þ b4 SocFilter þ b5 Size þ b6 Kruman Index



Size þ e

(6.5)

from which, when considering the partial effect of absolute size on patent growth rate (holding all other variables fixed), one obtains:
D




Pai;t Pai;tT DSize

1 T ln

 ¼ b5 þ b6 Krugman Index

(6.6)

this implies, with b6 <0, that, ceteris paribus, the more a cluster is “specialised” (high Krugman index score), the more an increase in its absolute size reduces its productivity in terms of new patents. This can be interpreted as showing that in the US the absolute size of clusters exerts a positive impact on innovation, but such impact is reduced when large absolute size is coupled to high specialisation. Specialisation is not inherently negative for innovative performance (Krugman index not significant), but larger clusters need to be “specialised in more than one sector” in order to fully exploit the benefit from their absolute economic size. In the EU, the absolute size of clusters always exerts a positive influence on innovative

X

X

Observations 145 145 R-squared 0.24 0.24 F-stat 11.12*** 10.31*** Robust standard errors in parentheses *significant at 10%; **significant at 5%; ***significant at 1% ^joinlty significant at 10%; ^^significant at 5%; ^^^significant at 1% (Wald test)

Geographical Dummies (North–East, South, Great Lakes and Planes)

145 0.24 11.22***

X

145 0.24 10.31***

X

145 0.24 11.21***

X

145 0.24 10.30***

X

Table 6.4 H-C OLS estimation of the empirical model: interaction terms. Annual Patent growth rate 1990–1999, US MSAs with R&D (145 Obs.) (1) (2) (3) (4) (5) (6) Constant 0.210*** 0.145 0.200*** 0.150 0.198** 0.150 (0.077) (0.136) (0.072) (0.129) (0.084) (0.145) Natural Log of patents per million 0.029*** 0.029*** 0.029*** 0.029*** 0.029*** 0.029*** inhab. (0.007) (0.007) (0.007) (0.007) (0.007) (0.008) Private R&D expense (% of total 0.012** 0.012** 0.012** 0.012** 0.012** 0.012** personal income) (0.005) (0.005) (0.005) (0.005) (0.005) (0.006) Social Filter CP 1 0.016*** 0.015*** 0.016*** 0.015*** 0.016*** 0.015*** (0.003) (0.003) (0.003) (0.003) (0.003) (0.003) Krugman Index 0.044 0.430 0.048 0.340 0.048 0.330 (0.045) (0.590) (0.045) (0.559) (0.045) (0.620) Total Population (Ln) 0.000 0.005 (0.005) (0.009) Interaction term Krugman Index*Total 0.030 Population (0.044) Total employment (Ln) 0.001 0.005 (0.005) (0.009) Interaction term KrugmanIndex*Total 0.024 Employment (0.044) Total Income (Ln) 0.001 0.004 (0.005) (0.008) Interaction term KrugmanIndex*Total 0.018 Income (0.038)

6.5 Results of the Analysis 109

X

X

Observations 266 266 R-squared 0.24 0.24 F-Stat 8.61*** 7.73*** Robust standard errors in parentheses *significant at 10%; **significant at 5%; ***significant at 1% ^joinlty significant at 10%; ^^significant at 5%; ^^^significant at 1% (Wald test)

Geographical Dummies (North–East, South, Great Lakes and Planes)

266 0.24 8.58***

X

266 0.24 7.73***

X

266 0.24 8.48***

X

Table 6.5 H-C OLS estimation of the empirical model: interaction terms. Annual Patent growth rate 1990–1999, US MSAs (full sample) (1) (2) (3) (4) (5) Constant 0.039 0.001 0.050 0.011 0.021 (0.047) (0.095) (0.043) (0.088) (0.050) Natural Log of Patents per million 0.025*** 0.026*** 0.026*** 0.026*** 0.026*** inhab. (0.006) (0.006) (0.006) (0.006) (0.006) Social Filter 0.017*** 0.016*** 0.016*** 0.016*** 0.016*** (0.003) (0.003) (0.003) (0.003) (0.003) Krugman Index 0.028 0.255 0.028 0.245 0.027 (0.041) (0.438) (0.040) (0.399) (0.041) Rate of NET DOMESTIC migration 0.002*** 0.002*** 0.002*** 0.002*** 0.002*** (0.001) (0.001) (0.001) (0.001) (0.001) Total Population (Ln) 0.010*** 0.014** ^^^ (0.003) (0.007) Interaction term Krugman Index*Total 0.018^^^ Population (0.034) Total employment (Ln) 0.010*** 0.014** ^^^ (0.003) (0.007) Interaction term KrugmanIndex*Total 0.019^^^ Employment (0.032) Total Income (Ln) 0.010*** (0.003) Interaction term KrugmanIndex*Total Income

266 0.24 7.65***

X

0.013** ^^^ (0.006) 0.017 ^^^ (0.031)

(6) 0.026 (0.106) 0.026*** (0.006) 0.016*** (0.003) 0.288 (0.492) 0.002*** (0.001)

110 6 Innovation In an Integrated Framework

6.5 Results of the Analysis

111

Table 6.6 H-C OLS estimation of the empirical model: interaction terms. Annual Patent growth rate 1990–2002 and 1995–2002, EU Regions (1) (2) (3) (4) 1990–2002 1990–2002 1995–2002 1995–2002 Constant 0.091 0.026 0.001 0.327 (0.088) (0.065) (0.572) (0.411) Natural Log of Patents 0.025*** 0.025*** 0.048** 0.052** per million inhab. (0.007) (0.007) (0.019) (0.020) R&D Expenditure (% Regional GDP) 0.105 0.145 3.200 3.240 (0.718) (0.752) (2.149) (2.082) Spatially weighted average 8.268** 8.008** 41.907* 44.060* of neighbouring regions’ R&D (3.654) (3.687) (22.461) (22.918) Social Filter 0.008 0.008 0.017 0.006 (0.006) (0.006) (0.055) (0.068) 2.737 1.235 Krugman Indexa (2.110) (1.192) Total Population (Ln) 0.012** 0.023 (0.006) (0.032) Interaction term Krugman Index*Total 0.187 Populationa (0.148) Total Income (Ln) 0.012** 0.001 (0.006) (0.027) Interaction term KrugmanIndex*Total 0.114 Incomea (0.117) National Dummies

X

X

X

Observations 96 96 96 R-squared 0.48 0.48 0.26 F-stat 5.90*** 5.57*** 2.40*** Robust standard errors in parentheses *significant at 10%; **significant at 5%; ***significant at 1% ^joinlty significant at 10%; ^^significant at 5%; ^^^significant at 1% (Wald test) a Data to calculate the Krugman index are available from 1995 only

X 96 0.26 2.36***

performance and the degree of specialisation has always a negative impact (Krugman Index negative and significant). However, there is no significant (negative or positive) interaction between the two terms. Thus, ceteris paribus, an increase in the size of a cluster improves its innovative performance whatever its degree of specialisation and, symmetrically, more specialisation would lead to less innovation whatever the absolute economic size of the region. These results reconfirm the “national” bias of the EU process of innovation. In Europe lower levels of economic integration and factor mobility make specialisation a handicap for innovative performance.26 In contrast, in the integrated US 26

The impact of the sectoral structure upon regional innovative performance cannot be limited to the overall degree of specialisation. On the contrary, it would be necessary to fully control for the specific regional patterns of specialisation: given the aggregate degree of regional specialisation, the true differential factor could stem from a region being specialised in high tech R&D-intensive vs traditional sectors (Smith 2007). Further investigation, in an EU-US comparative perspective, of the sectoral level territorial processes remains in our agenda for future research.

112

6 Innovation In an Integrated Framework

context, only very large clusters in highly specialised local economies seem to suffer from reduced capacity to exploit knowledge-base complementarities. Special care is called for in interpreting the significance of the coefficients, as one does not have to test separately the significance of b5 and b6 but rather the joint hypothesis Ho : b5 ¼ 0, b6 ¼ 0 (Wooldridge 2002). To test this joint hypothesis, we implement the Wald Coefficient Test whose results, in the case of the US 266 MSAs, permit to reject Ho at the 1 percent level, allowing to conclude that the estimated coefficients are significant in both cases. Additionally, the Wald Test is not affected by the intrinsically lower orthogonality of the two variables. This picture is further enriched when accounting for the influence on innovative performance of the degree of specialisation of the local economy in the two continents. In the United States, the proxy for the degree of specialisation of the local economy (the Krugman Index) is not statistically significant (Table 6.1 Regressions 11 and 12). In the European case (Table 6.3 Regression A), in contrast, more specialised regions seem to be persistently disadvantaged in their capability to produce innovation. Specialisation is not a constraint for the process of innovation of US regions, where there seems to be mix of MAR and Jacobs externalities, more so than in the EU. In the US context, the higher labour mobility and the relatively less constrained (by political, institutional, and cultural factors) location choices of firms and individuals allow each innovative actor to select the most advantageous location according to its own technological or organisational needs [e.g., according to the current stage of the life-cycle of their product, as in Audretsch, (2003) and Duranton and Puga, (2005)]. In this context the geography of innovative actors effectively accommodates the possibility, offered by each region, to benefit from either sectoral specialisation or diversity. By virtue of this mechanism, specialisation ceases to be an obstacle for the production of innovation, since internal factor mobility will allow specialised areas to attract agents able to benefit from MAR externalities while pushing the other agents towards more diversified areas. Furthermore the larger economic size of US agglomerations may also allow them to benefit from a certain degree of knowledge-base complementarities even in relatively more specialised contexts. This dynamism of the ongoing process of reorganisation of US economic activities in response to changing sectoral location advantages is confirmed by the evidence produced by Desmet and Fafchmps (2005) for county-level employment. They find both de-concentration of non-service employment and clustering of service jobs in high aggregate employment agglomerations: location patterns in certain sectors may be changing in response to shifts in the pattern of localised externalities. However, the de-concentration of manufacturing has benefited counties 20–70 km away from large agglomerations, while service activities have clustered in large agglomerations within a 20 km radius. This particular spatial scope of concentration/de-concentration dynamics seems to suggest the existence of fundamentally intra-MSA adjustments, along the lines of the “localistic” view of the US process geography. As the EU and USA show different rates of adjustment, through mobility of labour and capital, they generate different underlying patterns of specialization (even if the overall levels are similar). This clarifies the meaning of the observed negative impact of

6.6 Conclusions

113

specialisation on the innovative performance of the EU. It suggests that for the USA, it is not so much the level of specialisation that matters, as the ability to adjust the content of local factors to the changing needs of innovation. Ciccone (2002) suggests that the degree of agglomeration of the local economy is not substantially different in the two continents. Our results indicate that the comparative analysis of the processual aspects of the production of new knowledge cannot be “reduced” to the level of agglomeration. On the contrary, the analysis of agglomeration economies needs to be pursued in conjunction with other relevant territorial processes, in the context of the overall geographical dynamic of the economy and its factor markets.

6.6

Conclusions

Our empirical analysis of the geography of innovation in the US and Europe reveals that knowledge production in both continents is governed by different geographical processes. In the US the generation of innovation usually occurs in self-contained geographical areas that rely on their own R&D inputs, on favourable local socioeconomic environments, and on the training and attraction of highly skilled individuals. In Europe the process is much more linked not just to having an adequate local socio-economic context, but to proximity to other innovative areas and to the capacity to assimilate and transform inter-regional knowledge spillovers into innovation. Human capital mobility, in contrast to the US case, does not play a role. Specialization is also negatively associated with the genesis of innovation in Europe, where agglomeration is a better driver of innovation. Hence, while Europe relies on Jacobian externalities for innovation, US MSAs can count on both MAR and Jacobian externalities. It can then be inferred that the dynamic reorganisation of European innovation resources is severely limited by the EU’s lower levels of factor mobility and integration than in the US. Two key forces lie behind these diverse territorial dynamics of innovation. First is the “national bias” in EU innovation. Despite rapid economic integration, distinct national and regional systems of innovation persist in Europe and are propped up by lingering interest by national governments to preserve or improve local technological or innovative capacities. Countries across the EU maintain their own innovative strategies that may to a larger or lesser extent conform with the European-wide Lisbon Agenda. Since the US is a much more economically, culturally, and psychologically integrated market, innovation processes are shaped principally by national forces; the states have a minor role in financing major innovation programs. While some states do have major public research universities, the R&D in those institutions is financed by the federal government at 60% on average, with private funds contributing about 20% to their budgets, and states about 20%, mostly toward their teaching activities. At a national level, in any case, the emphasis tends to be much more on creating adequate general conditions for innovation to take hold and allowing geographical processes to bring the agents together.

114

6 Innovation In an Integrated Framework

A second force is the European concern with cohesion, even in the genesis of innovation. Whereas in the US the location of innovation is strongly influenced by market forces, in the EU the Lisbon Agenda shares the somewhat contradictory goals of “making Europe the most competitive knowledge-based economy in the world”, while, at the same time, promoting territorial cohesion. National innovation policies have also been sensitive to cohesion concerns, with large chunks of the public R&D effort in recent years having been spent – especially in peripheral countries – in territories with a weaker innovative base. A clear example of these two contrasting cultures of innovation can be found in the Aeronautics industry. In the US, Boeing has almost half of its workforce concentrated in one location, where it can benefit from economies of scale and the advantages of specialisation. Workers with the necessary skills to work in Boeing are expected to migrate. Airbus, Boeing’s main rival was originally a consortium of firms and, in practice, remains very much so. National interests, cultural differences, and the greater cost of migration in Europe, have made of Airbus a company with multiple centres, with only 28% of its workforce concentrated in its main plant in Toulouse, France. Different functions are performed in different plants located hundreds of kilometres from one another. Do these different territorial dynamics of innovation make a serious difference in outcomes? It is important not to rush to judgment from this analysis. At first glance, it might be tempting to echo many other analyses in calling for an “Americanization” of European process geography: greater factor mobility, bigger and more specialised agglomerations, more integration. The analysis pursued here, however, also sees some signs of a distinctly European pathway to integration: given these lower levels of mobility, and an historically more dispersed and less specialised urban system, and the persistence of national institutions and cultures, it may be that Europe is developing functional equivalents for American mobility and specialisation, in the form of greater inter-metropolitan knowledge exchange and cooperation. Certainly, the advances in the European transport system (high speed rail, cheap flights) are bringing metropolitan areas closer together than ever before, as are heightened levels of intra-firm, inter-firm, and inter-governmental cooperation. In the global aeronautics industry it is unclear whether the Boeing or the Airbus model will emerge victorious. Despite coordination glitches – such as the recent wiring problem in the new Airbus 380 – Airbus has shown that complex collaborative innovation arrangements can work and compete with more geographically integrated systems. And recent advances in telecommunications and transportations may help increase the viability of the more diffuse European systems of innovation. The question is whether, at some point, these represent viable functional alternatives to the American process geography, i.e., capable of helping Europe overcome the innovation gap. In any event, the present analysis suggests that, in both continents, the territorial dynamics of innovation, as well as its basic geographical foundations, are essential elements in considering the performance of these innovation systems.

Chapter 7

What Can We Learn From the “Integrated Approach” To Regional Development? The Impact of EU Infrastructure Investment

7.1

Introduction

Previous chapters have been devoted to the development of a consistent framework for the analysis of regional growth dynamics. Innovation has been identified as the key engine of regional growth. However, the linear approach to the analysis of the translation of innovation into economic growth has been progressively broadened in order to explicitly take into account institutional and geographical factors. The empirical analyses that tested this “integrated” model for the genesis of regional growth confirmed its explanatory power and shed new light on the drivers of regional growth in Europe. In addition Chap. 6 showed how an “integrated” approach can be successfully applied to the comparative analysis of regional innovative dynamics in Europe and the United States. This chapter aims at applying the “integrated” framework developed in the book to the analysis of the impact of infrastructure policies on regional economic performance. The analysis will show how a progressive departure from the neoclassical linear growth model can dramatically change our predictions on the regional growth impact of new infrastructure development. Infrastructure development – in particular transport infrastructure – is generally considered desirable in terms of achieving both economic efficiency and territorial equity for a series of reasons. First, a modern and efficient infrastructure endowment is supposed to be a necessary competitive asset for the maximization of the local economic potential and for allowing an efficient exploitation of resources. Second, it is often perceived that improvements in infrastructure endowments not only provide accessibility, but also contribute to a better market integration of peripheral and lagging regions, allowing them to catch up with the more advanced territories. In the EU, in particular, infrastructure development has been regarded as a necessary condition for the economic success of the regions, as well as a tool for an equitable distribution of the benefits of the process of European integration. Infrastructure thus not only contributes to enhance the benefits of integration, but is also considered to be the main means for spreading its benefits. Given this predominant view, it comes as no surprise that infrastructure, in general, and transport infrastructure, in particular, have acquired an important role in European Union policy.

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_7, # Springer-Verlag Berlin Heidelberg 2011

115

116

7 What Can We Learn From the “Integrated Approach” To Regional Development?

This chapter examines how infrastructure development, and especially investment in transport infrastructure, has affected regional development in the EU by adopting a broad theoretical perspective in which other relevant features with a bearing on regional economic performance are also considered, allowing for a more accurate assessment of the effect of infrastructure capital. These include the concentration of innovative activities, the presence of (un-)favourable social conditions, agglomeration economies, and inward mobility of individuals. We also assess the spatial interactions between each region and its neighbouring areas in the form of spillovers, by explicitly analysing the impact of both internal and external conditions on regional economic performance. The inclusion in the analysis of the infrastructure endowment (and other conditions) in neighbouring regions makes it possible to isolate the impact of a favourable geographical location of any given region not only in terms of its capacity to reap network externalities, but also to benefit from other growthenhancing conditions of interconnected regions. Finally, the chapter also aims at providing some policy lessons, including an assessment of the magnitude and policy implications of the spillover of infrastructure investment across regions.

7.2 7.2.1

Infrastructure and Regional Economic Development The Rationale for Public Infrastructure Development

The Treaty on the European Union explicitly puts forward (Article 154) “the establishment and development of trans-European networks in the area of transport, telecommunications and energy infrastructure” as a policy tool to help achieve the objectives of an integrated internal market (Article 14), on the one hand, and of “an overall harmonious development” in terms of economic and social cohesion, on the other. The European Commission’s (2005) “Integrated Guidelines for growth and jobs (2005–2008)” have re-asserted the role of infrastructure development in the micro-economic field for raising Europe’s growth potential and cohesion. As a consequence, the development of infrastructure in the Member States has been supported, integrated and coordinated by means of the trans-European networks (TENs) in the fields of energy, telecommunications and transport. In particular, the construction of the Trans-European Transport Network (TEN-T) has been driven by the Community Guidelines agreed by the Essen European council which led, in 1996, to the identification of 14 priority projects. This list of projects was extended in 2004 following the progressive enlargement of the EU to 25 and then 27 members, to comprise 30 priority projects to be completed by 2020 and a variety of smaller projects. The establishment of the TEN-T has provided an overall planning framework for the development of a European-wide transport infrastructure, superseding a previous system fundamentally based on the needs of individual regions, which lacked a supra-regional perspective (Vickerman 1995). Infrastructure development absorbs a significant percentage of the financial resources of the European Union. For the programming period 2000–2006,

7.2 Infrastructure and Regional Economic Development

117

€ 195 billion (at 1999 prices) were allocated to the Structural Funds (Puga 2002, p. 374), of which around two-thirds were allocated to Objective 1 regions. Roughly half of this Objective-1 allocation was earmarked for the development of new infrastructure (Rodrı´guez-Pose and Fratesi 2004). In addition, about half of the € 18 billion of the Cohesion Fund for the same period went into infrastructure and European Investment Bank (EIB) loans totalled € 37.9 billion (European Commission 2007a). The TEN-T alone have mobilized a substantial amount of financial resources for transport infrastructure both under the specific heading of the Common Transport Policy and by drawing upon Structural and Cohesion funds. In the 2000–2006 Structural Funds programming period, TEN-T not only received a budget of € 4.2 billion, but also benefited from the allocation of € 16 billion under the Cohesion Fund and from part of the € 34 billion of the European Regional Development Fund (ERDF) invested in transport infrastructure. In the 2007–2013 financial framework, about € 8 billion have been earmarked for TEN-T “but the ERDF and the Cohesion Fund will continue to be the main sources of community assistance for co-funding of the TEN-T” (European Commission 2007a, p. 5): About € 35 billion under the cohesion fund will be chiefly earmarked to the priority projects. The justification – in terms of growth and cohesion – for this significant amount of financial resources devoted by the EU to the development of infrastructure relies crucially on the underlying model of the functioning of the regional economy and on the factors assumed as drivers of economic growth. Infrastructure has traditionally been regarded from three different perspectives. First, as an “unpaid factor of production”, directly generating improvements in output; second, as an “augmenting factor”, enhancing the productivity of labour and capital; and, third, as an incentive for the relocation of economic activity (Lewis 1998). Aschauer (1989) introduced the idea – previously somewhat overlooked in the economic literature – that differences in the stocks of public infrastructure and private capital could provide an important explanation for differences in levels of national output. In Aschauer’s framework, other things being equal, the higher the stock of public infrastructure, the higher the capital productivity in the private sector: An increase in infrastructure endowment produces an increase in productivity and, albeit at a lower rate, in labour costs (Biehl 1991). The resulting labour cost/productivity ratio is a proxy for the region’s competitive position: The regions where productivity exceeds labour costs will benefit from higher income and more jobs, which will stimulate inward migration and capital inflows. Hence, improvements in infrastructure endowment will benefit the regional economy in excess of its potential GDP, as productivity will outstrip labour costs. This approach produced a stream of empirical literature based on regressions a` la Aschauer that provided significant evidence in support of the positive impact of infrastructure investment: The yearly rate of return to public investment was estimated to be higher than 100% (Holtz-Eakin 1993; Glomm and Ravi-Kumar 1994). From a different perspective, the classical location theory, by emphasising the advantages of locations benefiting from better accessibility and lower transport costs, also supported the view of infrastructure investment as a means for better economic performance. Seitz and Licht (1995) suggest – on the basis of the duality theory – a third separate approach to the

118

7 What Can We Learn From the “Integrated Approach” To Regional Development?

problem of the relationship between transport costs and regional growth: “investing in public infrastructures can be considered an instrument to improve the competitiveness of cities, regions and nations by reducing production and transport costs” (p. 239). The economic policy implications of these approaches seemed sufficiently straightforward to justify the emphasis placed by EU policy makers on their programme of infrastructure development. These clear-cut conclusions have been challenged by a variety of theoretical and empirical studies. First, Gramlich (1994) not only questioned the direction of causality of Aschauer’s regressions and highlighted how the lack of an agreed definition about the concept of infrastructure may have led to significant measurement inconsistencies. He also suggested that the way in which infrastructure is managed and priced is relevant when assessing their impact. Furthermore, the implausibly high rate of return to public investment of the Aschauer-style analyses soon started to contrast markedly with the evidence produced by micro-level impact analyses (e.g., Munnell 1990; Evans and Karras 1994; Button 1998; Vanhoudt et al. 2000). Growth models that determine the stock of capital endogenously have partially overcome such methodological limitations. When applied to analyse the impact of infrastructure, they deliver completely different results. In this vein, Vanhoudt et al. (2000, p. 102) not only find that “causality does not run from public investment to growth, but rather the opposite way”, but also suggest that public investment “can hardly be considered as an engine for long-run structural growth”. Second, the direct relationship between improvement in the general level of accessibility and economic development (and, by implication, greater cohesion) proved weak when the simplified spatial model of the classical location theory was replaced by a more realistic view of the European territory and its economic geography. On the one hand, the actual importance of transport costs and accessibility per se is changing: If it is true that new forms of transport (such as high speed trains) have created new locational advantages and disadvantages and that the volume of freight movements and travel has generally increased, it is also true that overall transport costs and the significant element of fixed costs they contain (packaging, shipping etc.) are a small (and decreasing) fraction of total industrial cost (Glaeser and Kohlhase 2004; Vickerman et al. 1997). Furthermore telecommunications have had an important and not yet fully understood impact on the mobility of goods and people and new location factors (e.g., quality of life), rather than mere accessibility, have emerged as significant sources of competitive advantage. On the other hand, the change in accessibility induced by the development of TEN-T, in fact, widen (rather than reduce) regional disparities for a series of reasons that include: (a) Providing central and peripheral regions with a similar degree of accessibility may damage firms in lagging regions by opening their local market, unless other advantages are developed (Puga 2002); (b) The problem of peripheral regions seems to lay more in the absence of adequate intra-regional networks for the dispersion of traffic around major centres and the enlargement of the local market rather than in the inter-regional connectivity supported by TEN-T projects (Martin and Rogers 1995; Vickerman 1995).

7.2 Infrastructure and Regional Economic Development

119

In light of all these considerations, it seems realistic to conclude with Button (1998) that “the exact importance of infrastructure as an element in economic development has long been disputed (. . .) but the body of evidence available is far from conclusive” (pp. 154 and 156) thus making the economic justification for the EU infrastructure investment much weaker than in Aschauer’s framework. The potential ambiguity of the impact of transport infrastructure on economic development has been explicitly addressed – in an analytical framework with imperfect competition and increasing returns to scale – by the New Economic Geography (NEG). This approach enables us to address the specific nature of transport infrastructure more effectively when compared to other forms of capital given “its role in facilitating trade and in allowing individuals, companies, regions and nation states to exploit their various competitive advantage” (Button 2001, p. 278). The development of transport infrastructure, by increasing the accessibility of economically weaker regions, “not only gives firms in less developed regions better access to inputs and markets of more developed regions (. . .) but it also makes it easier for firms in richer regions to supply poorer regions at a distance, and can thus harm the industrialisation prospects of less developed areas” (Puga 2002, p. 396). By allowing even a priori identical regions to endogenously differentiate between an industrialised core and a backward periphery in response to changes in their degree of accessibility1 – NEG models have formally accounted for the potentially ambiguous effect of changes in the degree of accessibility (“two-way” roads effect) (Puga 2002). In addition, they have highlighted the differential effect of inter- and intra-regional connections and the hub-and-spoke effect generated by uneven access conditions to major infrastructures. This strand of literature has shown that the development of transport infrastructure needs to be treated in a geographical perspective in order to reveal the specificities of this form of public capital and its impact on economic development. However, from an operational point of view, this approach presents important limitations as it poses significant obstacles to any attempt to empirically test its analytical models. Furthermore, contributions in NEG lack a univocal conclusion on the actual determinants of the economic success of a region (apart from history and/or path dependence) in response to changes in its accessibility (Martin 1999a; Neary 2001). As a consequence the direct regional policy implications of this approach have so far been limited.

1 These changes in accessibility, in turn, alter the balance between the set of dispersion and agglomeration forces constantly at work in the economy. Indeed, in these models the equilibrium depends on the interactions between agglomeration forces (economies of scale, home market effect, backward and forward linkages, labour pool) and dispersion forces (prices for intermediates, wages, competition).

120

7.2.2

7 What Can We Learn From the “Integrated Approach” To Regional Development?

A Broader Theoretical Framework for the Assessment of the Impact of Transport Infrastructure Development

Any model trying to assess the full impact of the endowment and of new investment in infrastructure in any given region has to take into consideration the overall set of conditions that shape the relationship between accessibility and regional growth dynamics, which the NEG has fundamentally left unexplored (Cheshire and Magrini 2002). A variety of forces in any given economy exert an influence on how economic performance can react to changes in accessibility. These include a raft of education, innovation and institutional factors that determine the potential of any space to benefit or not from relative changes in accessibility (Rodrı´guez-Pose 1998a; Rodrı´guez-Pose and Crescenzi 2008). Local actors, factors, and institutions need to be taken into account as the implementation of successful infrastructure policies will depend on the dynamic interaction between accessibility and local conditions. Chapter 2 has shown that growth is a multivariate process, where not only infrastructure endowment and investment, but also innovative efforts in the form of R&D activities, human capital accumulation, the sectoral specialisation of the labour force, migration and geographical location, among other factors, exert a direct influence and interact with one another in order to determine the economic dynamism of any given space (Fagerberg et al. 1997; Cheshire and Magrini 2002). These factors, associated in unique ways in any given territory, respond and adjust to external changes in different ways (Rodrı´guez-Pose 1998a). While some economic factors (such as capital and technology) are more able to adjust in response to external challenges – such as EU integration – by virtue of their relatively higher mobility, social structures tend to be much less flexible. Consequently specific sets of structural conditions will be associated with diverse levels of economic performance. As discussed in Chap. 4, different local features of the labour force, the level of employment of local resources, the demographic structure and change or the accumulation and quality of human capital are among the factors that need to be taken into consideration when analysing the impact of infrastructure endowment and investment, as a similar investment on infrastructure in two different regions may lead to different outcomes as a consequence of the interaction with local economic conditions. Furthermore, as transport infrastructure is about connectivity any analysis of its economic impact needs to be placed in a spatial perspective by considering both the impact of endogenous conditions as well as those of neighbouring regions. Transport infrastructure endowment (and investment) may exert an influence on economic activity which is not limited by regional boundaries. The effect of transport infrastructure in one region may spill over into another, significantly affecting/benefiting its economic performance: As highlighted by Puga (2002, p. 400) “sometimes the project in a single region can have a strong welfare effect rippling through numerous regions”. However, this spillover effect is affected by distance decay and thus tends to be bounded in space, essentially benefiting/affecting neighbouring areas (Seitz 1995; Chandra and Thompson 2000). Since transport infrastructure’s “impact may be prone to leak outside (of) small economic areas” (Chandra and Thompson 2000, p. 458), the assessment of this spillover effect needs to be included in the empirical analysis as appraisals based on too tightly drawn study areas may

7.3 The Model

121

lead to biased estimations (Holl 2006). In this perspective, it is necessary not only to capture the shorter-run Keynesian effect of infrastructure expenditure or the effect of relocation of economic activities in response to the change in transport costs, but also to provide a full appraisal of the impact of the network benefits arising when transport infrastructure allows for closer interactions with economic agents from neighbouring regions, thus increasing their interactions and possibly spreading agglomeration benefits (Rosenthal and Strange 2003). For this reason, coherently with the conceptual framework developed in Chap. 2 – we extend the standard “new growth theory” perspective on externalities (see Vanhoudt et al. 2000) to account for the impact of transport infrastructure upon the possibility/probability/frequency of keeping in touch with different ideas/organizations/products from those locally available thus either directly affecting growth by providing an additional source of knowledge or producing an indirect effect (Crescenzi 2005). We address the issue of spatial externalities arising from transport infrastructure from this standpoint by explicitly considering the intensity of innovative efforts pursued in neighbouring regions. The spatial boundedness of knowledge spillovers singled out in Chap. 5 (Audretsch and Feldman 2004; Cantwell and Iammarino 2003; Sonn and Storper 2008) may – even in the presence of equally good inter-regional connectivity – allow highly-accessible core regions to benefit from innovative activities pursued in their proximity, while preventing spillovers from reaching peripheral remote regions.

7.3

The Model

In accordance with the framework developed in the previous chapters, the empirical investigation aims at integrating the role of infrastructure in shaping economic growth in Europe into a model that takes into consideration other endogenous and external factors. The choice of empirical variables to be included in the model is determined according to the following matrix: Internal Factors Infrastructure endowment Kilometres (kms) of motorways and investment (level and annual change) R&D Investment in R&D in the region Relative wealth GDP per capita Agglomeration economies Social filter

Human capital mobility National effects

Total regional GDP

External Factors (Spillovers) Infrastructure in neighbouring areas Investment in R&D in neighbouring regions GDP per capita in neighbouring regions Total GDP in neighbouring regions Same characteristics in neighbouring regions

Structural characteristics that would make a region more “innovation prone”, including: l Education l Sectoral composition l Use of resources (unemployment) l Demographics Migration rate Migration in neighbouring areas National growth rate

122

7 What Can We Learn From the “Integrated Approach” To Regional Development?

By developing the framework above, we obtain the following empirical model: yi;t ¼ ai þ b ln GDP0i;t þ gInfi;t þ dxi;t þ zSpillInfi;t þ #Spillxi;t þ k ln Nayi;t þ ei;t

(7.1)

where: y lnGDP0 Inf x Spill Nay e

Represents the growth rate of regional GDP per capita; Is the initial level of GDP per capita; Denotes infrastructure endowment and investment; Is a set of structural features/determinants of growth of region i; Indicates the presence of these factors in neighbouring regions; Represents the national growth rate of per capita GDP of the member state region i belongs to; Is an idiosyncratic error;

and where i represents the region and t time. In the following paragraphs we describe the variables included in the model in detail: Growth rate of regional GDP per capita: The annual growth rate of regional GDP is the dependent variable and is used as a proxy for the economic performance of the region. Level of GDP per capita: As customary in the literature on the determinants of regional growth performance, the initial level of GDP per capita is introduced in the model in order to account for the region’s initial wealth. The significance and magnitude of the coefficient associated to this variable will allow us to test the existence of a process of convergence in regional per capita income and measure its speed. Existing stock and annual variation of transport infrastructure endowment: Transport infrastructure may affect economic performance through a variety of mechanisms not only related to its contribution to the regional stock of public capital, but also associated to its influence on the spatial organisation of economic activities. In order to capture the direct impact of transport infrastructure on regional growth, the model includes a specific proxy for both the stock of transport infrastructure and the annual additional investment in this area. The former is proxied by the kilometres (kms) of motorways (Canning and Pedroni 2004; see Table A.1 in the Appendix for further detail on the definition of the variable) in the region, standardised by regional population,2 while the latter is proxied by its annual change. Although other indicators may be as useful in capturing the role of transport infrastructure, the length of regional motorways (and change thereof) is adopted as a proxy for regional infrastructure because of, first, the constraints in terms of regional data availability3 (which a priori prevented us from considering alternative 2 Dividing by population is to account for the different size of regions. The proxy for infrastructure endowment has also been standardised by the total surface of the region and by its total GDP. As shown in the appendix, the results of the analysis do not change significantly when these alternative proxies are used. 3 See Appendix A for a discussion of the source of the data and for a discussion of data availability and limitations.

7.3 The Model

123

accurate proxies) and, second, its capacity to capture in a direct way the impact of better accessibility irrespective of the differentiated and hard-to-quantify cost of accessibility in different regions and countries. Motorways have been preferred to other modes of transport (e.g., rail) for their greater use in the shipment of goods, which results in their stronger impact on the spatial allocation of economic activity (Button 2001; Puga 2002). Furthermore, the initial emphasis of the main EU support plan for transport infrastructure development (the TEN-T) was on motorways before shifting to high-speed trains. As a consequence this mode of transport has benefited from policy support for a long enough time span as to allow a meaningful policy assessment. In addition to infrastructure, our variable of interest, we include some additional drivers of regional economic performance as independent variables. These include: R&D expenditure: The percentage of regional GDP devoted to R&D is the main measure of economic input used to generate innovation in each region. From an endogenous growth perspective this variable is regarded – as one of the key factors behind long-run inter-regional differences in productivity and income. Local R&D expenditure is also frequently used as a proxy for the local capability to adapt to innovation produced elsewhere (Cohen and Levinthal 1990; Maurset and Verspagen 1999). There are, however, measurement problems associated with this variable that must be borne in mind as they may partially hide the contribution of R&D towards economic performance. First, the relevant time-lag structure for the effect of R&D activities on productivity and growth is unknown and may vary significantly across sectors (Griliches 1979). Second, as pointed out by Bilbao-Osorio and Rodrı´guez-Pose (2004) for the case of European regions, the returns from public and private R&D investment may vary significantly. Furthermore, the fact that not all innovative activities pursued at the company level are classified as formal “Research and Development” may be a source of further bias in the estimations. Having acknowledged these points, we assume R&D expenditure to be a proxy for “the allocation of resources to research and other information-generating activities in response to perceived profit opportunities” (Grossman and Helpman 1991, p. 6) in order to capture the existence of a system of incentives (in the public and the private sector) towards intentional innovative activities. Socio-Economic Conditions: As in the rest of this book, structural socio-economic conditions are introduced into the analysis by means of a composite index, which combines a set of variables describing the socio-economic dynamism of the region. In the framework discussed in the previous section the structural dynamism of the region is a crucial pre-condition for its capability to benefit from changes in accessibility due to investment in new transport infrastructure. The analyses pursued in previous chapters have shown that the socio-economic features that seem to be more relevant for shaping the reaction capabilities of a region are those related to two main domains: Educational achievement (Lundvall 1992; Malecki 1997) and the productive employment of human resources (Fagerberg et al. 1997; Rodrı´guez-Pose 1999). The first domain is measured by educational attainment, expressed by the shares of persons with completed tertiary education, both relative to the labour force and to the overall population (human capital accumulation in the labour force and in the

124

7 What Can We Learn From the “Integrated Approach” To Regional Development?

population respectively). The structure of productive resources is measured by the percentage of the labour force employed in agriculture and the percentage of longterm unemployment. As in previous chapters we deal with problems of multicollinearity, which prevent the simultaneous inclusion of all these variables in our model, by means of principal component analysis (PCA). PCA allows us to merge the variables discussed above into a single indicator (called “social filter index”) that preserves as much as possible of the variability of the source data (see Appendix C). When looking at the output of the PCA,4 the eigenanalysis of the correlation matrix shows that the first principal component alone is able to account for around 57 and 58% of the total variance for the EU-15 and EU-25. The first principal component scores are computed from the standardised value of the original variables by using the coefficients listed under PC1. These coefficients assign a large weight to educational achievement; this is a major component of the socio-economic fabric of the regions. A negative weight is assigned, instead, to the long term component of unemployment and to the percentage of agricultural labour. This first Principal Component (PC1) – which explains 58% of the total 4

Results of the Principal Component Analysis (Panel Data Datasets). Eigenanalysis of the Correlation Matrix EU 15 Component Comp1 Comp2 Comp3 Comp4

Eigenvalue 2.2744 0.96758 0.734233 0.0237857

Difference 1.30682 0.233347 0.710447 .

Proportion 0.5686 0.2419 0.1836 0.0059

Cumulative 0.5686 0.8105 0.9941 1

Comp1 Comp2 Comp3 Comp4

2.30323 0.964829 0.714565 0.0173775

EU 25 1.3384 0.250263 0.697188 .

0.5758 0.2412 0.1786 0.0043

0.5758 0.817 0.9957 1

Principal Components’ Coefficients EU 15 Variable Agricultural Labour Force Long Term Unemployment Education Population Education Employed People

PC1 0.3942 0.2551 0.632 0.6165

PC2 0.3369 0.851 0.233 0.3288

PC3 0.855 0.4537 0.1914 0.1627

PC4 0.0098 0.0698 0.7139 0.6967

Agricultural Labour Force Long Term Unemployment Education Population Education Employed People

EU 25 0.4009 0.2662 0.6271 0.6125

0.3471 0.8389 0.2478 0.3381

0.8478 0.4697 0.1912 0.1549

0.0046 0.0686 0.7133 0.6975

7.3 The Model

125

variance of the original indicators and constitutes what we call the “Social Filter Index”, introduced into the regression analysis as an aggregate proxy for the socioeconomic conditions of each region. Given its theoretical and empirical relevance (large weight in the PCA), a specific proxy for human capital accumulation will be introduced separately into the regression analysis. In addition to the endogenous variables, the model also includes variables representing the potential spillovers from neighbouring regions that may affect economic performance in the region of interest. These spillover variables are: Extra-Regional Infrastructure (Endowment and Investment): In order to assess the impact of infrastructure on regional economic growth in the most comprehensive way possible, the model needs to account for development both within each individual region and across the whole of Europe, as what matters is not only the relative density of infrastructure within the borders of the region, but also the endowment of infrastructure in neighbouring regions. Hence in our framework transport infrastructure is not reduced to mere components of the “aggregate” neo-classical concept of physical capital but includes the potential for networking and connectivity among individuals and firms. We thus introduce the endowment of transport infrastructure in neighbouring regions as a proxy for the degree of inter-regional connectivity. Where the internal infrastructure endowment is reinforced by a good endowment in neighbouring regions the most favourable infrastructural conditions are supposed to be in place. Where, instead, internal infrastructures are not complemented by adequate neighbourhood conditions, bottlenecks and criticalities may arise, negatively affecting the accessibility of the region. Following the same line of reasoning, changes in infrastructure endowment (through new investment) may exert an influence on the economic performance of neighbouring regions. Extra-regional infrastructure endowment is proxied by the average of infrastructure intensity in neighbouring regions. The extra-regional regional infrastructure endowment SpillInfi is calculated as: SpillInfi ¼

n X

Infj wij

(7.2)

j¼1

where Infj is a proxy for the infrastructure endowment of the j-th region and wij is a generic “spatial” weight. In order to minimize both the endogeneity induced by traveltime distance weight and the potential bias due to the different number of neighbours of central and peripheral regions, we consider the k nearest neighbours (with k ¼ 4)5:

5

Alternative definitions for the spatial weights matrix are possible: distance weights matrices (defining the elements as the inverse of the distances) and other binary matrices (rook and queen contiguity matrices). However, the k-nearest-neighbours weighting scheme seems the most appropriate to capture the neighbourhood effect while minimising the endogeneity due to higher infrastructure density in central regions. The use of different values for the parameter k generated similar results to those presented in the paper.

126

7 What Can We Learn From the “Integrated Approach” To Regional Development?

( wij ¼

1=k if j is one of the k nearest neighbours to i 0 otherwise

with i 6¼ j

(7.3)

Extra-Regional Innovation: The economic success of an area depends both on its internal conditions and on those of neighbouring interconnected regions. In particular, when innovative activities pursued in neighbouring regions are shown to exert a positive impact on local economic performance, there is evidence in favour of inter-regional spillover effects: Knowledge produced in one region spills over into another (through the mechanisms discussed in the previous section), thereby influencing its economic performance. The spillover variable captures the “aggregate” impact of innovative activities pursued in the neighbourhood. The significance of this indicator suggests that access to extra-regional innovation facilitates the interregional transfer of knowledge: Proximity enables the transmission of knowledge which, in turn, has an impact on regional growth. The measure of “accessibility” of extra-regional innovative activities is calculated in the same way as that of the accessibility of extra-regional infrastructure presented in (7.2). For each region i: SpillR&Di ¼

n X

R&Dj wij

(7.4)

j¼1

where R&D is our proxy for regional innovative efforts and w is as above. Agglomeration and absolute size of the local economy: Different territorial configurations of the local economy may give rise to different degrees of agglomeration economies. The geographical concentration of economic activity has an impact on productivity (Duranton and Puga 2003), which needs to be controlled for in order to single out the differential impact of infrastructure endowment. From this perspective, the relative concentration of wealth (the “scale” side of agglomeration economies) and the absolute size of clusters need to be considered. A useful proxy for these factors is the total GDP of each region. Migration: The degree of internal6 labour mobility is reflected by the regional rate of migration (i.e., the increase or decrease of the population due to migration flows as a percentage of the initial population). A positive rate of migration (i.e., net inflow of people from other regions) is a proxy for the capacity of the region to benefit from better accessibility and transport infrastructure by attracting new workers, increasing the size of its labour pool and its “diversity” in terms of skills and cultural background. 6 Migration data are provided by Eurostat in the “Migration Statistics” collection. However there are no data for Spain and Greece. Consequently, in order to obtain a consistent measure across the various countries included in the analysis, we calculate this variable from demographic statistics. “Data on net migration can be retrieved as the population change plus deaths minus births. The net migration data retrieved in this way also include external migration” (Puhani 2001, p. 9). Net migration was standardised by the average population, obtaining the net migration rate. Consequently, it is impossible to distinguish between national, intra-EU, and extra-EU migration flows.

7.4 Results of the Analysis

7.4 7.4.1

127

Results of the Analysis Estimation Issues, Data Availability, and Units of Analysis

We estimate the model by means of Fixed-Effect Panel Data regressions.7 The effect of spatial autocorrelation (i.e., the lack of independence among the error terms of neighbouring observations) is minimized by explicitly controlling for national growth rates. Furthermore, by introducing the “spatially lagged” variables SpillInf and Spillx in our analysis, we take into consideration the interactions between neighbouring regions, thereby minimizing their effect on the residuals. Spatial autocorrelation in the residuals has been checked for by using the Moran’s I test (Cliff and Ord 1972) for each year. The test statistics are not significant for the majority of the years covered by the regression. In all other cases the magnitude of Moran’s I is low (See Appendix F). Given that the Moran’s I test does not detect spatial autocorrelation in the residuals, we believe that the combination of “national” variables and spatially lagged explanatory variables are able to capture a significant part of the total spatial variability of the data. This allows us to produce robust estimates of the parameters without relying on a spatially autoregressive specification of the model (Elhorst 2010) that would not be fully justifiable in light of the theoretical analysis and, in any case would preclude any ceteris paribus interpretation of the estimated coefficients (Anselin 2003, p. 158). In addition the procedure for the estimation of spatially autoregressive panel data models with fixed effect is still debated in the literature (see, for example, Lee and Yu 2010) discouraging us from further pursuing this direction. Another concern is endogeneity, which we partially address by introducing all explanatory variables with a 1-year lag. In addition, in order to resolve the problem of different accounting units, explanatory variables are expressed, for each region, as a percentage of the respective GDP or population. The model is run for 1990–2004 for the EU-15 and for 1995–2004 for the New Member States in line with data availability. As a consequence, longer term effects can only be analysed for the EU-15. Instead, when the sample is extended to the EU-25, the analysis is necessarily more prone to the potential distortions ascribable to the economic cycle. This sort of analysis should ideally be focused on FURs (Functional Urban Regions)8 rather than on NUTS (Nomenclature des unite´s territoriales statistiques) administrative regions, as it would allow a more accurate separation of the impact of infrastructure provision within the borders of

7 According to the Breutsch and Pagan Test fixed effects estimation has to be preferred given the high significance of the individual effects. 8 The concept of FURs has been defined as a means to minimize the bias introduced by commuting patterns. A FUR includes a core city, where employment is concentrated, and its hinterland, from which people commute to the centre. For a detailed analysis of this concept see Cheshire and Hay (1989).

128

7 What Can We Learn From the “Integrated Approach” To Regional Development?

a functionally integrated area from that attributable to an increase in connectivity between different functional regions. Unfortunately, as discussed in Appendix A the lack of available data for many of the relevant explanatory variables has prevented contemplating functional regions. As a consequence – as in all chapters of this book – we have been forced to rely on the mix of NUTS1 and NUTS2 regions presented in the Appendix A, selected in order to maximise their homogeneity in terms of the relevant governance structure and also considering data availability. The unit of analysis with the greatest relevance in terms of the institutions that may be relevant for the decision of developing new transport infrastructure – or which may be taken as a target area for such investment by the national government and/or the European Commission – was selected for each country. Consequently, the analysis uses NUTS1 regions for Belgium, Germany9 and the United Kingdom and NUTS2 for all other countries (Austria, Finland, France, Italy, the Netherlands, Portugal, Spain, Sweden in the EU15, as well as in the Czech Republic, Hungary, Poland, and Slovakia, when the EU25 is considered). Countries without equivalent sub-national regions (Denmark,10 Ireland, Luxembourg for the EU15 and Cyprus, Estonia, Latvia, Lithuania, and Malta for the EU25) are, as a consequence of the need to control for national growth rates, excluded a priori from the analysis.11 Lack of regional data on infrastructure from either Eurostat or national authorities means that Greece cannot be taken into consideration either. The entire dataset is based on Eurostat Regio data with the exception of the statistics on educational achievement which are based on Labour Force Survey Data provided by Eurostat through the European Investment Bank. Where fully comparable data are available, missing data in Eurostat Regio have been complemented by data from National Statistical Offices. Appendix A provides a detailed definition of the variables included in the analysis and further detail on the sources used to complement Eurostat data. In a few cases where information for a specific year and region was missing in all sources, the corresponding value has been calculated by linear interpolation or extrapolation.

9 The NUTS2 level corresponds to Provinces in Belgium and to the German Regierungsbezirke. In both cases these statistical units of analysis have little administrative and institutional meaning. For these two countries the relevant institutional units are Re´gions and L€ander, respectively, codified as NUTS1 regions. The lack of correspondence between NUTS2 level and actual administrative units accounts for the scarcity of statistical information on many variables (including R&D expenditure) below the NUTS1 level for both countries. 10 Even if Denmark introduced regions above the local authority level on January 1 2007 in line with the NUTS2 classification, regional statistics are not available from Eurostat. 11 As far as specific regions are concerned, no data are available for the French De´partments d’Outre-Mer (FR9). Trentino-Alto Adige (IT31) has no correspondent in the NUTS2003 classification. Due to the nature of the analysis, the islands (PT2 Ac¸ores, PT3 Madeira, FR9 De´partements d’Outre-mer, ES7 Canarias) and Ceuta y Melilla (ES 63) were not considered due to the problems with the computation of the spatially lagged variables.

7.4 Results of the Analysis

7.4.2

129

Stylised Facts

Figures 7.1a, 7.2a and 7.3a (for the EU-15) as well as 7.1b, 7.2b and 7.3b (for the EU-25) provide a visual representation of the phenomena under analysis. Figure 7.1a and b plot the initial GDP per capita of each region against the corresponding growth rate over the 1990–2004 and 1995–2004 periods respectively. Both figures show some (weak) degree of regional convergence. The figures also confirm the positive economic performance of capital city regions (such as Brussels, Paris, and Stockholm) and of the regions where highly innovative activities are concentrated (such as Bavaria, Bremen, Utrecht, and South East England). Fast growth is recorded in some initially disadvantaged regions of the EU-15 in Spain and Portugal and in some regions of the new member states of the Union (in particular in the capital city regions of Bratislava, Budapest, and Warsaw, and, to a lesser extent, Prague). Conversely, a number of regions in the lower area of each figure are either unable to catch up with the rest of the EU (on the left hand side) or show a less dynamic economic performance. The regression analysis will provide some explanations for the observed regional growth pattern and shed some new light on the potential role of transport infrastructure (and its development) on these dynamics. Figure 7.2a, b represent a first picture of infrastructure endowment and its change over time in the regions of the EU-15 and EU-25, respectively. Figure 7.2a highlights the development of major transport connections in the Objective 1 regions, fundamentally in Spain and Portugal but also in Austria (Burgenland), Sweden (Norra Mellansverige) and in the Objective 2 region of Basse Normandie (France). Figure 7.2b also reveals the effort of some regions in the new member states of developing new transport infrastructure. This is particularly evident in the western regions of Poland, in Hungary (Nyugat-Dunantul and Eszak-Magyarorszag), and in Slovakia (Za´padne´ Slovensko and Vy´chodne´ Slovensko). Improvements in these regions contrast with the backwardness of a large number of EU-25 regions which still show a very low density of transport infrastructure (lower left-hand corner of the figure). In general, “so far as roads are concerned, there are continuing differences between the EU-15 countries and the new Member States in the density of motorways: With the exception of Slovenia and Lithuania, they all score under 50% of the EU average” (European Commission 2007a, p. 60). Figure 7.3a, b combine regional growth dynamics and infrastructure development in the same picture. The Figure plots information on initial regional GDP per capita (x-axis), the annual average real growth rate (y-axis), and the corresponding variation in transport infrastructure endowment with the area of the circles being proportional to the percentage increase in motorway density (kms per thousand inhabitants). In Fig. 7.3a (and, to a lesser extent, in Fig. 7.3b) transport infrastructure investment seems to be higher in the regions showing a marked “convergence” trend over the observation period. In other words, the figures suggest some correlation between infrastructure investment and regional convergence. However, the

130

7 What Can We Learn From the “Integrated Approach” To Regional Development?

0

Annual Average Real Growth Rate 1990-2004 .5 1 1.5 2 2.5

a

0

10000

20000

30000

40000

30000

40000

GDP per capita, 1990

Annual Average Real Growth Rate 1995-2004 0 2 4 6 8 10

b

0

10000

20000 GDP per capita, 1995

Fig. 7.1 (a) EU-15: Initial GDP conditions and regional growth rate, 1990–2004. (b) EU-25: Initial GDP conditions and regional growth rate, 1995–2004

7.4 Results of the Analysis

131

% Variation in Kms of motorways per '000 inhab., 1990-2004 0 200 400 600

a

0

.2 .4 Kms of motorways per thousand inhabitants, 1990

.6

0

.1 .2 .3 .4 Kms of motorways per thousand inhabitants, 1995

.5

% Variation in Kms of motorways per '000 inhab., 1995-2004 0 200 400 600

b

Fig. 7.2 (a) EU-15: Endowment and change in motorways per thousand inhabitants, 1990–2004. (b) EU-25: Endowment and change in motorways per thousand inhabitants, 1995–2004

132

7 What Can We Learn From the “Integrated Approach” To Regional Development?

Annual Average Real Growth Rate 1990-2004 0 .5 1 1.5 2 2.5

a

0

10000

20000 GDP per capita, 1990

30000

40000

Circular area proportional to percent variation in Kms of motorways per '000 inhab., 1990-2004

Annual Average Real Growth Rate 1995-2004 0 2 4 6 8 10

b

0

10000

20000 GDP per capita, 1995

30000

40000

Circular area proportional to percent variation in Kms of motorways per '000 inhab., 1995-2004

Fig. 7.3 (a) EU-15: Initial GDP conditions, regional growth and infrastructure development, 1990–2004. (b) EU-25: Initial GDP conditions, regional growth and infrastructure development, 1995–2004

7.4 Results of the Analysis

133

correlation is far from perfect, implying the need for more careful investigation of the factors conditioning such a relationship, i.e., the set of local conditions which allow infrastructure investment to foster regional economic performance.

7.4.3

Transport Infrastructure and Regional Growth in the EU

The results for the estimation of the model of empirical analysis are presented in Tables 7.1a, 7.2a and 7.3a for the EU-15 covering the 1990–2004 period and in Tables 7.1b, 7.2b and 7.3b for the EU-25 covering the 1995–2004 time span.12 In regressions 1–2 of Tables 7.1a and 7.1b the preferred proxies for infrastructural endowment and investment are introduced together with the controls for initial conditions and spatial autocorrelation (i.e., the level of GDP per capita and the national growth rate, respectively). In regressions 3–4 the impact of the same indicators in neighbouring regions is assessed. In regression 5 our proxies for local innovative efforts and knowledge spillovers are introduced, broadening the analysis from a “new growth theory” perspective. From regression 6 onwards the variables relative to the socio-economic conditions (“Social Filter Index”), the accumulation of human capital and the territorial organisation of the local economy (migration and agglomeration) are introduced sequentially. Controlling for the level of GDP per capita whose significantly negative (albeit small) coefficient suggests a weak trend towards regional convergence, regression 1 in Tables 7.1a and 7.1b seems to offer a result in line with analyses a` la Aschauer: The local endowment of transport infrastructure is an important and robust predictor of economic growth. Both for the EU-15 (longer term effect) and the EU-25 (shorter time-span) the density of local transport infrastructure shows a positive and highly significant coefficient and this coefficient is robust to the introduction of additional control variables in the equation in columns (2–8). However, this picture of the regional growth mechanics changes immediately when the impact of further investment in infrastructure is assessed. The annual change in infrastructure 12

The constraints in terms of data availability have forced us to rely on a relatively short time span, producing a “large N/small T” panel whereby the cross-sectional dimension of the dataset (N) is significantly larger than the time dimension (T). In this context, the low time-series variability of the dataset a priori prevents non-stationarity from affecting our estimates through spurious correlation. The hypothesis of stationarity is confirmed by three different unit root tests for panel data (the Im-Pesaran-Shin, the augmented Dickey-Fuller and the Phillips-Perron tests) which, as expected, reject the hypothesis of non-stationarity at conventional significance levels (Tables D.1 and D.2 in Appendix D). The R2 confirms the overall goodness-of-fit of all the regressions presented. Following Wooldridge (2002, pp. 275–276), the estimates are based on a “robust variance matrix estimator [which] is valid in the presence of any heteroskedasticity or serial correlation [. . .], provided that T is small relative to N”. The large sample size also allows us to rely on the asymptotic theory to consider the distribution of test statistics as standard also with non-normal residuals. Furthermore, in order to minimize spatial correlation the national growth rate has been included in all equations.

1680 120 0.14 0.14 0.04 0.13

Observations Number of groups (NUTS regions) R-squared R-squared within R-squared overall R-squared between

1560 120 0.16 0.16 0.05 0.10

0.468*** (0.074) 1560 120 0.18 0.18 0.05 0.13

0.708*** (0.084)

0.049*** (0.018) 0.076*** (0.009) 0.004*** (0.000) 0.054 (0.048) 0.169*** (0.023)

(3)

1560 120 0.19 0.19 0.05 0.13

0.732*** (0.084)

0.046** (0.018) 0.079*** (0.009) 0.004*** (0.000) 0.048 (0.046) 0.186*** (0.024) 0.180*** (0.067)

(4)

plus: Infrastructure network effects

1560 120 0.20 0.20 0.04 0.12

0.794*** (0.088)

0.058*** (0.018) 0.086*** (0.009) 0.004*** (0.000) 0.046 (0.045) 0.210*** (0.026) 0.194*** (0.067) 0.001*** (0.000) 0.001* (0.001)

(5)

1560 120 0.25 0.25 0.05 0.19

0.013*** (0.002) 1.626*** (0.140)

0.043** (0.017) 0.171*** (0.015) 0.004*** (0.000) 0.039 (0.043) 0.185*** (0.025) 0.162*** (0.060) 0.001*** (0.000) 0.000 (0.000)

(6)

1560 120 0.24 0.24 0.05 0.17

1.327*** (0.116)

0.045*** (0.017) 0.145*** (0.013) 0.004*** (0.000) 0.032 (0.042) 0.202*** (0.026) 0.151** (0.061) 0.001*** (0.000) 0.000 (0.001) 0.002*** (0.000)

(7)

1560 120 0.25 0.25 0.01 0.02

1.297*** (0.118)

0.042** (0.017) 0.220*** (0.034) 0.004*** (0.000) 0.035 (0.042) 0.213*** (0.027) 0.151** (0.060) 0.001*** (0.000) 0.000 (0.000) 0.002*** (0.000) 0.074*** (0.028) 0.001*** (0.000)

(8)

plus: Innovation activities plus: Further socio-economic control and spillovers variables

Robust standard errors in parentheses; *significant at 10-% level; **significant at 5-% level%; ***significant at 1-% level

0.447*** (0.075)

Constant

Social Filter Index

Migration rate

Log of total Gross Value Added (level)

Ratio of employed people with Higher education in percent

Spatial weighted average of Total R&D expenditure

Total intra-regional R&D expenditure (all sectors) in percent of GDP

Spatial weighted average of Change in kms of motorways/1,000 inhab.

0.117*** (0.017) 0.049*** (0.008) 0.004*** (0.000) 0.075 (0.052)

(2)

(1)

0.093*** (0.015) 0.047*** (0.008) 0.005*** (0.000)

Simple model: Infrastructure endowment and investment

Simple model: Infrastructure endowment

Spatial weighted average of kms of motorways/1,000 inhab.

Change in kms of motorways/1,000 inhab.

Annual national growth rate

Kilometres of motorways per 1,000 inhabitants Log of GDPpc

Dependent variable:Regional GDP per capita (annual growth rate)

Table 7.1a EU-15: Regional growth and transport infrastructure, 1990–2004 134 7 What Can We Learn From the “Integrated Approach” To Regional Development?

(1) 0.066*** (0.020) 0.070*** (0.009) 0.003*** (0.000)

1449 161 0.15 0.15 0.08 0.17

Observations Number of groups (NUTS regions) R-squared R-squared within R-squared overall R-squared between

1288 161 0.19 0.19 0.06 0.12

0.941*** (0.108)

(2) 0.107*** (0.025) 0.101*** (0.012) 0.003*** (0.000) 0.107*** (0.037)

1288 161 0.19 0.19 0.06 0.12

0.942*** (0.118)

(3) 0.107*** (0.030) 0.101*** (0.013) 0.003*** (0.000) 0.107*** (0.038) 0.001 (0.036)

1288 161 0.20 0.20 0.06 0.12

0.955*** (0.120)

(4) 0.104*** (0.030) 0.102*** (0.013) 0.003*** (0.000) 0.103*** (0.038) 0.018 (0.038) 0.100 (0.070)

1288 161 0.20 0.20 0.06 0.12

0.976*** (0.122)

(5) 0.110*** (0.029) 0.106*** (0.013) 0.003*** (0.000) 0.100*** (0.038) 0.023 (0.039) 0.079 (0.068) 0.002 (0.001) 0.004** (0.002)

Robust standard errors in parentheses; *significant at 10-% level; **significant at 5-% level; ***significant at 1-% level

0.659*** (0.082)

Constant

Social Filter Index

Migration rate

Log of total Gross Value Added (levels)

Ratio of employed people with Higher education in percent

Spatial weighted average of Total R&D expenditure

Total intra-region R&D expenditure (all sectors) in percent of GDP

Spatial weighted average of Change in kms of motorways/1,000 inhab.

Spatial weighted average of kms of motorways/1,000 inhab.

Change in kms of motorways/1,000 inhab.

Annual national growth rate

Log of GDPpc

Dependent variable:Regional GDP per capita (annual growth rate) Kilometres of motorways per 1,000 inhabitants

Table 7.1b EU-25: Regional growth and transport infrastructure, 1995–2004

1288 161 0.20 0.20 0.06 0.12

0.001 (0.002) 0.950*** (0.139)

(6) 0.111*** (0.029) 0.103*** (0.015) 0.003*** (0.000) 0.100*** (0.038) 0.031 (0.038) 0.083 (0.068) 0.002 (0.002) 0.005** (0.002)

1288 161 0.20 0.20 0.06 0.13

1.040*** (0.141)

(7) 0.109*** (0.029) 0.113*** (0.016) 0.003*** (0.000) 0.101*** (0.038) 0.016 (0.038) 0.069 (0.067) 0.002 (0.002) 0.004** (0.002) 0.000* (0.000)

1288 161 0.22 0.22 0.02 0.05

1.126*** (0.142)

(8) 0.104*** (0.028) 0.027 (0.037) 0.002*** (0.000) 0.098** (0.039) 0.061 (0.038) 0.093 (0.067) 0.002** (0.001) 0.004** (0.002) 0.001** (0.000) 0.142*** (0.033) 0.000 (0.000)

7.4 Results of the Analysis 135

136

7 What Can We Learn From the “Integrated Approach” To Regional Development?

Table 7.2a EU-15: Regional growth and transport infrastructure with annual lags, 1990–2004 (1) (2) (3) (4) (5) Dependent variable:Regional GDP per capita (annual growth rate) Number of annual lags in all 2 3 4 5 6 variables Kilometres of motorways per thousand inhabitants

0.046** (0.023)

0.019 (0.020)

0.028 (0.023)

0.053*** (0.015)

0.026 (0.020)

Change in kms of motorways/ 1,000 inhab. Spatial weighted average of kms of motorways/1,000 inhab. Spatial weighted average of Change in kms of motorways/1,000 inhabitants Log of GDPpc

0.014 (0.031) 0.233*** (0.033) 0.075 (0.049)

0.043* (0.024) 0.083*** (0.025) 0.108** (0.046)

0.048* (0.028) 0.048 (0.030) 0.114** (0.047)

0.012 (0.029) 0.038 (0.023) 0.002 (0.044)

0.049 (0.033) 0.073** (0.029) 0.017 (0.054)

0.256*** (0.043) 0.000** (0.000)

0.123*** (0.043) 0.000 (0.000)

0.161*** (0.035) 0.000 (0.000)

0.049 (0.051) 0.000** (0.000)

0.163*** (0.043) 0.000 (0.000)

0.001 (0.000) 0.002*** (0.000) 0.089** (0.036) 0.000*** (0.000) 0.001*** (0.000) 1.498*** (0.132)

0.002*** (0.000) 0.001*** (0.000) 0.011 (0.038) 0.000** (0.000) 0.000 (0.000) 1.061*** (0.106)

0.001** (0.001) 0.001*** (0.000) 0.042 (0.029) 0.000 (0.000) 0.000 (0.000) 1.122*** (0.128)

0.001*** (0.000) 0.000 (0.000) 0.029 (0.046) 0.000 (0.000) 0.001*** (0.000) 0.763*** (0.131)

0.001** (0.000) 0.001** (0.000) 0.184*** (0.039) 0.001*** (0.000) 0.000 (0.000) 0.332** (0.149)

Total intra-regional R&D expenditure (all sectors) in percent of GDP Spatial weighted average of Total R&D expenditure Ratio of employed people with Higher education in percent Log of total Gross Value Added (levels) Migration rate Annual national growth rate Constant

Observations 1440 1320 1200 1080 960 Number of groups 120 120 120 120 120 (NUTS regions) R-squared 0.18 0.13 0.13 0.13 0.15 R-squared within 0.18 0.13 0.13 0.13 0.15 R-squared overall 0.00 0.02 0.01 0.03 0.00 R-squared between 0.01 0.07 0.03 0.08 0.01 Robust standard errors in parentheses; *significant at 10-% level; **significant at 5-% level; ***significant at 1-% level

7.4 Results of the Analysis

137

Table 7.2b EU-25: Regional growth and transport infrastructure with annual lags, 1995–2004 (1) (2) (3) (4) (5) Dependent variable:Regional GDP per capita (annual growth rate) Number of annual lags in all variables Kilometres of motorways per thousand inhabitants Change in kms of motorways/ 1,000 inhab. Spatial weighted average of kms of motorways/1,000 inhab. Spatial weighted average of Change in kms of motorways/ 1,000 inhabitants Log of GDPpc Total intra-regional R&D expenditure (all sectors) in percent of GDP Spatial weighted average of Total R&D expenditure Ratio of employed people with Higher education in percent Log of total Gross Value Added (levels) Migration rate Annual national growth rate Constant

2

3

4

5

6

0.073** (0.033) 0.015 (0.034) 0.024 (0.044) 0.022 (0.067)

0.025 (0.028) 0.003 (0.041) 0.063 (0.048) 0.033 (0.061)

0.032 (0.032) 0.044 (0.051) 0.129*** (0.045) 0.128* (0.066)

0.010 (0.039) 0.036 (0.038) 0.022 (0.051) 0.018 (0.054)

0.111* (0.062) 0.118* (0.068) 0.063 (0.083) 0.080 (0.093)

0.048 (0.044) 0.001 (0.001)

0.075* (0.043) 0.000 (0.001)

0.079* (0.045) 0.002 (0.002)

0.178** (0.080) 0.002 (0.002)

0.060 (0.110) 0.003** (0.002)

0.006*** (0.001) 0.000 (0.000) 0.176*** (0.041) 0.000 (0.000) 0.002*** (0.000) 1.289*** (0.180)

0.010*** (0.003) 0.000 (0.000) 0.018 (0.039) 0.000*** (0.000) 0.003*** (0.000) 0.890*** (0.136)

0.009*** (0.003) 0.000 (0.000) 0.039 (0.038) 0.000** (0.000) 0.002*** (0.001) 0.399*** (0.128)

0.001 (0.003) 0.000 (0.000) 0.121* (0.072) 0.001*** (0.000) 0.001** (0.000) 0.451** (0.186)

0.002 (0.003) 0.000 (0.000) 0.104 (0.101) 0.000 (0.000) 0.001*** (0.000) 1.585*** (0.185)

Observations 1127 966 805 644 483 Number of groups (NUTS 161 161 161 161 161 regions) R-squared 0.22 0.21 0.12 0.12 0.37 R-squared within 0.22 0.21 0.12 0.12 0.37 R-squared overall 0.02 0.08 0.05 0.01 0.19 R-squared between 0.04 0.13 0.07 0.01 0.29 Robust standard errors in parentheses; *significant at 10-% level; **significant at 5-% level; ***significant at 1-% level

(1) 0.042** (0.017) 0.035 (0.042) 0.213*** (0.027) 0.151** (0.060) 0.220*** (0.034) 0.001*** (0.000) 0.000 (0.000) 0.002*** (0.000) 0.074*** (0.028) 0.001*** (0.000) 0.004*** (0.000)

Lag 4 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 4 Change in kms of motorways/1,000 inhab.

Lag 3 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 3 Change in kms of motorways/1,000 inhab.

Lag 2 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 2 Change in kms of motorways/1,000 inhab.

Annual national growth rate

Migration rate

Log of total Gross Value Added (levels)

Ratio of employed people with Higher education in percent

Spatial weighted average of Total R&D expenditure

Total intra-regional R&D expenditure (all sectors) in percent of GDP

Log of GDPpc

Spatial weighted average of Change in kms of motorways/1,000 inhab.

Spatial weighted average of kms of motorways/1,000 inhab.

Change in kms of motorways/1,000 inhab.

Dependent variable:Regional GDP per capita (annual growth rate) Kilometres of motorways per thousand inhabitants

(2) 0.044** (0.020) 0.028 (0.043) 0.203*** (0.032) 0.166*** (0.063) 0.205*** (0.040) 0.001*** (0.000) 0.000 (0.001) 0.002*** (0.000) 0.057* (0.033) 0.000*** (0.000) 0.004*** (0.000) 0.026 (0.027) 0.048 (0.048)

(3) 0.044** (0.022) 0.013 (0.036) 0.104*** (0.029) 0.024 (0.048) 0.124*** (0.039) 0.001 (0.000) 0.003*** (0.001) 0.001*** (0.000) 0.011 (0.034) 0.000 (0.000) 0.002*** (0.000) 0.012 (0.027) 0.068 (0.043) 0.037 (0.024) 0.099** (0.046)

Table 7.3a EU-15: Regional growth and the impact transport infrastructure over time, 1990–2004 (4) 0.078*** (0.026) 0.077** (0.038) 0.112*** (0.033) 0.059 (0.066) 0.154*** (0.039) 0.001 (0.001) 0.003 (0.002) 0.001*** (0.000) 0.013 (0.033) 0.000 (0.000) 0.002*** (0.000) 0.010 (0.039) 0.124** (0.052) 0.036 (0.025) 0.125*** (0.046) 0.041* (0.022) 0.114** (0.048)

(5) 0.109*** (0.026) 0.096*** (0.037) 0.079** (0.036) 0.075 (0.062) 0.071 (0.050) 0.004*** (0.001) 0.002** (0.001) 0.001*** (0.000) 0.072 (0.044) 0.000 (0.000) 0.002*** (0.000) 0.053 (0.036) 0.078 (0.059) 0.030 (0.024) 0.051 (0.057) 0.067*** (0.020) 0.175*** (0.045)

(6) 0.114*** (0.031) 0.083** (0.037) 0.105*** (0.036) 0.030 (0.064) 0.092** (0.039) 0.003* (0.002) 0.002* (0.001) 0.001*** (0.000) 0.110*** (0.033) 0.000 (0.000) 0.001*** (0.000) 0.075** (0.035) 0.120** (0.059) 0.052 (0.039) 0.069 (0.061) 0.056** (0.022) 0.167*** (0.048)

(7) 0.101*** (0.030) 0.063* (0.037) 0.123*** (0.043) 0.039 (0.072) 0.121** (0.048) 0.002* (0.001) 0.002 (0.001) 0.001*** (0.000) 0.144*** (0.045) 0.000 (0.000) 0.000 (0.000) 0.042 (0.034) 0.106 (0.068) 0.039 (0.041) 0.013 (0.064) 0.047 (0.036) 0.201*** (0.069)

138 7 What Can We Learn From the “Integrated Approach” To Regional Development?

1560 120 0.25 0.25 0.01 0.02

Observations Number of groups (NUTS regions) R-squared R-squared within R-squared overall R-squared between

1440 120 0.22 0.22 0.01 0.02

1.331*** (0.130) 1320 120 0.17 0.17 0.03 0.11

1.273*** (0.112) 1200 120 0.19 0.19 0.03 0.08

1.325*** (0.134) 1080 120 0.23 0.23 0.02 0.06

1.376*** (0.149)

0.020 (0.024) 0.045 (0.044)

Robust standard errors in parentheses; *significant at 10-% level; **significant at 5-% level; ***significant at 1-% level

1.297*** (0.118)

Constant

Lag 7 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 7 Change in kms of motorways/1,000 inhab.

Lag 6 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 6 Change in kms of motorways/1,000 inhab.

Lag 5 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 5 Change in kms of motorways/1,000 inhab.

960 120 0.36 0.36 0.02 0.05

1.976*** (0.150)

0.036 (0.023) 0.052 (0.044) 0.016 (0.026) 0.111** (0.047)

840 120 0.44 0.44 0.02 0.04

0.005 (0.033) 0.102** (0.049) 0.001 (0.021) 0.132*** (0.046) 0.011 (0.029) 0.023 (0.045) 2.595*** (0.162)

7.4 Results of the Analysis 139

(1) 0.104*** (0.028) 0.098** (0.039) 0.061 (0.038) 0.093 (0.067) 0.027 (0.037) 0.002** (0.001) 0.004** (0.002) 0.001** (0.000) 0.142*** (0.033) 0.000 (0.000) 0.002*** (0.000)

Lag 4 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 4 Change in kms of motorways/1,000 inhab.

Lag 3 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 3 Change in kms of motorways/1,000 inhab.

Lag 2 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 2 Change in kms of motorways/1,000 inhab.

Annual national growth rate

Migration rate

Log of total Gross Value Added (levels)

Ratio of employed people with Higher education in percent

Spatial weighted average of Total R&D expenditure

Total intra-regional R&D expenditure (all sectors) in percent of GDP

Log of GDPpc

Spatial weighted average of Change in kms of motorways/1,000 inhab.

Spatial weighted average of kms of motorways/1,000 inhab.

Change in kms of motorways/1,000 inhab.

Dependent variable:Regional GDP per capita (annual growth rate) Kilometres of motorways per thousand inhabitants

(2) 0.107*** (0.035) 0.079* (0.041) 0.012 (0.046) 0.017 (0.066) 0.024 (0.044) 0.001 (0.001) 0.003** (0.002) 0.001*** (0.000) 0.143*** (0.041) 0.000 (0.000) 0.003*** (0.001) 0.039 (0.036) 0.010 (0.067)

(3) 0.089*** (0.032) 0.094** (0.040) 0.008 (0.049) 0.032 (0.077) 0.025 (0.046) 0.001 (0.001) 0.002* (0.001) 0.001*** (0.000) 0.089** (0.044) 0.000 (0.000) 0.002*** (0.001) 0.020 (0.033) 0.074 (0.061) 0.020 (0.046) 0.057 (0.065)

Table 7.3b EU-25: Regional growth and the impact transport infrastructure over time, 1995–2004 (4) 0.078* (0.043) 0.086* (0.052) 0.047 (0.063) 0.072 (0.104) 0.107** (0.047) 0.000 (0.000) 0.000 (0.001) 0.001** (0.000) 0.000 (0.045) 0.000 (0.000) 0.000 (0.000) 0.044 (0.036) 0.104 (0.068) 0.029 (0.051) 0.114* (0.063) 0.052 (0.049) 0.153** (0.076)

(5) 0.100 (0.061) 0.157** (0.066) 0.131 (0.082) 0.256** (0.120) 0.167** (0.081) 0.001 (0.000) 0.000 (0.001) 0.002*** (0.000) 0.115 (0.082) 0.000 (0.000) 0.001** (0.000) 0.092 (0.058) 0.043 (0.101) 0.002 (0.052) 0.169** (0.081) 0.006 (0.047) 0.091 (0.073)

(6) 0.128 (0.103) 0.203* (0.115) 0.360** (0.145) 0.400** (0.188) 0.378*** (0.138) 0.000 (0.000) 0.001 (0.001) 0.001 (0.001) 0.437*** (0.134) 0.001*** (0.000) 0.000 (0.001) 0.212* (0.113) 0.378** (0.181) 0.083 (0.105) 0.018 (0.157) 0.059 (0.090) 0.063 (0.149)

(7) 0.219*** (0.080) 0.298*** (0.088) 0.587*** (0.110) 0.666*** (0.171) 0.386*** (0.116) 0.000 (0.000) 0.001* (0.001) 0.003*** (0.001) 0.166 (0.117) 0.000* (0.000) 0.008*** (0.001) 0.304*** (0.085) 0.607*** (0.151) 0.292*** (0.083) 0.536*** (0.133) 0.139* (0.077) 0.300* (0.153)

140 7 What Can We Learn From the “Integrated Approach” To Regional Development?

1288 161 0.22 0.22 0.02 0.05

Observations Number of groups (NUTS regions) R-squared R-squared within R-squared overall R-squared between

1127 161 0.22 0.22 0.02 0.05

1.168*** (0.179) 966 161 0.18 0.18 0.05 0.10

1.092*** (0.176)

Robust standard errors in parentheses, *significant at 10%; **significant at 5%; ***significant at 1%

1.126*** (0.142)

Constant

Lag 7 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 7 Change in kms of motorways/1,000 inhab.

Lag 6 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 6 Change in kms of motorways/1,000 inhab.

Lag 5 Spatial weighted average of Change in kms of motorways/1,000 inhab.

Lag 5 Change in kms of motorways/1,000 inhab.

805 161 0.13 0.13 0.10 0.15

0.996*** (0.213) 644 161 0.14 0.14 0.00 0.00

0.372 (0.267)

0.046 (0.041) 0.027 (0.066)

483 161 0.35 0.35 0.03 0.05

0.881** (0.377)

0.089 (0.093) 0.106 (0.128) 0.094 (0.094) 0.044 (0.127)

322 161 0.80 0.80 0.10 0.11

0.170** (0.077) 0.375** (0.146) 0.165* (0.088) 0.378** (0.149) 0.076 (0.051) 0.217*** (0.082) 1.804*** (0.380)

7.4 Results of the Analysis 141

142

7 What Can We Learn From the “Integrated Approach” To Regional Development?

endowment is not significant for the EU-15 (Table 7.1a, Regression 2) and has a negative and significant coefficient for the EU-25 (Table 7.1b, Regression 2). These results are partially in line with the existing literature: While there seems to be a clear correlation between the levels of GDP and infrastructure endowment, attempts to explain economic growth by transport investment have been much less successful (Vickerman et al. 1997). Our results find a correlation between the change in economic wealth and the level of infrastructure endowment but we fail to identify any evidence of a systematic relationship between further investment and economic growth. In interpreting this evidence it must be borne in mind that our proxy for infrastructure development – kilometres of motorways normalised by regional population – can only capture a limited amount of the Keynesian impact of the construction phase, as it is not based upon expenditure data. The proxy captures the “quantity” of infrastructure actually built irrespective of different costs under different natural and institutional conditions. Furthermore, since new infrastructure is recorded by official statistics only after final completion, our proxy is intrinsically better equipped to capture the ex-post impact of transport infrastructure on accessibility, the mobility pattern, and spatial re-organisation of economic activity. Once a satisfactory level of infrastructure endowment is in place, economic growth is made easier thanks to the “traditional” micro-economic impact on productivity. However, this level may well be the result – rather than the cause – of a dynamic local economy whose previous growth pattern may have supported and stimulated the enhancement of local infrastructural endowment (Vanhoudt et al. 2000), thus making infrastructure a factor that accompanies the process of regional development rather than being one of its engines. This idea is confirmed by the non significant or even negative impact of the development of further infrastructure: The impact of an additional kilometre of motorway depends on a variety of conditions related to both the nature of the connection developed and to the features of the areas actually involved in the project. As an example, Vickerman et al. (1997, p. 3) suggest that “transport improvements have strong and positive impacts on regional development only where they result in removing a bottleneck”, while Chandra and Thompson (2000) find evidence of a positive direct impact of the construction of a new interstate highway only for the US counties “that the highway directly passes through” (p. 487). The direct impact of further infrastructure development may be absent where the appropriate conditions are not met. It may be even negative where the additional infrastructure increases the exposure to external competition of a weak economic tissue as discussed in Sect. 7.2. The picture is not complete unless inter-regional spillovers are fully accounted for, as in regressions 3 and 4 where we consider the impact of the level and the change of infrastructure endowment in neighbouring regions, respectively. The empirical analysis suggests that the infrastructure endowment of neighbouring regions exerts a positive influence on local economic performance (Regression 3), and that this impact is highly statistically significant for the EU-15 (Table 7.1a), though not significant for the EU-25 sample (Table 7.1b). Where a good internal endowment of infrastructure is complemented by an equally good endowment in neighbouring regions, connectivity and accessibility are enhanced and bottlenecks

7.4 Results of the Analysis

143

and inefficiencies in inter-regional connections are prevented. Hence better economic performance ensues. Conversely – and symmetrically with the evidence discussed before – further investment in transport infrastructure in neighbouring regions has a negative impact on local economic performance: The coefficient of the proxy for infrastructure spillovers is negative and highly significant for the EU-15 (Table 7.1a, Regression 4) and not significant for the EU-25 (Table 7.1b, Regression 4). Transport infrastructure investment in neighbouring regions may negatively affect the local economy in the same way as internal investment does. The further development of transport infrastructure in neighbouring regions may not only increase the exposure of the local economy to external competitive forces but may also encourage the re-location of local economic activity towards better endowed neighbours as observed by Chandra and Thompson (2000) for the US case. Let us now consider the determinants of regional growth more accurately by introducing a proxy for local innovative efforts and knowledge spillovers, in line with the “endogenous growth” approach (regression 5). The results suggest that local innovative activities are important predictors of economic growth in the case of the EU-15 (Table 7.1a) but much less so in the EU-25 (Table 7.1b) while exposure to knowledge spillovers tends to be a positive factor in both samples. This piece of evidence on the relevance of knowledge flows is line with other analyses of the EU regional growth and innovation dynamics (Crescenzi et al. 2007; Rodrı´guez-Pose and Crescenzi 2008). The possibility of benefiting from knowledge flows is a differential source of competitive advantage for the EU regions, which allows them to compensate for the weaknesses of their internal innovative capacity. How can transport infrastructure contribute to these knowledge-based economic dynamics? The empirical results confirm the robustness of our previous conclusions: The internal and external infrastructural endowment plays a role but its changes do not. Other factors are probably more important for a knowledge-based economy. An important clue in this direction is provided by regressions 6 and 7 where the proxies for internal socio-economic conditions are introduced into the analysis, presenting the full specification of our empirical model. In the longer-run perspective provided by the EU15 sample the socio-economic conditions of the regions – as proxied by the social filter index discussed in Sect. 7.3 – are an important determinant of economic performance (Table 7.1a, Regression 6) while this effect is less accentuated in the EU25 case (Table 7.1b, Regression 6). However, where the most important component of regional socio-economic structure – i.e., human capital accumulation – is autonomously assessed its impact is positive and significant in both cases (Tables 7.1a and 7.1b, Regression 7). The local socio-economic conditions, in general, and of human capital accumulation, in particular, by avoiding the important side-effects of infrastructural development – which often more than offset any short-term positive impact – are much better predictors of economic growth and thus more convincing tools for regional development strategies. In order to test for the robustness of these results further control variables are introduced into the model (Regression 8). In particular, we control for some relevant proxies for the territorial organisation of the local economy: The magnitude of agglomeration economies and the capacity to attract migration inflows.

144

7 What Can We Learn From the “Integrated Approach” To Regional Development?

The absolute size of the local economy (our proxy for agglomeration) is synonymous with better economic performance in the EU-15 (Table 7.1a) while it turns out to be detrimental when the new members of the EU and a shorter time-span are considered (Table 7.1b). The net rate of migration is either negative (EU-15) or not significant (EU-25), suggesting that the inward mobility of people is not per se supportive of economic growth after controlling for the overall level of human capital accumulation. However, what matters more for the purpose of this paper is that both the magnitude and the significance of infrastructure endowment are not sensitive to the inclusion of controls. Tables 7.2a, 7.2b and 7.3a, 7.3b present a dynamic picture of the link between economic performance and the variables included in the full specification of our model (regression 8 of Tables 7.1a and 7.1b). Tables 7.2a and 7.2b present several annual lags in all variables, allowing for a maximum of 5 years between the base year of the variables and their impact on regional growth. This highlights the different time-span necessary for each factor to produce its impact on economic performance. In Tables 7.3a and 7.3b the model is estimated by sequentially introducing several annual lags of the proxies for infrastructure endowment and investment only in order to specifically capture their cumulative dynamics over time (holding all other variables fixed at time t-1 as in Tables 7.1a and 7.1b). The tables for the EU-25 are discussed only for comparison purposes since the 1995–2004 time span is too limited to allow for the introduction of several time lags without severely affecting the overall significance of the analysis, due to the sharp reduction in the number of observations. In any case, the results of the dynamic analysis for the EU-15 and the EU-25 are similar. Turning specifically to Tables 7.2a and 7.2b, the results underscore the interesting dynamics between the endowment with and additional investment in infrastructure, on the one hand, and economic growth, on the other. The results point to a short-lived impact of infrastructure endowment and investment in a region when other variables are controlled for. As in the case of Tables 7.1a and 7.1b, the endowment of motorways relative to the population in any given region is initially significant, but its significant association with growth disappears after the second annual lag, suggesting that after three years the current infrastructure endowment is no longer a valid predictor of economic performance (Tables 7.2a and 7.2b, Regression 2). Similarly, the negative and significant association between investment in new transport infrastructure and growth detected in Tables 7.1a and 7.1b above disappears after the fourth annual lag in the EU-15 sample (Table 7.2a). This means that any negative impact of infrastructure investment recorded in some cases quickly vanishes after the local economy begins to react to new competitive conditions, making infrastructure a less relevant predictor of economic growth than other variables included in the analysis (Rodrı´guez-Pose and Fratesi 2004). The effect of the spillovers from infrastructure in neighbouring areas on economic growth generally lasts slightly longer but also tends to wane with time: The positive and significant effect of transport infrastructure endowment in the rest of the EU on the economic growth of any given region disappears after 3 years, while the negative and robust association between investment in infrastructure in

7.5 Conclusions

145

neighbouring regions and economic growth after four (Table 7.2a). In contrast, other variables included in the model remain significant in time. This is, for example, the case of the spillovers from the total R&D investment in neighbouring regions, which not only remains significant throughout the period of analysis, but whose association with regional growth switches from negative in the relative short term to positive (Table 7.2a). Migration effects on growth suffer a similar change from negative to positive, while the coefficient of the level of education remains significant and positive throughout the period of analysis (Table 7.2a), although not in the case of the enlarged EU (Table 7.2b). This reliance on local human capital and on the innovation potential of neighbouring regions for economic development in the medium term implies that the regional convergence detected initially wanes and becomes divergence in the medium run (Table 7.2a). Regions with better endowments in human capital and surrounded by innovative regions tend not only to be richer, but also to perform better in the medium term. This overall picture of lack of medium term relevance of infrastructure investment for regional economic performance in Europe relative to human capital and R&D variables is further confirmed in Tables 7.3a and 7.3b, which analyse the dynamic association of investment in transport infrastructure within a region and in the rest of the EU with economic growth in a cumulative manner. The results indicate that while investment in infrastructure in any given region tends not to be robustly associated with economic performance over time, investment in infrastructure in the rest of the EU is generally negatively associated with growth in any given region. Such a result is an indicator of the existence of “two way road” effects, whereby regions with a weaker endowment in human capital and a lower capacity to generate innovation become increasingly exposed and vulnerable to investment in new transport infrastructure in the rest of Europe.

7.5

Conclusions

Inspired by favourable theoretical and empirical accounts about the impact of infrastructure on economic growth, the European Union has bet on investment in infrastructure in general, and transport infrastructure in particular, as its main development axis. The aim of this strategy is not only to achieve greater economic growth but also to generate greater social and economic cohesion. Infrastructure expenditure has, so far, represented roughly half of the structural and cohesion expenditure by the EU (Rodrı´guez-Pose and Fratesi 2004). The question is whether this important effort is paying off and whether it is delivering greater medium and long-term growth, especially in the periphery of the EU. This chapter has addressed this issue by analysing the role of transport infrastructure endowment and investment in regional growth by adopting an “integrated” approach to the analysis of the genesis of regional growth. This approach made it possible to directly contrast the effect on economic growth of infrastructure with that of other factors, most notably education and investment in

146

7 What Can We Learn From the “Integrated Approach” To Regional Development?

R&D, and by controlling for the spatial spillover effects linked to endowments and investment not just in any given region, but also in the remainder of the EU. It has adopted both a static and a dynamic approach in order to discern the impact of different factors on growth over time. The results highlight that a good infrastructure endowment is a precondition for economic development. Regions with adequate initial motorway networks tend to perform better than regions lacking this type of basic infrastructure. But the analysis has also revealed that the effects of infrastructure are much more complex than initially predicted by Aschauer-type analyses. Whereas regions benefit from having good initial levels of infrastructure and from being surrounded by regions with equally good endowments, new investment in infrastructure seems to be completely disconnected from growth performance, and investment in neighbouring regions is, on the whole, negatively associated with regional growth. Moreover, the positive effects of infrastructure endowment wane quickly in time, becoming insignificant shortly after the initial positive impact. In contrast, the negative effect on the regional growth potential of investment in transport infrastructure in the neighbourhood seems to be longer lasting. Infrastructure endowment and investment also appear to be less relevant for economic growth in the medium-term than human capital and innovation endowments. The level of education is, and remains in time, one of the main predictors of economic growth, while a region surrounded by others investing heavily in R&D tends to outperform other regions. The geographical concentration of human capital and R&D investment in Europe thus leads to a reversal of the regional convergence trend observed in the static regression analysis. Since the effects of human capital and knowledge spillovers on economic growth last longer than those of other factors, short-term regional convergence gives way to divergence in the medium term as a consequence of the better endowment of core regions with these factors. New investment in transport infrastructure has, by and large, contributed to enhance the centripetal effect as new roads linking the periphery with the core seem to be fostering the dynamism of the regions in the core, at the expense of regions in the periphery, through a “two way road” effect. The main policy implication that can be drawn from this analysis of European regions is the need to consider infrastructure policies within the framework of balanced strategies. If the aim is to maximise the regional economic return to any new infrastructure investment and to enhance economic cohesion, investment in infrastructure has to be coordinated with policies aimed at developing human capital and the innovative potential of regions. The timing of infrastructure investment is also crucial. Invest in transport infrastructure too early, and you may expose uncompetitive regions to stronger areas and markets, leading to even greater concentration. Invest too late, and you may prevent the development of the periphery. Only by paying attention to the complex relationship in time and space of the factors that influence growth will we be able to maximize the positive effects of delivering greater accessibility and connectivity of the regions in Europe, while minimising the economic and welfare risk of exposure of regions with weak economic conditions that are often illprepared to compete in more integrated markets.

Chapter 8

The EU Regional Policy and the Socio-economic Disadvantage of European Regions

8.1

Introduction

The empirical evidence presented in the previous chapters has highlighted the importance of underlying socio-economic factors in translating innovation into economic growth. The analysis showed that this is true for both endogenously produced innovation (by means of local innovative efforts) and exogenous knowledge flows. This evidence suggests that the locational disadvantage of peripheral regions (mainly due to the reduced exposition to knowledge flows) may be compensated for by reinforcing their capability to translate existing knowledge into economic growth. The empirical analysis showed that this result could be achieved by addressing the local sources of socio-economic disadvantage. When the socioeconomic sources of competitive disadvantage are effectively addressed, the capability to translate whatever kind of innovation into regional growth is enhanced. This chapter aims at investigating the capacity of the EU regional policy to pursue this function i.e., to address the sources of socio-economic disadvantage of the EU regions. The analysis will focus its attention upon the a priori structure of the EU policy in order to assess whether (and to what extent) it is consistent with the socio-economic factors that have been shown to hamper the local economy’s capability to grow and develop at an adequate pace. In order to achieve this objective the chapter sets out to combine the results of the empirical analysis on the role of underling socio-economic conditions as an explanation of differential regional growth performance with the analysis of the regional policies of the EU. This chapter will make a direct comparison between the socio-economic preconditions for successful regional development and the actual allocation of structural funds for this purpose. On the basis of the evidence provided by the empirical analysis hitherto pursued, if the EU regional funds are to maximise their chances of achieving success, they should be allocated in accordance with the geography of the sources of competitive disadvantage. In other words, given that a set of socioeconomic conditions have been demonstrated to be factors hampering the economic success of many EU regions, the EU funds should be allocated in such a manner as to “compensate” for the structural disadvantages of assisted areas.

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_8, # Springer-Verlag Berlin Heidelberg 2011

147

148

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

This chapter aims to assess whether the geographical allocation of the structural funds has been biased (Objective 1 and 2) under both the 1994–1999 and 2000–2006 programming periods for the purpose of shedding some light on the coherence of the policy hitherto pursued and draw some implications for the forthcoming programming period. More specifically, in this chapter: (a) The spatial concentration of structural expenditure is analysed. A low degree of spatial concentration of development funds would contradict the principle of territorial concentration introduced by the 1989 reform of the funds as an important prerequisite for their effectiveness; (b) The spatial concentration of EU funds is contrasted with a specifically developed indicator of the socio-economic disadvantage of the EU regions. This analysis will allow us to investigate the coherence of the existing EU regional policies in terms of the structural disadvantage of EU regions thus uncovering a potential inconsistency between policy objectives (favouring disadvantaged areas) and the beneficiaries of the funds; (c) An empirical model to assess to what extent regional funds are, in fact, associated (in a statistically significant way) with the above-mentioned sources of competitive disadvantage is developed; (d) A simple convergence analysis is pursued in order to show that increasing the concentration of the funds and investing in the most disadvantaged areas is the best strategy to promote cohesion. Low territorial concentration and a weak correlation between the geographical allocation of the funds and the structural disadvantage would suggest that, even before their operational translation into actual policies, the reasons for the weak impact of the funds may be found in their inability to correctly select their targets i.e., the regions where socio-economic disadvantage is more severe. This chapter is organized into four sections. In the first section the EU regional policy is analysed in the light of the existing literature to single out the potential causes for its limited impact and set the foundation for a subsequent analysis. In the second section the methodology followed to assess the spatial structure of both funds and socio-economic disadvantage is presented and the empirical model to measure the correlation between development funds and socio-economic disadvantage is outlined. In the third section the empirical results are discussed. The fourth section concludes with some proposals for the design of regional policies.

8.2 8.2.1

Regional Policy and Structural Disadvantage The EU Regional Policy

The European Community Treaty states that “(. . .) the Community shall aim at reducing the disparities between the levels of development of the various regions and the backwardness of the least favoured regions or islands, including

8.2 Regional Policy and Structural Disadvantage

149

rural areas” (Article 158). The same objective is included in the EU draft Constitution (article III-220). The financial resources devoted to pursue this objective have grown substantially over the years: from ECU 68 billion (at 1997 prices) allocated by the Brussels European Council in 1988 to the €195 billion (at 1999 prices) of the 2000–2006 programming period1(European Commission website). Altogether the expenditure for regional policy is particularly significant when assessed as a percentage of the GDP of many lagging regions: 2.7% (of national GDP) in Greece, 2.8% in Portugal, 1% in Spain, 0.7% in Ireland in the year 2000 (E.C. 2000). However, while the amount of resources devoted to the objective of promoting an “overall harmonious development” of the Union has not been negligible, the evidence of the influence of such resources on the actual level of territorial cohesion of the EU is rather mixed. In particular the literature has emphasized: (a) The lack of upward mobility of Objective 1 regions, which have remained almost the same from 1989 to 2005 (with a few exceptions2). (b) The absence of convergence across EU regions in contrast with the convergence observed across the member states which dominated the past 25 years of European growth (Boldrin and Canova 2001; Magrini 1999; Puga 2002). Rather, a process of “club convergence” would seem to be in place across the EU regions thus leading to the formation of clusters of regions with persistently different income levels (Canova 2004; Quah 1996 and 1997). On the basis of such evidence, which is undoubtedly the result of a complex set of forces in place in the EU economy, many of which not related to any policy action, some empirical studies have attempted to single out the link between structural funds and regional economic development in order to assess the impact (if any) of the funds upon regional economy. These contributions address the different factors that seem to prevent regional policy from delivering its intended benefits. Midelfart-Knarvik and Overman (2002)’s analysis highlights the distortion produced by structural funds on the location decision of R&D intensive firms. Structural funds provide an incentive for firms to locate in assisted regions with a poor endowment of human capital, producing an inefficient outcome for both firms (that cannot benefit from an adequate labour pool in the local area) and workers (who do not benefit from an increase in labour demand due to the skill mismatch). Thus, EU aid should be focused “on helping regions change their endowments and specialize according to the resulting comparative advantage” (p. 352). Though produced using different

1 In addition the Cohesion Fund distributes resources for about €2.5 billion per year from 2000 to 2006, for a total of €18 billion (at 1999 prices). 2 Abruzzo (Italy) lost its Objective 1 status in 1997. A few regions and areas lost their Objective 1 status with the 2000–2006 programming period but received transitional support under Objective 1 of the Structural Funds for the period from 1 January 2000 to 31 December 2005 or 2006 (Commission Decision 1999/502/EC).

150

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

theoretical frameworks,3 this evidence is not far away from the results of Cappelen et al. (2003) who concluded that the impact of structural funds is positive but “crucially dependent on the receptiveness of the receiving environment” (p. 640). These findings emphasize the role of relatively more favourable contextual conditions/endowment, which in their turn, lead to a paradoxical situation whereby the EU funds fail to work precisely where they are most needed. Rodrı´guez-Pose and Fratesi (2004) by more directly assessing the impact of structural funds on regional growth performance, find that such impact crucially depends on the distribution of resources across development axes. Where fund allocation more closely addresses such contextual conditions, i.e., by being channelled towards human capital enhancement, its effects tend to be positive and significant while this is not the case when other objectives are pursued (i.e., infrastructure). The evidence reviewed above suggests that a cause for the relatively weak impact of the investments pursued up to now might be the “operational” mismatches between policy targets and the real needs of the lagging regions when financial resources are divided among the different axes and then translated into concrete actions. In this chapter we contrast this explanation for the unsatisfactory performance of structural expenditure with an alternative explanation: potential “spatial” mismatches between areas where the factors of disadvantage are concentrated and areas where the resources being channelled by the policy design may a priori have prevented the funds from delivering the expected benefits.

8.2.2

Territorial Concentration and Correlation with Structural Disadvantage

Structural funds are designed to promote economic and social cohesion in the EU by fostering the economic development of lagging regions (Objective 1) and assisting economic and social conversion in areas experiencing structural difficulties (Objective 2). However, “since 1994 the connection between poor nations and structural spending has been greatly diluted (as) large parts of Finland and Sweden were designated as eligible, and even some Austrian regions, together with all of the former East Germany” (Baldwin and Wyplosz 2003; par.9.5). It was the pressure for setting aside budget resources aimed at financing the eastward enlargement of the EU that forced a reduction in both the areas eligible for assistance and community initiatives in the Agenda 2000 reform of the structural fund (Armstrong 2001). Such a reduction was explicitly inspired by the principle of territorial and financial concentration: i.e., the relatively scarce resources for the EU development policies should be channelled more specifically to where they are most needed in order to

3

While Midelfart-Knarvik and Overman (2002) focuses the determinants of firms’ location Cappelen et al. (2003) develop a “new growth theory” model with a Schumpeterian perspective.

8.2 Regional Policy and Structural Disadvantage

151

maximise their effectiveness. Over time the need for an increase in the geographical concentration of the structural funds expenditure has become progressively more apparent and “concentration” has been included, within the “framework for cohesion policy 2007–2013”, among the key leading principles for the new programming period.4 But why is geographical concentration so important for the impact of the policy? Intuitively a smaller number of beneficiaries may allow a larger amount of resources to flow in selected regions. However, not only is the level of expenditure in the objective region relevant in itself but also that in its neighbouring regions (Dall’Erba 2005). By this we mean that the spatial externalities produced by the implementation of regional development programmes of whatever nature need to be taken into account because an insufficient spatial “concentration” of the funds may decrease their impact by reducing the amount of such externalities “flowing” within the assisted areas. In addition, the spatial structure of the funds needs to be assessed in combination with the underlying socio-economic conditions of the assisted regions. In order to maximise their impact the funds should be directed where persistent factors of disadvantage prevent the local economy from fully expressing its potential i.e., the geography of the funds should reflect as much as possible the geography of the structural disadvantage of the EU regions.

8.2.3

Where Are the Funds Most Needed? The Empirical Evidence Produced So Far

The empirical analysis pursued in the previous chapters has shown that a specific set of factors act as structural sources of competitive disadvantage for the local economy. This is in line with the evidence that lagging regions in the EU, notwithstanding their profound differences under many respects, share a common set of analogous social conditions whose role is emphasized by the economic restructuring accelerated by the process of European integration (Rodrı´guez-Pose 1994; Rodrı´guez-Pose 1998a). While some economic factors (such as capital and technology) seem more able to adjust in response to the challenges of the EU integration (by virtue of their relatively higher mobility), social conditions tend to be much less flexible. Consequently it has been possible to identify a specific set of “structural” conditions that are persistently associated with poor economic performance and which are very slow to adjust themselves endogenously. However, the distinctive role of underling socio-economic conditions must be assessed in a theoretical framework where, in line with Lisbon Agenda objectives, innovation is explicitly 4 COMMUNICATION FROM THE COMMISSION, Brussels, 05.07.2005 COM(2005) 0299, “Cohesion Policy in Support of Growth and Jobs: Community Strategic Guidelines, 2007–2013”, p. 8.

152

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

considered the driving force of growth. The objective of an innovation based growth model for the Union has guided the implementation of the EU structural policies and the assessment of their results since the year 2000. However, with the drawing up of the Community Strategic Guidelines “Cohesion Policy in Support of Growth and Jobs: Community Strategic Guidelines, 2007–2013” which set out a framework for new programmes for the next programming period, “knowledge, innovation and the optimisation of human capital” are explicitly assumed as means for Europe to “renew the basis of its competitiveness, increase its growth potential and its productivity and strengthen social cohesion”. (Presidency conclusions, European Council, March 2005 and incipit of the above-mentioned draft Community Strategic Guidelines). In addition the role played by the cohesion policy in pursuing the Lisbon agenda has increased in 2007–2013 programming period as emphasized in 2007–2013 Financial Perspective, which concentrated expenditure on the Lisbon objectives (Presidency conclusions, European Council, December 2005). In this political framework the empirical evidence produced so far highlights the role of socio-economic conditions in the translation of innovation into regional growth which has been treated in a systematic way by the introduction of the concepts of “regional system of innovation” (Iammarino 2005) and of the “social filter” (Rodrı´guez-Pose 1999): the interaction of a complex set of economic, social, political and institutional features that makes some regions “prone” and others “averse” to innovation. In line with the previous analysis the multifaceted socio-economic conditions of the EU regions are introduced in our analysis by means of a set of variables describing the local socio-economic realm coherently with the previous chapters. Innovation averse socio-economic conditions, by persistently hampering the growth capabilities of some areas, trace out the geography of the structural disadvantage of the EU territories. As a consequence, it seems reasonable that in terms of both equity and efficiency, the geographical allocation of regional funds should follow the spatial structure of these factors. Thus, as regards equity such a distribution of resources across regions should compensate the residents of “disadvantaged” regions for unfavourable starting conditions. In terms of efficiency, giving adequate attention to the structural sources of competitive disadvantage of assisted regions seems the most effective way of promoting the full employment of local resources.

8.3

Where Do the Funds Actually Go?

In the previous section we discussed how the weak impact of the structural policies of the EU has been explained in terms of its translation into policies not correctly tailored to the needs of the assisted areas. However, we also discussed the importance of territorial concentration and the geographical distribution of the funds in

8.3 Where Do the Funds Actually Go? Assessing Their Territorial Concentration

153

relation to the structural disadvantage of the EU regions. This section sets out to outline an empirical strategy to assess this second hypothesis by investigating the spatial structure of the allocation of the EU structural funds and their relationship with the sources of structural disadvantage discussed in the previous section. The descriptive spatial analysis of both phenomena will be followed by an empirical analytical model that identifies the importance (statistical significance) of the socioeconomic factors in the distribution of the EU structural funds (Objective 1 and 2) under both the 1994–1999 and 2000–2006 programming periods, in order to shed some light on the coherence of the policy hitherto pursued. In this section the methodology followed to pursue such analysis is briefly presented together with the corresponding dataset. The empirical results will be discussed in the fourth section.

8.3.1

A Measure for Socio-economic Conditions

The variables which seem to be more relevant for describing the socio-economic disadvantage of a regional space – as discussed in Chap. 4 – are those related to three main domains: educational achievements, the productive employment of human resources and demographic structure. From the first domain, tertiary educational attainment (of both the population and the labour force) and participation in lifelong learning programmes are assumed as a measure for the accumulation of skills at the local level. In the second domain, the percentage of labour force employed in agriculture and the long-term component of unemployment are included in the analysis in order to capture the amount of human resources excluded from productive employment. Long term unemployment represents the incidence of people whose possibilities of being productively involved in the labour market is persistently hampered by inadequate skills. Agricultural employment is frequently synonymous with “hidden unemployment” 5 and a backward structure of the local economy. For third domain, the percentage of population aged between 15 and 24 is assumed as a proxy for the flow of new resources entering the labour force, thus “renewing” the existing stock of knowledge and skills. These factors are autonomously introduced into the analysis in order to assess their individual weight. However, in order to assess their “global” relationship with the allocation of structural funds, while minimising the problems of multicollinearity,6 the socioeconomic variables – coherently with procedure followed in previous chapters – are combined by means of Principal Component (PC) Analysis (Jolliffe 1986). Consequently, the set of variables discussed above is “reduced” to an individual 5 Where long term unemployment tends to be persistently high and labour mobility low, less skilled workers tend to move to the countryside to be employed, with a very low marginal productivity, in (frequently family owned) small farms thus allowing an easier access to primary goods. 6 Which prevents their simultaneous introduction into the regression equation.

154

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

variable that is able to preserve as much as possible of the initial information (variability) (see Appendix C for the results of the PC analysis and technicalities). Such procedure allows us to handle an individual variable that “summarizes” the multifaceted nature of the socio-economic conditions of each region.

8.3.2

The Empirical Model for the Allocation of Funds Across Regions

This section outlines the empirical model for the analysis of the role of socioeconomic disadvantage in determining the allocation of structural funds. The model aims at estimating a “hidden” decision function of the European policy maker in the allocation of the structural funds across regions. Such a “decision function” would reflect the “rationale” of the policy uncovering the coherence of the policy design with the identified sources of structural disadvantage. The estimation of the model, by regressing the per capita regional commitments of the structural funds on the sources of socio-economic disadvantage identified above, will allow us to “measure” the role of these factors in the actual allocation of the funds. The reduced weight of these factors in the allocation decision, which might be a possible explanation for their limited impact, can reflect: (a) The predominant role of “power” factors in the design of the policy where the present allocation of the funds might be the result of the political equilibrium reached in the bargaining process between the Commission, the national governments, the local governments and the various pressure groups; (b) The willingness of the European policy-maker to privilege, in the distribution of the funds, the relatively more advantaged regions on the basis of the (questionable, as we will discuss later) assumption that this category of regions would show a better potential for growth and development. Two models will be estimated in the empirical analysis. A first model analyses the allocation of Objective 1 and Objective 2 funds separately (8.1 and 8.2), while a second model considers the overall regional distribution of the structural funds (8.3). Consequently, the first part of the empirical analysis is based on a two-stage Heckman selection model (Heckman 1979; Green 2003). The first stage determines “eligibility” as an Objective 1 (Objective 2) area. Such a decision is based on specific criteria that should improve the territorial concentration of the funds and, a priori, select the most disadvantaged areas according to each objective’s “mission”. However, such a decision can, in fact, be biased for the reasons discussed above. Consequently, the first step of the Heckman selection model aims at assessing, through a probit model, how the factors of socio-economic disadvantage in fact influence the probability of a region of being assisted (or not). The model is estimated separately for Obj1 regions and for Obj2 regions in both the programming periods considered. The estimated model is the following:

8.3 Where Do the Funds Actually Go? Assessing Their Territorial Concentration 0

wi ¼ Zi g þ ei

155

(8.1)

where wi ¼ 1 if the region i is an assisted region and wi ¼ 0 if the region is not assisted;and 0

0

Prðwi ¼ 1Þ ¼ Fðg Zi Þ and Prðwi ¼ 0Þ ¼ 1  Fðg Zi Þ; where F(x) is the normal cumulative distribution function, Zi is a set of socio-economic explanatory variables described above, g is a vector of parameters, and ei is the error term. In a second step the level of support is regressed on its potential determinants while taking into account the selection bias introduced in the sample by the a priori selection of eligible areas. Consequently, the following second-step H–C OLS model is estimated: 0

yi ¼ a Xi þ ei

(8.2)

Where yi (>0) is the level of per capita commitment in region i, a is a parameter vector, X are the explanatory variables and ei is the error term. The set of explanatory variables includes: the socio-economic conditions, a set of national dummy variables (to estimate a potential “national” bias in the distribution of the funds) and the Inverse Mills Ratio (IMR). The IMR is calculated from the first stage probit model and is used in the second step as an instrument for the latent variable that determines whether an area is eligible or not. In other words the IMR links the participation of the regions to the distributions of the funds (1st step) with the amount of funds received (2nd step). The second part of the empirical analysis will focus on how socioeconomic factors drive the observed level of total regional expenditure per capita (under both Objective 1 and 2): the interaction of Objective 1 and 2 purposes might even further “dilute” the policy targets. Consequently, we will estimate an OLS model regressing the commitment level per capita under both Objective 1 and 2 on the socioeconomic variables and a set of national dummy variables: 0

yi ¼ a Xi þ ei

(8.3)

where yi (that this time includes all the regions included in the sample) is the level of per capita commitments in region i, a is a parameter vector, X are the explanatory variables (socio-economic factors + national dummies) and ei is the error term.

8.3.3

The Dataset

Since the objective of the analysis is to assess the coherence of the spatial allocation of structural funds with the sources of competitive disadvantage of the EU regions it is necessary to identify the most appropriate spatial scale of analysis in order consider

156

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

homogeneous and (to the extent possible) functionally “self contained” units in terms of both their capacity to receive funds (and exert political pressure for this purpose) and their socio-economic structure. Where funds are allocated to areas without any corresponding governance level and a reduced functional self-consistency, a leakage effect seems to prevail (due to the functional links of the area with the rest of the region) thus forcing us to assume that the entire region is a beneficiary of the funds. Consequently, given the constraint of data availability, but also for reasons of homogeneity and coherence in terms of the relevant institutional level discussed above, the analysis – as in the rest of this work – is based upon NUTS1 regions for Germany, Belgium and the UK and NUTS2 for all other countries (Spain, France, Italy, the Netherlands, Greece, Austria, Portugal, Finland). The data on the regional distribution of commitments7 for structural fund expenditure was collected on the basis of the information provided by the European commission on its website (Inforegio) and takes into account all structural funds.8 In addition, the analysis relied upon an Annex of the EC report “The impact of structural policies on economic and social cohesion 1989–1999”. For the sake of comparability between programming periods, Objective 1 and 6 data, on the one hand, and Objective 2 and 5b, on the other, are combined together for 1994–1999 commitments. The Operational Programmes (OP) and Single Programming Documents (SPD) for both programming periods have been associated to the appropriate NUTS region thus providing the total committed expenditure in each region. The total commitment has been divided by the average population of the region during the respective programming period in order to obtain per capita expenditure. The year 1994 is assumed as reference year for the socio-economic conditions variables in order to minimize any potential endogeneity between higher (lower) funds and better (worse) socio-economic conditions.

8.4 8.4.1

Empirical Results Spatial Concentration: Structural Funds Versus Socio-economic Disadvantage

The analysis of the spatial distribution of the variables is pursued by calculating the value of Moran’s I (see Appendix B for technicalities). Moran’s I is a measure for the global spatial autocorrelation of the variables (Cliff and Ord 1981). When Moran’s I is significantly different from zero the variable of interest exhibits a 7 Only data for commitments rather than expenditure are available. However the use of commitments data is coherent with our theoretical framework, as we aim at analysing the a` priori structure of the policy rather than estimating the impact of actual expenditure. 8 The European Regional Development Fund (ERDF), the European Social Fund (ESF), the Guidance section of the European Agricultural Guidance and Guarantee Fund (EAGGF-Guidance) and the Financial Instrument for fisheries guidance (FIGS).

8.4 Empirical Results

157

systematic spatial pattern. A positive value of this index means that areas with a high (low) level of per capita structural expenditure tend to be clustered close to other areas with high (low) expenditure. The same line of reasoning is valid for the factors of socio-economic disadvantage, where a positive value of the index means a pattern of clustering of regions with similar high/low values. The magnitude of the indicator provides a measure of the strength of the spatial pattern i.e., the extent of the clustering process of similarly high/low values. Table 8.1 shows the value of Moran’s I for regional expenditure under Objective 1 and 2 and for total structural fund expenditure. The table shows that a clear spatial pattern is identifiable in the distribution of both funds and indicators of socioeconomic disadvantage. Moran’s I is positive and significant in all cases, thus showing a positive spatial autocorrelation: regions with a high (low) level of expenditure (socio-economic disadvantage) tend to be clustered together. This result is in line with the principle of concentration of the funds repeatedly claimed by the European Commission. However, if the results are examined in further detail by considering the magnitude of the index, it is possible to note, as was expected, that Objective 1 tends to be more concentrated than Objective 2 expenditure where the latter seems to respond more weakly to this principle of concentration (in both the programming periods). It must be noted, though, that the overall territorial concentration of expenditure has increased after the Agenda 2000 reform of the structural funds: Moran’s I for Objective 1, Objective 2 and total expenditure has increased from one programming period to the other. However, as we discussed in the previous sections, the territorial concentration of the funds should be compared with that of the socio-economic sources of competitive disadvantage. This benchmark is provided by the last line of Table 8.1 which shows Moran’s I for the “Social Factors” variable which is calculated through the Principal Component Analysis from the whole set of socio-economic variables previously discussed. The comparison between the magnitudes of Moran’s I of the “Social Factors” and that of structural expenditure shows that social factors are more spatially concentrated Table 8.1 Spatial concentration of Objective 1 and Objective 2 Funds per capita, 1994–1999 and 2000–2006 and social factors Variables I E(I) sd(I) z p-value* Programming Period 1994–1999 Objective1

0.102

0.008

0.009

11.649

0

Objective 2 Total expenditure

0.039 0.095

0.008 0.008

0.009 0.009

5.061 10.929

0 0

Programming Period 2000–2006 Objective1

0.142

0.008

0.009

15.911

0

Objective 2 Total expenditure

0.094 0.149

0.008 0.008

0.009 0.009

10.781 16.658

0 0

0.009

24.329

0

Social Factors a

Social Factors

0.223

0.008

*1-tail test a This variable is the linear combination of the socio-economic variables described in the text and is calculated through the Principal Component Analysis (Appendix C)

158

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

than structural funding. Thus, even if the territorial concentration of expenditure increased with successive reforms of the structural funds it seems to be still insufficient when compared to the spatial pattern of the sources of structural disadvantage. This provides the first evidence in favour of our hypothesis of there being a “spatial mismatch” between the factor of structural disadvantage and development funds, thus encouraging further analysis of the geographical allocation of the funds. Territorial concentration of the funds per se might not be sufficient for the policy to deliver the expected benefits, a closer adherence to the regional sources of structural disadvantage might be necessary.

8.4.2

The Drivers of the Regional Allocation of Structural Funds

In the previous paragraph the spatial distribution of the structural funds has been analysed. In the following part we discuss the estimation results of our empirical model, whose aim is to highlight the weight of the observed socio-economic factors in the “implicit” decision function for the regional allocation of structural funds. Following the specification presented in par. 8.3.2 we estimate a two-stage Heckman selection model for the allocation of Objective 1 (Table 8.2) and Objective 2 (Table 8.3) funds. The tables show the estimations results for the programming periods 1994–1999 (on the left hand side of the table) and 2000–2006 (right hand side). For each programming period equations (8.1) and (8.2) are estimated by regressing the funds on the “Social Factors” variable (a) and on some of its individual components9 (b). When looking at the results for the Probit Selection Model (lower part of the tables) it should be born in mind that the magnitude of the parameters estimated by the probit technique does not have a direct meaning in terms of the extent of the corresponding effect. However, the parameters are informative as far as their signs and significance are concerned. As regards Objective 1 funds (Table 8.2), the social factors variable shows a negative sign and a high significance level in both the programming periods thus implying that favourable socio economic conditions (i.e., a high value of the social factors variable) reduce, as expected, the probability of being considered an eligible area (column a). This seems to confirm that the actual eligibility criterion, based on per capita income, is a good proxy for weak socio-economic conditions. However, if the factors influencing the probability of becoming an eligible region are considered in greater detail (column b), we shall notice that the “traditional” sources of disadvantage are more “rewarded” by this system: the “percentage of labour force concentrated in agriculture” and “long term unemployment” significantly increase the chances of being under the 75% of the EU average per capita income (thus 9

As noted previously multicollinearity prevents the simultaneous inclusion of all these variables into the regression. A detailed description of the individual variables is included in Appendix A).

1.4158*** (0.348857) 1.0370*** (0.329578) Social Factorsa Education Population 5.044067* (2.89385) 5.754955*** (2.826307) Agricultural Labour Force 17.32992*** (3.535073) 15.12283*** (3.218646) Long Term Unemployment 3.435833*** (1.171702) 2.609007*** (1.091462) Young Population 5.912144 (4.973609) 6.068956 (4.78766) Constant 0.265963 (0.17737) 4.737*** (1.13581) 0.16692 (0.172587) 4.25439*** (1.07249) rho 1 1 1 0.94973 sigma 4846.965 358.7948 2111.375 69.35247 lambda 4846.97 (23328.48) 358.795 (178.5998) 2111.37 (15897.1) 65.866* (41.52635) *, ** and ***denote significance at a 10%, 5% and 1% level respectively. SE in parentheses a This variable is the linear combination of the socio-economic variables described in the text and is calculated through the Principal Component Analysis (Appendix C)

Table 8.2 Heckman Selection model, Objective 1 Funds per capita, 1994–1999 and 2000–2006 Period 1994–1999 Period 2000–2006 Equation (8.2) Variables Coef. Coef. Coef. Coef. (a) (b) (a) (b) Social Factorsa 3622.424 (21602.14) 1218.957 (10951.03) Education Population 4988.11* (2562.976) 1913.78 (456.1678) Agricultural Labour Force 1348.16 (1043.342) 312.165 (222.0423) Long Term Unemployment 574.539 (588.8321) 89.498 (110.8817) Young Population 3218.96 (2456.867) 1067.57** (503.5399) National Dummies de 1286.602 (3153.09) 1044.413*** (362.087) 264.6077 (1293.069) 291.6251 (68.56178) it 10.02819 (2446.981) 119.275 (215.7996) 83.11813 (1066.923) 49.53745 (46.58662) at 198.3732 (3683.407) 309.7738 (279.0372) 142.7548 (1579.302) 180.4558*** (60.11469) be 498.6349 (3469.236) 281.757 (304.0943) 100.9242 (1514.511) 95.4871 (62.36345) pt 248.376 (2651.336) 362.557* (186.396) 157.058 (1134.62) 123.3903*** (38.62917) nl 512.8831 (3378.771) 369.2325 (316.798) 122.9396 (1487.263) 134.3599*** (66.7445) uk 745.6835 (3216.694) 398.8849* (227.0967) 193.8667 (1310.763) 129.0245*** (43.20416) es 621.0167 (2306.694) 634.0799** (288.4948) 252.0606 (997.5152) 319.0792*** (59.05076) gr 192.1769 (2456.519) 224.2701 (187.8398) 21.8073 (1054.395) 1.55839 (39.39773) fi 534.0902 (2926.159) 233.248 (286.6558) 0.204899 (1271.065) 32.9576 (57.13414) Constant 3561.73 (14885.26) 2025.47*** (659.4408) 1614.26 (11007.22) 574.4937*** (137.1147) Probit Selection Model (Equation 8.1)

8.4 Empirical Results 159

160

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

becoming an Objective 1 region). On the contrary, other factors are less accurately proxied by the actual income-based eligibility criteria. The “percentage of the young population” is not significant while “tertiary education attainments” shows a positive sign meaning that in many cases the regions selected for assistance are not those with a relatively poorer human capital endowment. In the second step of the model, the amount of funds received (by eligible areas) is analysed (8 2). The empirical results show that, while significant for the acquisition of the status of assisted region, the socio-economic factors are not significant for determining the level of the funds received by assisted regions (column a). In other words, the distribution of the funds across the eligible areas does not seem to reflect their actual differentiated socio-economic status. When considering specific socio-economic factors (column b) we notice that only the education level variable shows a high level of significance in 2000–2006: a relatively higher percentage of tertiary educational achievements seems to reduce the amount of funds received in favour of less well endowed regions. The national dummies highlight a certain degree of national bias in the allocation of the funds in favour of some member states (in particular Germany and Spain in 1994–1999 and Spain in 2000–2006), but this bias seems to disappear when the socio-economic conditions are fully accounted for by the Social Factors variable. Such national bias can thus be considered the effect of systematically higher disadvantage of the regions of these countries (which the distribution of the funds is able to reflect), rather than the result of a more favourable treatment in favour of these countries. Such evidence supports the idea that even if the present eligibility criterion is able to pursue a (rough) discrimination in favour of the relatively more disadvantaged regions, the amount of funds then transferred to assisted regions is not correlated to the extent of their actual socio-economic disadvantage. Table 8.3 presents, in the same way as in the previous table, the results for the estimation of the two-step Heckman selection model for Objective 2 funds. The results for the probit selection model show that, as expected, Objective 2 regions tend to present relatively more favourable socio-economic conditions: the socioeconomic factors variable is positive and significant. In addition, as expected, Objective 2 regions are mainly industrial regions (a high % agriculture labour force tends to reduce the probability of being “selected”) and the population is relatively younger in comparison with other areas. However, the present eligibility criteria seem unable to discriminate the areas with a relative scarcity of skilled labour, as shown by the non-significance of the education variable in 2000–2006. When we move on to the analysis of the determinants of the amount of funds allocated to the regions, we find no sign of any correlation with the underling socioeconomic conditions of the assisted areas (except for the education variable in 2000–2006). This evidence supports the idea of an overall weakening of the coherence between the structural funds and their ideal targets operated by means of the expenditure under the Objective 2. On the contrary, where aiming at favouring the socio-economic “restructuring” of declining regions, Objective 2 funds should follow the geography of socio-economic disadvantage.

Social Factorsa 1.121132*** (0.330526) 1.331961*** (0.343357) Education Population 7.02116** (2.844077) 3.15919 (2.750046) Agricultural Labour Force 16.0497*** (3.350845) 14.7694*** (3.387493) Long Term Unemployment 3.23574*** (1.131636) 3.56761*** (1.134586) Young Population 10.283*** (4.739716) 19.6541*** (5.100463) Constant 0.22104 (0.173643) 5.339909*** (1.114868) 0.38479*** (0.178404) 6.028806*** (1.164758) rho 1 1 1 0.11154 sigma 214.6384 363.2897 96.03772 13.05521 lambda 214.6384 (1720.033) 363.2897 (714.9973) 96.03772 (517.8416) 1.456141 (28.80728) a This variable is the linear combination of the socio-economic variables described in the text and is calculated through the Principal Component Analysis (Appendix C) *, ** and *** denote significance at a 10%, 5% and 1% level respectively. SE in parentheses

Table 8.3 Heckman Selection model, Objective 2 Funds per capita, 1994–1999 and 2000–2006 Programming Period 1994–1999 Programming Period 2000–2006 Equation (8.2) Variables Coef. Coef. Coef. Coef. (a) (b) (a) (b) 41.24806 (979.3314) 15.24312 (360.1518) Social Factorsa Education Population 1473.4 (2604.039) 219.959** (86.8514) Agricultural Labour Force 2313.08 (5708.642) 146.9052 (213.0774) Long Term Unemployment 292.403 (1097.94) 45.70872 (53.61375) Young Population 2649.94 (4296.254) 95.0998 (299.439) National Dummies de 14.1343 (61.11901) 21.8045 (131.9588) 15.2183 (25.85857) 16.5432*** (5.622292) it 13.79382 (83.21526) 18.6619 (147.3966) 41.2794 (38.36847) 43.8702 (7.736061) at 31.6908 (69.25755) 42.80739 (211.879) 20.1437 (27.39351) 5.56321 (9.046899) be 4.40015 (124.5079) 54.1565 (220.7587) 6.2263 (61.19157) 17.4202 (11.50549) nl 74.98787 (81.38781) 116.1177 (221.6512) 1.86291 (43.41586) 1.35525 (12.65517) uk 51.9274 (82.03706) 46.94875 (139.8897) 15.96409 (35.93839) 6.896866 (6.055499) 20.99423** (10.78373) es 151.6018** (72.02708) 123.0932 (218.1189) 25.25797 (30.96621) fi 77.1801 (113.6932) 70.01067 (235.5529) 28.5619 (49.59434) 33.2919*** (11.58116) Constant 66.0253 (1528.65) 726.9151 (291.69) 34.9188 (511.2596) 52.246 (67.34726) Probit Selection Model (Equation 8.1)

8.4 Empirical Results 161

162

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

In Table 8.4 the overall allocation of structural funds under both Objective 1 and 2 is assessed, thus focusing upon their interactions and “composition effect” as parts of a single EU policy action. The results for the regression of the level of total structural funds per capita on the socio-economic conditions (8.3) are presented. The overall amount of funds allocated to the EU regions partially reflects their underlying socio-economic conditions, even if the percentage of the overall variability explained by such factors is relatively small (the R-squared increases from 1994–1999 to 2000–2006 but it is still relatively small). When considering the specific socio-economic factors that influence the distribution of the funds, we notice that agricultural labour force, as a “traditional” source of disadvantage, still seems to be the main driver of the funds at the expense, for example, of the level of human capital accumulation which, instead, has been shown to be particularly relevant in the context of a knowledge based economy. The national dummies, while minimising the problem of spatial autocorrelation, highlight a certain degree of national bias in the distribution of the funds in favour of the “cohesion countries”. A bias for which, in the 1994–1999 period, Germany also received particular benefit. Overall this analysis of the “hidden” determinants of the allocation of the structural funds uncovers an even weaker association between the funds and the structural disadvantage of the EU territories. The introduction of the principle of territorial concentration has not only increased the spatial concentration of the funds but also improved their adherence to these factors of disadvantage. However, the analysis highlights that there is still much room for further improvement in both respects. In addition, while the general socio-economic structure of each region should be taken into account by the allocation mechanism of the funds, some specific factors deserve greater attention in the context of the knowledge based economy. This is especially true for human capital accumulation, whose deficiency has been shown insignificant to determine the amount of resources received by the regions but which has become a key source of competitive advantage for both the development of Objective 1 and the restructuring of Objective 2 regions.

8.4.3

Socio-economic Disadvantage and Regional Convergence

In the previous section it has been argued that a potential explanation for the lack of correlation observed between the factors of socio-economic disadvantage and the amount of funds received by the EU regions may be explained in terms of the desire to privilege, in the distribution of the funds, the relatively better endowed regions. This choice could find its theoretical justification in the emphasis on the receptiveness of the local economy as a prerequisite for successful development policies. In this perspective, developed in the framework of the neo-Schumpeterian literature, relatively more favourable socio-economic conditions are necessary for the investment to deliver (Cappellen et al. 2003) and, consequently, the policymaker may find it more cost-effective to target funds towards relatively better-off regions (those

Table 8.4 Heteroskedasticity-Consistent OLS model, Objective 1 and Objective 2 Funds per capita, 1994–1999 and 2000–2006 Programming period 1994–1999 Programming period 2000–2006 Variables Coef. Coef. Coef. Coef. Social Factorsa 327.894*** (129.8615) 162.214*** (42.01456) Education Population 771.8936 (863.6608) 10.0642 (231.26) Agricultural Labour Force 1846.892*** (566.4197) 703.0175*** (195.4019) Long Term Unemployment 363.4748 (264.9683) 119.7216 (81.18214) Young Population 3029.142** (1395.854) 1200.057*** (494.6487) National Dummies de 294.7922*** (111.1332) 205.139** (81.83613) 65.45534** (27.4801) 35.56319* (20.35761) it 57.38723 (80.60264) 46.11072 (96.23988) 9.09578 (27.36722) 22.1725 (26.60234) at 37.8744 (63.17935) 71.8916 (99.93928) 17.1091 (25.62074) 40.7265 (37.53585) be 153.1352 (100.7441) 15.7337 (119.9024) 54.42931* (26.19563) 2.24039 (30.53526) pt 58.9707 (73.48608) 69.3652 (93.02556) 179.3968*** (42.1867) 167.1739*** (52.87925) nl 91.98157 (61.66183) 194.286* (107.3449) 20.23761 (19.88387) 95.4172*** (36.32245) uk 214.5534*** (83.53881) 60.30519 (56.59665) 102.6423*** (27.09222) 33.96666 (22.9845) es 460.8256 (87.2242) 130.3368 (130.6492) 173.652*** (36.87841) 50.1997 (47.33312) gr 348.8422 (96.97734) 61.27249 (152.8804) 9.13357 (25.41967) 114.086** (52.04321) fi 233.367*** (83.44499) 82.88095 (102.4067) 15.2933 (10.75426) 78.7236*** (27.42229) Constant 247.3297 (60.25865) 596.29* (307.5034) 111.9031*** (18.47053) 178.189** (89.55031) R-squared 0.37 0.46 0.46 0.56 F-stat 8.71*** 5.47*** 17.38*** 7.62*** a This variable is the linear combination of the socio-economic variables described in the text and is calculated through Principal Component Analysis (Appendix C) *, ** and ***denote significance at a 10%, 5% and 1% level respectively. SE in parentheses

8.4 Empirical Results 163

164

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

Table 8.5 Testing sigma-convergence, all regions and Objective 1, 1994–2003 Test for sigma convergence 1994 2003 T1 All regions Sigma^2 33376383.85 43887527.32 0.760498 Objective 1 regions Sigma^2 9532911.765 11726050.54 0.812969

p 0.94 0.77

which show the better development potential) in order to maximise their impact. However, the empirical evidence on the economic performance of the Objective 1 regions over the 1994–2003 period (i.e., from the first year of implementation of the 1994–1999 programming to the most recent year for which regional GDP data are currently available) explicitly contradicts this assumption. When sigma-convergence is considered, by assessing the change in the total variance of the regional income per capita from 1994 to 2003, the lack of convergence for both the whole Europe and the subset of Objective 1 regions is apparent (Table 8.5). However, the comparison between the T1 statistic10 (i.e., the initial year variance/final year variance ratios) for all the EU regions and that for the Objective 1 only shows that dispersion for regional per capita income increased significantly more in the EU as a whole than in the Objective 1 regions, thus supporting the idea of there being a variety of “clubs” developing at different rates. The lack of a trend towards generalised (unconditional) convergence in the EU regions is confirmed by the simple beta-convergence analysis a` la Barro-Sala-i-Martin (1992) presented in Table 8.6. The regression shows a negative coefficient for the log of the initial level of the GDP per capita (8.1). However the evidence of unconditional convergence becomes much weaker and almost insignificant when a set of national dummies is introduced into the analysis (8.2) both controlling for the “national growth” effect and minimising the extent of spatial autocorrelation. The picture changes when the sub-sample of Objective 1 regions is considered separately: the degree of convergence is not only stronger (8.3) but also remains significant after the introduction of the national dummy variables (8.4). This confirms the idea of a process of “club convergence” (Quah 1996) among the Objective 1 regions which explicitly contradicts the idea of a better growth potential of the relatively more well-off regions. On the contrary, the initially more disadvantaged Objective 1 regions seem to grow faster than other potentially better endowed areas. The catching up of the former with the latter uncovers the growth potential of the poorest Objective 1 regions, a potential that would have been more effectively emphasized by a higher degree of concentration of the structural funds. In addition, as shown above, such reduced concentration has been coupled with a lack of correlation between the funds and the factors of structural disadvantage. The growth potential of more disadvantaged The T1 statistics is : T1 ¼

10

s ^21 ^2T s

^21 is the variance of regional income per capita at time 1; . Where s

^2T is the variance at time t. This statistic is distributed as a F with (n-1; n-1) degrees of freedom. s

no

yes

no

yes

4 Obj.1 0.1368** (0.054) 0.0128** (0.00565) 6.88E-05 (0.00056) yes

0.000966** (0.00041) no

58.20% 54.30% 15.04***

6 All 0.01273*** (0.00144)

5 All 0.017575*** (0.00066)

R-Sq 31.60% 59.5% 33.9% 60.5% 4.00% R-Sq (adj) 31.10% 55.7% 32.6% 49.9% 3.30% F 59.63*** 15.86*** 26.18*** 5.71*** 5.44** *, ** and ***denote significance at a 10%, 5% and 1% level respectively. SE in parentheses

National Dummies

Table 8.6 Regression analysis for beta-convergence Dependent variable: growth rate of regional GDP per capita, 1994–2003 1 2 3 Regions All All Obj.1 Constant 0.1207*** 0.0702*** 0.1582 (0.0133) (0.0202) (0.0267) LnGDP‘94 0.0108*** 0.00406* 0.01494*** (0.00140) (0.00208) (0.00292) Social Factors

18.4% 16.8% 11.51***

0.00179*** (0.00052) no

7 Obj.1 0.02049*** (0.00101)

60.5% 48.7% 5.11***

0.00017 (0.00129) yes

8 Obj.1 0.1323** (0.0645)

8.4 Empirical Results 165

166

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

Regional growth rate (1994-2003), per capita income

regions is confirmed when disadvantage is assessed in terms of socio-economic factors and becomes very apparent when considering the Objective 1 subset alone (compare Figs. 8.1 and 8.2 where regional growth rates are scattered against socioeconomic factors for all the EU15 regions and for the Objective 1 regions only).

GR21

1.0

0.8

FI 2

GR25

GR24

ES13 FIFI 1314 ES3 ES12 ES52 ES21 ES24 UKH ES22 UKJ GR23 FI 15 ES61 ES41 PT15 FR83 ES51 AT11 ES62 NL31 DEE DEG ES42 I T13 ES11 IPT11 T92 ES23 GR3UKN UKG UKK PT12 I T72 FR21 UKE DED FR72 BE2 AT31 GR11 DE5 UKF FR61 UKM ES53 NL42 FR26 FR52 FR63 PT14 I T93 I T51 UKL FR81 DE2 UKDNL33NL32 FR62 FR71 IAT22 T53 FR51 DE6 DE8 FR3 I TA I T8 AT21FR25 DE1 NL41 NL22 FR1 II TB BE1 I T52I T4 AT33 T91 I T11 AT32 NL21DE7 FR53 DE4 FR43 I T32 AT34 I T6 FR23 FR24 I T33 NL23 FR82 UKC I T71 FR22DECDE9 I T2 FR41 DEA FR42 NL12 AT13 BE3 DEB GR22 GR14 GR43 GR41

0.6

0.4

GR42 GR13

0.2

GR12 ES43 AT12 PT13

DEF NL13 NL34

I T12

DE3

NL11

0.0 0.0

UKI

0.2

0.4

0.6

0.8

1.0

Socio-economic Factors

Regional growth rate (1994-2003), per capita income

Fig. 8.1 Regional growth rate (94–03) versus socio-economic factors, all regions

GR21

1.0

0.8

GR25

GR24 GR22 GR14 GR43 GR41

0.6

GR42 GR13

GR23

PT15 FR83 AT11 IPT11 T92 ES11 I T72 PT12 GR11 PT14 I T93 I TA I T8 IITB T91

0.4

ES13 FI 13 ES12 ES52 ES41 FI 15 ES61 DEE DEG ES42 ES62 UKK GR3 UKN UKE DED UKM UKL UKD DE8 FR3 DE4 NL23 BE3

GR12 ES43PT13

0.2 DE3

0.0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Socio-economic factors

Fig. 8.2 Regional growth rate (94–03) versus socio-economic factors, Objective 1 regions

8.5 Conclusions

167

However, when convergence is assessed on the basis of socio-economic factors (Table 8.6; equations 8.5–8.8), the evidence suggests that, when national effects are controlled, many socio-economically disadvantaged regions are not able to catchup with the EU as whole (8.7) and with the Objective 1 “club”(8.8). In other words, in line with the literature on the socio-economic preconditions for regional growth, we find that such factors have hampered the capacity of Objective 1 regions to converge. Consequently, while there is no evidence to encourage the targeting of resources towards relatively better endowed regions (the contrary is in fact true) there is plenty of evidence to support the necessity for the EU development funds to tackle structural disadvantage. In consequence, the geographical correlation between such disadvantage and the allocation of the funds is confirmed to be a necessary condition for their effectiveness.

8.5

Conclusions

This chapter set out to compare the socio-economic pre-conditions for successful regional development (as indentified in previous chapters of this book) with the correlated allocation of structural funds in order to investigate the reasons for the reduced impact of the EU structural funds expenditure. A large part of the existing literature has referred such unsatisfactory performance back to the policy implemented within the EU development policy framework. Some contributions highlighted the distortive effects on the location of R&D intensive firms others the misallocation of the resources across development axes. This chapter suggests an additional explanation that the origin of the EU development policy problems in delivering the expected benefits may have arisen at a more upstream phase i.e., in the allocation mechanism of the funds to the regions. This mechanism might have caused not only an insufficient territorial concentration of the expenditure, but also an insufficient correlation between the funds and the set of socio-economic conditions which the empirical analyses pursued in previous chapters have proven responsible for hampering the economic success of many EU regions. Our empirical analysis investigated both these issues in order to test this potential explanation for the weak impact of structural funds. The results reveal that the regional distribution of the structural funds shows a degree of spatial concentration in compliance with the principle of concentration. However, while the theoretical discussion supported the idea that the EU funds should be allocated in order to “compensate” the structural disadvantage of the assisted areas (maximising their effectiveness), the empirical results suggest that the disadvantage is more spatially concentrated than the associated funds: in this perspective the present degree of funds’ concentration can be judged insufficient. Furthermore, the empirical model uncovered the weak association between the amounts of regional funds and the above-mentioned sources of competitive disadvantage, especially as far as the problem of human capital accumulation is concerned.

168

8 The EU Regional Policy and the Socio-economic Disadvantage of European Regions

Such a spatial allocation of the EU funds is likely to have reduced their capability to impact the regional growth performance of assisted regions and has inevitably produced a bias in the allocation of national resources as well, due to the co-financing mechanism.11 The policy analysis suggests that this geographical allocation of the funds may be either the result of the political dilution of the policy objectives (required by EU political equilibria) or the effect of an intentional focus on relatively better endowed regions. However, the empirical evidence casts doubt on the rationale of such a bias in favour of the areas supposed to offer a better receptiveness for the funds. Among the Objective 1 regions precisely the most socio-economically disadvantaged have shown a relatively better growth potential over the past years. Consequently, every effort should be produced not only to promote the territorial concentration of the expenditure (which is a necessary but not sufficient condition for increased effectiveness) but also to increase its capability to target the factors of socio-economic disadvantage. Such critical issues (and geographical concentration in particular) have been explicitly considered by the European Commission when assessing the weaknesses of the past programming periods. However, when the Commission’s analysis has to be balanced against both the individual countries’ claims in terms of budget equilibriums and/or inaccurate diagnoses on where it is most worth investing, the implementation of concrete corrective measures becomes a very gradual process.

11

“Each euro spent at the EU level by cohesion policy leads to further expenditure, averaging 0.9 euros, in less developed regions (current Objective 1) and 3 euros in regions undergoing restructuring (current Objective 2)” COMMUNICATION FROM THE COMMISSION, Brussels, 05.07.2005 COM(2005) 0299, “Cohesion Policy in Support of Growth and Jobs: Community Strategic Guidelines, 2007-2013”, p. 7.

Chapter 9

Conclusions

9.1

What We have Learnt on Regional Innovation and Growth Dynamics: Synopsis of Empirical Results

This book has focused on the relationship between innovation and regional growth in order to uncover the features (related to both physical space and the socioeconomic realm) that have enabled some areas to benefit more than others from the process of globalisation and technological change. In this perspective the objective of this work has been to analyse the role of geography (accessibility) and socio-economic conditions (social filter) in the translation of innovative efforts into regional economic growth. This analysis has been made possible by the cross fertilisation of different theoretical approaches to the process of innovation and its economic impact. The combination of different theoretical strands has resulted in the development of a joint empirical model, which, in its turn, made it possible to systematically investigate phenomena previously treated on a case-study basis. Indeed this “contamination” has been shown to be very promising, as it has enabled a large variety of phenomena, which were hitherto previously “confined” to the “grey areas” between different approaches, to be rigorously analysed on a quantitative basis. Our empirical results can be summarized as follows: (a) Our findings support the possibility that an increase in innovative efforts (where achieved) may boost growth performance even at the regional level. However, an increase in innovative effort is not necessarily likely to produce the same effects in all regions. On the contrary, the analysis of the EU regional growth dynamics suggests that a variety of local factors have an influence on this process; (b) The reduced physical accessibility of a territory (i.e., peripherality) curbs its capability to successfully translate innovation into regional growth due to both the influence of the reduced interconnectedness on the network structure of the local economy and the lower exposure to knowledge flows. Consequently one Euro invested in R&D in a peripheral region is less conducive to regional economic growth than the same Euro spent in a central, highly accessible

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3_9, # Springer-Verlag Berlin Heidelberg 2011

169

170

(c)

(d)

(e)

(f)

9 Conclusions

region. From this point of view peripheral regions suffer from an intrinsic source of competitive disadvantage which needs to be taken into account when innovation-based development policies are implemented; Socio-economic contextual factors (as proxies for regional systems of innovation) are fundamental for the process of innovation as they determine the productivity of local innovative efforts. Empirical analysis has identified a specific set of factors which seem to effectively proxy the system of innovation conditions of the both the EU and the US regions. These conditions are those related to three main domains: educational achievement, the productive employment of human resources and demographic structure. When the endogenous socio-economic conditions of an area prevent the establishment of a functioning system of innovation, the impact of R&D on regional growth is significantly curbed. In addition the analysis does not detect the presence of a minimum threshold effect: even small incremental improvements in contextual conditions may prove helpful. Knowledge spillovers are spatially bounded. Proximity is important for the transmission of economically productive knowledge, as spillovers show a strong distance-decay effect. In the EU 25 context, only innovative efforts pursued within a 180 minute travel radius have a positive and significant impact on regional growth performance. This evidence contradicts the idea of knowledge as being public and instantaneously available to everybody. The 180-minute limit for interregional knowledge flows is in line with the idea of a “human-embodied” transmission technology, since it allows the maximization of face-to-face contacts between agents. In the US knowledge spillovers tend to be even more localised remaining constrained to the functional borders of metropolitan areas. When the evidence outlined under the previous points is considered in a “systemic” way (as our theoretical framework allows us to do), it is possible to uncover the interrelations between the various mechanisms. Since knowledge spillovers are spatially bounded (d) they tend to create localised pools of knowledge in core areas (where the innovative efforts are also concentrated (a)) thus only marginally benefiting peripheral areas (b). The reduced exposition to knowledge flows can be compensated by a more efficient exploitation of existing knowledge where the correct system of innovation is in place locally (c). When looking at Europe and the United States in a comparative perspective we find that in the US the generation of innovation usually occurs in relatively selfcontained geographical areas that rely on their own R&D inputs, on favourable socio-economic environments and on the formation and attraction of highly skilled individuals. In Europe the process is much more linked to not just having an adequate socio-economic context, but to proximity to other innovative areas and to the capacity to assimilate and transform knowledge spillovers into innovation. Human capital mobility, in contrast to the US case, does not play a role. In the European Union, imperfect market integration, and

9.1 What We have Learnt on Regional Innovation and Growth Dynamics

171

institutional and cultural barriers across the continent prevent innovative agents from maximising the benefits from external economies and localised interactions, but compensatory forms of geographical process may be emerging in concert with further European integration. In contrast, the higher mobility of capital, population, and knowledge in the US not only promotes the agglomeration of research activity in specific areas of the country but also enables a variety of territorial mechanisms to fully exploit local innovative activities and (informational) synergies. The evidence produced so far, by uncovering some relevant mechanisms of the regional growth dynamics of the EU (also in a comparative perspective with the US), has provided us with a set of “diagnostic tools” for the understanding of the potential limitations of the EU regional policy to influence such dynamics. In particular: (g) The analysis of the impact of transport infrastructure investment – the key item of the EU regional policy budget – shows that regions with a good transport infrastructure endowment and well connected to regions with similar good endowments tend to grow faster. However, investment in infrastructure within a region or in neighbouring regions seems to leave many peripheral regions more vulnerable to competition. Furthermore, the positive impact of infrastructural endowment on growth tends to wane quickly and is weaker than that of, for example, the level of human capital. (h) The empirical analysis of the allocation of all EU regional policy funds shows that they are actually unable to adequately address the socio-economic factors that negatively affect the chance of economic success of any region. The analysis has shown that the sources of disadvantage are more spatially concentrated than the funds devoted to compensating such disadvantage and uncovers a weak association between structural disadvantage and EU funds, which provides a potential alternative explanation for the limited impact of EU regional policy. Consequently, structural policies could prove helpful to promote development in the EU’s lagging regions. provided that the necessary corrections are introduced in their allocation mechanism, so that the geographical concentration of the funds is increased and the available resources correctly earmarked to the most socio-economically disadvantaged regions. The discussion of the results achieved together with the policy implications derived from them show just how fruitful and promising the territorial “integrated” approach is with respect to the process of innovation and regional economic development. An in-depth understanding of the EU regional growth trajectories in response to the everlasting process of technological change is not only a fascinating research task but of fundamental importance for the design of adequate development policies at all levels. And the capacity of the European Union to deliver (cost) effective development policies is, in its turn, a necessary condition for an equitable distribution of the costs and benefits of the process of integration across territories and individuals.

172

9.2

9 Conclusions

What Are the Implications for Local and Regional Economic Development Policies?

In the previous paragraph we have discussed what the territorial “integrated” approach has allowed us to conclude on the regional policies implemented so far in the European Union. But how can we improve from here? Is there any general lesson for local and regional economic development policies that can be learnt from the “integrated” conceptual framework developed (and empirically tested) in this book? In order to answer this question let us briefly discuss how the “integrated” framework can work in practice as a guide for local and regional economic development policies and as a link between top-down and bottom-up approaches.

9.2.1

The Territorial “Integrated” Framework as a Common Platform for Top-Down and Bottom-Up Policies

When looking at the literature that has inspired and guided local and regional economic development policies, the separation between macro and micro economic theories inspiring top down development policies on the one hand and “meso-local” theories informing local bottom-up development policies on the other is immediately apparent. Micro-economic analysis has shaped top-down micro-policy options aiming to influence the allocation of labour and capital, while macro-economic approaches have provided the rationale for top-down macro-policy options targeting aggregate regional income and expenditure. In accordance with the predictions of their underlying theories the aim of these policies “is to induce capital and labour to locate in areas which would not normally have been chosen by those making the location decision” (Armstrong and Taylor 2000, p. 234). Conversely, bottom-up policies have addressed “the naturally occurring sources of economic potential growing from within localities and regions” (Pike et al. 2006, p. 155) targeting the drivers of economic performance brought to light by meso-theories of innovation and growth looking at “territorial” and “relational” assets including local institutions and networks (e.g., Regional Systems of Innovation), social capital and localised tacit knowledge. As far as the empirical informative basis for the design (and evaluation) of these policies is concerned, top-down policies have been traditionally informed and evaluated by means of quantitative/econometric analyses, while bottom-up policies by means of (almost exclusively) qualitative case-study-based evidence. The upper and lower parts of Fig. 9.1 show precisely this situation whereby macro and micro economic literature provides the theoretical framework and quantitative analysis the informative input for top-down policies while meso-level regional analysis and more qualitative studies constitute the basis for bottom-up approaches. However, in practice top-down (macro and micro) and bottom-up development policies co-exist, interact with, and impact upon the same agents (individuals and

9.2 What Are the Implications for Local and Regional Economic Development Policies?

Underlying Theories

Use of theories

Macro and Micro Economic Theories

Locally suited remedies based onthe Territorial ‘Integrated’ Framework

Bottom-Up

Informative basis / Evalutation Quantitative Analyses

Top-Down

Growth Diagnostic Approach

MesoLevel/Regional Theories

Policy approach Use of data

173

Regional Benchmarking and quantitative analysis of LED policies Qualitative Analyses/Case Studies

Fig. 9.1 Theoretical and informative basis of local and regional economic development policies

firms) and territories but – notwithstanding the unanimous call for coordination by academics, practitioners and policymakers – they show surprisingly limited synergies and osmosis. Our tenet is that this separation is the result of the lack of a common theoretical and conceptual ground in terms of the underlying economic development theories. In Chap. 2 we have extensively documented a similar division between different theories of regional innovation and growth. However, we also showed how different streams of macro, micro and meso-level literature can be cross-fertilised and combined into an “integrated framework”, developing a powerful analytical tool for the understanding of “real-world” innovation and growth dynamics. In the same vein, the “integrated approach” – as a combination of their underlying theories – forms a common conceptual background for both topdown and bottom-up policies. The explicit conceptualisation of both inter-regional external processes (in the form of spillovers) and internal indigenous factors – justified in light of either macro linear approaches (R&D efforts) or meso-level theories (Systems of innovation) – allows us to account for bottom-up policies as being part of an interactive and interconnected geography of localities (giving rise to spillovers) and, at the same time, for top-down policies as being rooted into differentiated heterogeneous territories (in terms of their indigenous characteristics). Following this line of reasoning the integrated approach developed in this book allows us to identify developmental factors as targets for both top down and bottom-up regional development policies, offering a common ground for their coordination and synergic convergence. Quantitative results from regression

174

9 Conclusions

analyses, as repeatedly pointed out in this book, should be seen as stylised representations of the processes they address and the “regularities” they highlight, and while robust, they need to be complemented by an understanding of the residual (unexplained) heterogeneity of each individual case. As an example, the lessons of the innovation systems approach go precisely in this direction and leads us to consider the complexity of the relationship between innovation and growth. Our quantitative analysis has been able to identify only some common features of the social filter conducive to the present geography of the European regional systems of innovation while additional qualitative analysis is needed to understand their interaction with “eminently contextualized conditions” (Cantwell and Iammarino 2003, p. 11) as “untraded interdependencies” and informal knowledge flows (Storper 1995).

9.2.2

Translating the Territorial “Integrated” Approach into a Diagnostic/Policy Tool

Once we assume the “integrated” conceptual framework as a common ground for both top-down and bottom-up policies, how to translate it into practical policy prescriptions? Existing economic literature has extensively attempted to identify practical targets for development policies, “translating” different theories into policy-guidance frameworks. In international macro-development literature, the “growth diagnostics” (Hausmann et al. 2008; Rodrik 2010) approach develops “practical” framework for growth-enhancing policies by adopting a similar perspective to the integrated framework developed in this book which is the result of the “eclectic” cross-fertilisation of different theories of regional innovation and growth. “Growth diagnostics is based on the idea that not all constraints [to economic growth and development] bind equally, and that a sensible and practical strategy consists of identifying the most serious constraint(s) at work (p. 6) (. . .) and remove them with locally suited remedies. Diagnostics requires pragmatism and eclecticism, in the use of both theory and evidence. It has no room for dogmatism, imported blueprints, or empirical purism” (Rodrik 2010, 7). However, existing works based on the “growth diagnostics” approach, even if extremely attentive to local characteristics, are still grounded into macro theories of development that are silent on the role of territorial assets, remaining relatively uninformative for bottom-up policies. Conversely, in the planning studies and regional development literature a number of works have explicitly attempted to provide informative (both qualitative and quantitative) and analytical support for bottom-up local development policies. Works on “regional benchmarking” (Huggins 2009a, b) “seek to understand regional contexts and promote improved regional innovation and competitiveness outcomes” (Huggins 2009a, 275) by comparing their performance, processes and policies in a systematic fashion. These analyses have accumulated an impressive stock of knowledge on various aspects of regional economies that can

9.2 What Are the Implications for Local and Regional Economic Development Policies?

175

certainly support the diagnosis of local economic conditions making inter-regional comparisons possible (as with the integrated framework developed in this book) together with the definition of “measurable” policy targets. However, the capability of these exercises to support policy-making is heavily constrained by the lack of a sound underlying conceptual framework justifying a particular choice of indicators and accounting not only for their qualitative differences and relative importance in different contexts but also for their functional relationships. Other contributions have moved their focus from “benchmarking” into the explicit search for the “drivers” of “regional competitiveness” looking for the theoretical mechanisms linking these drivers with the genesis of regional growth. Kitson et al. (2004) critically review this literature that spans from Porter’s competitive advantage diamond (and subsequent variations on the theme produced by private consultancies, national governments or the European Commission) to the progressively more sophisticated academic attempts by, of example, Budd and Hirmis (2004) or Wong (2002) and highlight a number of weaknesses that hamper their suitability to devise policy interventions. In a “growth diagnostic” perspective the lack of coherence in the theoretical justification of the various drivers of competitiveness – Kitson et al. (2004) suggest that “different theories seem to be implicit in different drivers” (p. 996) – is not per se problematic. However, this does become misleading “whereby it is assumed that the same ‘drivers’ are equally important everywhere, and hence the same basic policy model is applicable” (p. 996), as this contradicts the growth diagnostic logic and paves the way to imitation of best practices that overlook the “(often subtle) interdependencies that exist between the different factors contributing to a successful model” (Boschma 2004, p. 1011). The translation of the “integrated framework” for regional and local innovation and growth developed in the book into a conceptual toolkit for (bottom up and topdown) development policies is an attempt to show how it is possible to fill this double gap in the literature (central section of Fig. 9.1). On the one hand we claim that the “integrated approach” can successfully bring the “growth diagnostic approach” into the regional and local policy-making by grounding our diagnosis into the literature directly focused on regional-territorial processes (central, lefthand side section of Fig. 9.1). On the other hand, by pragmatically and eclectically cross-fertilising theories (as we attempted to do in this book in line with the growth diagnostic approach) rather than simply linking-up indicators (as in most of the literature on regional competitiveness) we can identify clear, measurable, theoretically-grounded targets for local and regional policies and, at the same time, “grasp the logistics of the relationships between (these) different socioeconomic factors in the development process” (Wong 2002, p. 1833), producing a more accurate picture of the local economy for locally suited remedies (central, right-hand side section of Fig. 9.1). The arrows in Fig. 9.1 show the convergence of all these factors in the integrated framework: a regional “growth diagnostics” approach allows us to use this framework as a tool for the identification of policy targets and locally-suited remedies while our theory-driven “regional benchmarking” approach enables the collection of information simultaneously relevant for both top-down and bottom-up policies

176

9.2.3

9 Conclusions

A Diagnostic Policy Tool for Locally-Suited Economic Development Policies

By looking at regional innovation and growth through the lenses of the “integrated approach” developed in this book, the “growth diagnostics” approach allows us to indentify a set of developmental factors that act as targets simultaneously for bottom-up and top-down economic development policies. The five keystones of regional and local economic development grounded into the “integrated” framework, brought into light by a “growth diagnostics” approach to regional dynamics and operationalised as in the regional benchmarking literature are visualised in Fig. 9.2 and include: Innovative activities, socio-economic (social filter) conditions, geographical factors/accessibility, international (trans-local) linkages and local and regional policies. These factors can be seen as key regional assets/liabilities that benefit/hamper local firms and business, and hence are major aspects of regional competitive advantage/disadvantage (Kitson et al. 2004). Local innovative activities are the engines of regional economic performance. In quantitative terms they work as inputs in the Knowledge Production Function for the generation of new ideas to be translated into economic growth. However, in a more qualitative perspective, innovative activities can be pursued in different contexts – as discussed in Chap. 6 when comparing Europe and the US – with different roles being played by private firms, research centres and universities. The developmental impact of innovative activities crucially depends upon two Innovative Activities

r ilte lF cia So

d ise

Accessibility to innovationprone space

ies olic s Sy

es ntiv Inc e

Local and regional policies and investment

s

tive

en

Inc

Incentives / infrastructure

ers

G se lob un arc al as tra h f se da or ts ble

Synergy Compensation Hindrance

Geography (Density and Accessibility)

Sp illo v

ic p tem

ge wled Kno vers / spillo ledge w Kno ing k see

Socio-Economic Conditions

l oca to l rs re e osu illov Exp sp

(Innovative and absorptive capacity)

International Local-Global Linkages (Pipelines/Networks)

Fig. 9.2 The diagnostic policy tool for locally-suited development strategies and its five keystones

9.2 What Are the Implications for Local and Regional Economic Development Policies?

177

other factors: socio economic conditions and geography. In this book the concept of “Social filter” – the structural socio-economic pre-conditions for the development of a well-functioning Regional System of Innovation – has been adopted as a quantitative proxy for local institutions supportive/detrimental to the process of innovation, making inter-regional and inter-temporal comparisons and benchmarking possible. However, in a bottom-up perspective, the assessment of the “social filter” conditions can be complemented by qualitative considerations capturing its institutional and relational underpinnings. In Fig. 9.2, the arrows show the functional links between the various components of our diagnostic framework. The social filter conditions should be assessed in relation to innovation as they shape the translation of innovative efforts into new economically-viable knowledge and economic growth. The same line of reasoning applies to geography: in Chaps. 3 and 5 we showed that different quantitative proxies (accessibility index and spatially lagged variables respectively) can capture the impact of geographical distance from economic and innovative activities on local economic performance. The exposure to knowledge spillovers is an important predictor for the economic success of the EU regions, highlighting two potential mechanisms for this interaction to take place: highly localised intra-regional contacts (density) or inter-regional connections (accessibility). However, in Chap. 6 we showed that not only established patterns of proximity to other agents matter but also the dynamic (often qualitative) process by which such densities and proximities are achieved and how they adjust over time. As a consequence, the effects of changes in density are influenced by the qualitative features of the underlying inflows of productive factors. For example, the attraction and agglomeration of highly skilled individuals has very different implications for local economic performance when compared to the sedimentation of low-skilled individuals. Similarly, the impact of improvements in regional accessibility resulting from transport infrastructure investments – as highlighted in Chap. 7 – is heavily influenced not only by other local characteristics (i.e., other keystones in Fig. 9.2) but also by the qualitative features of specific infrastructural projects. Micro-analyses have emphasized how insufficient accessibility may not only be the result of “physical” barriers or lack of infrastructure but also the consequence of bottlenecks or network failures that can only be identified by means of a qualitative assessment of the constraints to local accessibility. Geography – as suggested by the arrows in Fig. 9.2 – is directly linked to innovative activities and socio-economic conditions in shaping innovation and growth performance. Total accessibility to innovative activities (Chap. 5) determines the potential exposure to knowledge spillovers while accessibility to innovation prone space (Chap. 4) enables spillovers from innovation prone regions towards their neighbourhood. As extensively discussed in this book, innovation and growth dynamics are not only influenced by physical accessibility: the position of each region in “global networks” and its exposure to global knowledge flows is also important for the diagnosis of local conditions. The capability of local actors to develop organisational, institutional and social proximity relations with other agents determines the position of the local economy in global networks. In this case, while the “degree of centrality” of each region in these networks can be assessed in quantitative terms

178

9 Conclusions

(e.g., by looking at patent citations or FDI flows) the design of bottom-up policies will require a qualitative case-by-case understanding of the nature of this nonlocal links and of their supportive forces. Participation in global networks will exert an influence on all of the developmental “keystones” mentioned above. Global nodes/actors (e.g., Multinational Firms) located in the regions will benefit from and contribute to the Regional System of Innovation and, more generally, to the socioinstitutional environment in their global search for untradeable assets. By means of similar mechanisms they would benefit from (but also contribute to) local innovative efforts: spillovers will take place between local and global actors channelling “global” knowledge into the local environment and “pumping” the results of local innovative efforts into global knowledge pipelines. Geographical accessibility is also a key catalyser for the localisation of global networks given that good accessibility will attract the location of these “nodes” by reinforcing exposure to internal localised spillovers generated by indigenous activities with the accessibility to inter-regional broader knowledge flows. Finally, any top-down or bottom-up assessment of the regional economy would be incomplete without taking into account (pre)existing development policies that, initiated and implemented at different levels (from national policies to communitylevel initiatives), impact upon the local economy. These policies can be targeted towards one or more of the “keystones” of local economic development: policies at all levels can address “systemic” social filter conditions, incentivize/support R&D activities or try to increase the “centrality” of the region in geographical (e.g., by means of improvements in transport infrastructure) or social/organisational/institutional terms (e.g., by promoting international cooperation among firms and/or universities or by means of incentives for the location of Multinational Firms or other “global” actors). The simultaneous and reciprocal interaction of all these five factors determines – as we have shown in this book – how regions produce innovation and growth. When these interactions are synergetic (i.e., all drivers of local economic development work in the same direction), the growth potential of the local economy is maximised. However, good economic performance is also possible when some factors work in order to compensate for the weaknesses of others (e.g., our empirical analyses showed that innovation can be achieved even in the presence of suboptimal R&D efforts provided that social filter conditions and exposure to external spillovers are strong enough). Conversely, innovation and growth are hampered in the regions where one or more of these factors are persistently weak without some sort of compensation or where they tend to check each other. This happens, for example, when accessibility is increased without a prior reinforcement of internal socio-economic conditions, leaving the local economy exposed to external competition without fully developing its indigenous potential. Good governance and wellfunctioning institutions allow synergies to prevail “endogenously”. However, local and regional development policies can certainly help the establishment of balanced patterns where this is not the case. If these five factors (or keystones) are given a quantitative meaning (as we did in the econometric analyses included in the book), they may serve as targets

9.3 Concluding Remarks

179

for top-down policies by allowing inter-regional comparisons (e.g., for opportunity/ cost assessment in the use of available resources) and a clear pinpointing of policy targets. Conversely, where the “keystones” in Fig. 9.2 are differentiated in qualitative terms (as the underlying literature covered in the book allows us to do), they work as diagnostic tools for local economic conditions and as drivers for bottom-up policies. In this perspective the “integrated framework” constitutes a common ground for both top-down and bottom-up policy actions that makes their coordination easier. It is certainly true that bottom-up and top-down policies are governed by different political processes and collective action dynamics involving a different balance between efficiency and equity issues: while top-down regional policies have been traditionally concerned with a mixture of aggregate efficiency and territorial equity, bottom-up approaches have been essentially concerned with local efficiency. However, increasing constraints in terms of public finance have emphasized efficiency considerations in top-down policies while increasing interconnectedness between local areas (and their communities) has favoured local actors’ awareness of the impact of external conditions on local performance, making coordination between different policy actions increasingly relevant.

9.3

Concluding Remarks

Let us conclude with two final remarks on the subject matter and geographical scope of the conceptual framework developed in previous sections of this book and directly linked to local and regional economic development policies in this final chapter. As far at the subject matter of our analysis is concerned, both our theoretical and empirical analyses looked at the territorial processes underlying the generation of innovation and its translation into economic growth. In some sections of the book our dependent variable was the growth rate of regional GDP per capita, in others we narrowed down our focus directly into the generation of new ideas (the engine of economic growth) choosing patent intensity as our outcome variable. How and to what extent are economic growth and innovation intensity informative on local and regional economic development? “Economic growth” and “economic development” are certainly not synonyms and cannot be used interchangeably. As highlighted by Pike et al. 2007, “what constitutes development changes over time, shaped by critique, debate, experience and evaluation (and it is) geographically differentiated, varying within and between places over time. (Furthermore) the historically dominant focus upon economic development has broadened, albeit highly unevenly, to include social, ecological, political and cultural concerns” (p. 1255). As a consequence economic development is a broader concept involving local prosperity and wellbeing that are only partially the result of increases in productivity, income and employment (Storper 1997). These quantitative characters of economic development – explicitly addressed in this book – need to be coupled (pretty much as per the keystones discussed above) with the assessment of

180

9 Conclusions

their qualitative dimension by looking at their nature in terms of economic, social and environmental sustainability and “qualities” (Pike et al. 2007). In light of all this, should we care about income differences across regions? Following what Acemoglu (2009) concluded for cross-country differences,1 the answer is definitely yes: “high income levels reflect high standards of living” (p. 6) however this “implies neither that income per capita can be used as a ‘sufficient statistic’ for the welfare of the average citizen nor that it is the only feature that we should care about. (. . .) Economic growth is generally good for welfare but it often creates winners and losers. Joseph Schumpeter’s famous notion of creative destruction emphasizes precisely this aspect of economic growth; productive relationships, firms, and sometimes individual livelihoods will be destroyed by the process of economic growth, because growth is brought about by the introduction of new technologies and creation of new firms, replacing existing firms and technologies” (p. 7). It is precisely for this reason that the promotion and support of local and regional innovation-based growth should be addressed in the broader framework of “holistic” local and regional development policies able to target simultaneously and systematically the varied (social, ecological and political) dimensions of the local development process. The “participatory process that encourages and facilitates partnership between the local stakeholders, enabling the joint design and implementation of development strategies” (Canzanelli 2001, p. 9) at the basis of Local Economic Development policies serves precisely the scope of identifying the local “socially optimal” balance between all these dimensions assuring the long-term term sustainability of local and regional economic growth. The second concluding remark concerns the geographical scope of our analysis. Large part of the theoretical and empirical literature cited in this book was developed with reference to countries, regions and territories in the so called “developed” world. In addition, the vast empirical material analysed in the various sections of the book covers Europe and the United States. As a consequence the answer to the question “What can we learn for developing countries?” remains largely unexplored. Even if increasingly popular in the developing world (Nel 2001, Rodrı´guez-Pose 2002a), it is in Europe and the United States that local and regional development policy tools have been applied for a sufficiently long period of time and with enough quantitative and qualitative information to allow a consistent assessment of their effects and potential drawbacks. As a consequence, by abandoning a “best practice”/case study approach and developing sound conceptual framework for the analysis of regional and local innovation and growth dynamics, this book provides us with the necessary (though not sufficient) conceptual tool to understand what of the “developed world” policy experience can be successfully transferred into “developing” and emerging economies (and under what conditions). The re-design and re-conceptualisation of the framework developed in this book in line with the specific features of emerging economies is in our agenda for

1

Or, in Paul Krugman’s (1990) words: “A country’s ability to improve its standard of living over time depends almost entirely on its ability to raise its output per worker” (p. 9).

9.3 Concluding Remarks

181

future research together with the necessary validation by means of quantitative empirical analysis. We share Scott and Storper’s (2003) view that: “as globalization and international economic integration have moved forward, older conceptions of the broad structure of world economic geography as comprising separate blocs (First, Second and Third Worlds), each with its own developmental dynamic, appear to be giving way to another vision. This alternative perspective seeks to build a common theoretical language about the development of regions and countries in all parts of the world, as well as about the broad architecture of the emerging world system of production and exchange. . . it recognizes that territories are arrayed at different points along a vast spectrum of developmental characteristics.” (p. 582). In this perspective we are convinced that a pragmatic and eclectic use of both theory and evidence (a` la Rodrik) within the integrated framework developed in this book will provide us with useful directions for locally-tailored policy transfer.

.

References

Abramovitz M (1956) Resources and output trends in the United States since 1870. Am Econ Rev 46:523 Abramovitz M (1985) “Catching up and falling behind”, Economic research report n.1. Stockholm: Trade Union Institute for Economic Research Abramovitz M (1986) Catching up, forcing ahead and falling behind. J Econ Hist 46(2):386–406 Acemoglu D (2009) Introduction to modern economic growth. Princeton University Press, Princeton, NJ Acs ZJ (2002) Innovation and growth in cities. Edward Elgar, Northampton, MA ´ cs Z, Audretsch DB (1989) Patents as a measure of innovative activities. Kyklos 42:171–181 A Acs ZJ, Audretsch DB, Feldman MP (1992) Real effects of academic research: comment. Am Econ Rev 82:363–367 Adams JD (2002) Comparative localization of academic and industrial spillovers. J Econ Geogr 2:253–278 Adams JD, Jaffe AB (2002) Bounding the effects of R&D: an investigation using matched firm and establishment data. Rand J Econ 27:700–721 Aghion P, Howitt P (1992) A model of growth through creative destruction. Econometrica 60 (2):323–351 Amin A, Thrift N (1995) Institutional issues for the European regions: From markets and plans to socioeconomics and powers of association. Economy and Society 24:41–66 Andersson R, Quigley JM, Wilhehnsson M (2005) Agglomeration and the spatial distribution of creativity. Papers in Regional Science, 84:445–464 Anselin L (2003) Spatial externalities, spatial multipliers and spatial econometrics. Int Reg Sci Rev 26(2):153–166 Anselin L, Varga A, Acs Z (1997) Local geographic spillovers between University research and high technology innovations. J Urban Econ 42:422–448 ´ cs ZJ (2000) Geographic and sectoral characteristics of academic knowAnselin L, Varga A, A ledge externalities. Pap Reg Sci 79:435–443 Archibugi D, Coco A (2004) A new indicator of technological capabilities for developed and developing countries (ArCo). World Dev 32(4):629–654 Armstrong HW (1995) An appraisal of the evidence from cross-sectional analysis of the regional growth process within the European union. In: Vickerman RW, Armstrong HW (eds) Convergence and divergence among European regions. Pion, London Armstrong HW (2001) European union regional policy. In: El-Agraa AM (ed) The European union, 6th edn. Prentice Hall, Harlow Armstrong HW, Taylor J (2000) Regional economics and policy. Blackwell, Oxford Aschauer DA (1989) Is public expenditure productive? J Monetary Econ 23(2):177–200 Asheim BT (1999) Interactive learning and localised knowledge in globalising learning economies. Geo J 49:345–352 Audretsch DB (2003) Innovation and spatial externalities. Int Reg Sci Rev 26(2):167–174

R. Crescenzi and A. Rodrı´guez-Pose, Innovation and Regional Growth in the European Union, Advances in Spatial Science, DOI 10.1007/978-3-642-17761-3, # Springer-Verlag Berlin Heidelberg 2011

183

184

References

Audretsch DB, Feldman MP (1996a) R&D spillovers and the geography of innovation and production. Am Econ Rev 86:630–640 Audretsch DB, Feldman MP (1996b) Innovative clusters and the industry life cycle. Rev Ind Organ 11:253–273 Audretsch DB, Feldman M (2004) Knowledge spillovers and the geography of innovation. In: Henderson JV, Thisse JF (eds) Handbook of urban and regional economics, vol 4. Elsevier, Amsterdam, pp 2713–2739 Baldwin R, Wyplosz C (2003) The economics of European integration. McGraw-Hill, London Barro RJ (1991) Economic growth in a cross section of countries. Q J Econ 106(2):407–443 Barro RJ, Sala-i-Martin X (1992) Convergence. J Pol Econ 100:223–251 Bathelt H (2001) Regional competence and economic recovery: Divergent growth paths in Boston’s high technology economy. Entrepreneurship and Regional Development 13: 287–314 Batty M (2003) The geography of scientific citation. Environ Plann 35(5):761–765 Becattini G (1987) Mercato e forze locali. Il distretto industriale. Il Mulino, Bologna Becattini G (2000) Lo sviluppo locale nel mercato globale: riflessioni controcorrente. QA-La Questione Agraria 1:3–27 Beeson PE, DeJong DN, Troesken W (2001) Population growth in U.S. counties, 1840–1990. Reg Sci Urban Econ 31:669–699 Beugelsdijk S, de Groot H, van Schaik T (2004) Trust and economic growth, a robustness analysis, Oxford Economic Papers, 56: 118–134 Biehl D (1991) The role of infrastructure in regional development. In: Vickerman RW (ed) Infrastructure and regional development. Pion, London, UK Bilbao-Osorio B, Rodrı´guez-Pose A (2004) From R&D to innovation and economic growth in the EU. Growth Change 35:434–455 Boldrin M, Canova F (2001) Inequality and convergence in Europe’s regions: reconsidering European regional policies. Econ Policy 16:207–253 Borra´s S (2004) System of innovation theory and the European Union. Sci Public Pol 31 (6):425–433 Borts GH, Stein JL (1964) Economic growth in a free market. Columbia University Press, New York Boschma RA (2004) Competitiveness of regions from an evolutionary perspective. Reg Stud 38(9):1001–1014 Bottazzi L, Peri G (2003) Innovation and spillovers in regions: evidence from European patent data. Eur Econ Rev 47:687–710 Breschi S, Lissoni F (2001) Localised knowledge spillovers vs. innovative milieux: knowledge ‘tacitness’ reconsidered. Pap Reg Sci 80:255–273 Budd L, Hirmis AK (2004) Conceptual framework for regional competitiveness. Reg Stud 38(9):1015–1028 Burroni L (2001) Allontanarsi crescendo: Politica e sviluppo locale in Veneto e Toscana. Turin, Italy:Rosenberg & Sellier Bush V (1945) Science: the endless frontier. Ayer, North Stanford Button K (1998) Infrastructure investment, endogenous growth and economic convergence. Ann Reg Sci 32(1):145–162 Button K (2001) Transport policy. In: El-Agraa AM (ed) The European union: economics and policies. Prentice Hall Europe, Harlow, UK Cairncross F (1997) The Death of Distance. Cambridge, Ma: Harvard Business School Press Camagni R (1995a) The concept of innovative milieu and its relevance for public policies in European lagging regions. Pap Reg Sci 74:317–340 Camagni R (1995b) Global network and local milieu: toward a theory of economic space. In: Conti G, Malecki E, Oinas P (eds) The industrial enterprise and its environment: spatial perspectives. Aldershot, Avebury

References

185

Camagni R, Capello R (2003) La citta` come “milieu” e i “milieux” urbani: teoria e evidenza empirica. In: Garofoli G (ed) Impresa e territorio. Il Mulino, Bologna Canning D, Pedroni P (2004) The effect of infrastructure on long-run economic growth. Harvard University, Mimeo Canova F (2004) Testing for convergence clubs: a predictive density approach. Int Econ Rev 45:49–78 Cantwell J, Iammarino S (1998) MNCs, Technological innovation and regional systems in the EU: Some evidence in the Italian case. Int J Econ Bus 5:383–408 Cantwell J, Iammarino S (2003) Multinational corporations and European regional systems of innovation. Routledge, London Canzanelli G (2001) Overview and learned lessons on local economic development, human development, and decent work. ILO, Geneva Capello R (2004) Economia regionale. Il mulino, Bologna Cappelen A, Castellaci F, Fagerberg J, Verspagen B (2003) The impact of EU regional support on growth and convergence in the European Union. J Common Market Stud 41:621–644 Carlino G, Chatterjee S (2002) Employment deconcentration: a new perspective on America’s postwar urban evolution. J Reg Sci 42(3):445–475 Carlino G, Chatterjee S, Hunt R (2001) Knowledge spillovers and the new economy of cities. Working Paper n.01–14 Caselli C, Coleman J (2001) The US structural transformation and regional convergence: a reinterpretation. J Pol Econ 109(3):584–616 Castro EA, Jensen-Butler C (1999) Regional economic inequity, growth theory and technological change, Discussion paper 9903, Department of Economics. University of S.Andrews, Scotland Chandra A, Thompson E (2000) Does public infrastructure affect economic activity? Evidence from the rural interstate highway system. Reg Sci Urban Econ 30(4):457–490 Charlot S, Duranton G (2006) Cities and workplace communication: some quantitative French evidence. Urban Stud 43:1365–1394 Cheshire PC (2002) The distinctive determinants of European urban growth: Does one size fit all?, Research Papers in Environmental and Spatial Analysis N. 73, Department of Geography and Environment, London School of Economics Cheshire PC, Carbonaro G (1995) Convergence-divergence in regional growth rates: an empty black box? In: Armstrong HW, Vickerman RW (eds) Convergence and divergence among European regions. Pion, London Cheshire PC, Hay DG (1989) Urban problems in Western Europe: an economic analysis. Unwin Hyman, London Cheshire PC, Magrini S (2000) Endogenous processes in European regional growth: convergence and policy. Growth Change 31:455–479 Cheshire PC, Magrini S (2002) The distinctive determinants of European urban growth: does one size fit all? In: Research papers in environmental and spatial analysis no. 73. Department of Geography and Environment, London School of Economics Cicciotti E, Rizzi P (2003) Capacita` innovativa e sviluppo reginale: alcune evidenze delle regioni italiane negli anni novanta. In: Garofoli G (ed) Impresa e territorio. Il Mulino, Bologna Ciccone A (2002) Agglomeration effects in Europe. European Economic Review, 46:213–227 Ciccone A, Hall RE (1996) Productivity and the density of economic activity. Am Econ Rev 86(1):54–70 Cliff A, Ord JK (1972) Testing for spatial autocorrelation among regression residuals. Geogr Anal 4:267–284 Cliff AD, Ord JK (1981) Spatial processes: models and applications. Pion, London Coe NM, Bunnell TG (2003) ‘Spatialising’ knowledge communities: towards a conceptualisation of transnational innovation networks. Glob Networks 3(4):437–456 Cohen W, Levinthal D (1990) Absorptive capacity: a new perspective on learning and innovation. Admin Sci Q 35:128–152

186

References

Cooke P (1997) Regions in a global market: the experiences of Wales and Baden-Wurttemberg. Review of International Political Economy 4 Cooke P (1998) Origins of the concept. In: Braczyk H, Cooke P, Heidenreich M (eds) Regional innovation systems. UCL Press, London Cooke P, Gomez UM, Etxeberria G (1997) Regional innovation systems: institutional and organizational dimensions. Res Policy 26:475–491 Cooke P, Morgan K (1998) The associational economy: Firms, regions and innovation. Oxford, U. K.: Oxford University Press Crescenzi R (2004) Le differenziazioni regionali dell’agricoltura in Polonia di fronte alla Pac. QA/ La Questione Agraria 1:87–120 Crescenzi R (2005) Innovation and regional growth in the enlarged Europe: the role of local innovative capabilities, peripherality and education. Growth Change 36:471–507 Crescenzi R (2009) Undermining the principle of territorial concentration? EU regional policy and the socio-economic disadvantage of European regions. Reg Stud 43(1):111–133 Crescenzi R, Rodrı´guez-Pose A (2008) Infrastructure endowment and investment as determinants of regional growth in the European union. Eur Investment Bank Pap 13(2):62–101 Crescenzi R, Rodrı´guez-Pose A (2009) Systems of innovation and regional growth in the EU: endogenous vs. external innovative efforts and socioeconomic conditions. In: Fratesi U, Senn L (eds) Growth and innovation of competitive regions. Springer, Berlin, pp 167–192 Crescenzi R, Rodrı´guez-Pose A, Storper M (2007) The territorial dynamics of innovation: a Europe-United States comparative analysis. J Econ Geogr 7(6):673–709 Criscuolo P, Verspagen B (2006) Does it matter where patent citations come from? Inventor versus examiner citations in European patents, ECIS Working Papers 05.06 D’Antonio M, Scarlato M (2004) Trent’anni di trasformazioni dell’economia italiana: verso la ripresa dello sviluppo? Economia Italiana 2:277–331 Dall’erba S (2005) Distribution of regional income and regional funds in Europe 1989–1999: an exploratory spatial data analysis. Ann Reg Sci 39:121–148 Dall’erba S, Hewings GJD (2003) European regional development policies: the trade-off between efficiency-equity revisited. Discussion Paper REAL 03-T-02, University of Illinois at Urbana Champaign De Blasio G (2006) Production or consumption? Disentangling the skill-agglomeration connection, Bank of Italy. Tema di discussione n. 571 De Bondt R (1996) Spillovers and innovation activities, International Journal of Industrial Organization 15:1–28 De la Fuente A, Dome´nech R (2001) The redistributive effects of the EU budget. J Common Market Stud 39:307–330 Delmas MA (2002) Innovating against European rigidities. Institutional environment and dynamic capabilities. J High Technol Manage Res 13:19–43 Demissie E (1990) Small-scale agriculture in America. Westview, San Francisco Desmet K, Fafchmps M (2005) Changes in the spatial concentration of employment across US counties: a sectoral analysis 1972–2000. J Econ Geogr 5(3):261–284 Dicken P (1994) The Roepke lecture in economic geography global-local tensions: firms and states in the global space-economy. Econ Geogr 70(2):101–128 D€ oring T, Schnellenbach J (2006) What do we know about geographical knowledge spillovers and regional growth?: a survey of the literature. Reg Stud 40(3):375–395 Dosi G, Freeman C, Nelson R, Silveberg G, Soete L (eds) (1988) Technological change and economic theory. Pinter, London Dosi G, Llerena P, Sylos LM (2006) The relationships between science, technologies and their industrial exploitation: an illustration through the myths and realities of the so-called ‘European Paradox’. Res Policy 35(10):1450–1464 Drennan M, Lobo J (2007) Specialization matters: the knowledge economy and United States cities. Los Angeles: UCLA School of Public Affairs, unpublished manuscript Duntenam GH (1989) Principal component analysis. Sage, London

References

187

Duranton G, Puga D (2001) Nursery cities: Urban diversity, process innovation, and the life cycle of products. American Economic Review, 91:1454–1477 Duranton J, Puga D (2003) Micro-foundation of urban agglomeration economies. In: Henderson VJ, Thisse JF (eds) Handbook of regional and urban economics Vol. 4 cities and geography. Elsevier, Amsterdam Duranton J, Storper M (2006) Agglomeration and growth: a dialogue between economist and geographers. J Econ Geogr 6(1):1–7 Easterly W, Levine R (1997) Africa’s growth tragedy: Politics and ethnic divisions. Quarterly Journal of Economics 112:1203–1250 Edquist C (1997) Systems of innovation approaches – their emergence and characteristics. In: Edquist C (ed) Systems of innovation: technologies, institutions and organizations. Pinter Publishers/Cassell Academic, London Elhorst JP (2010) Spatial panel data models. In: Fischer MM, Getis A (eds) Handbook of applied spatial analysis, Part 3. Springer-Verlag, Berlin, New York, pp 377–407 Engelbrecht H-J (1997) International R & D spillovers, human capital and productivity in OECD economies: an empirical investigation. European Economic Review 41 (8), 1479–1488 Ergas H (1987) Does technology policy matter? In: Guile B, Brooks H (eds) Technology and global industry. National Academy Press, Washington, pp 191–245 European Commission (2000) Real convergence and catching-up in the EU, European Economy 71, Office for Official Publications of the EC. Luxembourg European Commission (2001) European Innovation Scoreboard. Office for Official Publications of the EC, Brussels European Commission (2002) European report on quality indicators of lifelong learning. Office for Official Publications of the EC, Brussels European Commission (2005a) Integrated guidelines for growth and jobs 2005–2008. COM(2005) 141 final2005/0057 (CNS) European Commission (2005b) Towards a European research area: science, technology and innovation, key figures 2005. Office for Official Publications of the European Communities, Luxembourg European Commission (2005c) EU’s higher education achievements and challenges: frequently asked questions (FAQ), MEMO/05/133, Brussels European Commission (2006) The demographic future of Europe – from challenge to opportunity, COM(2006) 571 final, Brussels European Commission (2007a) Commission staff working document accompanying the green paper “The European Research Area: New Perspectives” COM(2007)161, Brussels European Commission (2007b) “Communication from the commission – Trans-European networks: Towards an integrated approach”. COM/2007/0135 final EUROSTAT (1996) The regional dimension of R&D and Innovation statistics. Regional manual. Official Publication of EC, Luxemburg EUROSTAT (2006a) Population statistics. Official Publication of EC, Luxemburg EUROSTAT (2006b) Patent applications to the EPO at national level. Statistics in focus n.3 EUROSTAT (2006c) Patent applications to the EPO in 2002 at regional level. Statistics in focus n.4 Evans P, Karras G (1994) Are government activities productive? Evidence from a panel of United States states. Rev Econ Stat 76(1):1–11 Fabiani G (ed) (1991) Letture territoriali dello sviluppo agricolo. Franco Angeli, Roma Fageberg J (1988) Why growth rates differ. In: Dosi G, Freeman C, Nelson R, Silveberg G, Soete L (eds) Technological change and economic theory. Pinter, London Fagerberg J (1994) Technology and international differences in growth rates. J Econ Lit 32:1147–1175 Fagerberg J, Verspagen B, von Tunzelmann N (1994) The economics of convergence and divergence: an overview. In: Fagerberg J, Verspagen B, von Tunzelmann N (eds) The dynamics of technology, trade and growth. Edward Elgar, Cheltenham

188

References

Fagerberg J, Verspagen B, Caniels M (1997) Technology, growth and unemployment across European regions. Reg Stud 31(5):457–466 Faulconbridge JR (2006) Stretching tacit knowledge beyond a local fix? Global spaces of learning in advertising professional service firms. J Econ Geogr 6:517–540 Feldman M (1994) The geography of innovation. Kluwer, Boston Feldman M, Audretsch DB (1999) Innovation in cities: science-based diversity, specialisation and localised competition. Eur Econ Rev 43(2):409–429 Feldman MP, Florida R (1994) The geographic sources of innovation – technological infrastructure and product innovation in the US. Ann Assoc Am Geogr 84(2):210–229 Fischer MM (2010) A spatial Mankiw-Romer-Weil model. Ann Reg Sci. doi:10.1007/s00168010-0384-6 Fischer MM, Varga A (2003) Spatial knowledge spillovers and university research. Ann Reg Sci 37:303–322 Fischer M, Scherngell T, Jansenberger E (2009a) Geographic localisation of knowledge spillovers: evidence from high-tech patent citations in Europe. Ann Reg Sci 43(4):839–858 Fischer MM, Bartkowska M, Riedl A, Sardadvar S, Kunnert A (2009b) The impact of human capital on regional labor productivity. Lett Spat Resour Sci 2(2–3):97–108 Fischer MM, Scherngell M, Reismann T (2009c) Knowledge spillovers and total factor productivity: evidence using a spatial panel data model. Geogr Anal 41(2):204–220 Freeman C (1987) Technology policy and economic performance: lessons from Japan. Pinter, London Freeman C (1994) Critical survey: the economics of technical change. Camb J Econ 18:463–512 Freeman C, Soete L (1997) The economics of industrial innovation. MIT, Cambridge (MA) Fritsch M (2002) Measuring the quality of regional innovation systems: a knowledge production function approach. Int Reg Sci Rev 25(1):86–101 Fritsch M (2004) Cooperation and the efficiency of regional R&D activities. Camb J Econ 28(6):829–846 Fujita M, Thisse J-F (2002) Economics of agglomeration. University Press, Cambridge Furman JL, Porter ME, Stern S (2002) The determinants of national innovative capacity. Res Policy 31(6):899–933 Gambardella A, Malerba F (1999) The organization of innovative activity in Europe: toward a conceptual framework. In: Gambardella A, Malerba F (eds) The organization of economic innovation in Europe. Cambridge University Press, Cambridge, pp 1–2 Gertler MS, Wolfe DA, Garkut D (2000) No place like home? The embeddedness of innovation in a regional economy. Review of International Political Economy 7:688–718 Glaeser E (1998) Are cities dying? J Econ Perspect 12:139–160 Glaeser E, Kohlhase J (2004) Cities, regions and the decline of transport costs. Pap Reg Sci 83(1):197–228 Glaeser E, Kallal H, Scheinkman J, Schleifer A (1992) Growth in cities. J Polit Econ 100 (6):1126–1152 Glomm G, Ravi-Kumar B (1994) Public investment in infrastructure in a simple growth model. J Econ Dyn Cont 18(6):1173–1187 Gordon IR (2001) Unemployment and spatial labour markets: strong adjustment and persistent concentration. In: Martin R, Morrison P (eds) Geographies of labour market inequality. Routledge, London Gramlich E (1994) Infrastructure investment: a review essay. J Econ Lit 32(3):1176–1196 Green WH (2003) Econometric analysis. Prentice Hall, Upper Saddle River Greenbaum RT, Bondonio D (2004) Losing focus: a comparative evaluation of spatially targeted economic revitalisation programmes in the US and the EU. Reg Stud 38(3):319–334 Gregersen B, Johnson B (1996) Learning economies, innovation systems and European integration. Reg Stud 31:479–490 Greunz L (2003) Geographically and technologically mediated knowledge spillovers between European regions. Ann Reg Sci 37:657–680

References

189

Griliches Z (1979) Issues in assessing the contribution of research and development to productivity growth. Bell J Econ 10(1):92–116 Griliches Z (1986) Productivity, R&D, and basic research at the firm level in the 1970s. Am Econ Rev 76:141–154 Griliches Z (1990) Patent statistics as economic indicators: a survey. J Econ Lit 28:1661–1707 Grossman GM, Helpman E (1991) Innovation and growth in the global economy. MIT, Cambridge (MA) Grossman GM, Helpman E (1994) Endogenous Innovation in the Theory of Growth. Journal of Economic Perspectives 8:23–44 Guerrieri P, Iammarino S, Pietrobelli C (eds) (2001) The global challenge to industrial districts: small and medium-sized enterprises in Italy and Taiwan. Edward Elgar, Cheltenham Guiso L, Sapienza P, Zingales L (2004) The role of social capital in financial development. American Economic Review 94:526–556 Hart DM (2001) Antitrust and technological innovation in the US: ideas, institutions, decisions, and impacts, 1890–2000. Res Policy 30:923–936 Hausmann R, Rodrik D, Velasco A (2008) Growth diagnostics. In: Stiglitz J, Serra N (eds) The Washington consensus reconsidered: towards a new global governance. Oxford University Press, New York Heckman J (1979) Sample selection bias as a specification error. Econometrica 47:153–161 Held D, McGrew A, Goldblatt D, Perraton J (1999) Global Transformations: Politics, Economics and Culture, Stanford: Stanford University Press Henderson JV (1999) Marshall’s economies National Bureau of Economic Research. Working Paper 7358 Henderson JV (2003) Marshall’s scale economies. Journal of Urban Economics, 53:1–28 Henry N, Pinch S (2000) Spatialising knowledge: Placing the knowledge community of Motor Sport Valley. Geoforum 31:191–208 Hirch F (1977) Social limits to growth. Routledge, London Holl A (2006) A review of the firm-level role of transport infrastructure with implications for transport project evaluation. J Plann Lit 21(1):3–14 Holtz-Eakin D (1993) Solow and the states. Capital accumulation, productivity, and economic growth. Natl Tax J 46(4):425–439 Howells J (1999) Regional systems of innovation? In: Archibugi D, Howells J, Michie J (eds) Innovation policy in a global economy. Cambridge University Press, Cambridge Huggins R (2009a) Regional benchmarking in a global context: knowledge, competitiveness and economic development. Econ Dev Q 23(4):275–293 Huggins R (2009b) Regional competitive intelligence: benchmarking and policy-making. Reg Stud, First Published On Line Jan 2009 Iammarino S (2005) An evolutionary integrated view of regional systems of innovation: concepts, measures and historical perspectives. Eur Plan Stud 13(4):497–519 Institute for Higher Education (2006) Academic Ranking of World Universities – 2006, Shanghai Jiao Tong University (http://ed.sjtu.edu.cn/rank/2005/ARWU%202005.pdf) IRPUD (2000) European peripherality indicators (E.P.I). IRPUD GIS database. Institute of Spatial Planning, Dortmund Jaffe AB (1986) Technological opportunity and spillovers of R&D: Evidence from firms’ patents, profits and market share. Am Econ Rev 76:984–1001 Jaffe AB (1989) The real effects of academic research. Am Econ Rev 79(5):984–1001 Jaffe AB, Lerner J (2004) Innovation and its discontents. Princeton University Press, Princeton Jaffe AB, Trajtenberg M, Henderson R (1993) Geographic localisation of knowledge spillovers as evidenced by patent citations. Q J Econ 108:577–598 Jolliffe IT (1986) Principal component analysis. N Y J Int Bus Stud 32(4):641–665 Kaldor N (1978) The case for regional policy. In: Targetti F, Thirlwall AP (eds) Further essays on economic theory. Duckworth, London

190

References

Keeble D, Offord J, Walker S (1988) Peripheral regions in a community of twelve member states. Office for Official Publications of the European Community, Luxembourg Kendall M (1990) Rank correlation methods. Edward Arnold, Oxford Kitson M, Martin R, Tyler P (2004) Regional competitiveness: an elusive yet key concept? Reg Stud 38(9):991–999 Knack S, Keefer P (1997) Does social capital have an economic impact? A cross-country investigation. Quarterly Journal of Economics 112:1252–1288 Kristensen PH (1992) Industrial districts in West Jutland, Denmark. In Industrial districts and local economic regeneration, ed. F. Pyke and W. Sengenberger, 122–73. Geneva:International Institute for Labour Studies, International Labour Organization Leamer EE, Storper S (2001) The economic geography of the Internet age Lee L, Yu J (2010) Estimation of spatial autoregressive panel data models with fixed effects. J Econometrics 154:165–185 Lewis BD (1998) The impact of public infrastructure on municipal economic development: empirical results from Kenya. Rev Urban Reg Dev Stud 10(2):142–155 Lucas R (1988) On the mechanics of economic development. J Monetary Econ 22(1):3–42 ˚ (1992) National systems of innovation: towards a theory of innovation and interacLundvall BA tive learning. Pinter, London ˚ (2001) Innovation policy in the globalising learning economy. In: Archibugi D, Lundvall BA ˚ (eds) The globalising learning economy. Oxford University Press, Oxford Lundvall BA Maclaurin WR (1953) The sequence from invention to innovation and its relation to economic growth. Q J Econ 67(1):97–111 Maggioni MA, Nosvelli M, Uberti E (2006) Space vs networks in the goegraphy of innovation: a European analysis, Working paper 2006.153, Fondazione Eni Enrico Mattei Magrini S (1999) The evolution of income disparities among the regions of the European union. Reg Sci Urban Econ 29:257–281 Malecki E (1997) Technology and economic development: the dynamics of local, regional and national competitiveness, 2nd edn. Addison Wesley Longman, London Malerba F (2000) Economia dell’innovazione. Carocci, Roma Mankiw NG, Romer D, Weil D (1990) A contribution to the empirics of economic growth. NBER Working Paper 3541 Mariani M (2002) Next to production or to technological clusters? The economics and management of R&D location. J Manage Governance 6:131–152 Marshall A (1948) Principles of economics. Macmillan, London Martin P (1998) Can regional policies affect growth and geography in Europe? World Econ 2:757–774 Martin P (1999a) Are European regional policies delivering? EIB Papers 4(2):10–23 Martin R (1999b) The new geographical turn in economics: some critical reflections. Cambridge J Econ 23(1):65–91 Martin P, Rogers CA (1995) Industrial location and public infrastructure. J Int Econ 39(3–4):335–351 Maurset PB, Verspagen B (1999) Europe: one or several systems of innovation? An analysis based on patent citations. In: Fagerberg J, Guerrieri P, Verspagnen B (eds) The economic challenge for Europe. Edward Elgar, Cheltenham Midelfart KH, Overman HG, Redding S, VenablesA J (2002) The location of European industry. Eur Econ 2:216–273 Midelfart-Knarvik H, Overman HG (2002) Delocation and European integration: is structural spending justified? Econ Pol 17(35):322–359 Moreno R, Paci R, Usai S (2005a) Spatial spillovers and innovation activity in European regions. Environ Plann 37:1793–1812 Moreno R, Paci R, Usai S (2005b) Geographical and sectoral clusters of innovation in Europe. Ann Reg Sci 39(4):715–739 Morgan K (1997) The learning region: institutions, innovation and regional renewal. Reg Stud 31:491–503

References

191

Morgan K (2004) The exaggerated death of geography: learning, proximity and territorial innovation systems. J Econ Geogr 4:3–21 Mowery DC (1992) The US national innovation system: origins and prospects for change. Res Policy 21:125–144 Mowery DC (1998) The changing structure of the US National Innovation System: implications for International Conflict and Cooperation in R&D Policy. Res Pol 27(6):639–654 Munnell AH (1990) How does public infrastructure affect regional economic performance? New England Economic Review, September, pp. 11–32 Myrdal G (1957) Economic theory and underdeveloped regions. Duckworth, London Neary JP (2001) Of hype and hyperbolas: introducing the new economic geography. J Econ Lit 39(2):536–561 Nel E (2001) Local economic development: a review and assessment of its current status in South Africa. Urban Stud 38(7):1003–1024 Nelson RR, Rosenberg N (1993) Technical innovation and national systems. In: Nelson RR (ed) National systems of innovation: a comparative study. Oxford University Press, Oxford Nelson RR, Winter SG (1982) An evolutionary theory of economic change. Belknap, Cambridge (MA) NTSC National Science and Technology Council (1999) Annual report 1998. The White House, Washington OECD (2001) Using patent counts for cross-country comparisons of technology output. STI Rev 27:129–146 OECD (2006) Compendium of patent statistics. OECD, Paris Olson M (1982) The rise and decline of nations: economic growth, stagflation and social rigidities. Yale University Press, New Haven Oo˜huallachain B, Leslie TF (2007) Rethinking the regional knowledge production function. J Econ Geogr 7:737–752 Ottaviano G, Peri G (2006) The economic value of cultural diversity: evidence from US cities. J Econ Geogr 6(1):9–44 Peri G (2005) Skills and talent of immigrants: a comparison between the European Union and the United States. Institute of European Studies, UC, Berkeley Mimeo Pike A, Rodriguez-Pose A, Tomaney J (2006) Local and regional development. Routledge, London Pike A, Rodrı´guez-Pose A, Tomaney J (2007) What kind of local and regional development and for whom? Reg Stud 41(9):1253–1269 Piore M, Sabel C, (1984) The second industrial divide. New York: Basic Books Psaltopoulos D, Thomson KJ, Efstratoglou S, Kola J, Daouli A (2004) Regional social accounting matrices for structural policy analysis in lagging EU rural regions. Eur Rev Agric Econ 31:149–178 Puga D (2002) European regional policy in the light of recent location theories. J Econ Geogr 2:373–406 Puhani AP (2001) Labour mobility – an adjustment mechanism in Euroland? Empirical evidence for Western Germany, France, and Italy. Ger Econ Rev 2(2):127–140 Putnam R, (1993) Making democracy work: Civic traditions in modern Italy, Princeton: Princeton University Press Quah D (1996) Regional convergence clusters across Europe. Eur Econ Rev 40:951–958 Quah D (1997) Empirics for growth and distribution: stratification, polarisation and convergence clubs. J Econ Growth 2:101–120 Rebelo ST (1991) Long-Run Policy Analysis and Long-Run Growth, Journal of Political Economy, 99(3): 500–521 Richardson HW (1973) Regional growth theory. Macmillan, London Rodrı´guez-Pose A (1994) Socioeconomic restructuring and regional change: rethinking growth in the European community. Econ Geogr 70(4):325–343 Rodrı´guez-Pose A (1998a) The dynamics of regional growth in Europe: social and political factors. Oxford University Press, New York

192

References

Rodrı´guez-Pose A (1998b) Social conditions and economic performance: the bond between social structure and regional growth in Western Europe. Int J Urban Reg Res 22:443–459 Rodrı´guez-Pose A (1999) Innovation prone and innovation averse societies. Economic performance in Europe. Growth Change 30:75–105 Rodrı´guez-Pose A (2000) Economic convergence and regional development strategies in Spain: the case of Galicia and Navarre. EIB Papers 5(1):89–115 Rodrı´guez-Pose A (2001) Is R&D investment in lagging areas of Europe worthwhile? Theory and Empirical evidence. Pap Reg Sci 80:275–295 Rodrı´guez-Pose A (2002a) The European Union. Economy society and polity. OUP, Oxford Rodrı´guez-Pose A (2002b) The role of the ILO in implementing local economic development strategies in a globalised world. ILO, Geneva Rodrı´guez-Pose A, Crescenzi R (2008) R&D, spillovers, innovation systems and the genesis of regional growth in Europe. Reg Stud 42(1):51–67 Rodrı´guez-Pose A, Fratesi U (2004) Between development and social policies: the impact of structural funds in objective 1 regions. Reg Stud 38(1):97–114 Rodrı´guez-Pose A, Storper M (2006) Better rules or stronger communities? On the social foundations of institutional change and its economic effects. Economic Geography 82(1): 1–25 Rodrik D (2010) Diagnostics before prescription. J Econ Perspect 24:33–44 Romer PM (1986) Increasing returns and long-run growth. J Polit Econ 94(5):1002–1037 Romer PM (1990) Endogenous technological change. J Polit Econ 98(5):97–103 Romer PM (1994) The Origins of Endogenous Growth. The Journal of Economic Perspectives, 8(1):3–22 Rosenberg N (1994) Exploring the black box: technology, economics, and history. Cambridge University Press, New York Rosenthal S, Strange WC (2003) Geography, industrial organisation, and agglomeration. Rev Econ Stat 85(2):377–393 Rossert B (2000) Contributing to regional development through project selection. EIB Papers 5(1):137–148 Sch€urmann C, Talaat A (2000) Towards a European Peripherality Index. Final Report. Report for General Directorate XVI (Regional Policy) of the European Commission. Institute of Spatial Planning, Dortmund Scott A, Storper M (2003) Regions, globalization, development. Reg Stud 37:579–593 Sedgley N, Elmslie B (2004) The geographic concentration of knowledge: scale, agglomeration and congestion in innovation acress US states. Int Reg Sci Rev 27(2):111–137 Seitz H (1995) The productivity and supply of urban infrastructures. Ann Reg Sci 29(2):121–141 Seitz H, Licht G (1995) The impact of public infrastructure capital on regional manufacturing cost. Reg Stud 29(3):231–240 Semlinger K (1993) Economic development and industrial policy in Baden-W€urttemberg: Small firms in a benevolent environment. European Planning Studies 1: 435–463 Smith K (2007) Does Europe perform too little corporate R&D? In: Paper presented at the DRUID Summer Conference 2007. Copenhagen CBS, Denmark Solow R (1957) Technical change and the aggregate production function. Rev Econ Stat 39:312–320 Sonn JW, Storper M (2008) The increasing importance of geographical proximity in technological innovation: an analysis of US patent citations, 1975–1997. Environ Plann A 40(5):1020–1039 Stein JA (2004) Is there a European knowledge system? Sci Public Policy 31(6):435–447 Stiglitz JE (1986) Economics of the public sector. Norton, New York, London Storper M (1995) Regional technology coalitions. An essential dimension of national technology policy. Res Pol 24:895–911 Storper M (1997) The regional world: territorial development in a global economy. Guilford, New York Storper M, Venables AJ (2004) Buzz: face-to-face contact and the urban economy. J Econ Geogr 4:351–370

References

193

Trajtenberg M (1990) Economic analysis of product innovation. Cambridge University Press, Cambridge Trigilia C (1992) Sviluppo senza autonomia.Effetti perversi delle politiche nel Mezzogiorno Bologna: Il Mulino Vandamme F (2000) Labour mobility within the European union: findings, stakes and prospects. Int Labour Rev 139(4):437–455 Vanhoudt P, Math€a T, Smid B (2000) How productive are capital investments in Europe? EIB Papers 5(2):81–106 Varga A (1998) University research and regional innovation. Kluwer, Boston Varga A (2000) Local academic knowledge spillovers and the concentration of economic activity. J Reg Sci 40:289–309 Verspagen B (1991) A new empirical approach to catching up and falling behind. Struct Change Econ Dyn 12:374–397 Vickerman RW (1995) Regional impacts of Trans-European Networks. Ann Reg Sci 29(2):237–254 Vickerman R, Spiekermann K, Wegener M (1997) Accessibility and economic development in Europe. Reg Stud 33(1):1–15 Waters R, Lawton SH (2002) Regional development agencies and local economic development: scale and competitiveness in high-technology Oxfordshire and Cambridgeshire. Eur Plan Stud 10(5):633–649 Wieser R (2005) Research and development productivity and spillovers: empirical evidence at the firm level. J Econ Surv 19(4):587–621 Wong C (2002) Developing indicators to inform local economic development in England. Urban Stud 39:1833–1863 Wooldridge JM (2002) Econometric analysis of cross section and panel data. MIT, Cambridge (MA), USA Wooldridge JM (2003) Cluster-sample methods in applied econometrics. American Economic Review 93:133–138 Woolgridge JM (2003) Introductory econometrics: a modern approach. Thomson, Mason Zak P, Knack S (2001) Trust and growth. Economic Journal 111:295–321 Zimmermann K (1995) Tackling the European migration problem. J Econ Perspect 9:45–62 Zimmermann K (2005) European labour mobility: challenges and potentials. De Economist 127(4):425–450

.

Appendix A: Data Availability and Description of the Variables

The lack of statistics for the EU regions, as pointed out by almost all the scholars in this research field, seriously constrains the possibility of acquiring a deeper understanding of socio-economic dynamics. Even if over the last decade remarkable improvement have been recorded in response to increasing political and academic attention on regions, the situation is still critical on at least three counts. First, most statistics only go back a few years thus impeding long-term analyses. A second issue involves the spatial coverage of existing data as many variables are not available for some regions or countries (e.g., Swedish R&D expenditure data are not collected at the regional level). Thirdly the number of variables collected at regional level is very limited thus making the analysis of broader socio-economic processes often impossible. Moreover, the statistical picture of the EU regions has become even more fragmented following the accession of the New Member states on May 1 2004. For the new ten members of the Union very few data are available and, in the large majority of the cases, only from 1995 onwards. When the theoretical concepts outlined in the book have to be empirically tested, the selection of an appropriate scale of analysis, under the constraint of data availability, becomes crucial. In the context of our analysis the focus is upon the “institutionally defined region”, the sub-national level which maximises the level of internal coherence in terms of socio-institutional features while being associated with a meaningful political decision-taking level. By coherently applying such criteria to the EU-25 regions, in our empirical analyses, we will focus upon NUTS1 regions for Germany, Belgium and the UK and NUTS2 for all other countries (Spain, France, Italy, the Netherlands, Greece, Austria, Portugal, Finland, Czech Republic, Hungary, Poland, Slovakia). Countries without a relevant regional articulation (Cyprus, Denmark, Estonia, Ireland, Latvia, Lithuania, Luxemburg, Malta and Slovenia) were necessarily excluded from the analysis.1 Overall nine countries out of 25 are excluded from the analysis, as 1 As far as specific regions are concerned, no data are available for the French De´partments d’Outre-Mer (Fr9). Uusimaa (Fi16) and Etela-Suomi (Fi17) were excluded from the analysis due to the lack of data on socio-economic variables. Etela-Suomi (Fi17) and Trentino-Alto Adige (IT31) were excluded from the analysis as they have no correspondent in the NUTS2003 classification, thus preventing us from matching data available only in the new NUTS classification.

195

196

Appendix A: Data Availability and Description of the Variables?

inevitably happens in all empirical studies focusing upon EU regions that apply the same methodology. However, even if such a limitation should be explicitly acknowledged, we do not believe this affects the generality of our results, as the countries included in the analysis cover 95.6% of the total EU population, 95.8% of total GDP, and 96.9% of total R&D expenditure in the EU (1999 Eurostat data). In our EU analysis EUROSTAT data (stored in the REGIO databank on which we largely relied for our empirical analysis) have been complemented with Cambridge Econometrics (CAMECON) data for GDP. Table A.1 provides a detailed definition of the variables included in the analysis. A final consideration concerns the distortions produced by the use of NUTS2 regions as a unit of analysis. As pointed out by Cheshire and Magrini (2000), NUTS regions may bias regression analyses as their boundaries are often arbitrary and non homogeneous. This bias should be effectively addressed, as suggested by these authors, by focusing the analysis on F.U.R. (Functional Urban Regions)3 rather than on NUTS thereby capturing the functional structure of the regions. Unfortunately, the lack of available data for many of the relevant explanatory variables – a priori – has prevented us from considering functional regions in our analysis. Table A.1 Description of the variables, European Union Variable Definition Dependent Variables Dependent variable Annual growth rate of regional GDP (For the EU-15 1990–2004; for the EU-25 1995–2004). Dependent variable Annual growth rate of regional patents applications (Chap. 6) Innovation R&D Social Filter Life-Long Learning Education Employed People Education Population Agricultural Labour Force

Expenditure on R&D (all sectors) as a % of GDP Rate of involvement in Life-long learning – % of Adults (25–64 years) involved in education and training % of employed persons with tertiary education (levels 5–6 ISCED 1997) % of total population with tertiary education (levels 5–6 ISCED 1997) Agricultural employment as % of total employment (continued)

Islands (PT2 Ac¸ores, PT3 Madeira, FR9 Departments d’Outre-Mer, ES7 Canarias) and Ceuta y Melilla (ES 63) were excluded from the analysis as time-distance information, necessary for the computation of spatially lagged variables, is not available. 2 The Nomenclature of Territorial Units for Statistics (NUTS) was established by Eurostat more than 25 years ago in order to provide a single uniform breakdown of territorial units for the production of regional statistics for the European Union and the definition of these regions mainly served administrative purposes. 3 The concept of Functional Urban Regions (FURs) have been defined by the literature in order to minimise the bias introduced by commuting patterns. A FUR thus includes a core city, where employment is concentrated, and its hinterland, from which people commute to the centre. For a detailed analysis of this concept see Cheshire and Hay (1989).

Appendix A: Data Availability and Description of the Variables Table A.1 (continued) Variable Long Term Unemployment Young People Social Filter Index

197

Definition Long term unemployed as % of total unemployment People aged 15–24 as % of total population The index combines, by means of Principal Component Analysis, the variables describing the socio-economic conditions of the region (listed above)

Structure of the local economy Migration rate Net migration was calculated from population change plus deaths minus births and then standardised by the average population thus obtaining the net migration rate Population density Calculated as Average Population (units) in the base year/Surface of the region (Sq Km) % regional of national Total regional GDP as a percentage of national GDP GDP Krugman index of The index is calculated as discussed in the text on the basis Regional specialisation employment data classified according to the “Classification of economic activities – NACE Rev. 1.1 A17” branches Transport Infrastructure (Chap. 7) Kms of motorways per thousand inhabitants Motorways4 (Inhab.)5 Motorways (GDP) Kms of motorways per million EUR of GDP Motorways (Region Kms of motorways per square-kilometre area) D Motorways (Inhab.) Annual change in Kms of motorways per thousand inhabitants D Motorways (GDP) Annual change in Kms of motorways per million EUR of GDP D Motorways (Reg. Annual change in Kms of motorways per square-kilometre Area) Other Control Variables Log of GDPpc Natural logarithm of regional GDP per capita at time t National growth Annual growth rate of national GDP (for the EU-15 1990–2004; for the EU-25 1995–2004).

4

Definition of Motorway (Eurostat Regio Guide Book 2006): “Road, specially designed and built for motor traffic, which does not serve properties bordering on it, and which: is provided, except at special points or temporarily, with separate carriageways for the two directions of traffic, separated from each other, either by a dividing strip intended for traffic, or exceptionally by other means; does not cross at level with any road, railway or tramway track, or footpath; is specially signposted as a motorway and is reserved for specific categories of road motor vehicles. Entry and exit lanes of motorways are included irrespectively of the location of the sign-posts. Urban motorways are always included.” 5 Italy: missing data for all regions after the year 2000. Missing have been replaced by means of comparable ISTAT dataGreece: data are missing from 1996. Greece has been excluded from the analysisPoland: data are missing in the Eurostat databank for some regions without any explanatory note. Data are also missing from the Polish National Statistical Institute databank. By inspecting a map of motorways in Poland (2004) the Kms of motorways in these regions appears to be zero.Portugal: missing data for Centro, Lisboa and Alentejo from 1990 to 2002Regional surface in 2003 has been used to calculate the density of transport infrastructure to avoid

198

Appendix A: Data Availability and Description of the Variables

The US analysis is based upon 266 MSA/CMSAs6 covering all continental US States (and the District of Columbia), while MSAs in Alaska, Hawaii, or in other non mainland territories of the US are excluded from the analysis. The lack of substate level data for R&D expenditure was addressed by relying upon Standard & Poor’s Compustat7 North American firm-level data which provide a proxy for private R&D expenditure in 145 MSAs out of the total of 266. The proxy was calculated by summing up firms’ R&D expenditure in each MSA. Though rough, this is the only measure available and similar proxies have been commonly used in the literature on the MSA innovative activities (e.g., Feldman 1994). All other US variables are based on US-Census data included in the USA Counties 1998 CD-Rom Table A.2 Description of the variables, United States (Chap. 6) Variable Definition Innovation R&D Private expenditure on R&D as a % of GDP was calculated from Standard & Poor’s Compustat North America firmlevel data Social Filter Education: bachelor’s, graduate or professional degrees Education: some college level education Agricultural Labour Force Unemployment Rate Young People Structure of the local economy Domestic migration Population density % regional of national GDP Krugman index of specialisation

Persons 25 years and over – some college or associate degree as a percentage of total population Persons 25 years and over – bachelor’s, graduate, or professional degree as a percentage of total population Agricultural employment as % of total employment Rate of unemployment People aged 15–24 as % of total population Rate of net domestic migration Calculated as Average Population (units) in the base year/ Surface of the region (Sq Km) Total regional GDP as a percentage of national GDP The index is calculated on the basis of the 13 major industry groups reported by 1990 census classification and developed from the 1987 Standard Industrial Classification (SIC) Manual.

generating noise in the density variable due to changes in the calculation of the regional surface. Regional GDP and average population in 1990 and 1995 have been used to standardize the variables included in the EU-15 and EU-25 regressions respectively. 6 The MSA/CMSA list is based on Metropolitan Areas and Components, 1993, with FIPS Codes, published by the Office of Management and Budget (1993). 7 Standard & Poor’s Compustat North America is a database of financial, statistical, and market information covering publicly traded companies in the U.S. and Canada. It provides more than 340 annual and 120 quarterly income statements, balance sheets, flows of funds, and supplemental data items on more than 10,000 active and 9,700 inactive companies.

Appendix B: The Weight Matrix and the Moran’s I

The Moran’s I is calculated on the basis of the following formula: n P n P

I¼

ðxi  xÞwij ðxj  xÞ

i¼1 j¼1 n P

ðxi  xÞ

i¼1

where wij is a sequence of normalised weights that relate observation i to all the other observations j in the data. Values of I larger (smaller) than the expected value E(I) ¼ 1/(n1) signal the presence of positive (negative) spatial autocorrelation. In our empirical application the element wij of the matrix of the normalised weights is: 1 dij

wij ¼ P j

1 dij

where dij is the average trip-length (in minutes) between region i and j calculated by the IRPUD (2000) for the computation of the Peripherality Indicators and made available by the European Commission.

199

.

Appendix C: Technicalities of the Principal Component Analysis and Results for the EU and the US

The principal component analysis (PCA) is “a statistical technique that linearly transforms an original set of variables into a substantially smaller set of uncorrelated variables that represents most of the information in the original set of variables: (. . .) a smaller set of uncorrelated variables is much easier to understand and use in further analysis than a larger set of correlated variables” (Duntenam 1989 p. 9). Through the PCA the original variables (in the case of our analysis the variables shown in literature as representative of the socio-economic disadvantage of the EU regions) are linearly combined by means of a set of “weights” (a1, a2, . . ., ak) calculated in order to maximise (under the constraint of that the sum of the squared weights is equal to one) the variability of the resulting indicator, i.e., of the principal component (our Social Factors variable). Consequently the i-th principal component is: yi ¼ ai1 x1 þ ai2x2 þ    þaip xp where (ai1, ai2 aip) are the wights and x1, x2, . . . ,xk are the k variables. It is possible to calculate as many PCs as the original variables under the constraint of non-correlation with the previous ones. Anyway the PCs are able to account for a progressively decreasing amount of the total variance of the original variables. Consequently, the procedure allow us to concentrate our attention on the first and limited number of PCs, which are the most representative of the phenomenon under analysis. Table C.1 shows the Eigenanalysis of the Correlation Matrix. The first PC alone accounts for around 43% of the total variance with an Eigenvalue significantly larger than 1, the second PC accounts for an additional 22% of the total variability with an Eigenvalue still larger than 1. The first two principal components therefore explain a significant part of total variability (65%). Table C.1 EU regions: Eigenanalysis of the Correlation Matrix Eigenvalue 2.566 1.3311 0.8847 0.6542 Proportion 0.428 0.222 0.147 0.109 Cumulative 0.428 0.65 0.797 0.906

0.5381 0.09 0.996

0.0259 0.004 1

201

202

Appendix C: Technicalities of the Principal Component Analysis and Results

The coefficients of the first PC (Table B.2) assigns a large weight to the educational achievements of the population (0.576) and the labour force (0.551) and to the participation in Life Long Learning Programmes (0.383). A negative weight is, as expected, assigned to the agricultural labour force (0.446) and, with a smaller coefficient, long-term unemployment (0.139). The weight of the young population (0.006) is much smaller but positive. This first principal component provides us with the “joint measure” for each region’s socio-economic conditions. Consequently, the first principal component’s scores are computed from the standardised8 value of the original variables by using the coefficients listed under PC1 in Table C.2. Table C.2 EU regions: principal components’s coefficients Variables PC1 Education population 0.576 Education labour force 0.551 Life-long learning 0.383 Agricultural labour force 0.446 Long term unemployment 0.139 Young people 0.006

PC2 0.218 0.318 0.326 0.227 0.505 0.662

Table C.3 US MSAs : Eigenanalysis of the correlation matrix US Eigenvalue 1.6979 1.0514 1.0306 Proportion 0.34 0.21 0.206 Cumulative 0.34 0.55 0.756 Table C.4 US MSAs: principal components’ coefficients Variable PC1 US People with any college level degree 0.413 People with bachelor degree 0.682 Rate of unemployment 0.203 Agricultural labour force 0.174 Young people 0.542

8

Standardised in order to range from 0 to 1.

0.9499 0.19 0.946

PC3 0.043 0.05 0.355 0.068 0.802 0.471

0.2702 0.054 1

PC2 0.491 0.105 0.856 0.119 0.04

Appendix D: List of the Regions Included in the Analysis

Table D.1 EU NUTS Regions Country AT AT AT AT AT AT AT AT AT BE BE BE CZ CZ CZ CZ CZ CZ CZ CZ DE DE DE DE DE DE DE DE DE DE DE DE DE DE DE DE ES

NUTS code AT11 AT12 AT13 AT21 AT22 AT31 AT32 AT33 AT34 BE1 BE2 BE3 CZ01 CZ02 CZ03 CZ04 CZ05 CZ06 CZ07 CZ08 DE1 DE2 DE3 DE4 DE5 DE6 DE7 DE8 DE9 DEA DEB DEC DED DEE DEF DEG ES11

Name Burgenland Niederosterreich Wien Karnten Steiermark Oberosterreich Salzburg Tirol Vorarlberg Bruxelles-Brussel Vlaams Gewest Region Walonne Praha Stredni Cechy Jihozapad Severozapad Severovychod Jihovychod Stredni Morava Ostravsko Baden-Wurttemberg Bayern Berlin Brandenburg Bremen Hamburg Hessen Mecklenburg-Vorpomm. Niedersachsen Nordrhein-Westfalen Rheinland-Pfalz Saarland Sachsen Sachsen-Anhalt Schleswig-Holstein Thuringen Galicia (continued) 203

204 Table D.1 (continued) Country ES ES ES ES ES ES ES ES ES ES ES ES ES ES ES FI FI FI FI FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR FR GR GR GR GR GR GR GR GR GR GR

Appendix D: List of the Regions Included in the Analysis

NUTS code ES12 ES13 ES21 ES22 ES23 ES24 ES3 ES41 ES42 ES43 ES51 ES52 ES53 ES61 ES62 FI13 FI14 FI15 FI2 FR1 FR21 FR22 FR23 FR24 FR25 FR26 FR3 FR41 FR42 FR43 FR51 FR52 FR53 FR61 FR62 FR63 FR71 FR72 FR81 FR82 FR83 GR11 GR12 GR13 GR14 GR21 GR22 GR23 GR24 GR25 GR3

Name Asturias Cantabria Pais Vasco Navarra Rioja Aragon Madrid Castilla-Leon Castilla-la Mancha Extremadura Cataluna Com. Valenciana Baleares Andalucia Murcia Ita-Suomi Vali-Suomi Pohjois-Suomi Aland Ile de France Champagne-Ard. Picardie Haute-Normandie Centre Basse-Normandie Bourgogne Nord-Pas de Calais Lorraine Alsace Franche-Comte Pays de la Loire Bretagne Poitou-Charentes Aquitaine Midi-Pyrenees Limousin Rhone-Alpes Auvergne Languedoc-Rouss Prov-Alpes-Cote d’Azur Corse Anatoliki Makedonia Kentriki Makedonia Dytiki Makedonia Thessalia Ipeiros Ionia Nisia Dytiki Ellada Sterea Ellada Peloponnisos Attiki (continued)

Appendix D: List of the Regions Included in the Analysis Table D.1 (continued) Country GR GR GR HU HU HU HU HU HU HU IT IT IT IT IT IT IT IT IT IT IT IT IT IT IT IT IT IT IT NL NL NL NL NL NL NL NL NL NL NL NL PL PL PL PL PL PL PL PL PL PL

NUTS code GR41 GR42 GR43 HU01 HU02 HU03 HU04 HU05 HU06 HU07 IT11 IT12 IT13 IT2 IT32 IT33 IT4 IT51 IT52 IT53 IT60 IT71 IT72 IT8 IT91 IT92 IT93 ITA ITB NL11 NL12 NL13 NL21 NL22 NL23 NL31 NL32 NL33 NL34 NL41 NL42 PL01 PL02 PL03 PL04 PL05 PL06 PL07 PL08 PL09 PL0A

205

Name Voreio Aigaio Notio Aigaio Kriti Kozep-Magyarorszag Kozep-Dunantul Nyugat-Dunantul Del-Dunantul Eszak-Magyarorszag Eszak-Alfold Del-Alfold Piemonte Valle d’Aosta Liguria Lombardia Veneto Fr.-Venezia Giulia Emilia-Romagna Toscana Umbria Marche Lazio Abruzzo Molise Campania Puglia Basilicata Calabria Sicilia Sardegna Groningen Friesland Drenthe Overijssel Gelderland Flevoland Utrecht Noord-Holland Zuid-Holland Zeeland Noord-Brabant Limburg Dolnoslaskie Kujawsko-Pomorskie Lubelskie Lubuskie Lodzkie Malopolskie Mazowieckie Opolskie Podkarpackie Podlaskie (continued)

206

Appendix D: List of the Regions Included in the Analysis

Table D.1 (continued) Country PL PL PL PL PL PL PT PT PT PT PT SK SK SK SK UK UK UK UK UK UK UK UK UK UK UK UK

NUTS code PL0B PL0C PL0D PL0E PL0F PL0G PT11 PT12 PT13 PT14 PT15 SK01 SK02 SK03 SK04 UKC UKD UKE UKF UKG UKH UKI UKJ UKK UKL UKM UKN

Table D.2 US MSAs Code MSA Name 40 Abilene, TX MSA 120 Albany, GA MSA 160 Albany-Schenectady-Troy, NY MSA 200 Albuquerque, NM MSA 220 Alexandria, LA MSA 240 Allentown-Bethlehem-Easton, PA MSA 280 Altoona, PA MSA 320 Amarillo, TX MSA 450 Anniston, AL MSA 460 Appleton-Oshkosh-Neenah, WI MSA 480 Asheville, NC MSA 500 Athens, GA MSA 520

Atlanta, GA MSA

Code 4080 4100 4120

Name Pomorskie Slaskie Swietokrzyskie Warminsko-Mazurskie Wielkopolskie Zachodniopomorskie Norte Centro Lisboa e V.do Tejo Alentejo Algarve Bratislavsky´ Za´padne´ Slovensko Stredne´ Slovensko Vy´chodne´ Slovensko North East North West Yorkshire and the Humber East Midlands West Midlands Eastern London South East South West Wales Scotland Northern Ireland

MSA Name Laredo, TX MSA Las Cruces, NM MSA Las Vegas, NV-AZ MSA

4150 Lawrence, KS MSA 4200 Lawton, OK MSA 4240 Lewiston-Auburn, ME MSA 4280 4320 4360 4400

Lexington, KY MSA Lima, OH MSA Lincoln, NE MSA Little Rock-North Little Rock, AR MSA

4420 Longview-Marshall, TX MSA 4472 Los Angeles-Riverside-Orange County, CA CMSA 4520 Louisville, KY-IN MSA (continued)

Appendix D: List of the Regions Included in the Analysis Table D.2 (continued) Code MSA Name 600 Augusta-Aiken, GA-SC MSA 640 Austin-San Marcos, TX MSA 680 Bakersfield, CA MSA 730 Bangor, ME MSA 740 Barnstable-Yarmouth, MA MSA 760 Baton Rouge, LA MSA 840 Beaumont-Port Arthur, TX MSA 860 Bellingham, WA MSA 870 Benton Harbor, MI MSA 880 Billings, MT MSA 920 Biloxi-Gulfport-Pascagoula, MS MSA 960 Binghamton, NY MSA 1000 Birmingham, AL MSA 1010 Bismarck, ND MSA 1020 Bloomington, IN MSA 1040 Bloomington-Normal, IL MSA 1080 Boise City, ID MSA 1122 Boston-Worcester-Lawrence, MA-NH-ME-CT CMSA 1240 Brownsville-Harlingen-San Benito, TX MSA 1260 Bryan-College Station, TX MSA 1280 Buffalo-Niagara Falls, NY MSA 1305 Burlington, VT MSA 1320 Canton-Massillon, OH MSA 1350 Casper, WY MSA 1360 Cedar Rapids, IA MSA 1400 Champaign-Urbana, IL MSA 1440 Charleston-North Charleston, SC MSA 1480 Charleston, WV MSA 1520 Charlotte-Gastonia-Rock Hill, NC-SC MSA 1540 Charlottesville, VA MSA 1560 Chattanooga, TN-GA MSA 1580 Cheyenne, WY MSA 1602 Chicago-Gary-Kenosha, IL-IN-WI CMSA 1620 Chico-Paradise, CA MSA 1642 Cincinnati-Hamilton, OH-KY-IN CMSA 1660 Clarksville-Hopkinsville, TN-KY MSA 1692 Cleveland-Akron, OH CMSA 1720 Colorado Springs, CO MSA 1740 Columbia, MO MSA

207

Code 4600 4640 4680 4720 4800 4880 4890 4900 4920 4940 4992

MSA Name Lubbock, TX MSA Lynchburg, VA MSA Macon, GA MSA Madison, WI MSA Mansfield, OH MSA McAllen-Edinburg-Mission, TX MSA Medford-Ashland, OR MSA Melbourne-Titusville-Palm Bay, FL MSA Memphis, TN-AR-MS MSA Merced, CA MSA Miami-Fort Lauderdale, FL CMSA

5082 5120 5160 5170 5200 5240 5280

Milwaukee-Racine, WI CMSA Minneapolis-St. Paul, MN-WI MSA Mobile, AL MSA Modesto, CA MSA Monroe, LA MSA Montgomery, AL MSA Muncie, IN MSA

5330 Myrtle Beach, SC MSA 5345 5360 5520 5560 5602

Naples, FL MSA Nashville, TN MSA New London-Norwich, CT-RI MSA New Orleans, LA MSA New York-Northern New Jersey-Long Island, NY-NJ-CT-PA CMSA 5720 Norfolk-Virginia Beach-Newport News, VA-NC MSA 5790 Ocala, FL MSA 5800 Odessa-Midland, TX MSA 5880 Oklahoma City, OK MSA 5920 Omaha, NE-IA MSA 5960 5990 6015 6020

Orlando, FL MSA Owensboro, KY MSA Panama City, FL MSA Parkersburg-Marietta, WV-OH MSA

6080 Pensacola, FL MSA 6120 Peoria-Pekin, IL MSA 6162 Philadelphia-Wilmington-Atlantic City, PA-NJ-DE-MD CMSA 6200 Phoenix-Mesa, AZ MSA 6240 Pine Bluff, AR MSA 6280 Pittsburgh, PA MSA (continued)

208

Appendix D: List of the Regions Included in the Analysis

Table D.2 (continued) Code MSA Name 1760 Columbia, SC MSA 1800 Columbus, GA-AL MSA 1840 Columbus, OH MSA 1880 Corpus Christi, TX MSA

Code 6320 6400 6442 6480

1900 1922 1950 1960

6520 6560 6580 6640

MSA Name Pittsfield, MA MSA Portland, ME MSA Portland-Salem, OR-WA CMSA Providence-Fall River-Warwick, RI-MA MSA Provo-Orem, UT MSA Pueblo, CO MSA Punta Gorda, FL MSA Raleigh-Durham-Chapel Hill, NC MSA

6660 6680 6690 6720 6740

Rapid City, SD MSA Reading, PA MSA Redding, CA MSA Reno, NV MSA Richland-Kennewick-Pasco, WA MSA

6760 6800 6820 6840 6880 6895 6922 6960 6980 7000 7040 7120 7160 7200

Richmond-Petersburg, VA MSA Roanoke, VA MSA Rochester, MN MSA Rochester, NY MSA Rockford, IL MSA Rocky Mount, NC MSA Sacramento-Yolo, CA CMSA Saginaw-Bay City-Midland, MI MSA St. Cloud, MN MSA St. Joseph, MO MSA St. Louis, MO-IL MSA Salinas, CA MSA Salt Lake City-Ogden, UT MSA San Angelo, TX MSA

2000 2020 2030 2040 2082 2120 2162 2180 2190 2200 2240 2290 2320 2330 2335 2340 2360 2400 2440 2520 2560 2580 2650 2655 2670 2700 2710 2720 2750 2760 2840 2880 2900 2975

Cumberland, MD-WV MSA Dallas-Fort Worth, TX CMSA Danville, VA MSA Davenport-Moline-Rock Island, IAIL MSA Dayton-Springfield, OH MSA Daytona Beach, FL MSA Decatur, AL MSA Decatur, IL MSA Denver-Boulder-Greeley, CO CMSA Des Moines, IA MSA Detroit-Ann Arbor-Flint, MI CMSA Dothan, AL MSA Dover, DE MSA Dubuque, IA MSA Duluth-Superior, MN-WI MSA Eau Claire, WI MSA El Paso, TX MSA Elkhart-Goshen, IN MSA Elmira, NY MSA Enid, OK MSA Erie, PA MSA Eugene-Springfield, OR MSA Evansville-Henderson, IN-KY MSA Fargo-Moorhead, ND-MN MSA Fayetteville, NC MSA Fayetteville-Springdale-Rogers, AR MSA Florence, AL MSA

7240 San Antonio, TX MSA 7320 San Diego, CA MSA 7362 San Francisco-Oakland-San Jose, CA CMSA

7460 San Luis Obispo-Atascadero-Paso Robles, CA MSA Florence, SC MSA 7480 Santa Barbara-Santa Maria-Lompoc, CA MSA Fort Collins-Loveland, CO MSA 7490 Santa Fe, NM MSA Fort Myers-Cape Coral, FL MSA 7510 Sarasota-Bradenton, FL MSA Fort Pierce-Port St. Lucie, FL MSA 7520 Savannah, GA MSA Fort Smith, AR-OK MSA 7560 Scranton–Wilkes-Barre–Hazleton, PA MSA Fort Walton Beach, FL MSA 7602 Seattle-Tacoma-Bremerton, WA CMSA Fort Wayne, IN MSA 7610 Sharon, PA MSA Fresno, CA MSA 7620 Sheboygan, WI MSA Gadsden, AL MSA 7640 Sherman-Denison, TX MSA Gainesville, FL MSA 7680 Shreveport-Bossier City, LA MSA Glens Falls, NY MSA 7720 Sioux City, IA-NE MSA (continued)

Appendix D: List of the Regions Included in the Analysis Table D.2 (continued) Code MSA Name 2980 Goldsboro, NC MSA 2985 Grand Forks, ND-MN MSA 3000 Grand Rapids-Muskegon-Holland, MI MSA 3040 Great Falls, MT MSA 3080 Green Bay, WI MSA 3120 Greensboro–Winston-Salem–High Point, NC MSA 3150 Greenville, NC MSA 3160 Greenville-Spartanburg-Anderson, SC MSA 3240 Harrisburg-Lebanon-Carlisle, PA MSA 3280 Hartford, CT MSA 3290 Hickory-Morganton, NC MSA 3350 Houma, LA MSA 3362 Houston-Galveston-Brazoria, TX CMSA 3400 Huntington-Ashland, WV-KY-OH MSA 3440 Huntsville, AL MSA 3480 Indianapolis, IN MSA 3500 Iowa City, IA MSA 3520 Jackson, MI MSA 3560 Jackson, MS MSA 3580 Jackson, TN MSA 3600 Jacksonville, FL MSA 3605 Jacksonville, NC MSA 3610 Jamestown, NY MSA 3620 Janesville-Beloit, WI MSA 3660 Johnson City-Kingsport-Bristol, TN-VA MSA 3680 Johnstown, PA MSA

Code 7760 7800 7840

209

MSA Name Sioux Falls, SD MSA South Bend, IN MSA Spokane, WA MSA

7880 Springfield, IL MSA 7920 Springfield, MO MSA 8000 Springfield, MA MSA 8050 State College, PA MSA 8080 Steubenville-Weirton, OH-WV MSA 8120 Stockton-Lodi, CA MSA 8140 8160 8240 8280

Sumter, SC MSA Syracuse, NY MSA Tallahassee, FL MSA Tampa-St. Petersburg-Clearwater, FL MSA

8320 Terre Haute, IN MSA 8360 8400 8440 8520 8560 8600 8640 8680 8750 8780 8800

Texarkana, TX-Texarkana, AR MSA Toledo, OH MSA Topeka, KS MSA Tucson, AZ MSA Tulsa, OK MSA Tuscaloosa, AL MSA Tyler, TX MSA Utica-Rome, NY MSA Victoria, TX MSA Visalia-Tulare-Porterville, CA MSA Waco, TX MSA

8872 Washington-Baltimore, DC-MD-VA-WV CMSA 3710 Joplin, MO MSA 8920 Waterloo-Cedar Falls, IA MSA 3720 Kalamazoo-Battle Creek, MI MSA 8940 Wausau, WI MSA 3760 Kansas City, MO-KS MSA 8960 West Palm Beach-Boca Raton, FL MSA 3810 Killeen-Temple, TX MSA 9000 Wheeling, WV-OH MSA 3840 Knoxville, TN MSA 9040 Wichita, KS MSA 3850 Kokomo, IN MSA 9080 Wichita Falls, TX MSA 3870 La Crosse, WI-MN MSA 9140 Williamsport, PA MSA 3880 Lafayette, LA MSA 9200 Wilmington, NC MSA 3920 Lafayette, IN MSA 9260 Yakima, WA MSA 3960 Lake Charles, LA MSA 9280 York, PA MSA 3980 Lakeland-Winter Haven, FL MSA 9320 Youngstown-Warren, OH MSA 4000 Lancaster, PA MSA 9340 Yuba City, CA MSA 4040 Lansing-East Lansing, MI MSA 9360 Yuma, AZ MSA

.

Appendix E: Unit Root Tests (Chap. 7)

Table E.1 EU15: Unit root tests IPS

Regional GDP per capita (Annual Growth Rate) Kms of motorways per thousand inhabitants Change in Kms of motorways per thousand inhabitants Spat.Weigh.Ave of Kms of motorways/thousand inhab. Spat.Weigh.Ave of Change in Kms of motorways per thousand inhabitants Log of GDPpc Total intramural R&D expenditure (all sectors) as % of GDP Spat.Weigh.Ave of Total R&D expenditure Social Filter Index % Employed people with Higher education, ISCED76 Levels 5–7 Log of Total GDP (Levels) Migration Rate Annual National Growth Rate

IPS-trend

17.683*** 12.595*** 13.291

1.237*

ADF

ADF-trend

PhillipsPerron

PhillipsPerron Trend

888.473***

782.099*** 1089.491***

807.405***

416.324***

623.802***

438.065***

377.252***

15.674*** 14.025*** 1145.003*** 1054.442*** 1697.867*** 1454.49***

16.138

4.132

206.563

249.137

299.115***

447.128***

9.474***

8.494***

714.773***

733.721*** 1547.743*** 1323.908***

4.081*** 11.139***

9.101*** 4.071***

38.722 260.287*

925.186*** 359.048***

50.357 187.576

263.707* 293.751***

18.341***

8.39***

263.937*

379.222***

198.743

272.432***

7.123 5.506

3.898*** 0.727

144.34 96.514

311.765*** 286.352***

115.158 115.94

328.813*** 362.169***

8.662*** 1.042 4.715***

29.039 448.617*** 519.446***

897.83*** 258.53* 385.279***

65.681 392.791*** 734.582***

266.386* 269.98* 522.976*

2.716*** 2.606*** 7.393***

*Significant at 10%; ** significant at 5%; *** significant at 1%

211

212

Appendix E: Unit Root Tests (Chap. 7)

Table E.2 EU25: Unit root tests IPS Regional GDP per capita (Annual Growth Rate) Kms of motorways per thousand inhabitants Change in Kms of motorways per thousand inhabitants Spat.Weigh.Ave of Kms of motorways/ thousand inhab. Spat.Weigh.Ave of Change in Kms of motorways per thousand inhabitants Log of GDPpc Total intramural R&D expenditure (all sectors) as % of GDP Spat.Weigh.Ave of Total R&D expenditure Social Filter Index % Employed people with Higher education, ISCED76 Levels 5–7 Log of Total GDP (Levels) Migration Rate Annual National Growth Rate

10.192***

10.319

IPS-trend 6.75***

4.524V

ADF 749.3579***

1043.642***

ADF-trend

PhillipsPerron

832.0173*** 1429.499***

1032.489***

676.3293***

PhillipsPerron-Trend 1138.108***

440.2294***

15.594

10.331***

859.8299***

913.1505*** 1385.309***

0.9

2.842***

550.951***

845.0921***

10.863***

9.132***

975.392***

887.1509*** 1372.871***

2.505*** 0.995

4.561*** 0.131

491.6397*** 552.1283***

714.3495*** 648.7965***

615.6541***

677.3552*** 1262.95V

771.9879***

2.667***

1.037

699.6489***

355.1378* 809.5998V

1062.309***

608.3887***

1157.184***

293.5088 532.2063***

3.395 0.999

2.854*** 3.196***

271.1387 274.5315

520.4754*** 462.6828***

228.2082 338.92

458.8831*** 549.1543***

5.3***

5.143***

455.2618***

780.0859***

349.1037

330.5161

1.772 3.76

474.7355*** 460.9955*** 470.7466 1143.498

497.2357*** 951.6454

394.0539 688.7744

0.781 10.666

*Significant at 10%; ** significant at 5%; *** significant at 1% IPS – Im-Pesaran-Shin test for unit roots; the W[t-bar] test statistic is standard-normally distributed under the null hypothesis of non-stationarity ADF – Augmented Dickey-Fuller Test; combines N independent unit root tests under the null hypothesis of non-stationarity of all series Phillips-Perron – Combines N independent unit root tests under the null hypothesis of nonstationarity of all series

Appendix F: Spatial Autocorrelation Test for the Regression Residuals (Chap. 7)

The Moran’s I test does not detect spatial autocorrelation in the residuals of all regressions included in this book: the combination of “national” variables and spatially lagged explanatory variables are able to capture a significant part of the total spatial variability of the data. In all Chapters spatial autocorrelation has been discussed together with the results of the Moran’s I test. As an additional check we include a sample of the Moran’s I Scatter Plots computed for all equations and in the panel data analyses (for each year t): all figures have not been included in the book as they would take up too much space. The weight matrix for the computation of the Moran’s I is based on the same weighting scheme adopted for the calculation of the spatially lagged variables included in the model (spillovers and social filter conditions of neighbouring regions). In addition to this weighting scheme (based on distance), first order contiguity has been also tested delivering similar results.

213

214

Appendix F: Spatial Autocorrelation Test for the Regression Residuals (Chap. 7)

Table F.1 EU-15: Spatial autocorrelation, Moran’s I for the residuals (Eq. 8, Table 7.1a) Moran’I = 0.0644

2

W_UFE1997

1

0

-1

-2

-3

-2

-1

0

1

2

3

UFE1997 Moran’I = 0.0548

W_UFE1999

2

0

-2

-4

-2

0

UFE1999

2

4

Appendix F: Spatial Autocorrelation Test for the Regression Residuals (Chap. 7)

215

Table F.2 EU-25: Spatial autocorrelation, Moran’s I for the residuals (Eq. 8, Table 7.1b) Moran’I = 0.0260

4

W_UFE1999

2

0

-2

-4

-6

-4

-2

0

2

4

6

2

4

6

UFE1999 Moran’I = 0.0138

4

W_UFF2002

2

0

-2

-4

-6

-4

-2

0

UFF2002