Road Pricing, the Economy and the Environment (Advances in Spatial Science) 3540771492, 9783540771494

Citation preview

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

Titles in the Series

M. M. Fischer, J. Revilla Diez and F. Snickars Metropolitan Innovation Systems VIII, 270 pages. 2001. ISBN 978-3-540-41967-9 L. Lundqvist and L.-G. Mattsson (Eds.) National Transport Models VIII, 202 pages. 2002. ISBN 978-3-540-42426-0 J. R. Cuadrado-Roura and M. Parellada (Eds.) Regional Convergence in the European Union VIII, 368 pages. 2002. ISBN 978-3-540-43242-5 G. J. D. Hewings, M. Sonis and D. Boyce (Eds.) Trade, Networks and Hierarchies XI, 467 pages. 2002. ISBN 978-3-540-43087-2 G. Atalik and M. M. Fischer (Eds.) Regional Development Reconsidered X, 220 pages. 2002. ISBN 978-3-540-43610-2 Z. J. Acs, H. L. F. de Groot and P. Nijkamp (Eds.) The Emergence of the Knowledge Economy VII, 388 pages. 2002. ISBN 978-3-540-43722-2 R. J. Stimson, R. R. Stough and B. H. Roberts Regional Economic Development X, 397 pages. 2002. ISBN 978-3-540-43731-4 S. Geertman and J. Stillwell (Eds.) Planning Support Systems in Practice XII, 578 pages. 2003. ISBN 978-3-540-43719-2 B. Fingleton (Ed.) European Regional Growth VIII, 435 pages. 2003. ISBN 978-3-540-00366-3 T. Puu Mathematical Location and Land Use Theory, 2nd Edition X, 362 pages. 2003. ISBN 978-3-540-00931-3 J. Bröcker, D. Dohse and R. Soltwedel (Eds.) Innovation Clusters and Interregional Competition VIII, 409 pages. 2003. ISBN 978-3-540-00999-3 D. A. Griffith Spatial Autocorrelation and Spatial Filtering XIV, 247 pages. 2003. ISBN 978-3-540-00932-0

Y. Okuyama and S. E. Chang (Eds.) Modeling Spatial and Economic Impacts of Disasters X, 323 pages. 2004. ISBN 978-3-540-21449-6 L. Anselin, R.J.G.M. Florax and S. J. Rey Advances in Spatial Econometrics XXII, 513 pages. 2004. ISBN 978-3-540-43729-1 R.J.G.M. Florax and D. A. Plane (Eds.) Fifty Years of Regional Science VIII, 400 pages. 2004. ISBN 978-3-540-22361-0 D. Felsenstein and B.A. Portnov (Eds.) Regional Disparities in Small Countries VI, 333 pages. 2005. ISBN 978-3-540-24303-8 A. Reggiani and L.A. Schintler (Eds.) Methods and Models in Transport and Telecommunications XIII, 364 pages. 2005. ISBN 978-3-540-25859-9 H.W. Richardson and C.-H.C. Bae (Eds.) Globalization and Urban Development X, 321 pages. 2005. ISBN 978-3-540-22362-7 G. Arbia Spatial Econometrics XVII, 207 pages. 2006. ISBN 978-3-540-32304-4 B. Johansson, C. Karlsson, R. Stough (Eds.) The Emerging Digital Economy X, 352 pages. 2006. ISBN 978-3-540-34487-2 H. Westlund Social Capital in the Knowledge Economy X, 212 pages. 2006. ISBN 978-3-540-35364-5 A.E. Anderssson, L. Pettersson, U. Strömquist (Eds.) European Metropolitan Housing Markets VI, 363 pages. 2007. ISBN 978-3-540-69891-3 A.T. Murray, T.H. Grubesic (Eds.) Critical Infrastructure VIII, 311 pages. 2007. ISBN 978-3-540-68055-0

J. R. Roy Spatial Interaction Modelling X, 239 pages. 2004. ISBN 978-3-540-20528-9

R. Cooper, K. Donaghy, G. Hewings (Eds.) Globalization and Regional Economic Modeling XIII, 475 pages, 2007. ISBN 978-3-540-72443-8

M. Beuthe, V. Himanen, A. Reggiani and L. Zamparini (Eds.) Transport Developments and Innovations in an Evolving World XIV, 346 pages. 2004. ISBN 978-3-540-00961-0

R. Capello, R. Camagni, B. Chizzolini, U. Fratesi Modelling Regional Scenarios for the Enlarged Europe: European Competiveness and Global Strategies XX, 337 pages, 2008. ISBN 978-3-540-74736-9

Chris Jensen-Butler † Birgitte Sloth · Morten Marott Larsen Bjarne Madsen · Otto Anker Nielsen Editors

Road Pricing, the Economy and the Environment

123

Professor Chris Jensen-Butler †

Professor Birgitte Sloth Department of Business and Economics University of Southern Denmark Campusvej 55 DK-5230 Odense Denmark [email protected]

Bjarne Madsen Centre for Regional and Tourism Research Stenbrudsvej 55 DK-3730 Nexø, Bornholm Denmark [email protected]

Morten Marott Larsen Danish Institute of Governmental Research Nyropsgade 37 DK-1602 Copenhagen V. Denmark [email protected]

Professor Otto Anker Nielsen Centre for Traffic and Transport Technical University of Denmark Building 115, Ground Floor, Office 010 Bygningstorvet DK-2800 Lyngby Denmark [email protected]

ISBN 978-3-540-77149-4

e-ISBN 978-3-540-77150-0

DOI 10.1007/978-3-540-77150-0 Advances in Spatial Science ISSN 1430-9602 Library of Congress Control Number: 2007942884 © 2008 Springer-Verlag Berlin Heidelberg 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. Production: LE-TEX Jelonek, Schmidt & Vöckler GbR, Leipzig Cover design: WMX Design GmbH, Heidelberg Printed on acid-free paper 987654321 springer.com

In Memory of Professor Chris Jensen-Butler

A main editor of this book was Professor Chris Jensen-Butler, School of Economics and Finance at the University of St. Andrews, Scotland. In March 2006 – just before finalising the editorial work of the book – it was discovered that Chris had cancer and sadly died on 17 May 2006. Chris was born on the 5 June 1945 in Derby, England. He was the son of Florence and Bill Butler. Chris attended the University of Durham, where he graduated in 1968 with a PhD in Economic Geography. In 1969, he moved to Denmark and was appointed Lecturer and later Associate Professor in the Department of Geography, University of Aarhus. In the late 1980s, he was the driving force behind the creation of the Institute of Mathematical Planning at the University of Aarhus. Chris returned to the UK in 1995 and was appointed Professor in the Department of Geography at the University of St. Andrews in Scotland. Subsequently, he moved to the Department of Economics and in 1998 was appointed Head of School. As an avid traveller in the early part of his career, Chris firmly believed that “a rolling stone gathers no moss”. In 2000, Chris returned to Denmark for a brief period to take up a Professorship at the University of Copenhagen, Department of Geography. He returned to St. Andrews a year later as Departmental Chairman in the Department of Economics. In 2004, Chris was appointed Head of School in the newlyconstituted School of Economics and Finance. He continued in this role until his death, leading the School to excellence in research and teaching. Throughout his active professional life, Chris had special relationships with Denmark and Portugal. In addition to his association with researchers in Denmark, he also had close research collaboration with the Universities of Lisbon and Aveiro. Chris participated actively in the Centre for Transport Research on Environmental Impacts and Policy (TRIP). In particular, he was involved in research projects on the modelling of regional economic impacts of road pricing and on teleworking and transport. Editorial work for this book was part of his association with the centre. In many ways, the book reflects Chris’ life, his approach to life, the way he worked and the kind and noble way he interacted with colleagues and friends. On the one hand, Chris had his own research interests and his contributions to the book demonstrate important new insights. First, Chris had special interest in the research and modelling of regional development, particularly the interaction between the regional economy and transport activities. One of his final contribu-

vi

Friends and colleagues

tions to this field of research is Chapter 12 of the book, co-authored with Jacob Kronback, Steen Leleur and Bjarne Madsen, on a systems approach to modelling regional economic effects of road pricing. The model system is very general and represents his ambition to understand and describe the regional and local economy, where distance plays an important role. Specifically, the article presents a looselycoupled two-sector model, for transport and agriculture, where a basic regional model is combined with regional sector models for the two sectors. Vertically, the model system includes international, national as well as regional and local levels. The article presents an application of the regional impacts of road pricing in Denmark using the model system. Second, the book includes an article, jointly written with Lasse Møller Jensen and Arne Hole Riis, studying the relationships between teleworking and transport. The article demonstrates Chris’ vivid interest in the role of new technology and its importance for regional development, especially the potential for interaction with others. In this case, the potential for home working while remaining professionally engaged is examined both from the perspective of peripheral areas and with respect to transport activity. The article is a good example of Chris’ research in this field. Chris acknowledged the potential of new technologies, but at the same time he was critical and did not believe in easy solutions. Although teleworking seems to have enormous potential, the impacts are in fact shown to be very limited. Even though teleworking has a role in peripheral areas, the positive impacts seem to be concentrated in urban areas. Chris was eager to find the true potential of new technology and not just headlines for a newspaper. On the other hand, this book also reflects Chris’ skills as an advisor, a mentor, a colleague and a friend, inspiring them to improve the results of their research activities. Not only as a researcher, but also as a native-speaking colleague, he made a major contribution to this book, giving good advice and corrections to the manuscript. As a researcher, friend and mentor, Chris will be remembered and greatly missed.

Friends and colleagues

Table of Contents

In Memory of Professor Chris Jensen-Butler ...................................................v 1.

Introduction ..................................................................................................1 Birgitte Sloth

2.

Road Pricing in Europe – A Review of Research and Practice ................5 Esko Niskanen and Chris Nash 2.1 Introduction.............................................................................................5 2.2 The Current Situation..............................................................................6 2.2.1 Background .................................................................................6 2.2.2 Current Pricing Policies at National Level..................................7 2.2.3 EU Pricing Policy........................................................................8 2.2.4 The Need for Pricing Policy Reform...........................................9 2.3 Key Issues, Concepts and Approaches..................................................10 2.3.1 The Goals of Road Pricing ........................................................10 2.3.2 Theoretical Approaches to Road Pricing...................................11 2.3.3 Practical Approaches and Instruments ......................................12 2.4 Road Pricing in Practice in Urban Transport ........................................14 2.4.1 Current Schemes in Urban Transport Pricing............................14 2.4.2 Barriers on Urban Road Pricing ................................................15 2.4.3 The Next Steps in Urban Road Pricing .....................................18 2.5 Road Pricing in Practice in Interurban Transport..................................19 2.5.1 Current Schemes in Interurban Transport .................................19 2.5.2 Barriers in Interurban Road Pricing ..........................................21 2.5.3 The Next Steps in Interurban Road Pricing...............................22 2.6 Conclusions...........................................................................................22 2.6.1 Lessons from Research..............................................................22 2.6.2 Lessons from Practical Experience: What Next? ......................24

3.

Road Pricing: Consequences for Traffic, Congestion and Location......29 Lars-Göran Mattsson 3.1 Introduction...........................................................................................29 3.2 Application of a Zone-Based Road Pricing System to Stockholm........31 3.2.1 Scenarios ...................................................................................31 3.2.2 Modelling Approach .................................................................32 3.2.3 Transport Effects .......................................................................33 3.3 Application of a Distance-Based Road Pricing System to Stockholm ..37

viii

Table of Contents

3.3.1 Scenarios ...................................................................................37 3.3.2 Modelling Approach .................................................................38 3.3.3 Transport Effects .......................................................................38 3.4 Application of Optimal Congestion Pricing to a Generic City..............40 3.4.1 Modelling Approach .................................................................40 3.4.2 Scenarios ...................................................................................42 3.4.3 Transport and Location Effects .................................................43 3.5 Discussion and Conclusions..................................................................45

4.

Implementation Paths for Marginal Cost-Based Pricing in Urban Transport: Theoretical Considerations and Case Study Results ...........49 Erik T. Verhoef, C. Robin Lindsey, Esko Niskanen, André de Palma, Paavo Moilanen, Stef Proost and Arild Vold 4.1 Introduction...........................................................................................50 4.2 Implementation Paths: Motivation and Theory.....................................50 4.2.1 Barriers......................................................................................51 4.2.2 Constraints on Pricing ...............................................................53 4.2.3 Correspondence Between Barriers and Constraints ..................54 4.3 Formulating Implementation Paths .......................................................56 4.4 Description of the MC-ICAM Urban Case Studies...............................58 4.4.1 The Paris Case Study.................................................................58 4.4.2 The Brussels Case Study ...........................................................60 4.4.3 The Helsinki Case Study ...........................................................61 4.4.4 The Greater Oslo Case Study ....................................................61 4.4.5 Assessment of the Case Studies ................................................61 4.5 Implementation Paths in the Case Studies ............................................62 4.5.1 Paris...........................................................................................65 4.5.2 Brussels .....................................................................................65 4.5.3 Helsinki .....................................................................................66 4.5.4 Greater Oslo ..............................................................................67 4.5.5 Role of Barriers and Constraints in the Case Study Implementation Paths................................................................68 4.6 Case Study Findings..............................................................................68 4.6.1 Main Results..............................................................................68 4.6.2 Some Specific Issues.................................................................73 4.7 Concluding Remarks.............................................................................75 4.7.1 Summary ...................................................................................75 4.7.2 Theoretical vs. Practical Approaches to the IPs ........................76 4.7.3 Questions of Priority and Timing..............................................76 4.7.4 Other Limitations of the Case Studies.......................................77

5.

The London Congestion Charging Scheme: The Evidence ....................79 John Peirson and Roger Vickerman 5.1 Introduction...........................................................................................79 5.2 Background to the LCCS ......................................................................79

Table of Contents

5.3 5.4 5.5 5.6 5.7 5.8 5.9

The Working of the LCCS ....................................................................82 Theory of Area Pricing..........................................................................83 Evidence on LCCS................................................................................84 Interpretation of the Evidence ...............................................................85 Extension of LCCS Area.......................................................................87 Relevance to Other Urban Road Pricing Schemes ................................87 Conclusions...........................................................................................88

6.

The AKTA Road Pricing Experiment in Copenhagen.............................93 Otto Anker Nielsen and Majken Vildrik Sørensen 6.1 Introduction...........................................................................................94 6.2 Design of the Experiment......................................................................94 6.2.1 Survey Setup .............................................................................95 6.2.2 Pricing Schemes ........................................................................96 6.3 Practical Issues ......................................................................................97 6.3.1 Problems with the GPS Technology .........................................98 6.3.2 Adding a Third Round...............................................................99 6.4 General Results ...................................................................................100 6.4.1 Socioeconomic Variables for AKTA Participants...................100 6.4.2 Experience and Attitudes Towards Road Pricing....................101 6.4.3 The General Population Survey of Attitudes Towards Road Pricing............................................................................103 6.5 Behavioural Changes and AKTA: The Main Results .........................105 6.6 Discussion and Conclusions................................................................107 6.6.1 GPS Technology .....................................................................107 6.6.2 The Main Experimental Design ..............................................107 6.6.3 Changes in Behaviour and Evaluation of Different Pricing Schemes ......................................................................108 6.6.4 Attitudes Towards Road Pricing .............................................108

7.

Experience with Measuring Equity and Efficiency: A Case from Oslo......................................................................................111 Farideh Ramjerdi, Knut Østmoe and Harald Minken 7.1 Introduction.........................................................................................111 7.2 Equity Measures..................................................................................113 7.2.1 Properties of Equity Measures ................................................114 7.2.2 Some Inequality Measures ......................................................115 7.3 Evaluation of Alternative Packages of Instruments for Oslo ..............117 7.4 A Sensitivity Analysis of MCF ...........................................................122 7.5 An Evaluation of the Equity Implications of an ”Optimal” Package ...............................................................................................123 7.6 Some Conclusions...............................................................................129

8.

Transport Costs in a Multiregional Equilibrium Job Search Model...133 Morten Marott Larsen, Ninette Pilegaard and Jos van Ommeren 8.1 Introduction.........................................................................................133

ix

x

Table of Contents

8.2 The Basic Model .................................................................................136 8.2.1 The Matching Model...............................................................136 8.2.2 Job Destruction........................................................................137 8.2.3 Equilibrium Employment and Unemployment........................137 8.2.4 Job Creation ............................................................................138 8.2.5 Workers...................................................................................138 8.2.6 The Spatial Wage Equation.....................................................138 8.2.7 Reservation Commuting Costs................................................139 8.3 Simulations with the Fundamental Model...........................................140 8.4 Theoretical Extensions ........................................................................141 8.4.1 Road Pricing and Regional Distributional Effects ..................144 8.4.2 The Case of Zealand................................................................154 8.5 Conclusion ..........................................................................................164 9.

Evaluation of the Introduction of Road Pricing Using a Computable General Equilibrium Model ...........................................167 Knud J. Munk 9.1 Introduction.........................................................................................167 9.2 The Theoretical Model........................................................................168 9.3 Application of the Theory of Public Economics to the Taxation of Transport and Investment in Transport Infrastructure ....................170 9.3.1 Taking into Account that Leisure Travel is Complementary to Leisure.................................................................................171 9.3.2 Taking into Account that Transport is Associated with Externalities.............................................................................172 9.3.3 Taking into Account that Leisure Travel is Predominately Consumed by Households with a Relatively High Income .....172 9.3.4 Reasons for Taxing Transport Higher than Other Goods........173 9.3.5 The Optimal Size of the Transport Infrastructure ...................173 9.3.6 The Effect of the Introduction of Road Pricing on the Optimal Taxation of Transport and the Optimal Provision of Transport Infrastructure ......................................................173 9.3.7 The Introduction of Road Pricing will be Associated with a Double Dividend ..................................................................174 9.4 The Parameterised Model ...................................................................175 9.4.1 Specification of Functional Forms for Free Road Capacity and Environmental Externalities .............................................175 9.4.2 Specification of Household Preferences..................................176 9.4.3 Real Income and Social Welfare .............................................177 9.5 The Specification of the Parameterised Model ...................................178 9.6 Presentation and Interpretation of Simulation Results ........................180 9.6.1 Consequence Analysis.............................................................180 9.6.2 Project Evaluation ...................................................................183 9.6.3 Optimality Analysis.................................................................184 9.7 Conclusion ..........................................................................................188

Table of Contents

10. Efficiency and Equity Considerations in Road Pricing.........................193 Harald Minken and Farideh Ramjerdi 10.1 Introduction ......................................................................................193 10.2 Aspects of Equity..............................................................................195 10.3 Indicators of Equity ..........................................................................197 10.4 Equity, Efficiency and Modelling.....................................................201 10.5 Optimal Road Pricing Subject to Equity Constraints........................203 10.6 Solving the Optimisation Problem....................................................204 10.7 Conclusions ......................................................................................204 11. Modelling the Economy, Transport and Environment Triangle, with an Application to Dutch Maglev Projects ......................................207 Jan Oosterhaven and J. Paul Elhorst 11.1 Introduction ......................................................................................207 11.2 The Different Nature of the Three Types of Interaction ...................208 11.3 Modelling Environmental Impacts ...................................................210 11.4 Modelling Transport-Economy Interactions .....................................211 11.5 An Application of Dutch Maglev Proposals .....................................214 11.5.1 Introduction ........................................................................214 11.5.2 Modelling the Interaction Between the Economy and the Transport System..........................................................217 11.5.3 A Welfare Evaluation of External Effects ..........................221 11.6 Conclusion ........................................................................................223 12. A Systems Approach to Modelling the Regional Economic Effects of Road Pricing .........................................................................................229 Bjarne Madsen, Chris Jensen-Butler, Jacob Kronbak and Steen Leleur 12.1 Introduction ......................................................................................229 12.2 A Systems Approach to Regional and Sub-Regional Economic Modelling..........................................................................................232 12.2.1 Data and Accounting Principles .........................................232 12.2.2 Modelling Principles ..........................................................232 12.2.3 Optimal Model Structure....................................................235 12.2.4 A Loosely Coupled Model for Transport and Agriculture .235 12.3 The Linking Procedures....................................................................237 12.3.1 Linking the Models Within the Transport Sector ...............237 12.3.2 Linking the Transport and Regional Economic Models .....240 12.4 LINE: the Full Model, a Graphical Presentation ..............................242 12.4.1 The Dimensions of LINE ...................................................245 12.5 Road Pricing and Modelling its Impacts...........................................247 12.5.1 Road Pricing .......................................................................247 12.6 Results from the Danish Road Pricing Toll Study ............................247 12.6.1 Changes in Transport Costs................................................248 12.6.2 Changes in Commodity Prices and Disposable Incomes....251 12.6.3 Changes in Demand, Production and Income.....................253 12.6.4 Changes in Employment and Income .................................253

xi

xii

Table of Contents

12.6.5

Recycling the Revenue from Road Pricing: a Balanced Budget .............................................................256 12.7 Limitations of the Model and Future Development Strategies .........259 12.8 Conclusion ........................................................................................260 13. External Effects and Road Charging......................................................267 Jeppe Rich and Otto Anker Nielsen 13.1 Introduction ......................................................................................267 13.2 External Effects: Background ...........................................................268 13.3 Demand Effect from Road Charging ................................................270 13.3.1 Charging and Demand: Experience from London, Stockholm and Copenhagen ...............................................270 13.4 Internalisation of External Effects Through Road Charging ............272 13.4.1 Road Safety ........................................................................272 13.4.2 Traffic Noise.......................................................................273 13.4.3 Emission .............................................................................274 13.5 Summary and Conclusions ...............................................................274 14. Assessing the Impacts of Traffic Air Pollution on Human Exposure and Health.................................................................................................277 Ole Hertel, Steen Solvang Jensen, Martin Hvidberg, Matthias Ketzel, Ruwim Berkowicz, Finn Palmgren, Peter Wåhlin, Marianne Glasius, Steffen Loft, Peter Vinzents, Ole Raaschou-Nielsen, Mette Sørensen and Helle Bak 14.1 Introduction ......................................................................................277 14.2 Air Pollution in Urban Areas ............................................................278 14.2.1 Urban Background and Street Pollution .............................279 14.3 Air Pollution Exposure Assessment..................................................280 14.4 Direct Methods for Exposure Assessment ........................................280 14.5 Indirect Methods for Exposure Assessment......................................281 14.5.1 Monitoring Networks .........................................................281 14.5.2 Application of Models for Exposure Assessment...............282 14.5.3 Modelling Long Range Transport of Pollution...................282 14.5.4 Modelling Pollution in the Urban Background...................284 14.5.5 Modelling Urban Street Pollution.......................................285 14.5.6 Modelling Nitrogen Oxide Chemistry in Street and Urban Background..............................................................285 14.5.7 Determination of Emissions and Emission Factors ............286 14.6 Particle Pollution ..............................................................................286 14.6.1 Source of Traffic Particles and Modelling..........................287 14.6.2 Other Particle Sources ........................................................289 14.7 Examples of Exposure Assessment in Danish Studies .....................290 14.7.1 Exposure of Bus Drivers and Postmen in Copenhagen ......290 14.7.2 Exposure of Danish Children to Traffic Air Pollution........292 14.7.3 Personal Monitoring of Air Pollution Exposure in Copenhagen ........................................................................292

Table of Contents

xiii

14.7.4

Using GIS in Street Pollution Modelling for Exposure Assessment .........................................................................293 14.8 Assessment of Health Effects of Air Pollution Exposure .................296 14.9 Perspectives ......................................................................................297 15. Car Use Habits: An Obstacle to the Use of Public Transportation?....301 Berit Møller and John Thøgersen 15.1 Introduction ......................................................................................301 15.2 Method..............................................................................................304 15.3 Data...................................................................................................304 15.4 Results ..............................................................................................307 15.5 Summary and Implications ...............................................................308 16. Road Pricing in Denmark – User Attitudes and User Reactions..........315 Mai-Britt Herslund 16.1 Introduction ......................................................................................315 16.2 User Studies of the FORTRIN project..............................................316 16.2.1 Pricing Strategy ..................................................................316 16.2.2 Test Population...................................................................317 16.2.3 Focus-Group Meetings .......................................................317 16.2.4 The Study by Questionnaires..............................................319 16.2.5 Trip Logs ............................................................................320 16.2.6 Conclusions of the FORTRIN Study..................................320 16.3 AKTA – the Road Pricing Study of Copenhagen .............................321 16.3.1 Design of the Study ............................................................321 16.3.2 Different User Studies ........................................................321 16.3.3 Results from Telephone Interviews ....................................322 16.3.4 Pricing Schemes .................................................................324 16.3.5 Focus Groups......................................................................326 16.3.6 Key Results from Questionnaires .......................................327 16.4 Overall Conclusions of the Two User Studies ..................................329 17. A Cost-Minimisation Principle of Adaptation of Private Car Use in Response to Road Pricing Schemes.........................................................331 Peter Loukopoulos, Tommy Gärling, Cecilia Jakobsson and Satoshi Fujii 17.1 Introduction ......................................................................................331 17.2 Market-Based Travel Demand Management (TDM) Measures .......332 17.3 Classification of Travel Demand Management (TDM) Measures....333 17.3.1 Coerciveness.......................................................................334 17.3.2 Tow-Down vs. Bottom-Up Processes.................................335 17.3.3 Time Scale..........................................................................335 17.3.4 Spatial Scale .......................................................................336 17.3.5 Market-Based vs. Regulatory Mechanisms ........................336 17.3.6 Impacting Latent vs. Manifest Travel Demand ..................337 17.4 Theoretical Framework.....................................................................338 17.5 Implications for the Effectiveness of TDM Measures ......................342

xiv

Table of Contents

17.6 Summary and Discussion..................................................................343 18. Car Users’ Trade-Offs Between Time, Trip Length, Cost and Road Pricing in Behavioural Models ................................................................351 Otto Anker Nielsen and Goran Vuk 18.1 Introduction ......................................................................................352 18.2 SP Design .........................................................................................352 18.3 Awareness of Travel Distance and Travel Time...............................354 18.4 SP Model Estimation ........................................................................357 18.4.1 Model Formulation (Utility Functions) ..............................357 18.4.2 Income Effect Models ........................................................361 18.4.3 Alternative Model Formulations ........................................362 18.5 The RP Route Choice Model ............................................................362 18.5.1 Utility Function ..................................................................363 18.5.2 Estimation...........................................................................363 18.5.3 Within Person Variation .....................................................364 18.5.4 Between Person Variation ..................................................365 18.5.5 Estimation of the Impact of Road Pricing on Traffic .........367 18.5.6 Income Effect .....................................................................369 18.5.7 Other Explanatory Variables ..............................................370 18.6 Comparison Between SP and RP Models .........................................371 18.7 Summary and Conclusions ...............................................................372 19. The Impacts of e-Work and e-Commerce on Transport, the Environment and the Economy...............................................................375 Andy Lake 19.1 Being Active Without Moving .........................................................375 19.2 The UK Study ...................................................................................376 19.3 E-Work .............................................................................................377 19.4 E-Business and e-Commerce ............................................................380 19.5 Rebounds ..........................................................................................383 19.6 Environment .....................................................................................385 19.7 Economic Growth, Transport Growth and European Policy ............387 19.8 Future Research Directions...............................................................388 19.9 Conclusion ........................................................................................389 20. A Web-Based Study of the Propensity to Telework Based on SocioEconomic, Work Organisation and Spatial Factors..............................395 Lasse Møller-Jensen, Chris Jensen-Butler, Bjarne Madsen, Jeremy Millard and Lars Schmidt 20.1 Introduction ......................................................................................395 20.2 Teleworking: Theoretical and Methodological Issues ......................396 20.2.1 Defining Teleworking ........................................................396 20.2.2 The Extent of Teleworking.................................................397 20.2.3 Teleworking and Savings in Transport Effort ....................397 20.2.4 Modelling the Take-Up of Teleworking.............................398

Table of Contents

20.3 20.4 20.5 20.6 20.7

xv

20.2.5 Technology Issues ..............................................................399 20.2.6 Long-Term Effects .............................................................399 The Determinants of Teleworking and Transport Substitution Effects...............................................................................................400 The Consequences for Travel of Working at Home .........................400 Determinants of Teleworking ...........................................................401 20.5.1 Multivariate Relations ........................................................404 Long-Term Effects of Teleworking ..................................................406 Conclusion ........................................................................................408

21. The Impact of Telecommuting on Households’ Travel Behaviour, Expenditures and Emissions....................................................................411 Andrea F. Glogger, Thomas W. Zängler and Georg Karg 21.1 Introduction ......................................................................................411 21.2 Model................................................................................................414 21.3 Method..............................................................................................416 21.4 Results ..............................................................................................417 21.4.1 Sample Characteristics .......................................................417 21.4.2 Travel Behaviour ................................................................417 21.4.3 Expenditures.......................................................................419 21.4.4 Emissions............................................................................420 21.5 Discussion and Conclusion...............................................................420

1

Introduction

Birgitte Sloth Department of Business and Economics, University of Southern Denmark

This volume is a collection of research papers that examine the possibility of reducing the negative impacts from traffic by using new information and communication technologies. The focus is on passenger transport and most papers study various aspects of road pricing as an instrument to decrease traffic. The volume was initiated by the Centre for Transport Research on environmental and health Impacts and Policy (TRIP), which was a “virtual” research centre financed by the Danish Environmental Research Programme. In TRIP, researchers from diverse institutions and fields collaborated on projects to increase the understanding of the role of transport in society and of the complex interplay among transport behaviour of individuals, the functioning of transport systems and effects on the environment and health. As it turned out that many projects in TRIP dealt with the impact of information and communication technologies and particularly road pricing, we organised a conference in 2003 for TRIP researchers and other researchers interested in these topics. Most of the papers in this volume were presented and discussed in some preliminary form at the conference. A few papers have been added later. Transportation demand and thereby traffic have been steadily increasing for a long time. This is a consequence of the general economic growth, which increases the demand for most commodities and services including transport. The rise in transportation demand is further sparked by globalisation, with increased international communication, trading and travelling and agglomeration, where specialisation of firms and labour implies that firms benefit from being located close to each other, such that cities and commuting districts become larger. Traffic explains a large part of our energy consumption and thereby contributes to CO2 emissions. It also has a direct negative impact on the environment and health in a variety of ways, most notably in the form of air pollution, accidents and noise. And traffic creates the congestion that seems to be an unavoidable characteristic of large cities and has negative impacts on public welfare since it increases travel times and costs. Objectives regarding sustainability and environmental improvements therefore naturally lead to traffic management objectives, which have proved difficult to fulfil. Traffic occurs because there is a need for transportation; hence it cannot be eliminated or restricted without cost. Various instruments can be used to regulate the demand for transport and thereby to regulate traffic itself: direct and indirect taxes on cars and petrol, prices and supply of public transport, traffic regulations and parking restrictions, incentives to facilitate the relocation of households and campaigns to influence the attitude and awareness of motorists about public transport. Developments within

2

B. Sloth

information and communication technology have improved the possibilities of using various forms of road pricing, where motorists pay for driving in specific areas or roads, possibly also depending on the time of day. This has been applied in some European cities and is under consideration elsewhere. Different instruments affect the different types of traffic differently and thus have different environmental effects and different consequences for the travellers. For example, in contrast to general taxes, road pricing differentiated according to specific trips has the potential of making the individual traveller pay for the negative external effects of his specific trip, such that resources are used efficiently. When the price of a particular trip reflects its cost, the external effect (externality) is taken into account by the traveller (internalised). Thus, road pricing is particularly relevant when considering external costs that vary with location or the time of day such as congestion. If motorists are to pay for the negative external effects they impose on others, motorists in congested cities in morning rush hours must pay more than other motorists. For local air pollution, for example, particulate pollution, and for noise, the costs that motorists impose on others naturally depend on how many people are exposed to the effect and therefore such costs are usually higher in dense city areas. Despite the technological development and the efforts of the EU to get road pricing implemented, it is still not widely used. In Chapter 2, Chris Nash and Esko Niskanen provide a comprehensive review of road pricing experiences and argue that the lack of political acceptability has been a major obstacle to wide implementation of road pricing. As an example of this, the survey results reported by Mai-Britt Herslund in Chapter 16 indicate that drivers are particularly adverse to the targeted use of road pricing, e.g. the additional rush hour charges and the additional charges in large cities that should be its main advantage. In Chapter 4, Eric Verhoef and his co-authors discuss further the obstacles and barriers to the introduction of road pricing, proposing that these obstacles are gradually removed such that road pricing can be introduced along a graduate path with different prices and pricing rules at various stages. In Norway, road pricing has been used for a very long period and has been politically feasible and even necessary because it has been seen as a means of financing transport infrastructure and not as an instrument to make the motorist pay for the costs imposed on others. In Chapter 7, Farideh Ramjerdi, Knut Østmoe and Harald Minken report the Norwegian experience and also recent developments in the use of road pricing to address transport externalities. To facilitate decision-making and thereby the possibility of extending the domain of road pricing, more research-based knowledge about the positive and negative consequences of road pricing and more ex ante studies are important. Several papers in this volume contribute to this. Lars Göran Mattsson in Chapter 3 presents different types of ex ante studies of road pricing using Stockholm as a case. He discusses how such studies can be combined to help decision-making and also how they can be improved as more evidence becomes available. In Chapter 5, John Peirson and Roger Vickerman report the first evidence from the road pricing introduced in Central London in February 2003. They find that road pricing has been very effective in reducing traffic and discuss if similar effects can be expected

Introduction

3

in larger areas or smaller cities. In Chapter 6, Otto Anker Nielsen and Majken Vildrik Sørensen report similar large effects found in an experimental introduction of different versions of road pricing in Copenhagen, which they compare to the behavioural effects predicted by models and surveys. To evaluate the effects of increased travel costs, one must study the interaction between the economy and the transport system and take into account other imperfections in the economy than the externalities in transport. In Chapter 8, Morten Marott Larsen, Ninette Pilegaard and Jos van Ommeren study the interaction of transport markets and a labour market in which there is unemployment because of search frictions and demonstrate how to quantify the detrimental effects of road pricing on employment and production costs. Distortionary taxation, in particular labour taxes, is an important instrument in redistributive policy. Thus, road pricing will have (re-)distributive effects, both directly and indirectly through the possibility of using the revenue to influence the income distribution. Knud J. Munk demonstrates in Chapter 9 that this implies that there is a possibility of obtaining a so-called double dividend, which should be taken into account when considering road pricing. In Chapter 10, Harald Minken and Farideh Ramjerdi emphasize the importance of distributional impacts and equity consideration for the appraisal of road pricing schemes and suggest a method to design the road pricing scheme, which simultaneously includes efficiency and distributional issues. The introduction of road pricing may also influence regional distribution through different impacts on costs and competitiveness in different regions. In Chapter 12, Bjarne Madsen, Chris Jensen-Butler, Jacob Kronbak and Steen Laleur demonstrate how such effects can be investigated in a model system which combines a transport model with a regional economic model. Including the interaction between the economy and the transport system when evaluating public policies and investments is not only relevant for road pricing, but also for traditional project evaluation. In Chapter 11, Jan Oosterhaven and J. Paul Elhorst apply a model structure with sectors, household types, transport modes and spatial zones to obtain evaluations of proposals for magnetic levitation rails in the Netherlands, taking into account indirect economic effects including labour migration and housing migration. As explained, road pricing is not only effective in reducing congestion, but is also suited to targeting other local external effects. In Chapter 13, Jeppe Rich and Otto Anker Nielsen compare road pricing experiences from different cities, considering not only the impact on travel demand but also on the local external effects: road safety, noise and air pollution. In principle, road pricing can be differentiated according to vehicle characteristics, specific route choices and time of the day, and thereby be targeted to ensure that travel costs reflect such local external effects. In order to do that, precise information on the size of external costs is required. The costs of local air pollution are particularly difficult to assess. To determine the exact associations between pollution exposure and health effects, it is necessary to be able to assess human exposure and therefore to be able to model both air pollution concentrations in different streets at different times and the time spent by individuals at different

4

B. Sloth

locations. In Chapter 14, Ole Hertel and co-authors describe a modelling structure that facilitates this and has been applied in a number of epidemiological studies to add to our knowledge of the health effects and thus costs of local air pollution. Transportation models and the economic models with which they are combined usually assume that choices of trips, modes and routes are made by rational, cost-minimising agents, such that the traffic reduction obtained from road pricing is explained by travellers basing their decisions on total travel costs including the value of travel time. Values of travel time are often estimated by stated preference experiments, where subjects are asked for their preferences regarding different time-cost combinations. In Chapter 18, Otto Anker Nielsen and Goran Vuk compare the findings from such a study to actual choices made in a road pricing experiment. They find that the model structures obtained correspond well, but their findings indicate that the value of travel time increases when road pricing is introduced, which may reflect that travellers have restrictions and inertia in their choices. One possible interpretation is that drivers use less rational and conscious reasoning, e.g. that choosing the car is merely a habit and thus not directly affected by prices. Peter Laokopoulos, Tommy Gärling, Cecilia Jakobsson and Satoshi Fujii argue in Chapter 17 that evaluation of the effectiveness of road pricing should be based on realistic behavioural assumptions allowing for e.g. habit formation. This is particularly important when comparing road pricing to alternative regulatory instruments affecting drivers differently, e.g. marketing programmes. The habitual choice of the private car is also studied in Chapter 15 by Berit Møller and John Thøgersen, who use a survey to demonstrate that car use habits act as an obstacle to the intentions to commute by public transportation. This indicates that road pricing may be less effective than expected as compared to instruments that influence the habits directly. Information and communication technology influence transport and the environment, not only through the feasibility of road pricing, but also through teleworking and e-commerce substituting trips for commuting and shopping. This may imply that people may substitute trips by other means of communication, such that road pricing does not just change travellers’ choice of route, mode or time, but may also lead to fewer trips. This has the potential of decoupling economic growth and transport growth, but much more experience and research is required before any firm conclusions can be reached. In Chapter 19, Andy Lake provides an overview of recent research on the impacts of telework and e-commerce on transport and the environment. The volume’s last two chapters, Chapter 20 by Lasse Møller Jensen, Chris Jensen-Butler, Bjarne Madsen, Jeremy Millard and Lars Schmidt and Chapter 21 by Andrea F. Glogger, Thomas W. Zängler and Georg Karg, both report survey results that clearly indicate that teleworking does reduce the use of the private car and the distance travelled and thus may be able to contribute to traffic reductions for the benefit of the environment.

2

Road Pricing in Europe – A Review of Research and Practice

Esko Niskanen1 and Chris Nash2 1 2

STA Research, Helsinki, Finland Institute for Transport Studies, University of Leeds, England

Abstract Despite the considerable benefits of road pricing that economists and other analysts have shown, and despite the efforts of the EU to get road pricing implemented, it has not been implemented on a broad scale. It is still widely considered to be a radical and controversial policy. This chapter provides a comprehensive review of experiences with road pricing in Europe and elsewhere. We consider pricing issues in urban road and interurban road transport separately. The policies in the two areas have typically been different focusing in urban transport on the private car and in interurban on freight transport. Also existing analyses, both theoretical and conceptual and those using real-world simulation models, have mostly dealt with these two topics separately. For both areas we consider the current pricing schemes, barriers to pricing and likely next steps. A central theme arising from this chapter is the importance of institutional implementation issues, including political acceptability. These questions have been given too little attention both in research and in actual attempts to implement road-pricing measures.

2.1 Introduction Road pricing as a policy to alleviate congestion, environmental and other problems related to urban and interurban road transportation has received much attention during the last decade. However, despite the considerable benefits of road pricing that economists and other analysts have shown, and despite the efforts of the EU to get road pricing implemented, it is not implemented on a broad scale. It is still widely considered to be a radical and controversial policy. This is due to many factors, including the fact that the public and politicians do not seem to be convinced of the benefits (either to society or to themselves). On the other hand, there are long-lasting disputes also among academics, especially around the application of the concept of marginal cost pricing. This chapter provides a comprehensive review of experiences of road pricing in Europe and elsewhere, both in research and practice. The main objective is to provide an overall picture of the recent developments in road pricing, whilst, at the same time, highlighting the most important issues and hopefully opening up new perspectives for research and policy. We consider pricing issues in relation to

6

E. Niskanen and C. Nash

urban road and interurban road transport separately. There are many reasons for a separate treatment. The policies in the two areas have typically been different, focusing in urban transport on the private car and in interurban on freight transport and they have been implemented by different levels of government. Also, existing analyses – both theoretical and conceptual and those using real-world simulation models – have mostly dealt with these two topics separately. The chapter starts (section 2.2) by briefly characterising the current situation and expected trends in the road transport industry and by reviewing the current policies and policy trends, especially at the EU level. Section 2.3 considers key issues of road pricing at a theoretical level, including the goals of road pricing and different approaches to road pricing. We proceed by considering the experience of road pricing in practice, in section 2.4 in urban transport and in section 2.5 in interurban road transport. In both cases, we review existing pricing schemes, identify the most important barriers and constraints to the implementation of road pricing and finally describe the foreseeable and most likely next steps. We conclude in section 2.6 by summarising the overall lessons from existing research and practice and present our view on what should happen next. This chapter is heavily based on the work and experiences in EU-funded Projects AFFORD, PETS, UNITE, MC-ICAM, CAPRI and IMPRINT-EUROPE. This means that, although we have aimed to provide a comprehensive review, our discussion necessarily is focused on the issues and perspectives addressed in these projects. It therefore seeks to draw general lessons for national and European policy, rather than providing detailed case studies. Results from these studies are reported in the relevant project reports as referred to in appropriate parts of this chapter. In addition, see for example a special issue of Transport Policy (de Palma et al., 2006), volumes Acceptability of Transport Pricing Strategies (Schade and Schlag, 2003) and La tarification des transports, Pourquoi? Pour Qui? (de Palma and Quinet, 2005).

2.2 The Current Situation 2.2.1 Background A long-standing cause of concern within the EU has been the continued growth of road (and air) transport, relative to rail, which has had a declining market share. According to CEC (2001), the market shares in freight transport in 2000 in the EU were the following: road 44%, short sea shipping 41%, rail 8%, inland waterways 4% and pipeline 3%. The corresponding numbers for passenger transport were: car 79%, air 5%, rail 6%, bus and coach 8.5% and tram and metro 1.5%. The share of road in freight transport had risen from 35% in 1970 and the share of car in passenger transport from 74%. The trend in favour of the road mode continues. The results have been growing problems of road congestion, road accidents and environmental damage. Such trends are even more pronounced in the accession countries, where under the former regime road transport was heavily

Road Pricing in Europe – A Review of Research and Practice

7

constrained. In its latest Transport Policy White Paper (CEC, 2001), the Commission has adopted the aim of modal shift from road to rail, initially aiming to return the rail share of the market to its 1998 level by 2010. Appropriate pricing and particularly the introduction of external costs into the charges, is seen as an important policy tool in achieving this. However, at the same time other policy developments in the European Union have tended to favour road transport. Perhaps the most important and visible has been the deregulation process that the main transport sectors and modes have gone through since the mid-1980s. Due to both the inter-modal differences and the lack of a common European approach, the degree of deregulation and opening up of the markets for competition varies across transport modes. The road transport mode has undergone extensive liberalisation (along with the air mode), whilst rail transport is still largely in the hands of national companies and is severely hampered in competition for international traffic. The most remarkable process of deregulation in the road freight sector was the liberalisation in 1998 of cabotage within the European Union for commercial road freight transport as well as for companies’ own transport. Currently there are great differences between modes and between different countries in the form and level of charges for the use of infrastructure. Comparing total costs and total charges (or average costs and charges) suggests that simply pricing to cover total social cost would certainly not typically raise charges for road transport as compared to those of rail and, thus, would not favour rail and would not facilitate the required modal shift. However, if the basis of prices is related to the costs imposed by an increase in traffic on each mode, the marginal social costs, there are important differences in relation to conclusions based on average costs. Whilst there may be significant differences between marginal and average values for all categories of cost, the crucial difference relates to congestion. Whilst all road users experience average congestion costs and thus do not need to be charged for it, the marginal cost of congestion imposed by extra traffic may be very much higher than the average. A marginal cost pricing philosophy puts congestion costs at the heart of road pricing policy. The overall picture of road pricing, as it is often cited including CEC (2001) – though great variations are possible – is: urban transport is underpriced, interurban car overpriced and interurban freight transport undercharged. A review conducted as part of the IMPRINT-EUROPE project found the evidence broadly to support this assertion (Nash and Matthews, 2001), although with wide differences between countries.

2.2.2 Current Pricing Policies at National Level During the last 10 years, a lot of effort has been put into attempts to reform pricing of European transport, with the aim being able to better reflect the user and polluter pay principles which, it is commonly believed, could promote more efficient use of the existing infrastructure and also could facilitate and encourage new forms of partnership with the private sector, also seen by many to be of vital importance.

8

E. Niskanen and C. Nash

However, the experience is that implementation of the suggested pricing principles and even agreeing on them can be extremely difficult compared to many other policies such as deregulation. Currently, in urban and interurban road transport in different European countries, road users are broadly speaking charged for their use of their vehicles and the road network by means of a three-part tariff comprising a vehicle purchase tax, an annual licence fee and a tax on fuel. Only in a few cases is this system complemented by direct road user charges – i.e. road tolls, area licences and similar systems. Less than 10% of road infrastructure costs are recovered by direct user charges (CEC, 1998, p. 12). Generally speaking, heavy reliance is placed on fixed charges as compared to variable road user charges as well as on fuel taxes. As a result, although the fixed fees often vary by vehicle type, there is a general lack of integration of external and infrastructure costs into the price of road transport. In particular, the current charges fail to differentiate adequately in time and space to allow for differences in congestion, environmental and other impacts. They fail to reflect adequately, even in the variable element of the charge, the differential impacts of different vehicle types. In short, the user and polluter pay principles are not applied; instead, users usually pay similar amounts irrespective of the infrastructure damage, congestion and pollution they cause. On the other hand, the overall picture is not uniform, as there are very different structures of transport-related taxation systems across member states and other European countries. There is also considerable national and local variation in the integration of external and infrastructure costs into the price of transport. The great inter-country variation is a combination of historical reasons and different economic and social development, but it also reflects the lack of a common European approach to these issues.

2.2.3 EU Pricing Policy The EU has a major influence on road pricing policies in Europe. Therefore, it is important to understand the trends and goals of EU policy. The current activity on transport pricing started with a Green Paper (CEC, 1995), which argued strongly for the inclusion of external costs in transport prices. The following White Paper (CEC, 1998) put forward a phased approach towards marginal cost pricing but still acknowledged total cost coverage (and fixed charges and subsidies) as an important goal and approach to be considered in certain situations. The most recent White Paper (CEC, 2001) argued for the need for efficient pricing, but avoided using the term 'marginal cost pricing'. Though this may partly be a matter of terminology, clearly the new White Paper adopted a more reserved approach to the idea of marginal cost pricing. For urban transport pricing, an important factor in EU policy is the subsidiarity principle. This means that the responsibility for implementing policies lies at the lowest appropriate administrative or governmental level. However, the EU has been active indirectly, by encouraging and supporting cities which are interested in road pricing. The EUROCITIES network and the PROGRESS project are such

Road Pricing in Europe – A Review of Research and Practice

9

examples. The White Paper (CEC, 2001) explains that this also means helping to develop and disseminate best practice throughout Europe. There is also a proposal to modify the regulation on public service obligations in order to open the market for the provision of urban public transport, either on a deregulated or a competitively tendered basis. For interurban road transport, the EU has taken a more active role, which makes sense as the transport and the problems often are international and crossborder by nature. Moreover, even for domestic traffic, the issue of competitive conditions between vehicles of different nationalities is important. Hence the EU's starting point is to distinguish between commercial transport and private vehicles. The focus is on commercial transport and, in the short term, on interurban freight in particular. Moreover, as stated above, the EU considers that on average road freight transport is underpriced, whilst private car transport is not. The EU also considers that acceptability barriers are lower in freight transport. Thus the EU's efforts have been dispersed, focusing on individual issues and cases. Currently there still exists no common European strategy, with comparable charging practices and inter-operable equipment.

2.2.4 The Need for Pricing Policy Reform In summary, we quote the White Paper (CEC, 2001) which states: "[Currently] while transport may be heavily taxed, it is above all badly and unequally taxed. Users are all treated alike, irrespective of the infrastructure damage, bottlenecks and pollution they cause. This failure to spread the burden fairly between infrastructure operators, taxpayers and users causes considerable distortion of competition both between transport operators and between modes of transport. For the modes to enjoy a level playing field, taxation should work according to the same principle regardless of mode and ensure a fairer distribution of the burden of transport costs." (pp 72-73) The White Paper emphasises that it is not the overall level of taxes and charges that needs to change significantly, but rather their structure needs to be altered radically: “The thrust of Community action should be gradually to replace existing transport system taxes with more effective instruments for integrating infrastructure costs and external costs [in the price of transport]. These instruments are, firstly, charging for infrastructure use, which is a particularly effective means of managing congestion and reducing other environmental impacts, and, secondly, fuel tax, which leads itself well to controlling carbon dioxide emissions. The introduction of these two instruments, which will allow greater differentiation and modulation of taxes and right of use, needs to be coordinated, with the first being backed up by the second.” (p 75)

10

E. Niskanen and C. Nash

Finally, the White Paper promised “a framework directive in 2002 to establish the principles of infrastructure charging and a pricing structure for all modes of transport. The proposal, which will leave each Member State wide scope in terms of implementation, will include a common methodology for setting price levels which will incorporate external costs…” ( p 78) However, this directive has never appeared. This may (at least partly) reflect the fact there remains substantial opposition to road pricing in different member countries. Individual member countries rarely make explicit reference to marginal social cost in the determination of pricing structures and levels; on the contrary, in some countries explicit policy statements arguing against it are made.

2.3 Key Issues, Concepts and Approaches 2.3.1 The Goals of Road Pricing When judging the different approaches to road pricing, it is a good idea to start with the question ‘Why road pricing?’ This question is important because clearly not all people or groups are convinced, either of the overall benefits of road pricing to society or of its favourable impacts on themselves. There are three explicit or implicit goals of road pricing, typically referred to: – to collect revenue  to reduce traffic and nuisance (externalities such as congestion, environmental damage, noise, etc.)  to promote efficiency. The interpretation of the first two goals is quite clear. Putting all kinds of possible measurement problems aside, these goals are concrete and easy for people and policymakers to understand. In particular, the revenues collected can be earmarked (hypothecated) or can contribute to general taxation. The goal of traffic reduction can also be related to the broader goal of promoting modal shift. However, the third goal, efficiency, is more problematic, and leaves room for interpretations. A change is said to be economically efficient when it is possible for the gainers to compensate the losers and still be better off themselves. In a first best world, where all such changes are made, a move to marginal cost pricing will be economically efficient. Where constraints prevent this situation from being achieved throughout the economy, pure marginal cost pricing ceases to be the ideal, but in general prices remain based on marginal cost and various second-best rules may be derived to identify the optimal adjustments from marginal cost pricing. Following the adoption of the term in the projects AFFORD and MC-

Road Pricing in Europe – A Review of Research and Practice

11

ICAM, this approach is now often referred to as marginal cost-based pricing.1 To economists, this approach to pricing is generally seen as the main goal of pricing reform, but the problem with using economic efficiency as a major policy goal is that in many cases it simply is far too general and abstract for the general public (including the media) and the politicians to be used as a relevant criterion. Equity concerns are also an important issue, whilst not a primary policy goal in the context of road pricing. Equity is not always considered when designing policies in market economies; on the contrary, it is typically emphasised that there are specific separate policies that will address such issues whether considered as goals or constraints. However, transportation, and transport pricing in particular, appears to be in this respect different from pricing in many other industries. The claims that equity issues should be reflected and included in the prices are repeated frequently. One reason for why this is the case may be that people have not been used to seeing road transport as a commodity subject to allocation through the market in the way that most commodities are.

2.3.2 Theoretical Approaches to Road Pricing During the last couple of decades, economists and other scientists have presented different theoretical approaches to road pricing. Over time, these approaches and frameworks have become standard reference also in practical policy discussions. Although policymakers typically are very practical, these theoretical approaches and traditions and adopted ways of thinking necessarily also affect their decisionmaking, in the background at least (see for example Faulhaber and Baumol, 1988). It is useful in any particular policy application to try to reveal these often hidden attitudes explicitly and transparently and to clarify their influence. This may partly help us to understand why different cities and countries may have adopted different approaches and solutions (to seemingly or theoretically similar problems). This is also important in order to be able to design optimal policy reforms under different institutional and cultural settings. Two long-standing issues which are the sources of both academic and policy disputes can be identified: 1. Should prices be based on short-run or long-run marginal costs? 2. Should the primary goal of pricing be internalisation of external costs (marginal cost pricing) or full cost recovery?

1

This term aims to reflect and highlight the following two aspects of the pricing system. First, the implementation of marginal cost pricing must not be considered in isolation but as part of a package of measures; this is a preferred approach because it is more effective under second-best conditions and because it is more acceptable (can incorporate compensatory measures). Second, the actual prices implemented may not strictly taken equal marginal costs but are typically second-best prices with optimal deviations from (direct) marginal costs to take account of constraints that prevent the achievement of the full theoretically optimal position.

12

E. Niskanen and C. Nash

Economic theory shows how short-run marginal cost pricing results in optimal use of existing capacity and is an appropriate and optimal policy for the internalisation of externalities or external costs (congestion, environmental, etc). Short-run marginal cost pricing by definition secures short-term efficiency. Where capacity is optimal, short run and long run marginal cost pricing are equivalent. It is therefore only where capacity is non-optimal that the issue arises. Furthermore, in standard models, short-run marginal cost pricing, when combined with the practice of social cost-benefit analysis to determine investment decisions, also leads to long-run efficiency. For instance, Vickrey (1969) argued long ago that short-run marginal cost pricing provides a reliable indicator of the marginal benefits from capacity expansion. However, not everybody is convinced and the situation may not be quite as simple in real life situations. The relevant issues here apply in particular when the pricing and investment decisions are decentralised or the relevant transport organisations are state or private enterprises operating on a commercial basis. For instance, according to one view, short-run marginal cost pricing gives the infrastructure manager an incentive to restrict investment in order to maintain high prices and therefore does not provide correct incentives for investment. Therefore, according to this argument, long-run marginal cost pricing should be used instead, as it will give an incentive to invest only if the operator can recoup the costs of the increased capacity in prices. Another much addressed issue in relation to short-run marginal cost pricing is to what extent it will be able to generate enough revenue to cover costs. According to a classical result by Harwitz and Mohring, the degree to which infrastructure costs are covered under short-run marginal cost pricing depends on returns to scale in investment and returns to scale in the level of usage. Some empirical evidence on returns to scale for European roads (also for rail and air travel) is presented in Quinet (1997). A different approach to or philosophy of road transport pricing is based on full cost recovery (by mode, etc.) as the target of road pricing (rather than general welfare maximisation). This principle is advocated by several authors and commentators and also emphasised in many countries as a basis for policymaking. This is a particularly relevant issue in regard to public transport and interurban road transport (and also in other modes like rail), where efficiency prices can be far from covering total costs. These issues have been addressed in detail in the UNITE project which also made a survey of approaches prevailing to road transport pricing in different countries in Europe. The results showed clear differences by country (Quinet, 2001).

2.3.3 Practical Approaches and Instruments The pricing instruments relevant for road pricing include: link based tolls, areabased charges, cordon tolls, electronic kilometre-based charges, vehicle taxation, fuel taxation and parking pricing. Public transport fares are also relevant. De Palma

Road Pricing in Europe – A Review of Research and Practice

13

et al. (2004) provide an overview of these instruments and their theoretical properties. The following text is based heavily on their discussion. Most applications (and studies) are based on point charging where a charge is levied to pass a point on a road, a form which is similar to conventional toll systems. A charge on a single road is likely to encourage traffic to divert to avoid the toll. Most point charging systems therefore involve cordon charging (or toll rings) where a series of charging points are established at all entries to a given area (often a city centre). We can distinguish between cordons that charge only inbound traffic, those that charge outbound traffic and those that charge in both directions. A variant of cordon charging is area charging (or area licensing) in which the charge is levied to use a vehicle within a defined area, rather than just to enter it. This will also include vehicle journeys wholly within the cordon, which are unaffected by cordon charging and might as a result increase if traffic flows entering from outside are reduced. Implementing tolls that vary over time in synchrony with fluctuations in congestion provides an inducement for people to travel at off-peak times. Various theoretical studies (see Small, 1992; Arnott et al., 1993; Chu, 1999) have shown that the efficiency gains from altering trip-timing decisions can dominate the gains from altering decisions on numbers of trips, route choice and other dimensions of travel behaviour. Furthermore, the efficiency gains can be expected to increase with the number of time steps incorporated into the toll schedule. Both cordon charging and area charging introduce boundary problems. Through traffic may reroute around the cordon and may increase congestion on the periphery. Those living just outside the cordon will pay to travel to the centre; those just inside will not. Drivers making long journeys across the cordon pay the same as those making short journeys. These discontinuities, which are both inefficient and inequitable, can be overcome by continuous charging systems, which charge for all travel within a defined area (such as a city). These can be based on distance travelled (distance-based charging) or time spent in delay (delaybased charging). Distance-based charging is particularly relevant for pricing interurban roads. A delay-based charging system was tested in the 80s in Cambridge (in the U.K.) involving charges for any 500m length journey which took longer than three minutes. However, subsequent research has tended to reject delay-based approaches as likely to lead to safety problems by giving a strong incentive for speeding and jumping red lights. It has been estimated in several studies (e.g. Schrank and Lomax, 1999; 2004) that a large fraction of time lost due to congestion is attributable to unpredictable fluctuations in capacity or demand on account of traffic accidents, bad weather, roadworks and strikes. Therefore, real-time or responsive pricing in contrast to time-dependent pricing according to a predictable schedule is important to consider. In order to track the externality costs that a driver actually imposes on a given trip (the current and envisaged systems charge drivers just before they have to make a decision), it is necessary to set tolls dynamically on the basis of the travel conditions actually experienced. In an ideal system, this would entail the collection of information, calculation of the appropriate tolls and dissemination of the updated information and tolls to drivers via on-board computers and in real-

14

E. Niskanen and C. Nash

time according to current or predicted travel conditions. Though infeasible in the past, this is becoming practicable through the use of Advanced Traveller Information Systems (ATIS). Responsive pricing presents some uncertainty to the driver regarding the actual price he/she has to pay. Of course this matters though it can be reduced by information provision. An important question here is: what is more important to a driver, uncertainty about monetary payment or uncertainty about travel time (time costs)? There is not much evidence on the value drivers place on reliability as opposed to journey time itself, though some advances have been made in recent years (see Small et al., 1995; Bates et al., 2001; Lam and Small, 2001; Brownstone et al., 2003; Bonsall et al., 2004). On various measurements of reliability, see Lomax et al.. (2003).

2.4 Road Pricing in Practice in Urban Transport

2.4.1 Current Schemes in Urban Transport Pricing Urban road charging schemes can cover the road network as a whole or can only be restricted to an individual road or on a certain stretch of a single road. Examples of the latter types of applications are in the U.S. and Canada. On the State Route 91 in Riverside County in California and on Highway 407 north of Toronto in Canada, the price system is time-dependent pricing according to a predictable schedule. On Interstate 15 north of San Diego in California, the system is real-time or responsive pricing. The two applications in California assume tolled and non-tolled lanes in parallel; the system is reported to work well in maintaining free-flow conditions on the tolled lanes. De Palma et al. (2004) provide detailed descriptions. This section focuses on ‘true’ urban road charging systems, covering the urban road network more broadly than just individual roads. Currently, such systems are in force in Singapore, Norway, Rome and London. The longest standing urban road-pricing scheme is that of Singapore which originally introduced an area licensing scheme in 1975 and replaced it in 1999 by an electronic road pricing scheme, followed by the Norwegian cities during years 1986-2001 (Bergen, Oslo, Trondheim, Kristiansand and Stavanger). Rome and London are much more recent cases; Rome has gradually introduced road tolling and London introduced a broadscale system in 2003. See Ramjerdi et al. (2004) on the Norwegian applications. Ricci and Fagiani (2003) and the references therein provide detailed description of the other applications. The aim of road pricing in Singapore is to alleviate huge congestion and pollution problems. By contrast, the Norwegian cities cannot be said to be amongst the most congested or polluted in Europe. Here, the original goal was to collect revenue to carry out specified investments which would directly benefit the drivers in question (and, as in Oslo, originally for a limited period only). However, more recently there has been a gradual shift to emphasise other goals. Legislation to

Road Pricing in Europe – A Review of Research and Practice

15

allow the use of toll rings for congestion management has been passed and there is now a debate whether to switch from revenue generation to congestion charging with time-of-day toll variation. The other two European cities with road pricing currently introduced, Rome and London, are known to suffer from severe congestion and environmental problems. In Rome, there has for long been attempts to introduce various measures to restrict access to the historical city centre by private cars, of which the introduction of pricing is just the latest measure. However, the pricing system here, which comprises charges for permits to access the central area, is still quite limited. More recently, and clearly the most important example in Europe, of course, is now London with its introduction of urban road pricing in central London on 17 February 2003. The London scheme is a simple area-licensing scheme, covering the area of central London. All road users within the cordon area on weekdays between 7 am and 6:30 pm are required to obtain a licence at a daily cost of £5 (€7.5); certain categories of vehicles and users are exempt and there is a 90% discount for residents in the area. Also, plans exist to extend the charge area and to raise the toll to ǧ8 in summer 2005. Checking is done through video cameras and by automatic number plate recognition technology that recognise whether or not each passing vehicle has obtained a licence. The principal objective is to alleviate congestion in the central city area, but also to raise revenue for other elements of the transport system, including public transport improvements. Initial impressions are that the charging system has been very successful, achieving a 20% reduction in traffic in the area with none of the major problems that were predicted.

2.4.2 Barriers on Urban Road Pricing Why is marginal cost-based road pricing not being implemented more generally in urban areas, despite arguments in favour of it by many economists and considerable welfare benefits demonstrated by various modelling research projects? Why do the public and politicians or the business community not want it? Why is road pricing introduced in the cities mentioned above rather than in some other cities? These and other related issues have been considered in a range of EU-funded research projects including AFFORD, IMPRINT-EUROPE and MC-ICAM. The discussion in this section, whilst aiming to summarise some of the results of these studies, largely builds on Niskanen et al. (2003) and Niskanen and Nash (2004). Evidently there are strong barriers to implementing pricing policies. Niskanen et al. (2003) made use of the following classification of potentially relevant barriers:  technological and practical barriers  legal and institutional barriers  acceptability-related barriers. According to the above-mentioned and other studies, the principal technological barrier to road pricing in an urban context is that there is no widely tested technology to permit fully differentiated pricing. Also, the existing technology may

16

E. Niskanen and C. Nash

be too expensive to justify its implementation on a broad scale. However, technological reasons alone cannot explain why there are so few working examples of road pricing on a smaller scale. The term ‘practical’ barrier in the above classification refers to issues such as: 1. quality and insufficiency of data on marginal and other costs as well as of the welfare benefits and other potential impacts of road pricing (land use impacts, impacts on business, etc); 2. complexity of transport networks (spatially) and related to this, geography of the city/country and the interaction between urban and interurban traffic; and 3. insufficient standardisation and harmonisation of technological solutions or pricing principles across different countries. The first two types of issue may generate important barriers to urban road pricing. (The third type is more relevant in the interurban road context, as discussed in Section 2.5.2 below.) The main legal barriers are the lack of powers in many countries to introduce road pricing and restrictions on the degree of prices that would be legally permissible. For instance in France urban road pricing is restricted to new infrastructure. In Sweden, road tolls are considered as taxes which must be decided by the Parliament and which by law cannot be earmarked to transport infrastructure investments (although the government has promised this in relation to the Stockholm road pricing plan, see section 2.4.3 below). In some countries, civil liberties legislation prevents certain forms of road pricing which require tracking of the location of individual vehicles. Major institutional barriers (beyond the legal ones) are related to the lack of co-ordination between responsible authorities or actors. This may apply to different parts of government at the same level (for example, those responsible for roads and those for public transport, or neighbouring communities or countries competing with each other for business and tax payers rather than co-operating). Second, co-ordination between different levels of government (for example, national interests versus local) may be insufficient. On a European scale, an important aspect of the problem here is the subsidiarity principle that restricts the development of pricing policies at the European level. (The principle of course may have some other virtues which the ‘negative’ commenting here in no way wants to deny.) And third, the lack of coordination between government and non-government institutions may also cause problems, particularly with regard to parking and public transport when they are privately owned or run so that the local public authority has only limited powers to control charges. However important, technological and practical barriers or legal and institutional barriers in themselves can evidently be no real reason for road pricing to be so rare in urban contexts. The technology either in principle exists or could be developed rather easily. Similarly, data problems in a modern world should not be insurmountable. And finally, new and more appropriate laws and institutions could in many cases be passed and built already in the short or medium term if desired. Instead, the key issue and barrier clearly is low acceptability – public, business and political acceptability. Public acceptability of course is a key factor in formulating and implementing policies in the western democracies. Public acceptability may be influenced by factors like the public dislike of new or higher charges as

Road Pricing in Europe – A Review of Research and Practice

17

well as complex charging methods and technology. But public acceptability of road pricing evidently is (negatively) affected also by the following fundamental asymmetry relating to the distribution of benefits versus that of costs: many of the benefits are indirect and small in per capita terms, whilst the costs are direct and relatively large in per capita terms. This means that those who would gain do not care much whereas those who would lose tend to be more active, forming interest groups. (Though intuitively plausible, however, no attempts are made to our knowledge to test this hypothesis empirically.) Another aspect of public acceptability is that it is generally lower for urban transport than for interurban road transport. In urban transport, the people are the key users and they use the car for their regular daily and often inflexible routines. Political acceptability is important because it is the politicians who ultimately take the necessary decisions, although their decisions are naturally affected by the attitudes of the general public (also the media) and the business community. Politicians (and other decision-makers) may naturally fear major mistakes, particularly if there are sunk costs or consequences such as driving away footloose industries that cannot be undone. But there is also a more personal aspect in the way they function: they typically worry about their careers; even those politicians who perceive that pricing could help to alleviate the problems, may fear the consequences of new or higher charges in the next election. The problem from their point of view is that (this is related to the asymmetry of costs and benefits above) many of the welfare benefits would be indirect and they might not be able to receive credit for them, whilst they have to take the blame for the costs and nuisances which typically are more direct. This, in particular, applies to local politicians in the cities where the pricing would be implemented and especially if many of the benefits would go elsewhere (also depending on earmarking arrangements) and not directly to their own voters. Therefore, as argued by Frey (2003) and many others, politicians do not generally benefit from pursuing road pricing. A final important observation is that the different types of barrier to pricing that we have identified may be strongly interrelated; for instance, which is the true or real barrier and which (at least) seemingly independent and relevant barrier is only a reflection of the other one, may not always be obvious. In the light of all these considerations, we might ask why road pricing has been adopted in Singapore and in the Norwegian cities, and why in Rome and in London? What are the key lessons to be learned from these cities? Or is this list of cities a pure coincidence? The fact that Singapore is a state comprising a single city has evidently made the implementation of road pricing easier there than in many other big cities which have suffered from equally acute problems. Clearly (at least) two aspects which complicate implementation in big European cities and which were referred to above, are missing here: 1. no need to consider interaction between urban and interurban traffic; and 2. no possibility of contradicting interests between the city level and the national government level. As for the Norwegian cities, Ramjerdi et al. (2004) mention two considerations that evidently have been affected here. First, the country has a long history (dates back to the 1930s) of toll financing of specific road improvements such as

18

E. Niskanen and C. Nash

tunnels and bridges. A second apparent factor has been the skill of the planners and politicians in packaging the pricing schemes for each toll ring with well-defined investment schemes for road infrastructure improvements in such a way that the benefits became apparent soon after the rings were introduced. The London case in turn demonstrates how much political will can matter. The political will was personified in Mayor Livingstone and was expressed during his election campaign. As a result, the road pricing system was implemented in a situation where national government was not supporting it (in this form) and the main political parties and much of the media were hostile (at the time of his initial election, Livingstone was not the candidate of any political party).

2.4.3 The Next Steps in Urban Road Pricing For urban road pricing in the short term, pricing schemes involving simple charges structures and technologies, and avoiding major changes in overall price levels, seem the most likely to be acceptable. The most likely systems are area-based charges and cordon tolls using simple electronic technology (such as in London), introduced only for the most congested central areas and on weekdays. In the medium term (3-10 years), it may be possible (for some cities) to move towards more sophisticated systems, as has happened in Singapore. Distance-based charges using GPS-based electronic road pricing (ERP) can replace area-based charges. Pricing may still be restricted to central areas and weekdays for technological, practical and acceptability reasons. In the long term, all charges may ultimately be fully differentiated according to true marginal social costs, as a result of adopting GPS-based ERP on all urban links and at all times. Also, the charging system could be integrated in the sense of covering both roads, parking charges and smart card ticketing for public transport. In London, the re-election of Mayor Livingstone has ensured the survival of the London scheme for the time being. However, at the moment, no other cities appear to be quickly following suit. This can be said although, in many European cities, different kinds of demonstration projects or experiments have been carried out or are underway. One such demonstration has been Project PROGRESS. The cities participating in the project are Bristol, Copenhagen, Edinburgh, Genoa, Gothenburg, Helsinki, Rome and Trondheim. Some of these cities (Rome, Trondheim) already have road-charging systems and some others (Edinburgh, Bristol) have fairly well-advanced plans for the introduction of road pricing. An interesting question is whether these demonstration projects and experiments have to any degree increased the likelihood of adopting true road pricing policies in those cities in near future. Continued success in London could evidently encourage these and other European cities to follow suit and with a gradual spread of the acceptability of road pricing through European cities as a whole. However, the defeat of road pricing proposals in the recent referendum in Edinburgh makes it unlikely that other British cities will implement road pricing in the near future. On the other hand, both the London and the Norwegian experiences suggest that opposition to road

Road Pricing in Europe – A Review of Research and Practice

19

pricing will be much less vocal after it has actually been implemented (for a related theoretical argument, see Glazer, 2003). Perhaps the most interesting European city considering urban road pricing, besides London, currently is Stockholm. In Stockholm, a full-scale trial with congestion taxes will be performed between 3 January and 31 July 2006. A planned referendum in September 2006 will then decide on whether the toll system should be made permanent or abolished. Because of a court process related to the procurement of the system, the trial has been delayed and it will be shorter than originally planned. The congestion tax scheme in Stockholm is based on a cordon toll charged with an automatic payment system using short-range microwave conveyance. The tolled area covers the whole central business district. Tolls will be charged from inbound and outbound vehicles during morning and afternoon peak hours (SEK20 or €2.18, one-way, maximum SEK80 or €8.73 per day; the toll is stepwise ascending and descending in the beginning and end of the toll periods; taxis and vehicles using non-fossil fuels are exempted). The national government decided on April 7 2006 to go on with the trial.2

2.5 Road Pricing in Practice in Interurban Transport 2.5.1 Current Schemes in Interurban Transport In interurban road transport, in contrast to urban transport discussed above, the question of the appropriate roles of the EU and the national governments assumes a particular importance. Another difference is the relative importance and treatment of freight transport (as compared to passenger transport and private car). The White Paper (CEC, 2001) indicates that in interurban road pricing, the need for change is the greatest in freight transport. The overall picture is quite mixed as there have been policy developments both at the EU level and in many countries at a national level and these have not necessarily always been consistent with each other. Currently in Europe there are: 1. Eurovignette countries with time-based charges; 2. countries with nationally implemented kilometre-based or distance-based charges; 3. countries with tolls on specific roads; and 4. countries with no direct road charging at all. Ricci and Fagiani (2003) and the references therein provide useful information on these cases. The Eurovignette system was originally adopted in six EU member countries: Germany, Belgium, Denmark, Luxembourg, the Netherlands and Sweden. The Eurovignette Directive of 1999/62/EC aimed to limit competition problems within the road freight sector caused by the existence of very different methods and levels of charging for infrastructure use in different countries. In particular, vehicles licenced in a country with a low annual licence duty plus supplementary tolls were 2

We thank Professor Lars Hultkrantz for providing updated information on the case.

20

E. Niskanen and C. Nash

felt to have an unfair competitive advantage when competing with a vehicle licenced in a country with a high licence duty and no supplementary tolls. Under the Eurovignette system, heavy goods vehicles (HGVs) for freight transport are required to buy a time-related supplementary licence (i.e. time-based toll) or ‘vignette’ in order to be able to use certain defined parts of the European motorway system for a certain period of time (usually a year). The Eurovignette Directive sets a limit for the maximum infrastructure access charges payable as such a licence, on the basis of average infrastructure costs, and with non-discrimination between goods vehicle operators of different nationalities. The Eurovignette system is limited to motorways and is only related to the cost of providing those roads, thus excluding external cost. It is based on time, rather than on the distance travelled. Indeed, as said, its original purpose was to set maximum charges and in this way to protect hauliers coming from other countries rather than paving the way for internalising external costs according to the user and polluter pay principles. The European Commission has proposed certain revisions to the Eurovignette Directive (CEC, 2003); these revisions would permit extension of the system to parallel main roads or indeed to all roads in the country in question thus correcting the feature of the existing system that it may divert some traffic to untolled roads, where marginal social costs are not necessarily lower. These revisions, whilst permitting differentiation of charges according to congestion and environmental costs, would still tie the average level of charges to average infrastructure costs. Even these modest changes have yet to be agreed at the time of writing. Switzerland was the first European country to introduce a kilometre-based charge on all heavy goods vehicles in the country (a phased introduction started in 2001). Moreover, the level of charges makes specific reference to external costs, although these are average rather than marginal. In Germany the new kilometrebased system was originally planned to come into effect in autumn 2003, but its introduction was delayed because of serious technical problems and it only came into operation in 2005. Austria also now has a kilometre-based charging system for heavy goods vehicles on motorways. Other European countries also are examining the case for kilometre-based charging of heavy goods vehicles and Britain plans to introduce such a system in 2008. Elsewhere, New Zealand has long had a kilometre-based charging system. In Southern Europe, France, Spain, Portugal and Italy have tolls on their motorways, which are differentiated by type of vehicle and sometimes by time of day or day of the week. These tolls are mainly seen as a way of raising money to pay for roads. Although these tolls may be of some value in improving allocative efficiency, they may also have negative effects in diverting traffic to untolled roads (like the existing Eurovignette system above), where marginal social costs are not necessarily lower.

Road Pricing in Europe – A Review of Research and Practice

21

2.5.2 Barriers in Interurban Road Pricing For pricing in interurban road transport, the same broad categorisation of potential barriers as was assumed for urban transport in Section 2.4.2 applies: 1. technological and practical barriers; 2. legal and institutional barriers; and 3. acceptability-related barriers. However, the above-mentioned short review of the current status of interurban road pricing in Europe suggests that barriers to pricing in interurban transport are generally lower than in urban transport. Three important exceptions to this general picture can however be concluded, as discussed below. These are the interoperability problems in cross-border traffic (a practical barrier), the limitations of the current EU legislation (a legal barrier), and the lack of a common European strategy (an institutional barrier). The discussion below on these and other issues is again based on Niskanen et al. (2003) and Niskanen and Nash (2004). See also Viegas (2002). Key technological barriers are similar to those in the urban context above: although appropriate technologies exist in principle, they are not widely tested and thus may not straightaway permit introduction of distance-based or kilometrebased pricing as full-scale fully differentiated marginal social cost-based pricing. (The delays in the introduction of the kilometre-based charging system in Germany showed that the significance of technology-related problems must not be underestimated.) As for practical barriers, in interurban road networks the links or routes to be priced are easier to define and isolate than in urban road networks (which can be, as discussed in Section 2.4.2, very complex). Similarly, the availability of good data on costs and of impacts of pricing does not seem to be as big an issue as in the context of urban transport pricing. Instead, as stated above, inter-operability problems in cross-border traffic are an important practical barrier. The main legal barrier (considering implementation of full-scale marginal social cost road pricing) is that currently the EU legislation does not allow the inclusion of externalities (charges can only be based on infrastructure costs) and does not allow charges to be applied except on main roads. The major institutional barrier to the European-wide (cross-border) implementation of the kilometre-based pricing system is the lack of a commonly agreed European strategy, including the lack of European-wide standards to ensure interoperability across national borders. Acceptability barriers, political, public or business though important, are less of a problem for interurban road pricing than for urban road pricing. There are (at least) two reasons for this. First, the tolls would be mostly paid not by voters, but by companies or commercial operators of lorries, which often can pass on their additional costs to their customers. Second, depending on the country many of these companies/lorries are foreign. Politicians in European countries where there is considerable through traffic have a natural incentive to use the tolls to collect revenue. Moreover, even the commercial operators themselves can support a charging system (as has happened in Germany) which they see as leading to a level playing field in relation to operators based in other countries (of course, as long as this, after possible compensations in the form of lowering other taxes, does not increase their charges overall).

22

E. Niskanen and C. Nash

2.5.3 The Next Steps in Interurban Road Pricing In the short term (up to 3 years), it can be expected that distance-based kilometre charges for heavy goods vehicles (HGVs) will continue to gain more ground. In particular, also other countries currently using the time-based Eurovignette system can be expected to follow Germany (and Austria and Switzerland) to adopt kilometre-based charging systems instead. But, as said, many other countries including the U.K are also considering this kind of charging. However, to secure acceptability and also for technological reasons, the kilometre-based charging systems potentially implemented in these and other countries may exhibit two features or restrictions: 1. no differentiation within a country in time or space, and 2. revenue neutrality i.e. no increase in total payments. To compensate for the no-differentiation condition and the fact that these charges apply to heavy goods vehicles only, in certain parts of Europe the kilometre-based pricing scheme may be complemented by a system of charges which would use simple technology and would cover certain roads (or stretches of roads) with particularly serious environmental and/or congestion problems or with particular financial needs related to those roads (e.g. when adopting public-private partnership types of solution). The revenue neutrality condition requires that the road charges will be accompanied by reductions in other taxes and/or vehicle licence fees. In the medium to long term, however, these restrictions may become irrelevant and fully differentiated distance-based charges for HGVs – and ultimately all vehicles may become possible, using GPS-based ERP technology and applying to all roads.

2.6 Conclusions 2.6.1 Lessons from Research Recently, a lot of progress has been made in making better quality data available and in providing more reliable, accurate and complete impact and policy analyses for designing optimal road pricing policies and for estimating their impacts. In particular, many EU-funded research projects have addressed these issues. Many interdisciplinary projects (such as EXTERN-E) have developed better methodologies for producing relevant data, for analysing it and for estimating relevant environmental and other external costs. Other projects (including PETS, TRENEN, UNITE and RECORDIT) have applied these methodologies to produce better estimated figures for external costs and have suggested optimal marginal cost prices based on these estimates as compared to the current price levels. Some other studies (such as AFFORD, TRENEN, MC-ICAM) have in turn used these estimated figures to estimate, by means of numerical simulation modelling analyses, welfare and other impacts of internalisation of externalities and hence of (short-run) marginal cost-based road pricing, and in different situations. Furthermore, the AFFORD and MC-ICAM modelling analyses also explicitly considered

Road Pricing in Europe – A Review of Research and Practice

23

second-best situations where the effective second-best constraints assumed reflect the different types of barriers to pricing discussed in Sections 2.4.2 and 2.5.2 above.3 The AFFORD and MC-ICAM studies demonstrated, among other things, the following two general results. First, when considering both first-best and secondbest pricing policies, in many instances, simple (second-best) systems could possibly do much of the job whilst waiting for more sophisticated (first-best) solutions. Second, much of the benefit of urban road pricing is indirect and is related to the use of revenues. These kinds of results can have important implications both for the formulation of effective policies and their acceptability. On the other hand, whilst presenting these and other results, the two projects emphasised the need for caution in interpreting them (and similar results, more generally): there may be potential biases involved in the analyses, either due to biases inherent to the models or deficiencies in the data (see Milne et al., 2000, and de Palma et al., 2004). Overall, despite the long tradition of research and despite the considerable benefits of road pricing that economists and other analysts have shown, the general public and majority of politicians still do not appear to be convinced of the available results. The benefits of road pricing and of marginal cost-based pricing in particular (as it often requires quite radical changes to current practices in many respects), to society at large and to individual stakeholders involved, still have to be demonstrated more convincingly. For instance, as argued, the analyses carried out thus far can still contain great uncertainties and may also contain serious biases. More research work is needed and, because of the great number of issues and dimensions to be allowed, this needs to be interdisciplinary work combining modelling, economic, institutional, etc. approaches. An important objective of such research, as discussed above, is to produce reliable estimates of welfare benefits and other impacts of road pricing. But a genuinely interdisciplinary approach to research may also contribute at a more general level, through affecting the overall thinking of people and policymakers’ of the relevant issues. Currently, there is still a clear gap between research and practice in this respect.

3

MC-ICAM went in this respect the furthest by considering the relationship between the secondbest constraints (as modelled in welfare economics) and the different types of barriers considered here. For that purpose, MC-ICAM made conceptual distinction between barriers and constraints. The constraints, as defined in second-best welfare economics, are limitations on the pricing system and measures used. The barriers, instead, represent factors or societal phenomena that are reasons for the constraints, they generate them. As contrasted to barriers that may be rather general and in some cases rather abstract, it should be possible to express the constraints in mathematical terms and quantify them for the purposes of theoretical and simulation modelling analysis. Another key principle is that a given barrier may give rise to or contribute to more than just one constraint or conversely different barriers may lead to similar constraints. MC-ICAM also investigated barriers and their relationship with the second-best constraints along implementation paths or transition paths ranging from a phased introduction of pricing with accompanying measures to a sudden widespread implementation (‘big bang’).

24

E. Niskanen and C. Nash

One reflection of this gap is that, from local politicians' viewpoint in particular, the economists' main criterion or (theoretical) justification in advocating marginal cost pricing, the promotion of efficiency, may be too general and abstract to communicate with voters or to support effective decision-making. Efficiencybased and thus rather abstract explanations of policies simply may not be regarded as important and relevant by people and politicians, who mainly tend to judge policies in terms of their impacts on their immediate surroundings including the fairness or equity types of issue. Local politicians are too close to their voters and their everyday problems to practise politics and to judge policy alternatives in terms of abstract efficiency criteria. Instead, more concrete and practical goals or targets such as reducing traffic congestion or environmental nuisances in certain parts of their city may appear more reasonable, but then there is the problem (from the point of view of applying congestion pricing) that there may be other more attractive measures for pursuing these targets.

2.6.2 Lessons from Practical Experience: What Next? Two key features of the current situation in European land transport are, on the one hand, the unbalanced growth of road transport as compared to rail and, on the other hand, the great differences between different countries in their approaches to road pricing. A third feature is the lack of common strategy or policy framework at the EU level. In this situation, although road pricing is most likely to be adopted in the long run as a major policy approach throughout Europe, both in urban and interurban context, the question of likely developments in the near future is very much open. Important preconditions to successful policy implementation on a broad scale, besides sufficiently great benefits shown by practical modelling analyses, are credible benchmark cases providing convincing practical experience. So far, the available experience and examples of best practice that could be transferred to other contexts are not too many. This, in particular, holds in urban transport where the Norwegian cities and Singapore, as examples of actual successful applications, are somewhat exotic to be considered as true benchmarks considering urban road pricing in bigger and smaller cities throughout Europe. Indeed, if they were such examples, one could expect (given that they have been around for some time) that the effects would already be more visible in the form of similar applications elsewhere. (This suggests, as conjectured in Section 2.4.2, that, excluding pure coincidence, there must be some specific features in these applications that do not necessarily hold elsewhere.) The really significant benchmark applications for urban road pricing still have to come. London certainly can be such and continued success there will no doubt speed up the progress (or alternatively possible setbacks will slow it) in urban road pricing in other countries and cities in the near future. By contrast in interurban road pricing, the situation appears more clear-cut: the kilometre-based charging of heavy goods vehicles in Austria, Germany and Switzerland most likely will show the way for other countries to follow, especially

Road Pricing in Europe – A Review of Research and Practice

25

those in Central Europe and with a lot of through traffic. However, what will happen in the peripheral countries in the north and south, whether the Central European examples make them to follow and within what time frames, is harder to predict. The new EU countries in the east complicate the picture even further. Clearly, the role of the EU, which is somewhat unclear at the moment, will be critical here. Future policy developments in the area of road pricing will inevitably be influenced by technological development (in communication technology and car making and thus largely exogenous to the transport sector). In particular, in the longer run, new technology, especially GPS-based electronic road pricing (ERP) or electronic fee collection (EFC), will make many of the arguments of opponents to road pricing obsolete. Whilst acceptability problems appear to be more difficult in urban transport than in interurban, it can be expected that developments in interurban transport, when ERP systems are seen to work, will over time facilitate the broad-scale application of similar systems in urban contexts, too. Similarly, in urban transport, increased application of smart card technology in public transport and of route guidance and other information technology in cars (the relevant accessories on board will become standard) will make ERP systems more acceptable. Perhaps the most important single lesson from actual attempts to implement road pricing so far is that institutional aspects of the implementation problem need to be given particular attention. Many issues need to be allowed here; one is the importance of getting the general public and stakeholder groups to be involved in developing the proposals from the very beginning. Another, and related key lesson, is that politics matters; this is because many stakeholders are involved (with widely-diverging interests) and because many of the issues involved are politically highly sensitive. As the London case shows, political will and skill is extremely important and can make a big difference.

References AFFORD. http://www.staresearch.fi/final-report1.pdf Arnott, R., de Palma, A., Lindsey, R. (1993). A structural model of peak-period congestion: A traffic bottleneck with elastic demand. American Economic Review, 83(1): 161-179. Bates, J., Polak, J., Jones, P., Cook, A. (2001). The valuation of reliability for personal travel. Transportation Research E, 37: 191-229. Bonsall, P., Shires, J. Matthews, B., Maule, J., Beale, J. (2004). Road User Charging – Pricing Structures, Draft Final Report for the Department for Transport, July, University of Leeds: The Institute for Transport Studies. Brownstone, D., Ghosh, A., Golob, T.F., Kazimi, C., Van Amelsfort, D. (2003). Drivers’ willingness to pay to reduce travel time: evidence from the San Diego I-15 congestion pricing project. Transportation Research A, 37A: 373-387.

26

E. Niskanen and C. Nash

CAPRI. http://www.its.leeds.ac.uk/projects/capri/index.html CEC (1995). Towards Fair and Efficient Pricing in Transport - Policy Options for Internalising the Eexternal Costs of Transports in the European Union. Green Paper, COM(95) 691 European Commission, Brussels. CEC (1998). Fair Payment for Infrastructure Use: A Phased Approach to a Common Transport Infrastructure Charging Framework in the EU. White Paper, COM(98) 466 European Commission, Brussels. CEC (2001). European Transport Policy to 2010: Time to Decide. White Paper, COM(01) 370 European Commission, Brussels. CEC (2003). Proposal for a Directive of the European Parliament and of the Council Amending Directive 1999/62/EC on the Charging of Heavy Goods Vehicles for the Use of Certain Infrastructures. Chu, X. (1999). Alternative congestion pricing schedules. Regional Science and Urban Economics, 29: 697-722. De Palma, A., Kilani, K., Lindsey, R., Moilanen, P., Proost, S., Sen, A., Vold, A., Niskanen, E. (2004). Welfare Effects – Urban Transport. MC-ICAM Deliverable 7. Project MC-ICAM. http://www.its.leeds.ac.uk/projects/mcicam/ De Palma, A., Quinet, E. (eds.) (2005). La tarification des transports. Pourquoi? Pour Qui? Les defis d’aujourd’hui et de demain. Economica, Paris. DESIRE. http://www.tis.pt/proj/desire.htm EXTERN-E. http://www.externe.info/ Faulhaber, G.R., Baumol, W.J. (1988). Economists as innovators: Practical products of theoretical research. Journal of Economic Literature, 26(2): 577-600. Frey, B.S. (2003). Why are efficient transport policy instruments so seldom used? In Schade, J., Schlag, B. (eds.). Acceptability of Transport Pricing Strategies. Elsevier, North-Holland: 63-75. Glazer, A. (2003). How to make unpopular policies popular after adoption. Paper presented at the third IMPRINT-EUROPE seminar, Leuven, May 2004. IMPRINT-EUROPE. http://www.imprint-eu.org/ Lam, T.C., Small, K.A. (2001). The value of time and reliability: measurement from a value pricing experiment. Transportation Research E, 37: 231-251. Lomax, T., Schrank, D., Turner, S. Margiotta, R. (2003). Selecting travel reliability measures. http://mobility.tamu.edu/ums/resources.stm (Accessed April 1, 2005) MC-ICAM. http://www.its.leeds.ac.uk/projects/mcicam/ Milne, D., Niskanen, E., Verhoef, E. (2001). AFFORD Final Report. Nash, C.A., Matthews, B. (2001). Why reform transport prices? Paper presented at the first IMPRINT-EUROPE Seminar, Brussels, 21-22 November 2001.

Road Pricing in Europe – A Review of Research and Practice

27

Niskanen, E., de Palma, A., Lindsey, R., Marler, N., May, T., Nash, C., Schade, J., Schlag, B., Verhoef, E. (2003). Pricing of Urban and Interurban Road Transport: Barriers, Constraints and Implementation Paths. MC-ICAM Deliverable 4. Niskanen, E., Nash, C. (2004). MC-ICAM Final Report. PETS. http://www.cordis.lu/transport/src/pets.htm PROGRESS. http://www.progress-project.org/ Quinet, E. (1997). Full social cost of transportation in Europe. In: Greene, D.L., Jones, D.W., Delucchi, M.A. (eds.). The Full Costs and Benefits of Transportation. Springer, Berlin: 69-111. Quinet, E. (2001). European pricing doctrines and the EU reform. Paper presented at the first IMPRINT-EUROPE Seminar, Brussels, 21-22 November 2001. Ramjerdi, F., Minken, H., Ostmoe, K. (2004). Norwegian urban tolls. In: Santos, G. (ed.), Road Pricing: Theory and Evidence, Research in Transportation Economics 9, Elsevier Science, 237-249. RECORDIT. http://www.recordit.org Ricci, A., Fagiani, P. (2003). Identifying Mode-specific Issues for Pricing Reform. TIPP Deliverable 2. Schrank, D., Lomax, T. (1999). The 1999 Annual Mobility Report Information for Urban America, Texas Transportation Institute. Texas A&M University System, College Station, Texas http://mobility.tamu.edu Schrank, D., Lomax, T. (2004). The 2004 Annual Mobility Report, Texas Transportation Institute. Texas A&M University System, College Station, Texas http://mobility.tamu.edu Schade, J. Schlag, B. (eds.) (2003). Acceptability of Transport Pricing Strategies. Elsevier. Small, K.A. (1992). Urban Transportation Economics. In: Lesourne, J., Sonnenschein, H. (eds.). Fundamentals of Pure and Applied Economics. Harwood Academic Publishers, Chur, Switzerland. Small, K.A., Noland, R.B., Koskenoja, P.M. (1995). Socio-economic Attitudes and Impacts of Travel Reliability: A Stated Preference Approach. PATH Report, MOU-117, California Partners for Advanced Transit and Highways. TIPP. http://www.strafica.fi/tipp/ TRENEN. http://www.cordis.lu/transport/src/trenen1.htm UNITE. http://www.its.leeds.ac.uk/projects/unite/index.html Viegas, J. (2002). Tolling Heavy Goods Vehicles on European Roads: From a Diverse Set of Solutions to Interoperability? Paper presented at the second IMPRINT-EUROPE seminar, 14-15 May 2002, Brussels. Vickrey, W.S. (1969). Congestion theory and transport investment. American Economic Review (Papers and Proceedings), 59(2): 251-260.

3

Road Pricing: Consequences for Traffic, Congestion and Location

Lars-Göran Mattsson Department of Transport and Economics, the Royal Institute of Technology, Stockholm.

Abstract Congested roads seem to be an unavoidable characteristic of large cities. Transport economists and planners have regularly suggested that road pricing would be an appropriate and effective instrument in an overall policy to relieve congestion. Politicians and the public at large have usually been quite sceptical, however. In this paper, three ex ante studies of transport and location effects of alternative road pricing systems are presented and compared. Different models estimated with different data sets are applied to calculate the effects. The first two studies deal with the effects on the traffic pattern of a zone-based and a distance-based road pricing system for the Stockholm area, respectively. In the third study, location effects are also included in an analysis of optimal congestion charges in a stylised symmetric city adjusted to resemble Stockholm. All studies indicate a substantial reduction in vehicle distance travelled. For the zone-based system, traffic volumes in the inner city of Stockholm are predicted to decrease by 30% for charged hours at a charge level equivalent to 3 SEK/km. For the distance-based system, traffic volumes in the inner city are predicted to be reduced by 35 and 19% at charge levels of 4 and 2 SEK/km for peak and office hours, respectively. For the case of optimal congestion pricing, the reduction is 25% at an average charge level of 2 SEK/km. Additional effects in the first study are that speed might increase on inner city roads and arterials by around 20%. Moreover, accessibility to activities in the other half of the city will be reduced significantly. The most affected relation is the one between inner northern and inner southern suburbs. In that case, a reduction of the number of vehicle trips by around 30% is predicted. In spite of quite substantial transport effects, the location effects are predicted to be very limited.

3.1 Introduction Congested roads seem to be unavoidable in our large cities. Attempts to meet demand by a larger supply have often been unsuccessful. Addition of new roads to the existing network tends to induce new traffic that sooner or later neutralises the immediate relief of the situation. Transport researchers and planners have for a long time argued that road pricing has to be a vital element in a successful transport policy that could reduce the level of congestion and improve the efficiency of the transport system.

30

L.-G. Mattsson

This conviction among many researchers and planners has generally been without parallel among politicians and the public at large. It has been difficult to convince decision-makers that road pricing is a sensible policy and, accordingly, there are only a few implementations in the world of citywide road pricing schemes. The only operational systems for the moment are those in Singapore, in the Norwegian cities of Bergen and Oslo, and in London. Stockholm will probably also be added to this list. From January 3 to July 31, 2006, a full-scale test of congestion charging was carried out in the city of Stockholm. This test was followed by a referendum in connection with the general election in September 2006. In the city of Stockholm 51% voted yes and 46% voted no. In response to this outcome, congestion charging will be permanently re-introduced by August 2007. Road pricing is an ambiguous concept. It is used here as a general term for a pricing system that charges the road users for their right to use a certain road or to drive in a certain area. This terminology excludes parking fees and fuel and vehicle taxes from being seen as a kind of road pricing. Road pricing has been advocated for different reasons. Above, we referred to the efficiency objective or road pricing as a way of reducing congestion to an optimal level (going back to the seminal work of Vickrey, 1955). First-best congestion pricing would be to charge each road user a fee that is equivalent to the increased costs his/her presence on the road implies for all other road users. In such a system, the charge level should theoretically vary not only with the time of the day and the type of the road but also with the time values of the road users that happen to be present on the road. This would be far too complicated. Any second-best congestion pricing scheme that would be a candidate for actual implementation has to be simplified in many respects and also predictable for the road user as to the charge level. Road pricing has also been proposed as a way of reducing the environmental impacts of road traffic. The charge level in such a system should be related to the costs of pollution and noise that the traffic contributes to and which could be added to the congestion-related charge. A third, and among policy-makers more popular reason for road pricing, is that of raising funds to finance transport infrastructure. In fact, the Norwegian toll rings have all been set up with this objective in mind. These different objectives for road pricing are partly in conflict with each other. A revenue-oriented system should allow for season tickets and it would also be tempting to choose charging points to avoid reductions of traffic flows. A congestion pricing system, on the other hand, aims directly at reducing the traffic flows. However, in order to increase the political acceptability of road pricing, a combination of the objectives in an integrated strategy has also been considered (May and Roberts, 1995). One of the key issues for the political, and probably also public, acceptability is how the revenues from road pricing are spent (Goodwin, 1987; Small, 1992). A road pricing scheme might affect the road user in many ways. The direct effect is that the monetary cost of using charged road segments during charged hours would increase. As an indirect effect, the time cost for the same segments will be reduced, since some road users will leave the road. For the typical user, this will to some extent, but not fully, compensate for the increased monetary cost. Those who

Road Pricing: Consequences for Traffic, Congestion and Location

31

leave the charged road segment will change their route, destination, frequency, mode or timing of the trip. In the long run, car ownership and land use would also be affected. The purpose of this chapter is to discuss how road pricing might affect traffic, congestion and location. Assuming that each road user makes the adjustments that are most favourable to him/her, it is obvious that the consequences of a road pricing scheme will depend on how good the alternatives to the present travel decisions are. But these alternatives vary from city to city. If we should be able to say something more precise about the effects, we have to consider specific pricing proposals for specific cities. The most direct approach would be to compare the situation before and after a road pricing system has been implemented in a city (see Ramjerdi, 1995, for an excellent study of the Oslo toll system). An alternative approach is an ex ante study where the effects of alternative pricing schemes are simulated by the use of some travel demand analysis system (see May and Milne, 2000, for a very interesting study for Cambridge). Broad results from three studies will be presented here. The first two studies deal with traffic effects of a zone-based and a distance-based road pricing system for the Stockholm area. In the third study, location effects are also included. That study involves the use of an integrated transport and location model of a stylised symmetric city.

3.2 Application of a Zone-Based Road Pricing System to Stockholm In the first case study, we consider the application of a zone-based road pricing system to the city of Stockholm. The particular study was carried out for the city executive board as part of an assessment of alternative road pricing systems (Mattsson, 1995). To reduce the amount of data to present, we will concentrate on one road pricing system and compare it with a situation without road pricing.

3.2.1 Scenarios In the zone-based road pricing system to be analysed, the inner city of Stockholm is subdivided into five zones. A charge is imposed on each vehicle entering in an inbound direction the area covered by these zones, i.e. crossing the outer border of the five zones. In addition, the same level of charge is also imposed on each vehicle crossing any of the borders between the five zones independent of direction. This means that a vehicle will not only have to pay for entering the inner city, but also for travelling within the inner city. The total charge on a vehicle will hence depend on the amount of travel within the inner city, operationalised as the number of times the vehicle crosses the borders between any zones. For the main semiorbital route (which includes the Gröndalsbron bridge), which actually crosses the western part of the charged area, vehicles are only charged for entering the outer

32

L.-G. Mattsson

border of the charged area. This implies that a vehicle can go between the northern and southern suburbs of Stockholm for a charge of one crossing. Such a zone-based system would be fairly complicated to implement. Some kind of electronic system would probably be necessary. The topography of Stockholm facilitates the practical implementation of the system. It is to a great extent possible to use the natural subdivision of the inner city by water when constructing the zones. This means that there are relatively few road connections between the zones, often in the form of bridges. It would be necessary to have 12 charging points to form the outer cordon of the area covered by the five zones and in addition to that 29 charging points for the borders between the zones inside the covered area, i.e. in total 41 charging points. This zone-based road pricing system was intended to represent congestion pricing in so far that the charges are only imposed during Monday to Friday, 6 a.m. to 7 p.m. The charge level was set to 9.75 SEK for light vehicles1 . For heavy vehicles (over 3 tonnes) the charge level was three times as high. The analysis was carried out based on population, employment and car ownership forecasts for the year 2005. It was assumed that there would be no important upgrading of the transport network as compared to 1995. The zone-based road pricing scenario will be compared with a reference scenario that is exactly the same except that no charges are imposed.

3.2.2 Modelling Approach To predict the traffic effects of the scenarios, we have applied one of the most advanced traffic analysis systems that was available for Stockholm at the time – the Fredrik system. The Fredrik system combines a nested logit travel demand model with the network equilibrium assignment model of Emme/2. The logit-based travel demand module handles trip frequency, travel mode and destination choices with separate models for 8 different individual-based trip purposes, whereas the assignment module predicts route choices and traffic flows on the road network. In the assignment module, additional vehicle flows representing freight transport and distribution traffic are added to the individual-based flows on the road network. The Emme/2 system allows congestion on the road network to be modelled by specifying the travel time of each link on the network as an increasing function of the traffic flow on the same link. The pertinent property of the network equilibrium assignment is the Wardrop principle of user optimum (or Nash equilibrium). It implies that the predicted car travel demand for any origin-destination zone pair will be assigned to the different possible routes on the network between the origin and the destination so that all routes that are actually used, i.e. have positive flows, will also get the same generalised cost, and so that no unused route has a lower generalised cost. The generalised cost is then calculated as the travel time multiplied by an assumed value of time plus the monetary cost, where the latter one 1

To facilitate comparisons between the studies, all monetary values are expressed in constant value of money of the year 2000.

Road Pricing: Consequences for Traffic, Congestion and Location

33

includes possible road user charges. Different values of time are then applied to private trips, business trips and additional flows of freight transport and distribution traffic. In this way, it is possible to represent different sensitivities to costs for these different categories. In the assignment module of the Fredrik system, car travel times between the different zones are calculated with respect to the congestion levels that are implied by the assigned flows to the links on the road network. These car travel times are then fed back to the module calculating the travel demand between the zones and this new demand is in turn returned back to the assignment module. The Fredrik system iterates between the demand module and the assignment module until an overall equilibrium is achieved. The demand and supply of travel vary considerably over the time of the day. To handle this, the traffic effects have been calculated for three separate time periods, representing the morning peak, afternoon peak and the rest of the day. The results for the different time periods are then weighted together to represent different aggregate time periods. Road user charges may, in principle, affect a trip maker’s choice of route, departure time, travel mode, destination and frequency. It may also affect the propensity for trip chaining and in the longer run car ownership and activity pattern by relocation of the trip maker’s residence or of the activities to which the trip maker is travelling. Of these potential behavioural responses, the Fredrik system handles route, mode, destination and frequency choices. These adjustments probably represent the most important responses with one obvious exception – the choice of departure time. The likely consequence of this shortcoming is that the Fredrik system will underestimate the car traffic reduction effect of road pricing during the time periods when charges are in place, while overestimating the overall reduction effect.

3.2.3 Transport Effects Table 3.1 presents the effect of the zone-based road pricing system on the vehicle distance travelled subdivided by where the travelling takes place. The inner city coincides almost perfectly with the charged area with the exception of the semiorbital route, which is partly inside. Since the road pricing system is consciously designed to divert traffic to that route, its traffic volume is presented separately. Obviously, the proposed road pricing system has a substantial effect on the traffic volumes in the charged area, i.e. the inner city. The system is designed to have a certain congestion pricing profile. In fact, only 39% of the hours of a week are charged. Since the charges are in place during the most congested hours, however, the traffic reduction effect as an average over all hours of a week is as high as 19% compared to 30% when considering only charged hours. Some of the traffic that disappears from the inner city is diverted to the semi-orbital route, which will be more congested. It is also interesting to note that although only the inner city and to some extent the semi-orbital route are actually charged, there is a notable reduction of traffic in the inner suburbs by 8 % for charged hours. The reason is that vehicle

34

L.-G. Mattsson

trips that in the reference scenario go between one inner suburb and the inner city or another inner suburb on the other side of the inner city may in the road pricing scenario change destination or mode or not even take place. Finally, it should be remembered that in a county like Stockholm as much as 60% of the traffic is in the outer suburbs (in the reference scenario). Hence, the traffic reduction effect of the proposed road pricing system on a county scale is only 4% for the charged hours and 3% as a weekly average. Table 3.1. Total vehicle distance travelled by subregion Subregion

Inner city Semi-orbital route Inner suburbs Outer suburbs County

Percentage of total distance travelled in reference scenario 6 2 32 60 100

Percentage change in road pricing scenario with respect to reference scenario Mon-Fri, 6 a.m. Weekly average to 7 p.m. -30 -19 15 10 -8 -5 0 0 -4 -3

The Saltsjö-Mälar water strait subdivides the Stockholm region into a northern and a southern part connected by five bridges. Because of the way in which the road pricing system is designed, one cannot go between these parts without being charged. Trips that only use the semi-orbital route are only charged once, however. Table 3.2 shows how the road pricing system affects the vehicle volumes on these critical bridges during the charged hours. The results indicate a substantial reduction by more than 40 % of the vehicle flows across the central bridges in the inner city. To some extent, these flows are rerouted to the Gröndalsbron bridge (which is a section of the semi-orbital route). The net reduction of the flows across the Saltsjö-Mälar passage is 25%. Table 3.2. Vehicle traffic volumes across the Saltsjö-Mälar passage (thousand vehicles per workday, 6 a.m. to 7 p.m.) Bridge Bridges in Old Town Central bridge Western bridge Gröndalsbron Total

Scenario Reference 56 92 41 109 264

Road pricing 32 52 22 117 223

Percentage change -43 -43 -46 7 -25

The implementation of the proposed road pricing system will lead to less interaction between the different subregions in the County of Stockholm. This is clearly illustrated in Table 3.3, which displays percentage changes in individual vehicle trips for charged time periods. It should be remembered, however, that the reduction in vehicle trips to some extent is explained by a change of mode of transport. The inner city exhibits the largest overall change with a 15% reduction of the vehicle trips to and from this subregion. Also, the trips that are totally inside

Road Pricing: Consequences for Traffic, Congestion and Location

35

the inner city are reduced substantially, by 12%. This is, of course, a consequence of the design of the road pricing system with zonal boundaries inside the inner city. There are even more obvious reductions in the vehicle trips between suburbs on opposite sides of the inner city. This is particularly the case for trips between the inner suburbs for which the reductions are about or slightly below 30%. These trips are very dependent on priced links through the heart of the city centre. Trips between the outer suburbs can to a larger extent use the less heavily priced semiorbital route. As a consequence, they are not reduced to the same extent. Considering the county as a whole, the number of vehicle trips goes down by 5%, which is about the same reduction as was noted for the vehicle distance travelled (see Table 3.1). In sum, the road pricing system will have a clear effect on the spatial pattern of interaction. This is partly a consequence of the particular structure of the Stockholm road network. All vehicle trips between the northern and southern parts have to go through the inner city and will be more or less heavily priced. Table 3.3. Percentage change in vehicle trips between subregions for road pricing scenario with respect to reference scenario for workdays, 6 a.m. to 7 p.m. To From O. n. suburbs I. n. suburbs I. city I. s. suburbs O. s. suburbs Total

Outer northern suburbs 2 2 -10 -22 -16 -1

Inner northern suburbs 2 4 -17 -28 -21 -5

Inner city -9 -16 -12 -21 -13 -15

Inner southern suburbs -21 -30 -23 6 4 -5

Outer southern suburbs -15 -23 -13 4 3 -1

Total

-1 -5 -15 -5 0 -5

The next two tables present the effects of the proposed zone-based road pricing system on average vehicle speed and trip travel time. Table 3.4 indicates that the road pricing system has a substantial impact on the level of congestion in the inner city. The system leads to an increase in speed on inner city roads by 17% and on inner city arterials by 24%. As noticed, the road pricing system will cause some of the traffic to be rerouted from inner city roads to the semi-orbital route (cf. Table 3.1). Hence, this route will be more congested and speeds will actually go down. Although the road pricing system only charges car driving in the inner city, it has a perceivable effect on the speed in inner suburbs as well, an increase of 9%, whereas there is no such noticeable effect in outer suburbs. Table 3.4. Average vehicle speed (km/h) during workdays, 6 a.m. to 7 p.m.

Inner city roads except arterials Inner city arterials Semi-orbital route Inner suburbs Outer suburbs

Reference 24 29 41 42 63

Scenario Road pricing 28 36 37 45 63

Percentage change 17 24 -8 9 0

36

L.-G. Mattsson

Table 3.5 shows the impact on vehicle travel time per trip by subregion of the origin. The largest effect is in the inner suburbs, where travel time goes down by 78%. This is a result of a combination of shorter trip length (not shown here) and higher speed (as was shown in the previous table). For trips originating in the inner city, there is only a slight reduction in travel time per trip, though there was a substantial increase in average speed on inner city roads and arterials. This is a consequence of an increase in travel distance per trip in this subregion, which in turn may be an effect of rerouting to avoid passing charging points in the inner city. It is also of interest to estimate the level of revenue to which the road pricing system would give rise. Table 3.6 shows that the number of crossings of the outer border of the charged area is about of the same magnitude as that of the borders inside this area. This means that of the total daily revenue of 5.7 million SEK from the road pricing system, about half comes from the outer and half from the inner border crossings. This implies an average charge level of 3.1 SEK per workday and inhabitant in the Stockholm County. Table 3.5. Vehicle travel time per trip (min) during workdays, 6 a.m. to 7 p.m. Subregion of trip origin Outer northern suburbs Inner northern suburbs Inner city Inner southern suburbs Outer southern suburbs County

Scenario Reference Road pricing 26.0 25.2 22.5 20.8 28.7 28.4 22.4 20.9 23.3 22.6 24.5 23.4

Percentage change -3 -8 -1 -7 -3 -4

Table 3.6. Number of charging point crossings and revenue from charging for a workday, 6 a.m. to 7 p.m., in the road pricing scenario Place of charging Outer border of the charged area (inbound direction) Zone borders inside the charged area (both directions) Total

Number of crossings

Revenue (million SEK)

240,000

2.9

230,000

2.8

470,000

5.7

The total annual revenue can be estimated as 250 times the revenue for a workday. In the present case, this amounts to 1.4 billion SEK/year or 780 SEK per inhabitant per year in Stockholm County. Net of collection costs, the total revenue would amount to 1.2 billion SEK/year (see Mattsson, 1995, for the calculation of collection costs). If the total net revenue were to be refunded to the inhabitants in the County of Stockholm on a per head basis, the refund would be 680 SEK/year. The average charge level per crossing is 12.0 SEK, which is higher than the charge level for a light vehicle because of a certain share of heavy vehicles in the traffic flow. The charge per vehicle distance travelled in the county as a whole during charged hours is 0.21 SEK/km. Since the charge is actually imposed only in

Road Pricing: Consequences for Traffic, Congestion and Location

37

the inner city, it might be more appropriate to calculate the charge per vehicle distance travelled in the inner city including the semi-orbital route, which amounts to 3.0 SEK/km. This could be compared to other vehicle operating costs that are assumed to be 1.1 SEK/km for private trips. It is also interesting to calculate an approximate arc elasticity of vehicle distance travelled with respect to total vehicle operating costs 2 . This calculation leads to an elasticity for the proposed road pricing system of -0.23. According to a review by Goodwin (1992), the long-term elasticity with respect to fuel cost (which represents the main part of the operation cost) could be around -0.33 and the short-term elasticity around -0.16. The present value is therefore well inside this interval and does not represent any extreme assumption about the trip makers’ cost sensitivity.

3.3 Application of a Distance-Based Road Pricing System to Stockholm The next case study is of a more recent date. It concerns the application of a distance-based road pricing system to the city of Stockholm and its inner suburbs. This particular study was carried out by the Transek consultancy firm by order of the Swedish National Environmental Protection Agency (Lindqvist Dillén et al., 2001). The purpose was to illustrate to what extent some kind of congestion pricing, in this case operationalised as a distance-based road pricing system, could reduce congestion and improve passability in the city.

3.3.1 Scenarios In this distance-based road pricing system, the densely built-up area of Stockholm is subdivided into two charge areas – inner city and inner suburbs. Inner city is defined almost exactly as in the previous case study, whereas inner suburbs are only roughly defined in the same way3. When driving in these charge areas during peak hours or office hours, a distance-based charge is levied on each vehicle. Peak hours are defined as workdays, 7 a.m. to 9 a.m. and 4 p.m. to 6 p.m., and office hours as workdays, 9 a.m. to 4 p.m. Two different road pricing scenarios are studied – a high and a low scenario. The charge levels in the high scenario are 2 Let TR and T0 be the total vehicle distance travelled in the county as a whole during charged hours in the road pricing and the reference scenario, respectively. Furthermore, let S be the total revenue and c operating vehicle cost per unit of distance excluding possible road charges. The arc elasticity of total vehicle distance travelled with respect to total vehicle operating cost per unit of distance can then be approximated by (TR  T0 ) /(TR  T0 ) . S /(2TRc  S ) 3 The main differences are that the present definition of inner suburbs includes part of the municipalities of Huddinge, Sollentuna and Täby but not the municipality of Nacka.

38

L.-G. Mattsson

shown in Table 3.7. The levels in the low scenario are simply half of the levels in the high scenario. Table 3.7. Distance-based charge levels in the road pricing high scenario (SEK/km) Peak hours Office hours

Inner city 4 2

Inner suburbs 1 0

Compared to the previous case study, the period of charging is reduced by two hours. Moreover, the idea of congestion pricing is more pronounced, since peak hours are priced more heavily than office hours. The area that is charged is larger during peak hours, since the inner suburbs are also included. The previous system was found to be equivalent to a charge level of 3.0 SEK/km in the inner city, which is midway between the levels for peak hours and office hours in the present high scenario. To implement a distance-based road pricing system, an electronic system would be necessary. Such technique is under development but will not be discussed further here. The analysis will be carried out for the year 2015. Assumptions about population, employment and economic development are chosen in agreement with the present regional plan. As for the transport system, the semi-orbital route has been extended by the southern and northern links as compared to today. In addition, a number of other road and public transport investments are assumed to have been completed. The distance-based road pricing high and low scenarios will be compared to a reference scenario that is exactly the same except that no charges are imposed.

3.3.2 Modelling Approach Recently, a new national transport model, Sampers, has become operational (Beser and Algers, 2002). Sampers is a comprehensive forecasting tool that combines an advanced application of nested logit models for travel demand and the network equilibrium assignment model Emme/2. The system includes a regional model system that covers the Stockholm region, and which is the specific model system that has been applied in the present study. It treats similar choice dimensions to the Fredrik system that was applied in the previous case study. This means that most of the important behavioural responses to road pricing are treated with the exception of the choice of departure time. Sampers is estimated on recently collected data and has been extensively validated. It is meant to be the reference model for Sweden (Widlert, 2002).

3.3.3 Transport Effects The effects of the distance-based road pricing system on vehicle distance travelled are displayed in Table 3.8. The first observation is that the proposed pricing system

Road Pricing: Consequences for Traffic, Congestion and Location

39

leads to a substantial reduction in travel distance. The reduction level varies in an expected way between peak and office hours and between the high and low charge scenario. Overall, the reduction is at the same level or possibly somewhat lower than in the previous case study. There, we observed a reduction of 30% for charged hours in the inner city at an equivalent charge of 3.0 SEK/km. Here, we have a reduction in the inner city of 35 and 19% for peak and office hours at a charge of 4 and 2 SEK/km, respectively. The high scenario has charges which are exactly twice as high as the low scenario. The reduction is almost proportional to the charge level in the inner suburbs, whereas it is less than proportional in the inner city. It can also be noted that there is a spill-over effect of the pricing system in so far that there is a discernible reduction of travel distance in the inner suburbs also during office hours when the charge in that area is zero. Table 3.8. Total vehicle distance travelled by subregion Subregion

Inner city Inner suburbs

Percentage change with respect to reference scenario Road pricing high scenario Road pricing low scenario Peak hours Office hours Peak hours Office hours -35 -19 -21 -13 -22 -5 -11 -3

The traffic volumes on the bridges across the Saltsjö-Mälar passage are also reduced substantially (see Table 3.9). The rerouting effect is less pronounced compared to the previous zone-based road pricing system, however. In the latter case, the volume on the central bridge was reduced by 43%, while the volume on the Gröndalsbron bridge was increased by 7%. Here, the volume on the Central bridge is reduced to a lesser extent, and the volume on the Gröndalsbron is reduced rather than increased. In the present system, the vehicles pay strictly for driven distance in the charged area, whereas in the previous system Gröndalsbron as part of the semi-orbital route was consciously lower priced to offer a rerouting option. The present system is less effective in reducing traffic through the city centre. If this is not desirable, pricing the semi-orbital route less heavily than inner city roads could probably modify the effect. In an electronic implementation of a distance-based road pricing system, such fine-tuning would be technically feasible. Table 3.9. Vehicle traffic volumes on bridges across the Saltsjö-Mälar passage Bridge

Central bridge Gröndalsbron

Percentage change with respect to reference scenario Road pricing scenario high Road pricing scenario low Peak hours Office hours Peak hours Office hours -24 -13 -20 -12 -12 -6 -10 -5

The revenues that would be generated by the present pricing system are 2.7 and 1.6 billion SEK/year for the high and low scenarios, respectively. This represents a charge level of 1,280 and 730 SEK per inhabitant and year, respectively. Of these revenues, 1.8 and 1.0 billion SEK/year, respectively, are from driving in the inner

40

L.-G. Mattsson

city. The corresponding value for the zone-based system was 1.4 billion SEK/year, all of which was collected in the inner city.

3.4 Application of Optimal Congestion Pricing to a Generic City So far, the analysis has been focussed on the most immediate behavioural responses to road pricing such as route, mode, destination and frequency choices. There are also long-term effects that may be important. The most prominent examples are changes in car ownership and in the location of traffic generating activities. How road pricing affects location is an issue that has been subject to varying opinions among researchers and public debaters. This is motivation for the fact that in this final case study we also include location effects of road pricing in our analysis. In the previous case studies, we have assessed systems for road pricing where the charge levels have been fixed in advance. Here, we take a normative approach and evaluate transport and location effects of a system of optimal congestion pricing, i.e. a system where the charge levels are determined endogenously to internalise the external effects of congestion and hence will vary with level of congestion on the different links in the road network. For the moment, such a system would not be technically possible to implement. It should rather be seen as a benchmark for other more feasible systems.

3.4.1 Modelling Approach To analyse optimal congestion pricing, a combined transport and location model of a generic city will be applied. The city is completely symmetric in all respects, which simplifies the analysis. The model will be described very briefly. For a full account, see Eliasson and Mattsson (2001). The city consists of 8 symmetric rays from the city centre, with 4 suburban zones on each ray at every 5th kilometre. There is a radial network, where links consisting of two lanes in each direction connect the zones on each ray. In addition, the zones in the ring immediately outside the city centre are connected by a ring road, also with two lanes in each direction. Since the city is completely symmetric in all respects, it will be sufficient to look at only one ray when presenting the results. The zones will then be denoted 1 to 5, from the city centre to the outermost suburbs, and the links A to D, from the innermost links to the outermost ones (see Fig. 3.1). There are four types of activities that are located in the zones in the city: households, workplaces, shops and service establishments. Logit models govern the location of activities, where the most important location factor for households is accessibility to shops, services and workplaces, for shops and service establishment it is accessibility to households (as customers) and workplaces (as deliverers) and for workplace it is accessibility to households (as workforce).

Road Pricing: Consequences for Traffic, Congestion and Location

zone 1 (centre)

ring ring A zone 2

(in)

A (out)

ring

B (in)

zone 3

C (in)

zone 4

D (in)

B

C

D

(out)

(out)

(out)

41

zone 5

ring

Fig. 3.1. A representative ray of the star-shaped city with the ring road added

These activities give rise to travel that can be carried out by three different modes: car, public transport and slow mode (the latter one is an aggregate of cycling and walking). All modes can use all radial links. Only cars can use the ring road. The generalised link travel costs for cars are increasing functions of the car flow on the link. For public transport, on the other hand, the generalised link travel costs are decreasing functions of the volume of passengers. The assumed mechanism behind this is that when passenger volumes go up, the density and/or frequency of the public transport lines can be increased. In both cases, the effective travel time will go down. The slow mode has generalised travel costs that are proportional to distance. We assume user equilibrium route choices based on generalised costs for each mode. With the assumed simple network in the city, there is always a unique cheapest route for public transport and slow mode. Because of the ring road for cars, there may be alternative cheapest routes for cars. Travel can take place during three different time periods: morning peak, office hours and afternoon peak. The worker of the household travels every workday to work during morning peak and returns home during afternoon peak. Shopping and service trips take place during office hours or afternoon peak. Logit models that take generalised travel costs into account determine the mode and destination choices of these trips. For shopping and service trips, there are also logit-based choices of trip frequency and time period (office hours or afternoon peak). In addition, there are deliveries from the workplaces to the shops and the service establishments. These deliveries have fixed frequencies, are all by car and only during office hours. A logit model that takes generalised transport costs into account determines the spatial pattern of the deliveries. Thus, the travel demand for different trip types depends on where different activities are located and what the generalised travel costs are on the different links in the transport network. The travel times (and, in the case of congestion pricing, also the travel costs) depend on the number of people choosing the different modes for the different links, i.e. on travel demand. The location of activities finally

42

L.-G. Mattsson

depends on the accessibility to other activities in the different zones, which in turn depends on the location of other activities and the generalised costs between the zones. All these relationships result in a number of equations that are solved for equilibrium (see Eliasson and Mattsson, 2001, for details). The model has been calibrated with the intention to replicate location and transport patterns of a “generic” symmetric European city. In the end, however, one has to make a choice of what a generic city is. In this paper, behavioural parameters and size variables have, when possible, been chosen so that the model should resemble the situation in the densely built-up area of Stockholm for the year 1996 (see Eliasson and Matttsson, 2001, for a complete specification). Since the consumer price index change is negligible between 1996 and 2000 (less than 1% increase), we may equally well think that 2000 is the year of application.

3.4.2 Scenarios Optimal (first-best) congestion pricing is achieved by charging each car user on a link the social marginal congestion cost of driving on that link4. If all car users have the same value of time, W, the optimal charge for a specific link as a function of the flow f on the link would be

charge( f ) W

dt ( f ) f , df

where t ( f ) is the travel time, which is assumed to be an increasing function of the car flow on the link5. This level of charge has an intuitive interpretation. To achieve optimal congestion pricing, each car user on a link should be imposed a charge that is equivalent to the additional cost his presence on the link imposes on all other users of the same link, which is the additional travel time caused by him, dt ( f ) / d f , times the number of car users affected, f, times their value of time, W. Since this level of charge depends on the actual car flow, it will be different for different links and different time periods. One additional difficulty should be noticed. The value of time varies, in fact, between different trip types, and hence the average time value on a link depends on the mixture of trip types. For computational reasons, we neglect this difficulty and consistently apply a common average value of time of W = 42.6 SEK/h. A congestion pricing scenario, where optimal congestion pricing according to the specification above is applied, will be compared to a reference scenario with no road pricing. Scenarios with an inner or outer cordon toll ring have also been analysed in Mattsson and Sjölin (2004), but will not be discussed here.

4 To be theoretically correct, economies of scale in the public transport system and land congestion should also be optimally priced. 5

We have assumed that only travel time and not operational cost varies with the traffic flow.

Road Pricing: Consequences for Traffic, Congestion and Location

43

3.4.3 Transport and Location Effects In the present study, the road user charges are endogenously determined as optimal congestion charges for each link and time period. Since all links have the same capacities, the charges will directly reflect the levels of congestion on the links. Table 3.10 presents the resulting congestion charges on a per kilometre basis. Since the differences between link directions are small, we only present averages over the two directions here and in forthcoming tables. Interestingly, the level of charge varies between 0 and 4 SEK/km, which is very much the same as was assumed in the previous two case studies. The charges and hence also the congestion levels are highest for the ring road, link B and A during morning and afternoon peaks. For the outermost link D during office hours, optimal congesting charge would be practically zero. Table 3.10. Optimal congestion charges by time period and link (SEK/km) Time period Morning peak Office hours Afternoon peak

A 3.1 1.1 2.1

B 3.5 1.7 2.2

C 1.9 0.4 1.2

D 0.8 0.1 0.5

Ring 4.3 1.7 3.2

With these levels of congestion charging, the total revenue collected per inhabitant would be 9.5 SEK per workday or 2,380 SEK per year6. This is a much higher annual charge for the inhabitants than was the case for the zone-based system, 780 SEK, or for the distance-based system, 1,280 SEK and 730 SEK for the high and low level of charge, respectively. One reason is that in the present model we have assumed that the population is much more densely located than actually is the case in the Stockholm region. In addition, the road system is sparser in the model than is the case in Stockholm, which leads to high levels of congestion and hence of congestion charges. Table 3.11 presents the effects of congestion pricing on total distance and time travelled by mode. Congestion pricing leads to a substantial reduction in car distance travelled by 25%, again of the same magnitude as for the congested parts of Stockholm in the previous studies. The effect on total car travel time is even more dramatic, which reflects the fact that the link speeds are much higher in the congestion pricing scenario than in the reference scenario (see Table 3.12). Part of the car traffic reduction is an effect of change of mode to both public transport and slow mode. Since we have assumed increasing returns to scale with respect to public transport travel time, total distance travelled by public transport goes up more than total time does. The car link speeds during morning and afternoon peaks on the most congested links (Ring, A and B) are quite low in the reference scenario, 9 to 14 km/h. This is too low to reflect the situation in Stockholm (cf. Table 3.4). The road network of 6 The model is expressed in workers rather than in inhabitants. We have applied the average inhabitant to worker ratio for Stockholm to obtain this result. In addition, we have assumed 250 workdays per year and that congestion charging is effective only on workdays.

44

L.-G. Mattsson

our generic city is too sparse realistically to represent the Stockholm network. Because of that, we also obtain unrealistically large speed increases in response to the introduction of congestion pricing. Some qualitative conclusions can still be drawn. By assumption, there are only work trips during the morning peak, while there are also shopping and service trips during the afternoon peak. Since the latter kinds of trips are elastic with respect to trip frequency and can also change time period, speed increases are greater for the afternoon peak than for the morning peak. We can also note that when the level of congestion is low (speed is high) in the reference scenario, the speed increase is low. Link D during office hours is the most extreme example. Table 3.11. Total distance and time travelled by mode. Percentage change in the congestion pricing scenario with respect to the reference scenario Car Distance -25

Public transport Distance Time 17 12

Time -59

Slow mode Distance Time 25 25

Table 3.12. Car link speed by time period and scenario (km/h) Link

A B C D Ring

Morning peak ReferConence gestion 14 24 13 23 21 29 33 37 11 20

% 72 79 40 10 83

Office hours ReferConence gestion 23 34 17 30 37 41 45 46 17 30

% 48 74 12 2 74

Afternoon peak ReferConence gestion 13 28 12 28 22 34 35 39 9 24

% 119 141 54 12 161

Table 3.13 shows the effects of congestion pricing on link mode shares. The model predicts fairly large decreases in car shares for all links, also the outer ones. The increases are slightly higher for slow mode than for public transport. The effects of congestion pricing on the location pattern are displayed in Table 3.14. The most striking result is that the effects are generally very small. Congestion pricing leads for most of the zones to a relocation of less than 1% of the activities. The most notable exception is a relocation of shops from the suburbs connected by the ring road, i.e. zone 2, to the city centre. Congestion pricing makes it more expensive to use the ring road for shopping trips. This makes it more attractive to locate shops in the city centre. In general, there is a slight tendency to move out activities from zone 2 to the next zone farther out. Congestion pricing hence seems to have a slight decentralising effect. This is in contrast to what was found when congestion pricing was applied to the same generic city but without the ring road (Eliasson and Mattsson, 2001). The location effects of congestion pricing were in general somewhat larger in that case. But more interestingly, they went in a centralising direction. The explanation for the present result may be that the ring road makes the interaction between the outer suburbs easier. When congestion pricing makes it more expensive to use the congested central links of the road network, the outer suburbs may become more attractive. The overall conclusion is that the location effects of congestion pricing are ambiguous, also in a highly stylised city as in our

Road Pricing: Consequences for Traffic, Congestion and Location

45

case. They seem to depend on the specific design of the road network and on the cost of interaction. Table 3.13. Absolute change in mode shares by link for the congestion pricing scenario with respect to the reference scenario (percentage units)a Mode

Link A -8 3 5

Car Public transport Slow a

B -11 5 7

C -10 4 6

D -7 3 4

For the Ring road, which is only open for cars, there are no mode changes by assumption.

3.5 Discussion and Conclusions The subject of this paper is the effect on transport – and to some extent on location – of road pricing. General answers to that question would definitely be very useful. One can raise doubts about whether it is possible, however. There are two different ways of investigating the question. The most direct way is to compare the situation before and after a road pricing system has been implemented. Table 3.14. Percentage change of the location of activities in the congestion pricing scenario with respect to the reference scenario 2

Zone 3

4

5

-0.8 -0.8 -1.6 -0.3

0.9 1.5 0.6 2.8

0.4 0.5 -0.1 -0.6

0.7 -0.5 -0.8 -2.1

Activity

Households Workplaces Shops Services

1 (city centre) 0.3 -0.3 8.4 -2.1

The other way is to carry out an ex ante study by analysing what happens in an urban transport – and possibly also a location – model. Such a model could be more or less realistic. In a very abstract model of a city, such as the monocentric model of urban economics, it may be possible to derive some general qualitative conclusions by mathematical arguments. An analysis of this type could indicate, under fairly general conditions, the direction of change of travel behaviour and residential choices as a result of, say, (optimal) congestion pricing. Such conclusions could be of some relevance for the design of transport policies for real cities. The standard monocentric model is an extremely crude representation of a real city, however. First real cities are not monocentric. Moreover, the effects of road pricing depend heavily on what alternatives the inhabitants have. Can they avoid paying the car user charges by changing route, mode, destination or even postponing some trips? The possibilities for extending an urban economics model to include such behavioural mechanisms, and still keep it mathematically tractable,

46

L.-G. Mattsson

are very limited. The possibility that remains is to simulate specific road pricing systems for specific cities by means of those detailed and often sophisticated computer-based travel demand analysis systems that are available in many cities. Although this may give us reasonably reliable results for the cities and pricing systems that are studied, we may still have problems to draw very general conclusions about the effects. By comparing different studies, it may be possible to draw more reliable conclusions about the effects. This is the philosophy behind this chapter. In the first two studies that have been presented, different road pricing systems for Stockholm have been analysed by two different travel demand analysis systems. In the third study, location effects were also included. In that case, the model was not applied specifically to Stockholm. It was rather applied to a stylised completely symmetric city, although inspired by the situation in Stockholm. It should be noted that all applied models are different and have been estimated using independent data sets. If they give similar results, this should increase the credibility of the conclusions. It must be admitted that though the models are different, they still have much in common. The theoretical foundation for all of them is a combination of nested logit models and principles of network equilibrium. Moreover, they are all based on crosssectional data. Panel data describing behavioural changes in response to travel cost increases would be a valuable complementary data source. It is also assumed in all studies that the road users perceive the road charge as a marginal cost and that they respond to this cost as to other marginal costs of driving. Comparing the results from the three studies, there are many striking similarities. First, the road pricing systems are not so different despite the fact that one is zone-based, one is distance-based and one is based on the principle of optimal congestion pricing. The distance-equivalent charge level varies between 0 and 4 SEK/km. It is particularly interesting that the optimal congestion charges varied around the levels that were applied in the other two studies. When it comes to the transport effects, all three studies indicate a substantial reduction in vehicle distance travelled at the assumed charge levels. For the inner city, the reduction is 30% for charged hours at a charge level equivalent to 3 SEK/km for the zone-based system as compared to reductions by 35 and 19% at a charge of 4 and 2 SEK/km for peak and office hours, respectively, for the distance-based system. In the third study based on the stylised city model, the reduction was 25% for an area that also included less congested suburbs. There are also some interesting differences between the studies. The first study predicts larger traffic reductions on the central bridge through the inner city than the second study which can be traced to the fact that the semi-orbital route is less heavily priced in the first study. This stimulates re-routing of trips from the inner city. This result indicates that the precise design of a pricing system is very important in order to achieve the effects that are desired. When there are fewer vehicles on the roads, congestion will go down and speed will go up. The first study predicts that this will result in an increase in speed for inner city roads and arterials during charged hours by around 20%. Both the zone-based and the distance-based road pricing systems that have been analysed would price all vehicle passages through the inner city. Since there are no

Road Pricing: Consequences for Traffic, Congestion and Location

47

alternative unpriced routes between the northern and southern halves of Stockholm, this will reduce the accessibility to the activities in the opposite half of the city in a substantial way. This is clearly illustrated in the first study. The most affected relation is the one between inner northern and inner southern suburbs. In that case, a reduction in the number of vehicle trips by around 30% is predicted. Location effects of transport policies are not so often considered (see Eliasson and Mattsson, 2001, for a brief review of the literature)7. They are included in the third study, however. The general conclusion is that the location effects are small, in spite of the fact that the transport effects are substantial. In the public debate, there has often been a fear that congestion pricing might negatively affect the attractiveness of locating shops into the city centre. This study rather indicates the opposite; congestion pricing increases the number of shops in the city centre. It should be remembered, however, that the location effects reported are based on a highly simplified stylised model of a symmetric city. On the other hand, limited location effects are reported in many other studies (see the review in Eliasson and Mattsson, 2001)8. Road pricing is a controversial transport policy that is much discussed among researchers, planners, policy makers and the public at large. One reason for continuing research is to provide these discussions with better information about the likely effects. Such information has to be as reliable, factual, unbiased and comprehensive as possible. The indicators that have been presented in this chapter are all fairly aggregate. Now, when congestion pricing is on the political agenda again in Stockholm and elsewhere, one of the main questions has turned out to be the distributional effects. Would congestion pricing benefit/harm rich or poor people, men or women, city centre or suburban inhabitants, single persons or families with children, business or private households? The answers to these questions are also related to how the charge revenues would be used. To elucidate these distributional effects of road pricing, including how the revenues could be used to compensate those who otherwise would lose on such a reform, is an important research task that requires much more effort in the future.

References Beser, M., Algers S. (2002). SAMPERS: The new Swedish national travel demand forecasting tool, in: Lundqvist, L., Mattsson, L.-G. (eds.), National Transport Models: Recent Developments and Prospects. Springer, Berlin: 101-118.

7

The EU project PROSPECTS: Procedures for Recommending Optimal Sustainable Planning of European City Transport Systems: http://www-ivv.tuwien.ac.at/projects/prospects.html and its sister projects within The Land Use and Transport Research Cluster: http://www.ess.co.at/LUTR/ represent other interesting research. 8

Wegener (1996) reports small location effects in a simulation of drastically increased petrol prices for Dortmund.

48

L.-G. Mattsson

Eliasson, J., Mattsson, L.-G. (2001). Transport and location effects of road pricing: A simulation approach. Journal of Transport Economics and Policies, 35: 417-456. Goodwin, P.B. (1987). The rule of three: A possible solution to the political problem of competing objectives for road pricing. Traffic Engineering and Control, 29: 495-497. Goodwin, P.B. (1992). A review of new demand elasticities with special reference to short and long run effects of price changes. Journal of Transport Economics and Policies, 26: 155-170. Lindqvist Dillén, J., Eliasson, J., Johansson, G., Sjöberg, B., Ingströmer, P. (2001). A system for improved passability in the Stockholm region. Report no. 5165, Published by the Swedish National Environmental Protection Agency (in Swedish). Mattsson, L.-G. (1995). Road pricing as an instrument in traffic management? An assessment of the proposal for inner city road pricing zones in Stockholm by the Swedish Society for Nature Conservation. Working Paper TRITA-IP AR 95-31, Department of Infrastructure and Planning, Royal Institute of Technology, Stockholm (in Swedish). Mattsson, L.-G., Sjölin, L. (2004). Transport and location effects of a ring road in a city with or without road pricing, in Lee, D.K. (ed.), Urban and Regional Transportation Modeling: Essays in Honor of David Boyce. Edward Elgar, Cheltenham: 113-133. May, A.D., Roberts, M. (1995). The design of integrated transport strategies. Transport Policy, 2: 97-105. May, A.D., Milne, D.S. (2000). Effects of alternative road pricing systems on network performance. Transportation Research A, 34: 407-436. Ramjerdi, F. (1995). Road pricing and toll financing with examples from Oslo and Stockholm. PhD Thesis, Department of Infrastructure and Planning, Royal Institute of Technology, Stockholm. Small, K.A. (1992). Using the revenues from congestion pricing, Transportation, 19: 359-381. Vickrey, W. (1955). Some implications of marginal cost pricing for public utilities. American Economic Review, 45: 605-620. Wegener, M. (1996). Reduction of CO2 emissions of transport by reorganisation of urban activities, in: Hayashi, Y., Roy, J. (eds.), Transport, Land-Use and the Environment. Kluwer Academic Publishers, Dordrecht: 103-124. Widlert, S. (2002). National models: The case of the Swedish national model system: SAMPERS, in: Lundqvist, L., Mattsson, L.-G. (eds.), National Transport Models: Recent Developments and Prospects. Springer, Berlin: 50-56.

4

Implementation Paths for Marginal CostBased Pricing in Urban Transport: Theoretical Considerations and Case Study Results

Erik T. Verhoef1, C. Robin Lindsey2, Esko Niskanen3, André de Palma4, Paavo Moilanen5, Stef Proost6 and Arild Vold7 1

VU University, Amsterdam, the Netherlands Department of Economics, University of Alberta, Canada 3 STA Research, Helsinki, Finland 4 University of Cergy-Pontoise, France 5 Strafica Ltd, Helsinki, Finland 6 Centre for Economic Studies, K.U. Leuven, Belgium 7 TØI, Institute of Transport Economics, the Norwegian Centre for Transport Research, Oslo, Norway 2

Abstract While economic theory provides grounds to believe that a rapid implementation of marginal cost-based pricing would be a sensible policy strategy, in the practice of policy-making more gradual implementation paths (IPs) appear to be favoured. Arguably the most important reason why this should be the case is the existence of barriers and constraints. Surprisingly, however, little or no attention to the design and performance of policy IPs has been given in the transport literature. The purpose of this paper is to develop a structured economic approach to the design and evaluation of such IPs. A second purpose is to apply the proposed approach to analyse implementation paths in the context of urban transport. There is evidently great tension or gap between an idealised approach and what is feasible in applied modelling work using large-scale empirical network models. Four urban case studies are presented, which seek an optimal compromise between the theoretically ideal approach and a pragmatic approach (of ‘fully arbitrary’ IPs). We believe the results are indicative of what one might expect to encounter along a realistic IP, but at the same time acknowledge that the IPs presented are unlikely to represent the best possible path given the local circumstances.

The research summarised in this chapter was carried out as part of the EU Fifth Framework research project MC-ICAM (DG TREN, Contract No. GRD1/2000/25475-SI2.316057). Financial support from the European Commission and helpful comments from Catharina Sikow and from MC-ICAM external experts and partners are gratefully acknowledged. Lindsey would also like to thank the Social Sciences and Humanities Research Council of Canada for funding of the project “Road pricing in urban areas”. The opinions expressed in this paper are those of the authors and do not necessarily reflect the views of the European Commission. This paper draws from earlier, unpublished, overviews that have been presented at various MC-ICAM meetings.

50

Erik T. Verhoef et al.

4.1 Introduction Since Pigou (1920) economists have endorsed pricing measures as a key component of transport policy. The basic idea is well-known: when prices are equal to marginal social costs including external costs, agents face socially efficient tradeoffs with respect to all their travel decisions: numbers of trips, destination, time of day, transport mode, whether to purchase an environmentally-friendly car and so on. Over the last few decades pricing policies based on marginal costs have gradually evolved from a purely academic theoretical construct to a realistic and widely-discussed policy option for many areas – both urban and non-urban – around the globe. The steady growth in transport-related problems such as congestion and CO2 emissions on the one hand, and the development of technologies for automated charging on the other, are important factors explaining this development. At least in theory, marginal cost pricing can be combined with rules for using the pricing revenues in a way that leaves everybody (or at least a large majority) better off. Yet, despite the large and growing assortment of plans to adopt pricing measures, the number of actual applications has remained small. Various types of barriers have been identified that may explain the difficulties that have been encountered in implementing marginal cost pricing (e.g. Niskanen et al., 2003). Awareness of these barriers has recently triggered interest in the design of implementation paths for pricing. If it is indeed not possible to introduce marginal cost-based pricing immediately and throughout the transport sector, is it at least feasible to identify a path along which pricing reform can be introduced gradually? Can a sequence of steps be identified such that each step individually passes the various barriers and together the steps lead to the desired end result? And what principles underlie the design of these steps? These and other similar questions have been addressed in the European Commission funded project MC-ICAM. This chapter will consider these questions in the context of pricing in urban transport, although much of the discussion in Sections 4.2 and 4.3 is also applicable to interurban transport. Section 4.2 starts by discussing barriers and constraints as the primary motivation or necessity for adopting implementation paths – as opposed to ‘big-bang implementation’ – in the introduction of transport pricing policies. Section 4.3 addresses how implementation paths can be formulated in terms of barriers and constraints and solved in a practically feasible way. Sections 4.4-4.6 then present four case studies from the MC-ICAM project that address implementation paths for urban transport pricing. Section 4.7 concludes.

4.2 Implementation Paths: Motivation and Theory Most studies of transport pricing have taken a comparative static approach in which the outcome of a pricing regime is compared with a benchmark or status quo (see, for example, various contributions in Button and Verhoef (1998)). The

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

51

process of transition from the status quo to the new equilibrium is typically ignored even though policymakers are often at least as concerned with this transition as with the eventual long-run design and impacts of pricing policies. Such a phased approach or implementation path (IP henceforth) may seem perfectly reasonable from a political standpoint. But the case for an IP is not logically obvious. Why not follow a ‘Big-Bang Implementation’ (BBI) strategy and implement marginal cost pricing immediately and throughout the transport sector in order to reap the largest possible net social benefits as early as possible? Why move towards optimal prices only gradually? Arguably the most important reasons why BBI is infeasible in practice are barriers and constraints. Project MC-ICAM made an important conceptual distinction between barriers and constraints. According to the definitions adopted by the project, barriers are underlying conditions that may limit the possibilities of designing and implementing a most desired pricing (and broader policy). By contrast, constraints refer to the functioning of the pricing system or the contents of the pricing policy itself. A barrier is relevant to the design of an IP if it brings about a constraint on pricing. To explain this bipartite approach we will first categorise barriers and constraints and offer examples of the causal linkage between each pair of barrier and constraint types. Then we will identify the conceptual and practical reasons for distinguishing between barriers and constraints.

4.2.1 Barriers Various types of barriers that can prevent BBI of optimal transport prices have been distinguished. Table 4.1 shows the main categories, with some examples of each. The entries are reasonably self-explanatory. It is worth noting, however, that barriers often do not exist in isolation, but rather are linked to other barriers in the same or other categories. At a general level acceptability depends on the existing technology and on the legal and institutional status quo. For example, a lack of legal safeguards to guarantee privacy may undermine acceptability. Such interdependencies between barriers make them more difficult to remove than if they were unrelated. Of the six entries under “A. Technological and practical barriers”, all but the first entry refer to practical considerations – although “Difficulty of computing optimal charges…” might be considered a technological barrier in the sense of computing speed or imperfect optimisation algorithms. “Costs of compliance for users” include the transactions costs incurred when dealing with an established and familiar pricing regime such as the costs borne by a driver of maintaining a positive credit balance on a transponder, keeping track of toll schedules and so on. “Adjustment costs” include the costs that households and firms incur in understanding new regulations and pricing instruments, and in adapting to them by changing travel patterns, location choices for residence, work and business, etc. Adjustment costs also include the time and money costs that regulators and administrators encounter while learning how to operate and administer new pricing technologies and systems, and how to enforce regulations.

52

Erik T. Verhoef et al.

While adjustment costs are listed here as a practical barrier, they may also be manifest as an acceptability barrier; for example due to lack of public understanding as per the first entry under “C. Acceptability barriers” in Table 4.1. Given this, and the potential quantitative significance of adjustment costs, it is useful to consider whether these costs provide any intrinsic reasons for preferring an IP to BBI or vice versa. The answer depends crucially on the nature of adjustments costs. If the adjustment costs rise more than proportionally with the size of the adjustment (i.e. they are convex, as is the case with quadratic formulations that have been favoured in theoretical studies) then a series of gradual changes would be less costly, ceteris paribus, than one big change. For example, if implementation is attempted quickly, design errors or system failures may occur that might have been avoided with a slow approach. Collecting data on costs, estimates of demand elasticities and other information required to derive optimal prices is also more difficult or costly in a short time span. Also, the costs and inconvenience of installing inductance loops, overhead tolling gantries and other infrastructure are probably smaller for system operators and users alike if the necessary roadwork is undertaken at a slow and (more or less) steady pace in order to minimise traffic disruptions. The case for BBI, in contrast, is strengthened if adjustment costs have a large fixed or lumpy component. Fixed costs are incurred in installing new types of transponders, in learning about changes in regulations and price schedules, in changing residential or job location in response to pricing, etc. Users are not obliged to keep themselves fully informed or to respond to every change, but they may still incur some costs in deciding not to do so, and if they choose not to reoptimise they may forego some benefits. Clearly, the shape of the adjustment cost function is an empirical question and unfortunately there is very little relevant evidence about it. There is a literature on adjustment costs incurred by firms in making new investments, replacing capital and hiring workers (see Hamermesh and Pfann (1996) for an insightful review). However, despite numerous studies, the nature of adjustment costs is still poorly understood, and the evidence is mixed on whether adjustment costs are convex, concave or discontinuous, and over what range of magnitude of adjustment. Moreover, it is doubtful that the adjustment costs incurred by firms are particularly indicative of the costs involved in the domain of transportation. There seems to be no systematic evidence on adjustment costs incurred from transport pricing reform, and obtaining information is severely complicated by the fact that adjustments to pricing reform are one-off changes rather than repeated adjustments, such as those made by firms in response to the business cycle that afford more opportunity for study. In short, little is known about adjustment costs with respect to transport pricing reform. Consequently, nothing definitive in relation to these costs can be said in favour of either phased implementation or BBI.

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

53

Table 4.1. Classification of barriers to marginal cost pricing A. Technological and practical barriers x Lack of technology capable of charging according to time, location, type of vehicle or user, etc. x Cost and time required to build tolling or other infrastructure x Lack of data on costs and/or demand needed to compute optimal charges x Difficulty of computing optimal charges for large transport networks x Costs of compliance for users x Adjustment costs B. Legal and institutional barriers Legal x Constitutional right to freedom of movement x Laws against discrimination x Privacy laws Organisational structures x Lack of institutions with mandates to introduce new pricing measures x Market power of private operators under Public-Private Partnerships or other arrangements Governments x Inability of government to commit itself or a successor government; e.g. to a particular use of revenues x Difficulties of horizontal or vertical policy coordination between levels of government Political x Resistance from political parties in opposition x Lobby groups exert undue influence on decisions C. Acceptability barriers (Public, political and business interests) x Lack of public understanding of the reasons for pricing or how it will operate x Costs of pricing measures are quantifiable, visible and immediate, whereas benefits are often uncertain and deferred x Costs of pricing measures are concentrated and large per capita, whereas benefits are widely distributed and small per capita

4.2.2 Constraints on Pricing As noted, a barrier becomes relevant when it implies a constraint on the pricing policies that can be implemented. We will distinguish the following five types of constraints on transport pricing policies: 1.

Coverage or ‘scope’ of the pricing system: This refers to which geographical areas, transport modes, user groups, externalities etc. can be priced.

2.

Composition and level of pricing measures: There may be constraints on which types of charges can be implemented concurrently (e.g. cordon tolls and

54

Erik T. Verhoef et al.

area licenses, fuel tax surcharges and gas guzzler taxes). Tolls may be subject to maximum or minimum levels, etc. 3.

Degree of differentiation of prices: Price differentiation may be constrained according to location, vehicle type, trip purpose, mode, time of day or season. Time differentiation may be restricted to a predetermined schedule rather than being responsive to weather, accidents or other circumstances.

4.

Rules and principles governing revenue use: Allocation of revenues may be earmarked to certain uses such as investments in the transport mode or jurisdiction in which they are generated, or to cross-subsidisation of another mode. Some uses of revenue may be precluded, such as lowering distortionary taxes elsewhere in the economy or providing lump-sum rebates to users.

5.

Use of supplementary non-price measures: Marginal cost pricing may be most beneficial if it is accompanied by non-pricing measures such as investments, regulations governing vehicle fuel economy, emissions and safety standards, information provision, and so on. Various constraints may impede deployment of such instruments.

4.2.3 Correspondence Between Barriers and Constraints As noted above, barriers underlie constraints on pricing. Examples of the causal linkages between each major type of barrier and each of the five types of constraint are shown in Table 4.2. The cells give (or directly imply) examples of concrete constraints that prevent optimal pricing from being viable. For example, the first cell A1 of the table pertains to the linkage between technological and practical barriers and the coverage of the pricing system. Acoustic or other engineering considerations may preclude efficient charging of drivers for noise on busy roads with many other noise sources – particularly if the marginal disutility from noise depends on the overall decibel level, acoustic frequency, weather conditions, etc. Practical considerations such as infrastructure costs may militate against comprehensive road pricing, and so on.

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

55

Table 4.2. Correspondence between barriers and constraints Constraints 1. Coverage on: or scope of the pricing Underlying system barriers: A. Technolog- Technologicical and al difficulty practical of pricing barriers noise according to source Excessive cost of tolling all links B. Legal and institutional barriers

C. Acceptability related barriers

2. Composition and levels of prices

Difficulty of charging negative tolls (potentially warranted if unpriced substitutes have high external costs) Freedom of Self-financmovement ing constraint necessitates implies lower toll-free bounds on array of alternatives Free parking charges at residences Price cap or workplaces regulations to prevent exploitation of market power by private infrastructure operators

Public demand for toll-free alternatives Lack of precedent for tolling inner city roads

Toll ceilings or floors

3. Degree of differentiation of prices

4. Rules and principles governing revenue use

5. Use of supplementary non-price measures

Inability to differentiate tolls by location, time, driver characteristics

Excessive administrative costs of targeting compensation to groups most affected by charging

Lack of space to expand road capacity

Law against price discrimination by time of day or trip purpose (e.g. commuting, business and leisure trips) Requirement for harmonisation of charges across jurisdictions

Inability of government to commit to revenue allocation. May require earmarking of revenues either to facility users or to users of the same mode Mandatory cross-subsidy payments from roads to public transport or rail New toll outlays to be offset by lower annual vehicle taxes

Public sentiment against discriminating between groups

Technology for real-time route guidance not ready or too expensive Environmental or other laws prohibiting acquisition of rights of way to build new roads

Local opposition prevents construction of new roads

Table 4.2 makes it clear that a single barrier may be the source of more than one constraint. For example, a public acceptability barrier to road pricing may create a demand for untolled routes (cell C1), impose upper limits on tolls (cell C2), and a requirement that toll outlays be offset by lower annual vehicle taxes (cell C4). Similarly, a given constraint may originate from multiple barriers. For example, differentiation of tolls according to time, location or driver type may be prevented by a technological barrier (cell A3) as well as an acceptability barrier (cell C3). The fact that the causal linkages between barriers and constraints can be many-to-one or one-to-many provides one reason for distinguishing between the

56

Erik T. Verhoef et al.

two concepts. A second overriding practical reason is that the constraints on pricing, use of revenues and supplementary measures are not predetermined and cannot be deduced from a blank slate, but rather must be derived from the particular economic and institutional setting – which is where barriers come in. In some cases the barriers are readily identified and concrete. For example, the capabilities and costs of a given pricing technology may be known more or less accurately. The same is true of laws, and whether or not particular institutions have given legal or administrative mandates. Other barriers, such as those related to public acceptability, may be vague and abstract; for example, the fact that people dislike being monitored while driving. Whether the barriers are concrete or abstract, for the purpose of simulation modelling the resulting constraints must be formulated (from the barriers) in quantitative and mathematical terms. The possibility of constructing an IP arises from the fact that barriers are generally not set in stone, but can be eased or removed given sufficient time and effort. Indeed, policy packages may include measures designed to do just this. Barriers are therefore relevant when considering broader or longer-term policy options, whereas constraints apply in the shorter run when choosing policies within an existing setting or framework. However, as explained in the next section, the formulation of IPs adopted here and in the case studies embraces a more limited ambition in which barriers are treated as given rather than endogenous.

4.3 Formulating Implementation Paths In a world with no constraints, policy instruments could be chosen freely. This is the world assumed in standard economic expositions of marginal cost pricing, and the resulting policy rules are often referred to as ‘first-best policies’. When binding constraints exist, we enter the world of second-best. The challenge is then to set policy instruments in such a way that the constraints are satisfied in the least distortionary way possible. Such policy rules are referred to as ‘second-best’ policies. One important feature of second-best prices is that they typically deviate from marginal costs. It is therefore more correct to speak of marginal cost-based pricing – as in the title to this chapter. From a theoretical perspective the appropriate way to formulate an optimal IP would be to set up an optimal control theory problem in which prices and any supplementary non-pricing instruments are the control variables, and the constraints are embodied in the equations of motion and inequality conditions. But given the inherent difficulties of economic dynamics, and the scale of urban transportation networks, it would generally be a futile task to try to formulate and solve such an optimal control problem. A more practical approach is to define an IP as a temporal sequence of secondbest optima along which the constraints evolve exogenously over time as the underlying barriers gradually erode or are removed in discrete steps. A phase in an IP can then be defined as the time period during which the constraints remain fixed, and hence the pricing rules and the instruments used remain fixed too. A

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

57

phase transition is the moment at which the constraints change. The IPs are thus defined in terms of pricing rules rather than in terms of price levels. For two reasons prices may change not only at a phase transition, but also during phases of implementation (i.e. for given pricing rules) – or indeed also after the final phase transition has taken place. First, the monetary value assigned to external costs may change as incomes rise or as scientific knowledge of health and other effects advances. Second, prices can change as agents adapt by purchasing new vehicles, changing places of residence or work, etc., and as investment and other supply-side policy measures take effect. Both these types of price changes would take place during the normal course of events even without pricing reform, and also if a BBI strategy were followed. Thus, price changes should not be assumed to occur only at times of phase transition along an IP. Four additional important points regarding IPs should be made. The first concerns the synergy or antagonism of deploying different policy instruments in combination. The concept of ‘additivity’ is useful here. Two or more instruments are said to be additive in their benefits if the welfare gain from implementing them together equals the sum of the gains from implementing them separately. Correspondingly, the benefits are sub-additive if the joint gains are smaller than the independent gains and super-additive if the joint gains are larger. When the welfare gains are super-additive (i.e. exhibit synergies) it is advantageous to deploy instruments at the same time rather than to postpone implementation of some to a later phase or not at all. This is all the more true if the adverse welfare-distributional effects of individual instruments tend to be offset by the other instruments. By contrast, if the gains are sub-additive then the case for implementing the instruments at the same time if at all is less clear – particularly if each instrument has large set up or fixed operating costs. A second point regarding IPs is that, even if the barriers underlying pricing constraints are exogenous, welfare at a given phase of implementation may depend not only on the contemporaneous values of the policy instruments, but also on the values chosen for the instruments in earlier phases. If so, the IP exhibits path dependency. Path dependency may arise because of the durability and immobility of buildings and road infrastructure, as well as from inertia in location choices. In principle, this intertemporal linkage between phases should be taken into account in solving the second-best optima, but for computational and other reasons this may not be easy to do in practice. The third point concerns the benefits from eliminating barriers (and hence constraints) to the use of instruments. One obvious approach to this question is to compare welfare at a given stage of an IP with welfare in the BBI in which all relevant instruments are fully deployed. When an IP exhibits path dependency, this comparison becomes conceptually more complicated, as a distinction can be made between welfare at a given stage under BBI when performed from stage 1 onwards, and when performed from the stage under consideration onwards. When multiple barriers contribute to the same constraint, one can expect to find that the benefits of removing barriers become interdependent. The final point is that in applied modelling work it may not be possible to identify strict second-best optima for given sets of constraints. This is especially

58

Erik T. Verhoef et al.

true for large-scale network models, in which the number of potential instruments (e.g. link-based tolls) may cause full optimisation or equilibrium algorithms to be prohibitively time consuming to implement. Likewise, it may not be possible to translate each and every barrier into an exact, quantitative constraint. For example, it is reasonable to assume that, for acceptability reasons, transport taxes would often have some upper limit – but it is hard or impossible to identify ex ante which exact tax level distinguishes ‘acceptable’ from ‘unacceptable’ taxes. In such cases, one could still try to design IPs according to the philosophy discussed above (e.g. one could still aim to identify taxes that are ‘close to’ the second-best optimal ones while being ‘likely to’ satisfy the relevant constraints that are ‘likely to’ apply), but some degree of arbitrariness of course becomes unavoidable. Such ‘motivated arbitrariness’ will also be present in the case studies below, to which we will now turn.

4.4 Description of the MC-ICAM Urban Case Studies The objective of the MC-ICAM project was to investigate how marginal costbased pricing should be implemented in transportation. A theoretical framework for analysing IPs based on barriers and constraints as described above was developed in Niskanen et al. (2003). Building on this framework, case studies were conducted for urban transport (de Palma et al., 2003) and interurban transport (Henstra et al., 2003). Attention is limited in this chapter to the four urban case studies that were carried out for Paris, Brussels, Helsinki and Greater Oslo. This section provides overviews of the studies and the simulation models that were used to implement them. Selected features are identified in Table 4.3.

4.4.1 The Paris Case Study The Paris case study covers Ile-de-France: an area that includes Paris Intra Muros, the suburbs, the new town around Paris, and a dispersed distribution of residences and employment. The number of people making morning trips in the area was 6.8 million in 2002, and is forecast to grow to about 8.5 million by 2012. Computations are performed with the software package METROPOLIS which is designed for dynamic traffic simulation and tracks the movements of individual vehicles in one-second time increments. It also accounts for travellers’ trip-timing preferences by computing the costs incurred by each traveller of arriving at the destination earlier or later than desired (schedule delay cost) and adding this to travel time cost and toll payments to compute generalised travel cost. The road network consists of nearly 18,000 links.

None Fixed at unity

Fixed at 2 per day

Deterministic Continuous time Modelled with exogeneous Origin-Destination matrices

None

Congestion, emissions, accidents, noise, social cost of public funds Flat and time-varying link and cordon tolls, travel time-based tax

Number of links No. time periods/days Auto ownership

Vehicle technology Vehicle occupancy

Number of trips

Route choice Departure time choice Exogeneous longtime dynamics

Endogeneous longrun dynamics

Externalities included

Types of pricing

Modes covered

Population size

Paris 2002: 6.8 m. travellers 2012: 8.5 m. travellers Auto, exog. generalised cost for public transport Approx. 18,000 road links Continuous over a day Fixed

Medium-run equilibrium with endogeneous vehicle stock (private and public) Congestion, emissions, accidents, noise, unpaid parking Full electronic road pricing, single cordon, parking, fuel tax, various public transport fares

Size and fuel type (exog.) Function of mode choice (solo driver or carpool) Fixed per resident and per commuter None Peak/off-peak Exogeneous destinations and locations

Brussels 2005: 1.0 m. residents 0.7 m. commuters Car, taxi, bus, rail, non-motorised One Peak and off-peak Fixed

Table 4.3. Selected features of case study settings and models

Link-based, toll ring, distance-based, zone specific parking charges, fuel tax, public transport fares

Congestion, emissions, accidents

Function of trip type & auto availability (exog.) Stochastic None Study-wide population/employment growth, building regulations, transport infrastructure, GDP growth, prices Building stock, household/ employment relocation

Emissions (exogeneous) By mode (exogeneous)

Helsinki 2000: 1.6 m. residents 2020: 2.0 m. residents Auto/public transport/slow modes, freight 7,500 roads, 440 transit lines 1 hour for peak & interpeak Function of accessibility

Congestion, environmental costs including VOC emissions, accidents Link-based, toll ring, zone specific parking charges, fuel tax, annual car tax, public transport fares, housing rent (endogeneous)

Short run with fixed car ownership, residential and workplace locations; longrun with all variables endogen.

Deterministic None

Thousands (for 7 modes) 1 hour for peak & interpeak Function of fuel costs and annual car taxes Emissions (exogeneous) Fixed for car, endogenous for transit Function of accessibility

Greater Oslo 2002: 1.0 m. residents 2015: 1.1 m. residents All urban modes

Implementation Paths for Marginal Cost-based Pricing in Urban Transport 59

60

Erik T. Verhoef et al.

Public transportation is treated at a rudimentary level by specifying an exogenous and time-independent generalised travel cost that depends on the origin and destination, the type of traveller and idiosyncratic preferences. Each traveller is assumed to make two trips per day and to travel alone if they drive. Trip origins and destinations, the vehicle stock and vehicle technology are treated as fixed. In all, three travel decisions are therefore modelled: a choice of mode and – if auto is selected – a choice of departure time and route. In addition to congestion, the external costs of noise, accidents, pollution and greenhouse costs are accounted for in the welfare assessment. All external costs except congestion are assumed to be proportional to distance travelled. The marginal cost of public funds is set at 0.14 cents per Euro which is the value used for government planning in France. The consequent 14% premium on government revenue is tallied with external costs in the welfare assessment. Pricing instruments include flat (time-independent) and time-varying tolls paid either on individual road links or to pass inside a cordon, and a tax proportional to travel time that approximates the effect of a fuel tax and could be implemented in a few years using GPS.

4.4.2 The Brussels Case Study The case study for Brussels encompasses the area within the outer ring road of Brussels. Unlike the other case studies all pricing scenarios are conducted for a single year (2005). Impacts are assessed with the TRENEN II URBAN model. Unlike METROPOLIS, in TRENEN the road network is modelled as a single link and segmentation of time is limited to peak and off-peak periods. However, it features (exogenous) differentiation of vehicles according to size, fuel type and passenger occupancy. In the medium run the stock of automobiles is adjusted in response to changes in the number of automobile trips taken. Two types of public transport are featured: buses and trams, and metro. Buses and trams use the same road network as cars and suffer from congestion, whereas metro runs on a separate right of way. Service frequencies for both types of public transport are optimised in the medium run, and waiting times at stops therefore decrease with passenger volumes. The set of policy instruments includes comprehensive road pricing (i.e. tolling of the single link), as well as parking fees and time-differentiated (peak/off-peak) public transport fares.1

1

In addition to studying implementation paths the Brussels study examined a scenario in which control of these policy instruments is divided between a city government and a regional government with different constituents and corresponding welfare objectives. The results of this exercise are described in de Palma et al. (2003).

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

61

4.4.3 The Helsinki Case Study The Helsinki case study covers the Uusimaa Region of Finland: an area of 14,400 square kilometres that includes the Helsinki Metropolitan Area, two surrounding cities, and 1.6 million of the approximately 5 million inhabitants of Finland in 2000. Computations are performed with an interactive land-use and transport model based on the MEPLAN modelling framework. MEPLAN incorporates five dimensions of behaviour: workplace location, residential location, automobile ownership, travel mode and route. The transport network includes about 10,000 road links and 200 transit lines. As in the Paris study, the level of transit service is assumed to be independent of ridership. As in the Brussels study, time is divided into two periods, but unlike either Paris or Brussels there is no substitution of demand across time. Auto ownership and numbers of trips are both endogenous. The land-use component of MEPLAN includes residential and workplace location decisions and the building stock, and the re-calculation of equilibria in five-year increments enables analysis of temporal phasing of policies. The suite of pricing instruments is similar to that for Brussels except that parking fees are differentiated by zone, and Brussels does not include a spatial dimension.

4.4.4 The Greater Oslo Case Study The Greater Oslo case study area contains the city of Oslo and surrounding municipalities in Akershus county. A land-use and transport model for Greater Oslo (RETRO) is used to study passenger transport and land use. RETRO assesses four dimensions of travel behaviour: trip frequency, destination choice, mode choice and route choice, for periods with peak and off-peak traffic load. The first three dimensions are described with a nested logit model. RETRO includes sub models for residential and workplace location and for car ownership. Although land use only accounts for periods of peak traffic load, the travel demand sub model can subsequently be used to calculate the resulting off-peak transport costs and demand. In the short run auto ownership and location decisions are fixed, whereas in the long run these and all other variables are endogenous. Besides a detailed road network, network representations are included for buses, tramways, subways, trains, water transportation, bicycling and walking. Service frequency increases with transit volumes so that, as in Brussels, there is a positive ridership externality. In other respects the RETRO model is similar to the MEPLAN model used for Helsinki.

4.4.5 Assessment of the Case Studies From these brief descriptions it should be evident that the case studies differ in a number of respects: their geographical areas and populations, representation of transport networks, dimensions of behaviour modelled, the sets of policy instru-

62

Erik T. Verhoef et al.

ments examined, and other aspects. Paris is distinct in terms of its large population, its abstract representation of public transportation, and a non-zero value for the marginal cost of public funds. Brussels is unique in specifying only a single link for each transport mode and in evaluating IPs for a single year.2 Only the Helsinki and Greater Oslo studies account for land use and residential and workplace location decisions. While the case study areas and models differ in various respects, the studies employ a common framework that facilitates description and comparison. First and foremost, all studies adopt the philosophy behind an IP as described in Section 4.2, i.e. consider it as a sequence of second-best optima. In cases where constraints cannot be formulated exactly, or second-best optima cannot be computed precisely, the choice of policy instruments is based on what was considered a likely result that formal second-best optimisation might have yielded. In principle, a single level of government therefore undertakes to maximise social surplus while accounting for the impacts on all affected parties/stakeholder groups. A second element of the common framework is the monetary values assigned to the external effects of transport. Local values are used where available and where appropriate (e.g. for externalities with localised impacts, such as noise). National or EU-wide values are employed where estimates at a more disaggregated level are lacking, or where external effects are global in nature (e.g. climate change). The IPs are described in the next section.

4.5 Implementation Paths in the Case Studies Table 4.4 summarises the main features of the IPs adopted in each study. The three columns in Table 4.4 correspond to the first three columns of Table 4.2. To economise on space, the last two columns of Table 4.2 are not included in Table 4.4 but relevant considerations are mentioned in the text. Each IP consists of the base year, and 2 to 4 subsequent phases of implementation. For Paris the base year is 2002 and no form of pricing is yet in use. The first stage of implementation is introduced in the same year, and three more phases ensue in 2005, 2008 and 2012. For Brussels all phases are evaluated for the year 2005. Parking fees and public transport fares are set at current levels for the base case. Two alternative scenarios are considered for Phase 1: adjustment of parking fees (Phase 1a) or adjustment of fares (Phase 1b). Both instruments are adjusted in Phase 2, and road pricing is added in Phase 3. For Helsinki the base year is 2000. Pricing instruments are adjusted in 2005, 2010 and 2015, and long-run impacts of the final phase are evaluated in 2020.

2

Since the Brussels study lacks an explicit temporal dimension it would be more accurate to call the implementation path an implementation “plan”. But for ease of reference the term path will be used for all the studies.

Brussels (2005)

Paris

Phase 3

Parking fees as in Phase 1a. Public transport fares as in Phase 1b. Optimal space and time differentiated congestion tolls.

Parking fees as in Phase 1a. Public transport fares as in Phase 1b.

Parking fees + public transport fares Parking fees + public transport fares + congestion pricing

Phase 2

By travel time

Travel time tax set at approximately 50% of equivalent existing fuel price. Current levels. 70% of parking is free.

Public transport fares set at 2nd best levels

By time of day

Time-dependent (single-step) cordon toll

Phase 1b Public transport fares

As 2002

Flat tolls

Parking fees as Phase 1a + fares as Phase 1b. As Phase 2 + tolls differentiated by space and time

No time or spatial differentiation of parking fees Peak/off-peak fares

Paid and unpaid parking

By link

Degree of differentiation of prices

Flat tolls.

Composition and levels of prices

Free parking abolished. Parking fees set at resource cost.

Coverage or scope of pricing system None

Subsets of radial highways & a ring road 2005 All selected highways and ring road 2008 Replace link tolls with inner cordon toll 2012 Cordon toll is replaced with a travel time tax on auto travel Base Parking fees + public transport case fares Phase 1a Parking fees

2002

Year/ Phase 2002

Table 4.4. Implementation paths for the urban case studies

Implementation Paths for Marginal Cost-based Pricing in Urban Transport 63

Radial/cordon fixed road peak charges added. Alternative paths include road, transit or road + transit investments

Cordon replaced by distancebased charge

Parking charges + fuel taxes + toll ring charges + public transport fares As 2002

2010

2015

2002

As 2015 + road link charges

As 2015+

2015+

2030

2015

As 2000. Alternative paths include road or road + transit investments

2005

As 2015+. Instruments and land use adjust gradually from 2015+ to 2030.

First-best measures: road link charges. Second-best measures: as 2015.

As 2015

As 2015

By mode

None

Current levels

Parking charges as in 2002. Fuel taxes either at current levels or with added CO2 charge. Peak toll ring charges & public transport fares set at 2nd best levels with varied constraints (upper & lower limits, equity impacts, and city budgets). Instruments and land use adjust gradually from 2002 to 2015.

As 2010

By mode, link/ location & peak/off-peak

By mode and vehicle type

Degree of differentiation of prices None/minimal differentiation

Towards (practical) maximisation of social welfare. All relevant benefits and costs considered.

Fuel tax & public transport fares adjusted further with more emphasis on social welfare while maintaining fiscal neutrality

Fuel tax & public transport fares adjusted slightly to enhance welfare while maintaining fiscal neutrality & focus on demand management and environment

Coverage or scope of pricing Composition and levels of prices system Fuel tax + public transport fares Fuel tax & public transport fares set at approximately current values

Year/ Phase 2000

Table 4.4. Continued

Helsinki

Greater Oslo

64 Erik T. Verhoef et al.

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

65

Finally, Greater Oslo features two phases of implementation. In the first phase from 2002 to 2015, existing price instruments are used. Instruments and land-use adjustments are assumed to take place gradually over the 13-year-period. At the end of 2015 (2015+) road link charges become available. During the second phase from 2015+ to 2030, all instruments and land-use adjustments are again assumed gradually to adjust. More detailed descriptions of the individual study paths follow.

4.5.1 Paris The IP for Paris consists of four phases in the years 2002, 2005, 2008, and 2012. In 2002, flat tolls are levied on subsets of five candidates: inbound tolls on four radial highways leading into Paris, and tolls (clockwise and counter-clockwise) on a ring road (A86) around the city centre.3 A86 yields a higher welfare gain than all four radial highways together, and on this basis is chosen for the first phase of pricing in 2002. In 2005, the four highways are tolled as well. By 2008, it is assumed that political acceptability constraints have relaxed enough for tolling to be introduced for the first time within the centre of Paris. Tolling of the radial highways and A86 is abandoned, and an inbound cordon around the city centre is introduced. The charge incorporates limited time variation in the form of a single time step (peak and off-peak).4 Finally, in 2012 the cordon toll is replaced by a toll proportional to travel time on auto travel throughout the case-study area.

4.5.2 Brussels In contrast to the other studies, the Brussels case study considered an IP for a single year, 2005. The effects of alternative policy designs are analysed relative to a given forecast traffic equilibrium for 2005. Three policy instruments are considered: parking fees, public transport fares, and road tolls. In the current Brussels setting it is not obvious whether it would be easier for transport policymakers to adjust parking fees or transit fares first, and two alternative choices for the first phase of implementation are therefore considered, called here Phase 1a and Phase 1b (their joint implementation occurs in Phase 2). In Phase 1a, free parking is abolished and parking fees are set at resource cost while public transport fares are kept at their current levels.5 This is a third-best pricing rule because it does not account for congestion and other external costs of vehicle travel. For the 3

Because the only tolls presently in Ile-de-France are on Highway A14 west of Paris Intra-Muros, it was considered that additional tolling would be undertaken first on highways in outlying regions such as these.

4 A peak period of 7:00-10:00 a.m. was chosen on the basis of the timing of congestion for simulations without tolls. 5

Resource cost is the long-run cost of the resources used to provide the parking space, which was estimated to be 1.9 ECU per trip for 1996.

66

Erik T. Verhoef et al.

alternative, Phase 1b, public transport fares are adjusted to second-best levels while parking is kept free. In Phase 2, parking fees and public transport fares are implemented in tandem, with parking fees still set at resource cost and fares set at second-best levels conditional on the parking fees. Finally, in Phase 3 a timedifferentiated (peak/off-peak) road toll is added, with the other two instruments set according to the same rules as in Phase 2.

4.5.3 Helsinki The IP chosen for Helsinki consists of the base case in 2000 and three subsequent phases in 2005, 2010 and 2015.6 Land-use impacts of the final phase are evaluated over a further five-year interval so that the time horizon extends to 2020. A basecase IP was constructed as well as several variants for sensitivity analysis (see below). The base-case IP incorporates four instruments: a fuel tax, public transport fares, peak-period radial/cordon tolls (introduced in 2010), and a distance-based charge (in 2015 to replace the radial/cordon tolls). In 2000 only the first two instruments are in place. Given the high level of transport taxes (e.g. fuel taxes comprise more than 75% of the market price of petrol) it was assumed that further increases in fuel taxes would have to be phased in gradually and accompanied by a reduction in public transport fares. Thus, fuel taxes and fares are set in the first phase at approximately 2000 values, and adjusted progressively – in discrete steps – over the IP. In 2005, the existing instruments are adjusted towards second-best levels while maintaining overall fiscal neutrality. The practical justification emphasises demand management in order to reduce congestion and auto usage. In 2010, peak-period radial and cordon tolls are introduced, dividing the metropolitan area into zones and the instruments are further adjusted to increase welfare. Finally, in 2015 it is assumed that the technology for distance-based tolls becomes available and this form of pricing replaces the cordon tolls.7 For sensitivity analysis several alternative IPs were also evaluated, including a path in which reductions of public transport fares are constrained to the metropolitan area instead of the whole study area, and a path to enhance acceptability with additional investments in road and/or transit infrastructure. The big-bang implementation (BBI) path discussed in Section 4.2 was evaluated too.

6

The IP was influenced by the findings of a committee struck by the Ministry of Transport and Communications in Finland to determine possible changes to transport charges and taxes. The Committee identified as barriers to change many of the elements listed in Table 4.1: technological capabilities, fiscal needs of the transport sector, the current high levels of transport user charges and taxes, a demand for identifiable linkages between charges and levels of service, and more general concerns for equity and distributive justice.

7

Investment costs for the cordon tolls and the distance-based tolls were each estimated at € 50 million.

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

67

4.5.4 Greater Oslo Similar to the Helsinki study, a base-case IP and several variants of it were constructed for Greater Oslo. In the base year 2002 the pricing instruments assumed to be available are parking charges, fuel taxes, toll ring charges, and public transport fares. Parking charges are maintained at 2002 levels throughout the IP. The other instruments are set at their actual levels for 2002 and at second-best levels for 2015 by which time land-use changes are assumed to have run their course. To account in the welfare analysis for gradual changes in transport behaviour and land usage during the intervening 13-year-period, linear interpolation is used to determine values of the pricing instruments and welfare effects between 2002 and 2015. Road link charging is assumed to become technologically and legally feasible at the end of 2015 (denoted 2015+) and is used together with the other instruments. Long-term land-use changes in response to this new package of instruments are assumed to end in 2030. Linear interpolation is again used to determine values of the pricing instruments and welfare effects in the intervening years. All welfare impacts are evaluated relative to the do-min path. Along the do-min path it is assumed that prices are real and that base-case levels of policy instruments resemble today’s situation during the whole period and that the instruments remain the same as in 2002 except that off-peak toll ring charges are set at 10% of the 2002 level throughout the IP. Road and public transport investments in the IP are undertaken following the same schedule as in the do-min scenario. Public transport service frequency is adjusted automatically in response to increases in ridership8. For sensitivity analysis two variants on the base case are considered. In one variant an extra CO2 tax is imposed on fuel. In the other variant additional constraints are imposed that preclude changes in peak tolls outside the range (-100%, +200%) relative to the do-min situation, preclude changes in public transport fares outside the range (-75%, +100%), and prevent either equity or net present value of finance from deteriorating relative to the do-min situation. The CO2 tax and additional constraints variants are considered both independently and jointly, so that three additional IPs are analysed.

8

Data for the base case are based on available statistics and official forecasts, used in several projects led by the City authorities for demographic development, household and employment location for 2015, land-use regulation and infrastructure development towards 2015. Given the lack of information about future infrastructure investments, and the lack of detailed forecasts beyond 2015, the 2015 infrastructure and demographic data were also used for the year 2030. Model representations of the transport networks in 2002 are based on the real world situation, whereas the transport network for 2015 is based on the current network plus planned infrastructure projects and service provision on the PT infrastructure. The RETRO model was calibrated to obtain a perfect fit between model output and data for residential and employment locations in the base case situation in years 2002 and 2015.

68

Erik T. Verhoef et al.

4.5.5 Role of Barriers and Constraints in the Case Study Implementation Paths The structures of the four case study IPs reflect all three types of implementation barriers in Table 4.1 as well as many of the linkages between barriers and constraints in Table 4.2. Technological/practical barriers: Technological and practical barriers play a role in all studies by precluding certain types of charging until later stages of the paths; i.e. the travel time tax for Paris in 2012, road tolling for Brussels in Phase 3, distance-based tolling for Helsinki in 2015, and link charging for Greater Oslo in 2015+. Since these advanced types of charging are assumed to be comprehensive and to permit differentiation of charges according to time and/or distance and/or link, the barrier-constraint relationship corresponds to cells A1 and A3 in Table 4.2. For Greater Oslo an extra CO2 tax on fuel was considered for sensitivity purposes, but not as part of the base-case IP. Such a tax would be practically difficult to implement at a local level because of opportunities for drivers to buy cheaper fuel in neighbouring regions. Legal and institutional barriers: In Brussels free parking is assumed to be widespread in the status quo, but it is abolished in Phase 1a (cell B1). In the Helsinki metropolitan area an agreement governed by a special law and institutional arrangements to coordinate public transport over the borders of the four main cities into a single ticketing system creates a natural institutional barrier to extending the pricing policy outside the area (cells B1). In Greater Oslo, road link charging becomes legally (as well as technologically) feasible in 2015+ (cells B1, B3) Acceptability barriers: For acceptability reasons tolling for Paris begins in 2002 on highways away from the city centre (cell C1). In Helsinki fuel tax increases are accompanied by reductions in public transport fares (cell C4), and investments are considered as a supplementary measure to enhance acceptability (cell C5). In Greater Oslo one of the IPs considered features upper and lower bounds on changes in peak tolls and public transport fares (cell C2), a constraint on equity (cells C2, C3), and a constraint on finances (cell B2).

4.6 Case Study Findings This section first summarises the main modelling results and then discusses some specific issues that were raised in the case studies.

4.6.1 Main Results Table 4.5 presents some of the main characteristics of the IPs. Change in mean automobile speed: Mean auto travel speeds generally rise as pricing reform proceeds, with larger increases realised when congestion is initially severe (i.e. during peak periods) and for pricing measures that are wide in scope.

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

69

The one exception is Paris where mean speed drops slightly in 2002 when a flat toll is applied on the inner ring road A86, and in 2005 when flat tolls are levied on A86 and on the four radial highways (A1, A6, A4, A13). Speeds fall despite the fact that the tolled roads become less congested because these roads sustain higher average speeds than the network as a whole so that a reduction in traffic volumes on the tolled roads brings down the overall average. Nevertheless, user benefits grow (and welfare gains accrue) in both years as discussed below. Pricing revenues: Pricing reform boosts annual transport revenues in most cases. Revenues rise monotonically from phase to phase in Paris and Helsinki. But in Brussels and Greater Oslo revenues drop in one of the phases. In Brussels revenues rise initially if Phase 1a is selected, but then decline in Phase 2 because second-best transit fares are below base-case levels and ridership is price inelastic. If Phase 1b is selected instead, the decline in fares and in revenues is even more precipitous because without elimination of free parking auto travel is priced even further below marginal costs. In Greater Oslo the revenue drop occurs in the last phase after road pricing is introduced in 2015+ as residents and firms relocate in order to reduce their toll and fuel tax outlays. User benefits: Unlike revenues (and price levels) user benefits grow monotonically in every study, except for Greater Oslo in the final phase. 9 The user benefits derive from reductions in travel costs as well as from improvements in service. As explained above, average auto speeds rise in most of the cases. In Brussels, buses and trams speed up because of less road congestion. Both metro service in Brussels and transit service in Greater Oslo improve as service improvements are made. Gains also accrue in Paris from reductions in early and late arrival costs because with less traffic motorists are able to arrive closer to their preferred times. External cost reductions: Benefits accrue to the public from lower external costs of accidents, local pollution, greenhouse gas emissions, noise and dust. In Paris benefits also arise from the marginal cost of public funds premium on extra transport revenues. The benefits of external cost reductions are appreciable, although in every phase of every study they fall well short of user benefits. Welfare gains: Welfare gains (the sum of user benefits and external cost reductions) increase monotonically in every study, both in total and per-capita terms. This is consistent with the definition of an IP as a sequence of second-best optima in which constraints become less binding over time. But the rate of growth of benefits differs across studies.

9

Information required to compute user benefits was not recorded for Brussels except in Phase 3. The same applies to external cost reductions, considered below.

Base year Flat toll on inner ring road Flat tolls on ring road + 4 radial highways One time-step cordon toll Two time-step cordon tolls + travel time tax Base case Parking fees

Policy measures

43 297

9.6%

35

-0.4% 1.4%

20

Additional pricing revenue per capita (€/yr)1

-0.3%

Change in auto speed

253

35

26

13

User benefits per capita (€/yr)

66

9

10

7

External cost reduction per capita (€/yr)

Phase 1a

13% peak, 42 —2 —2 0.3% off-peak Public transport fares 6% peak, -33 —2 Phase 1b —2 0.3% off-peak Parking fees + public 18% peak, 21 —2 Phase 2 —2 transport fares 0.3% off-peak Parking fees + public 71% peak, 148 157 15 Phase 3 transport fares + link tolls 0.3% off-peak 1) Includes toll revenues and fuel taxes with VAT. 2) Not calculated. * Travellers monetary costs and time savings.

2005

2012

2008

2005

2002 2002

Year

Table 4.5. Comparison of selected case-study results

Paris

Brussels

16% 48% 100%

82 172

36%

100%

14%

11%

6%

Welfare gains per capita as % of last phase

27

62

319

44

36

19

Welfare gains per capita (€/yr)

70 Erik T. Verhoef et al.

Policy measures

Helsinki

Greater Oslo

244 (353)

2% (1%)

286*

240 (247)

154

124

User benefits per capita (€/yr)

4.9

161 (156)

92

17

External cost reduction per capita (€/yr)

39% peak, 749 -404* 21 19% off-peak 42% peak, 710 -339* 23 18% off-peak 2) Not calculated. * Travellers monetary costs and time savings.

28

156

1%

9% peak, 0% off-peak

81

Additional pricing revenue per capita (€/yr)1

4%

Change in auto speed

Base year Flat radial/cordon tolls + raise fuel tax + lower public 2005 transport fares Peak-period radial/cordon 2010 tolls + raise fuel tax + lower public transport fares 2015 Distance-based link tolls + (2020) raise fuel tax + lower public transport fares 2002 Base year Parking fees + public trans2015 port fares + fuel tax + toll ring As 2015 with fewer 2015+ constraints + link tolls As 2015+ with land-use 2030 changes 1) Includes toll revenues and fuel taxes with VAT.

2000

Year

Table 4.5. Continued

-6% 100%

233

27%

99% (100%)

61%

35%

Welfare gains per capita as % of last phase

246

62

401 (403)

246

141

Welfare gains per capita (€/yr)

Implementation Paths for Marginal Cost-based Pricing in Urban Transport 71

72

Erik T. Verhoef et al.

The time pattern is quantified in the last column of Table 4.5 by expressing the gains per capita in each phase as a percentage of the gains achieved in the final phase. For Helsinki the gains are fairly equally distributed across the three phases with no phase standing out as a key contributor. The other three studies display less even trajectories, i.e. greater benefits are reaped at the later stages of the IPs considered. For Paris the gains in the first three phases are relatively modest because only a small part of the road network is tolled. Most of the gains accrue in 2012 with the travel time tax – which is paid for all trips. Comprehensive road pricing is also responsible for the bulk of the welfare gains in Brussels (Phase 3) and Greater Oslo (2015+). Costs of delaying full-scale adoption of pricing reform: As discussed in Section 4.3, the costs of delaying full-scale adoption of pricing reform can be determined by comparing the base-case IP with the big-bang implementation (BBI). This was done in the Helsinki study by assuming a BBI in 2005. In presentvalue terms the BBI yields an annual per capita welfare gain of € 227, which is 16% above the € 191 from the base-case IP. However, the BBI is more or less a theoretical construct as it involves large immediate changes in prices, and it is unlikely that the various technological, legal and acceptability barriers could all be overcome at the same time. The case for implementing instruments together as a package: As discussed in Section 4.3, the case for implementing instruments together as a package depends in part on whether the welfare gains are additive, sub-additive or super-additive. The case studies provide examples of additive and sub-additive benefits. 10 For Paris the welfare gains from tolling the four radial highways are approximately additive because they are far apart geographically, and serve outlying areas at different points of the compass, so that traffic spillover effects between the highways are minimal. By contrast, parking prices and public transport fares have sub-additive benefits in Brussels – as is evident from the fact that the welfare gains from Phase 2 fall short of the combined gains from Phase 1a and Phase 1b (see Table 4.5). In Helsinki and Greater Oslo the benefits from the cordon toll and distance-based road pricing are sub-additive. Indeed, by 2015+ and 2030 when land use has responded to the introduction of link tolls in Greater Oslo, the optimal toll ring charges are reduced nearly to zero because link tolls, fuel taxes and public transport fares together are more effective at internalising congestion and environmental externalities. Welfare-distributional effects and acceptability: As is shown in Table 4.5, in most cases user benefits fall short of additional transport payments or revenues, so that users are left worse off. Nevertheless, except for Greater Oslo (see below) user benefits amount to a large proportion of revenues so that in principle only a fraction of the revenues would have to be spent on rebates or useful services to

10

The case studies did not provide notable examples of super-additive benefits although such benefits are logically possible. For example, tolling just one of two roads in parallel is likely to yield a small fraction of the benefits from tolling both roads because it exacerbates congestion on the untolled route (Verhoef et al., 1996; Liu and McDonald, 1998).

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

73

leave users better off.11 In Helsinki the package of pricing instruments deployed results in a relatively balanced distribution of benefits and costs across users, transport operators and government. Higher fuel taxes and peak-period tolls reduce congestion and other externalities, and generate revenues for government, while the benefits for the new and existing public transport users from the fall in fares exceed the increase in car users’ driving costs. The situation is somewhat different for Greater Oslo, where persistent losses from pricing reform are incurred by travellers (because charges exceed the value of travel time savings), public transport operators (because of costly capacity expansions) and parking operators (because of reduced demand). Still, operators of the ring road and road-link tolling systems gain at each stage, and benefits also accrue to government, housing providers and members of the public who suffer external costs. In addition to carefully designed packages of pricing instruments, various supplementary non-pricing instruments have a potential role in enhancing acceptability of pricing reform by lowering acceptability barriers. This possibility was examined in the Helsinki study by considering variants of the basic IP that feature accelerated road and/or public transport infrastructure investments. The transit investments proved to be marginally beneficial in net present-value terms, but the road investments yielded a loss. In part this is explained by the fact that the pricing measures boosted transit ridership at the expense of auto travel. This illustrates the familiar lesson from transport economics that capacity investments are not warranted on strictly economic grounds unless there is adequate demand for new capacity, and consequently that investments can be an expensive way to gain political support – if indeed they succeed at all.

4.6.2 Some Specific Issues Reversals of Direction along Implementation Paths Reversals in the movements of prices or other dimensions of an IP may create financial or other administrative difficulties and endanger acceptability. Reversals of various sorts do in fact appear in all the studies. For example, in Paris tolling of road links is initiated in 2002 and expanded in 2005, but abandoned in 2008 when cordon pricing is introduced. In Brussels transit fares and revenues drop in Phase 1b (or Phase 2 if Phase 1a is implemented first), but rise in Phase 3 when road pricing is launched. In Helsinki introduction of the cordon toll in 2010 induces some residents to locate inside the cordon in order to avoid paying the toll. But the course switches to an out-migration in 2015 when distance-based tolls are brought in and the “boundary effect” created by the cordon disappears. In Greater Oslo there is also a reversal of migration: many people move outwards between 2002 and 2015 in order to avoid crossing the toll ring, and then move back towards the centre after the cordon toll is reduced nearly to zero and road link charges are introduced in 2015+. 11

To assess the welfare-distributional effects properly it would be necessary to identify gainers and losers at a more disaggregated level than was done in the case studies.

74

Erik T. Verhoef et al.

Path Dependency IPs can exhibit path dependency in the sense that the optimal policy to adopt in one phase may depend on policy decisions taken in earlier phases. Such an intertemporal dependency arises if barriers are endogenous, but this possibility was not considered in the case studies. Path dependency can also occur if policy changes induce behavioural adjustments with long-lasting effects. This phenomenon is allowed for in the Helsinki and Greater Oslo cases.12 For Helsinki complete adjustment within the five-year periods of the MEPLAN model is assumed to take place for location choices, but not for building stocks or transport infrastructure. For Greater Oslo land-use decisions by households and firms play out with RETRO over a number of years. The Helsinki and Greater Oslo studies therefore embody “inertia” in their land-use models, and this introduces the possibility of path dependency. 13 The impacts of path dependency were demonstrated for Helsinki by considering an IP in which pricing reform is abandoned after completing the second phase of implementation in 2010. Because the reform measures undertaken in 2005 and 2010 trigger enduring changes in the locations of infrastructure, employment and residences, the transport system does not revert exactly to its state in the donothing scenario by 2015 or 2020. Indeed, welfare ends up lower in 2020. This illustrates rather dramatically how erratic policy initiatives can have adverse economic effects over and above any psychological upsets or political fallout that may result. Welfare Comparison across Case Studies Returning, finally, to Table 4.5 it may seem anomalous that the per capita welfare gains in the final phases of the IPs are smaller for Paris than for Helsinki even though Paris has approximately four times the population. Since congestion is the largest external cost in the studies, and since per-capita congestion tends to increase with city size, one might expect the opposite pattern. A number of differences between the case study areas and models could be responsible for the anomalous result, including differences in 1) population density, road networks and modal splits; 2) public transport ridership and land-use effects; 3) the sets of pricing instruments considered, their coverage and the degree to which they are differentiated by location, time and vehicle type; and 4) the time horizons over which implementation is assumed to take place and so on. 14 It should also be emphasised that the analysis in the case studies was largely exploratory in nature and the quantitative results are sensitive to assumptions. For example, the welfare gains in Paris vary widely with the level of the time-based tax, and the gains for 12

This is not the case in the Paris and Brussels studies because none of the relevant dimensions of behaviour are subject to adjustment lags and (in the case of Brussels) because the stocks of private and public transit vehicles adjust fully to price changes in each phase.

13

These two studies also take into account adjustment costs – as discussed in Section 4.2.1 – to the extent they are “built into” the simulation models used.

14

These and other factors are discussed in de Palma et al. (2003).

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

75

Greater Oslo in the base-case path differ from the gains in the variants with either a supplementary CO2 tax or additional constraints imposed on the policy instruments.

4.7 Concluding Remarks 4.7.1 Summary While economic theory provides grounds to believe that a rapid implementation of marginal cost-based pricing would be a sensible policy strategy, in the practice of policy-making more gradual implementation paths (IPs) appear to be favoured. Arguably the most important reason why this should be the case is the existence of barriers and constraints. Surprisingly, however, little or no attention to the design and performance of policy IPs has been given in the transport literature. We have argued that a rigorous approach to these questions would have to account for a number of theoretical and empirical complications, which, as yet, have not been solved. When the motivation for applying an IP – as opposed to big-bang implementation (BBI) – indeed lies in the existence of barriers and implied constraints, such complications include (1) the identification of dynamic adjustment cost functions (in order to be able to optimise the time pattern of barriers); (2) the identification of the correspondences between barriers and quantitative constraints; and (3) the determination of exact second-best optima for given sets of constraints. Each of these steps becomes even more complex to handle if the IPs exhibit path dependency – as is likely to be the case in reality. One purpose of this paper has been to develop a structured economic approach to the design and evaluation of such IPs. A second purpose has been to apply the developed approach to analysing implementation problems in the context of urban transport. There is evidently great tension or a gap between an idealised approach and what is feasible in applied modelling work using large-scale empirical network models. The four urban case studies presented in Sections 4.4-4.6 have sought an optimal compromise between the theoretically ideal approach and a pragmatic approach (of ‘fully arbitrary’ IPs). We believe the results are indicative of what one might expect to encounter along a realistic IP, but at the same time acknowledge that the IPs presented are unlikely to represent the best possible path given the local circumstances. We see our contribution – and the contribution of the underlying MC-ICAM project – as only a first step towards a more rigorous approach to the economic analysis of transport policy implementation issues. We therefore prefer to devote the remainder of the conclusion to the identification of weaknesses in our approach, and therefore implicitly to chart an agenda for future research.

76

Erik T. Verhoef et al.

4.7.2 Theoretical vs. Practical Approaches to the IPs Numerous types of IPs are possible in practice. The nature of the initial barriers and constraints (the rows and columns in Table 4.2) may differ across applications, and the order in which barriers/constraints can be relaxed may differ too. Furthermore, unless the constraints are particularly tight, prices are not uniquely defined within each phase. For example, even if no differentiation of taxes is allowed there is still a continuum of tax levels to choose from. There are many ways to select subsets of links on a large network to be tolled. For smaller-scale models it may be possible to solve the second-best optimum; i.e., to identify the least distortionary way to satisfy the relevant constraints and – equivalently – to find the highest possible level of social welfare given the constraints on the policy instruments. But for large-scale network models it may be very computationally time-consuming, if feasible at all. In that event, it may be necessary to resort to testing a few plausible policy scenarios and choosing the one that performs best. This was in fact the case for the MC-ICAM urban case studies addressed in Sections 4.4-4.6 above. Such a scenario approach does not, strictly speaking, adopt the theoretical definition of an IP as a temporal sequence of second-best optima along which the constraints evolve exogenously over time, as given in Section 4.3. It is even further away from the theoretical ideal embodied in the optimal control problem with endogenous constraints also identified in Section 4.3. The fact that the case studies nevertheless provided some useful insights suggests that the scenario approach is a useful first step, but methodological improvements on it should be placed high on the research agenda.

4.7.3 Questions of Priority and Timing A number of questions about the IP remain largely unanswered by the case studies. Perhaps the most important is the question of priority: in what order should marginal cost pricing be implemented by geographical region, transport mode, links on a network, times of day, etc? There are obvious advantages to beginning with policies that yield the largest benefits, with BBI as the polar extreme. But because such policies tend to be comprehensive and to feature differentiation by time, vehicle type and so on, they are also more costly and difficult to realise. The general lesson from the case studies is that very little can be said in this respect without detailed information on the time pattern of both the induced costs and benefits. A related question concerns timing: should one implement whatever is feasible now, or should one wait until a more comprehensive, and possibly more gainful, implementation is possible? The answer of course depends not only on the time patterns of costs and benefits, but also on political uncertainties and ‘windows of opportunity’. Moreover, the answer also depends on whether or not the technology used for early-phase implementation would lend itself for future refinements (i.e., will sunk costs be lost when moving to a subsequent phase along the IP?). The approach being taken in the US under the Value Pricing Initiative is to fund a

Implementation Paths for Marginal Cost-based Pricing in Urban Transport

77

series of small demonstration projects in order gradually to gain public acceptance. In the case of London’s congestion charging scheme, in contrast, the authorities attached great importance to initiating some form of pricing quickly while a window of opportunity remained open. Because of the limited size of the charge area, the high costs of toll collection, and exemptions that have been granted to a large fraction of vehicles, the net benefits from the existing scheme appear to be relatively modest (Transport for London, 2003; Turner, 2004). If the charged zone is expanded (as it is planned at the time of writing) exemptions are reduced or eliminated, and electronic toll collection technology is introduced, the benefits may increase substantially.

4.7.4 Other Limitations of the Case Studies The case studies have various other limitations that reflect the early stages of research on the practical aspects of transport pricing reform. One such defect is that, with the exception of investment costs for Helsinki, the studies did not include estimates of the infrastructure and operating costs of charging systems. Obtaining accurate information on these costs is difficult – particularly for technologies that have not yet been implemented anywhere in the world and may yet be under development. Second, with the partial exception of the land-use models used for the Helsinki and Greater Oslo studies, the studies did not account for the fact that physical and institutional investments in pricing reform are partially or completely sunk, and that private decisions regarding land use, location choice, automobile ownership and so on are long-lived and costly to alter or undo. A third limitation is that market failures outside the transport sector are disregarded, and consideration of how transport revenues are used is limited to the inclusion in the welfare assessment for Paris of an exogenous marginal cost of public funds. The studies also ignore any policy impacts on regions outside the areas studied. Yet another limitation is that the theoretical approach underlying the case studies does not address how various uncertainties should be accounted for in the choice of IP. Of particular significance are uncertainties about the barriers and implied constraints to pricing. For example, it is far from clear how high tolls could be raised on a particular road or category of road without provoking widespread resistance from the affected population of users. It is even less apparent how rapidly the constraints will evolve (and, one can hope, diminish) with time as users and administrators gain experience with pricing. All these drawbacks notwithstanding, we believe that the approach to implementation of transport pricing reform described in Sections 4.2-4.3 and illustrated in Sections 4.4-4.6 is promising. Better understanding of barriers and constraints at both a theoretical and empirical level will be needed to take it forward. It is to be hoped that researchers will be encouraged by the initial efforts described here to take up the challenge.

78

Erik T. Verhoef et al.

References Button, K.J., Verhoef, E.T. (eds.) (1998). Road Pricing, Traffic Congestion and the Environment: Issues of Efficiency and Social Feasibility. Edward Elgar, Cheltenham, UK. de Palma, A., Kilani, K., Lindsey, R., Moilanen, P., Proost, S., Sen, A., Vold, A., Niskanen, E. (2003). Welfare Effects – Urban Transport. MC-ICAM Deliverable 7. http://www.its.leeds.ac.uk/projects/mcicam/reports.html Hamermesh, D.S., Pfann G.A. (1996). Adjustment costs in factor demand. Journal of Economic Literature, 34(3): 1264-1292. Henstra, D., Andersen, J., de Borger, B., Bovenkerk, M., Calthrop, E., Ivanova, O., Jin, Y., Proost, S., Raha, N., Traa, M., Vold, A. (2003). Marginal Cost Pricing Implementation Paths in Interurban Transport: Evaluation of welfare effects. MC-ICAM Deliverable 8. http://www.its.leeds.ac.uk/projects/mcicam/reports.html Liu, L.N., McDonald, J.F. (1998). Efficient congestion tolls in the presence of unpriced congestion: A peak and off-peak simulation model. Journal of Urban Economics, 44: 352-366. Niskanen, E., de Borger, B., de Palma, A., Lindsey, R., Nash, C., Rouwendal, J., Schade, J., Verhoef, E.T. (2003). Phased Approach. MC-ICAM Deliverable 6. http://www.its.leeds.ac.uk/projects/mcicam/reports.html Pigou, A.C. (1920). The Economics of Welfare. Macmillan, London. Transport for London (2003). Congestion charging 6 months on. October, http://www.tfl.gov.uk/tfl/downloads/pdf/congestion-charging/cc-6monthson.pdf. Turner, D. (2004). Central London congestion charging scheme: Has it achieved its objectives? Presentation at Round Table on Road Pricing in Belgium, Catholic University of Louvain, February 3. http://www.econ.kuleuven.be/ete/downloads/Round_Table_Turner.pdf Verhoef, E.T., Nijkamp, P., Rietveld, P. (1996). Second-best congestion pricing: The case of an untolled alternative. Journal of Urban Economics, 40(3): 279-302.

5

The London Congestion Charging Scheme: The Evidence

John Peirson and Roger Vickerman University of Kent, Department of Economics, Canterbury, UK

Abstract In February 2003, the Mayor of London introduced road pricing for driving in a small area of Central London. The London Congestion Charging Scheme uses relatively simple technology, was implemented over two and a half years, faced political opposition and required the efforts of a determined political champion who refused to be put off. The £5 charge has reduced car movements by about 30%, and increased bus use and traffic speeds. The unexpectedly large reduction in car use is partly explained by the possibility of taking routes around the charging zone. For this reason, the implied elasticity of about -0.8 is greater than many other previous elasticity estimates. It is suggested that extending the Scheme to a larger area or applying a similar scheme to certain other cities may not be quite so successful.

5.1 Introduction On 17 February 2003, the Mayor of London introduced the first major urban road charging scheme in the United Kingdom. It is estimated that the London Congestion Charging Scheme (LCCS) has reduced traffic congestion delays in the charging zone by about 30%. The consensus opinion has been that the scheme is successful and has encouraged other major urban areas in the United Kingdom and the world to consider the introduction of road pricing. This chapter describes the LCCS, examines its impacts and considers the relevance of this evidence to extending the charging area and the use of road pricing in other urban areas.

5.2 Background to the LCCS In considering the successful setting up and operation of the LCCS, it is important to understand the political background to urban road pricing in the United Kingdom. It is common to find suggestions that traffic speeds in Central London are at the same level as those a century ago (House of Commons Committee on Transport, 2003). Since the early 1960s, it had become clear that the unrestricted

80

J. Peirson and R. Vickerman

rapid increase in the use of urban roads would result in such high levels of congestion that it would have serious impacts on the quality of urban life and economic activity. As part of a solution to such problems, reports to Government such as the Buchanan Report (1963) and the Smeed Report (1964) both proposed charges for the use of urban roads. However, this proposal was not pursued seriously until the mid-1990s. In 1995, the London Congestion Research Programme, see Government Office for London (1995), considered a number of different proposals to reduce congestion and predicted substantial benefits. In 1999, the Labour government set up the Greater London Authority (the local government authority for London) with the expectation that they would be the major political party in control of the authority. At the same time, the Labour government’s support for road pricing had been curtailed as a result of political pressure from groups representing motorists and freight hauliers following substantial rises in the price of fuel.1 In the Transport 10 Year Plan (Department for Transport, 2000), the government encouraged local authorities to bring forward road charging proposals where the revenue could be used to fund local transport expenditure. In London, under the legislation which set up the Greater London Authority, the Mayor was allowed to introduce congestion charging on roads.2 In order to inform the new Mayor, a technical report on Road Charging Options for London (ROCOL) (Government Office for London, 2000) had been published prior to the Mayoral Election. The election of the Mayor in 2000 became a major political event with a Labour politician, Ken Livingstone, standing against the official Labour candidate and winning with 58% of the votes cast. An important element of the new Mayor’s manifesto was to reduce road traffic by 15% and “consult widely about the best possible congestion charge scheme to discourage unnecessary car journeys in a small zone of central London, to commence during the middle of my term of office, with all monies devoted to improving transport” (Livingstone, 2000). The new Mayor strongly supported a London congestion charging scheme and brought forward proposals soon after being elected. This scheme and a later overall transport strategy were subject to a consultation process with the public, local councils, businesses and representatives of road users. The scheme was modified over the next 18 months, in particular with respect to discounts and exemptions. The final scheme was adopted in February 2002, only one year before its implementation. An early 2001 MORI opinion poll found that 51% of Londoners were in favour of the scheme and 35% against, but a 2002 poll by the Committee for 1

An example of the Labour government being unwilling to increase the cost of road use was the removal in 2000 of the fuel price escalator which increased the tax on fuel to make fuel prices increase by 3% per annum over the rate of inflation.

2

Similar powers were made available to other local authorities, although to date only one other, Edinburgh, has made any progress towards the introduction of such a charge, planned for introduction in 2006 subject to the outcome of a public inquiry and subsequent referendum in 2004.

The London Congestion Charging Scheme: The Evidence

81

Integrated Transport (2002) found an approval rating of only 30%. This figure was increased to 58% when respondents were told that all of the revenue would be used to improve public transport. In the run up to the introduction of the LCCS, the ruling Labour government did not positively support the Scheme and the Conservative Party had voiced strong objections to congestion charging in London from the initial proposals. The London Evening Standard, the major London newspaper, had objected to the LCCS and with Transport for London conducted a heated dispute over the benefits and costs of the scheme. Finally, many businesses and road user groups raised objections to the scheme (see, for example, the views of the Automobile Association, 2000). In conclusion, the political background to the introduction of the LCCS was that the Mayor of London took a political gamble on the Scheme working successfully and in his own words “it is bound to be messy, it is bound to involve mistakes, but in the end everyone will look back and agree it was a good idea and they won't be able to recall why they ever resisted the change….Yet changing transport patterns that have built up over 30 years will clearly take time and politicians with skins thick enough to live with short term unpopularity and lashings of criticism” (Livingstone, 1998). A conclusion to be taken from the experience of the London and most other road pricing schemes is that the successful implementation of road charging requires the support of a bold political champion; see the review by Short of the OECD evidence on road pricing (ECMT, 2004). The implementation of LCCS took account of the need for secure political backing, careful research of the feasible options, improvements in public transport and implementation of traffic management measures (Dix, 2002). The ROCOL research programme (Government Office for London, 2000) was a careful and detailed examination of road pricing through policies of area licensing and work place parking schemes. Different options were considered with regard to the geographical extent of possible schemes, levels of charges and issues with regard to those who would gain and lose from different schemes. The issue of technical feasibility was particularly important. It was suggested by ROCOL that a paper licensing scheme, enforced visually by inspectors, would result in a high level of non-compliance. An alternative of electronic road pricing using onboard equipment was believed to face technical operational difficulties; despite the existence of such schemes elsewhere, it was felt that London would present different problems. The LCCS implemented is very close to one of the policy options given in ROCOL. One month after election, the new Mayor of London set up a team to investigate congestion charging and develop the means to implement a charging scheme, e.g. specific project management functions were allocated. This team successfully implemented the LCCS in two and a half years. Part of the success of the perception of the LCCS is that it was combined with planned improvements to the bus, underground and rail services, changes to the road network and a major overhaul of traffic management in London (see Dix, 2002, and London Assembly, 2004). These developments were planned to be funded through an increased central government settlement, net revenues from the LCCS and increased spending on transport by the Greater London Authority.

82

J. Peirson and R. Vickerman

5.3 The Working of the LCCS3 The LCCS operates in a small area of Central London that covers only 21 square kilometres. This represents only 1.3% of the total area of Greater London and the road ringing the charging zone is not part of the charging area, see Fig. 5.1. The charge is £5 each weekday for road users who drive into or within this central area of the capital during the period from 07.00 to 18.30. Over 200 cameras positioned on the zone's entry points, and some within the zone, match car number plates against a database of vehicles whose drivers have paid the charge. Any motorist who has not paid by the end of the day is fined £80. The proportion of users issued with penalty fines is about 7% of the total who should pay.

Source: Transport for London (2004). Fig. 5.1. The London Congestion Charging Zone

Road users can pay through retail outlets, the Internet, call centre, mobile phone text messaging and the post (one year after operation the split of payments was 36%, 26%, 19%, 19% and 1%). The charge can be paid in advance or up to 22.00 on the day itself, but not in arrears. Regular drivers can register for fast track payment to speed up the process and there is a procedure for fleet registration. 3

This paper refers to conditions during the first year of the operation of the zone. The charge was increased and the zone extended in early 2007.

The London Congestion Charging Scheme: The Evidence

83

Taxis, motorcycles, buses, emergency vehicles and certain alternative fuel vehicles are exempt from paying the charge. Residents of the congestion charging zone can register for a 90 per cent discount and disabled road users are eligible for a 100 per cent discount. It should be noted that the charge may be claimed as a legitimate expense by some road users. The goals of the LCCS were, as reported in Transport for London’s (2001) Transport Strategy: x to reduce congestion in and around the charging zone x to improve journey time reliability for car users x to make radical improvements in bus services x to invest in the Underground x to make the distribution of goods and services more reliable, sustainable and efficient x to generate net revenues to improve transport in London more generally.

5.4 Theory of Area Pricing In theory, road users should pay for the full social cost they impose on society. Thus, road use prices should reflect the externalities of congestion, air pollution, accidents, road damage, noise, etc. A major problem is that there are no policy instruments available that can measure and make road users pay the full social costs as they are incurred. One is thus forced to adopt measures that will necessarily fail to achieve the full welfare optimum of this first-best solution (see De Borger and Proost, 2001). The LCCS is a scheme of area charging or licensing where road users pay for driving within the area as opposed to cordon pricing where users have to pay for crossing a boundary line. Other possible policies are workplace parking charging and distance-related charging (measured electronically). Important determinants of the full social costs are the distance travelled and the time of travel. It is shown in De Borger and Proost (2001) that a simple cordon pricing system adjusted for time of travel in the Greater London area would only achieve a small proportion of the welfare gain of a pricing system that fully differentiates between distance travelled and time of travel, etc. This result compares with smaller cities where the proportion of the welfare gain is much greater, e.g. in Brussels. In the exercises carried out in De Borger and Proost (2001), charging for the full resource costs of driving in the inner London area generates much of the full potential welfare benefit. It should be noted that such charging would cause important switches in use of transport modes, but the revenues raised would only be modest. Estimation of the full social costs is difficult and can only be taken as a guide. Santos and Shaffer (2004) suggest that the fixed congestion charge of £5 is consistent with their estimates of the marginal congestion cost, though this implies

84

J. Peirson and R. Vickerman

that the charge is only concerned with covering congestion costs within the charging zone and not those costs imposed in inner and outer London. Thus, the economic theory and evidence suggest that road pricing in London should focus on the most congested area and differentiate by time of use. Potentially important switches from private road use are achievable, but the amount of revenue is unlikely to be substantial.

5.5 Evidence on LCCS Transport for London is committed to long-term monitoring of the impacts of the LCCS (see Transport for London, 2003a) and making available the associated data. Nearly all the following statistics come from the subsequent monitoring reports (Transport for London, 2003b, 2003c, 2004). The major conclusions of these reports are: x Traffic flow in the charging zone and period has been reduced by about 15%. Car movements into the zone have been reduced by 30% with taxi, bus and coach movements all increasing by 20%. Van and lorry movements have been reduced by 10%. Pedal and two-wheeler movements have increased by 1020%. The average traffic delay has been reduced by about 30%, resulting in an increase in average speed of 1.7km/hr, with average speeds of the order of 1617 km/hr. Journey time reliability has improved by an average of 30%. x There are about 110,000 payments per day. There are about 65,000 fewer journeys crossing into the charging zone. Between 20 to 30% of this reduction is made up of car journeys diverting around the charging zone, 50-60% is from former car users using public transport and the remaining 15-25% comes from other adaptations in travel behaviour. It is estimated that there are 4,000 fewer individuals per day travelling to a destination in the charging zone. x Traffic flow on the Inner Ring Road bounding (but not in) the charging zone has increased by 4-5% with local variations. There are no major changes in orbital or radial traffic flows outside the charging zone, with increases and decreases reported at different sites. The changes in traffic behaviour settled down quickly within the first three months and there has been little further change over the rest of the first year of operation. x The fall in the number of accidents of different types has been greater than the reductions observed elsewhere in London and is in line with the reduction in traffic flows. x There has been a continued increase (from before the introduction of the LCCS) in the use of buses, with an estimated 29,000 additional bus passengers entering the zone in the morning peak since the introduction of charging, compared with a forecast increase of 15,000. Transport for London has provided more buses, but there has been an increase in the number of passengers per bus. Bus speeds have improved by 7%. Excess waiting time for buses has improved by about one-third in and around the charging zone.

The London Congestion Charging Scheme: The Evidence

x

x

x

x

85

The use of the Underground to travel to the charging zone has fallen by 5-10% which compares with an approximate 2-7% reduction in use across the entire network. This is believed to be related more to problems with the Underground service as during this period parts of the Central Line were closed following an accident, there was a reduction in visitors to London given fears over international security and a general decline in retail activity. Transport for London argued that congestion charging had made little contribution to reduced economic activity in Central London. However, the impact would appear to vary across sectors with the retail sector suffering the greatest loss of business.4 This may, however, reflect the general problems faced by the retail sector in Central London given the downturn in visitor numbers and problems with the Underground mentioned above. The set-up costs for the LCCS were reported as £58.2 million. The net annual revenue was initially estimated at £200 million, but this was reduced to £121 million just before the start of the LCCS. With the introduction of LCCS, there was an unexpectedly high reduction in the number of vehicles entering the charging zone, fewer fleet vehicles using the automated scheme than predicted and more evasion. The most recent forecast of net revenue is £68 million based on gross revenue of £165 million and operational costs of £97 million. In the future, net revenue is estimated to be £80-100 million per year. These estimates are based on lower running costs and greater income from stricter enforcement. Penalties for non-payment were increased in early 2004. The lower revenue does present a problem for Transport for London as it provides less than expected support for the improved bus services. Though analysis of the distributional impacts of the LCCS was carried out before the introduction of the scheme, no current analysis is available as such effects take time to emerge and require detailed survey data.

5.6 Interpretation of the Evidence The most surprising result of the LCCS is the remarkable reduction in car traffic. It is important to consider whether this reduction fits with the previous results about the responsiveness of the demand for car travel to change in the price. The car traffic outcomes from the LCCS are above the upper range of the predictions of the Scheme’s promoters and this suggests that the Scheme may require us to reappraise the responsiveness of car traffic to prices. The most commonly cited review of the

4

See the conclusion from the Royal Institute of Chartered Surveyors (2004) “whilst the introduction of the charge has been remarkably smooth and the overall impact on the residential and office sectors has been broadly neutral or even positive, the retail and leisure sectors do appear to have been adversely affected.” A more recent report (Royal Institute of Chartered Surveyors (2005) confirms that whilst nine out of ten retailers report some loss of turnover, there appeared to have been little or no impact on either moves to relocate outside the zone or on land values or rents.

86

J. Peirson and R. Vickerman

relation between money price and car travel demand is that of Goodwin (1992) who concluded that the money price elasticity is in the range -0.16 to -0.33. Santos and Shaffer (2004) suggest that the evidence from the LCCS is in line with the lower end of this range. The present analysis suggests that interpretation of the evidence is more complex than may be first thought. First, it is necessary to make an estimate of the price of car travel. The UK Department for Transport Economic Model gives a 2003 marginal price of 13.59p per vehicle-km. Alternative estimates of the average price that include fixed costs such as insurance and vehicle excise duty may double the price, see Santos and Schaffer (2004).5 This marginal price can be used with trip distance data to work out the price of an average car trip. This is complicated as car trips lengths depend on the journey that is being made, e.g. Outer to Central London journeys are much longer than Inner to Central London journeys. Additionally, nearly half of all journeys previously passing through Central London did not have the Central Area as the origin or destination (see Department of Transport, 1994). Thus, adjusting for this gives an average trip length of a journey that goes through the Central Area of nearly 20 kilometres compared to the 13 kilometres for journeys that start or end in the centre6. The appropriate distance to use in an elasticity calculation is double this number as on most days cars return to their original origin. Finally, one should note that cars may make multiple trips in a day through the Central Area, an omission that will result in underestimation of the price elasticity. Transport for London estimates that car vehicle kilometres have been reduced by 34%. However, many car owners are exempt or obtain discounts from the charge and presumably, with lower congestion levels, they will travel further than in the past. Using Transport for London estimates and amalgamating private and work cars, suggests that the true reduction in car vehicle-kilometres for nonexempt or non-discounted cars is in the region of 42%. Using a constant elasticity of demand function, these calculations imply a price elasticity of demand of -0.82 a much larger estimate than traditionally associated with private car travel. In examining the response of traffic to the LCCS, it is important to remember that, for nearly one-half of the trips which pass into the Central Area, an alternative to paying the charge, not travelling or using some other form of transport is to travel round the Inner Ring Road. In the charge period, traffic on the Inner Ring Road has increased by about 4-5%. On a simple assumption that such car traffic travels around one-quarter of the inner ring road, this increase will represent 28,000 displaced journeys through the Central area.7 In addition, other car users who previously drove through Central London may choose routes not including the Inner Ring Road and their presence cannot be detected in traffic flow data further

5

The use of a higher average price will give higher elasticity estimates as the relative price increase from a congestion charge is smaller. 6

Own calculation using data supplied by Department for Transport.

7

Transport for London suggests that such road users are about half this figure.

The London Congestion Charging Scheme: The Evidence

87

from the Central Area. This analysis suggests that changes in transport behaviour are complex and cannot always be examined with simple notions of elasticities. The use of the revenue from the LCCS to fund transport improvements was important in securing public approval or acquiescence to the initial proposal (see Automobile Association, 2000, and Committee for Integrated Transport, 2002). After the successful implementation of the LCCS, the major negative reporting by the media was on the failure to raise the predicted net revenues to fund planned investment in the London transport system, e.g. see London Assembly (2004).

5.7 Extension of LCCS Area After the successful introduction of the LCCS, Transport for London started to consider a proposal to enlarging the charging zone to the West (Transport for London, 2003d). The enlargement raises various issues. The analysis of the responsiveness of traffic to a change in the price might be taken to imply that a similar reduction in traffic will not take place as the diversion of traffic from a larger zone may not be so feasible. However, it is proposed that there will be an uncharged road (a section of the Inner Ring Road) going through the enlarged zone making it still possible to drive through Central London, though it should be remembered that this uncharged road will only allow traffic through on North-West to South-East axis. The proposal would also probably either abolish or reduce the residents’ discount as the additional area contains many more residents. Although it is suggested that the reduction in traffic may not be so substantial as that found in the initial LCCS, compensation for this will derive from the availability of larger revenues and sharing of the fixed costs of setting up and operating congestion charging in a larger area. The enlargement would not take place before 2006.

5.8 Relevance to Other Urban Road Pricing Schemes Many cities, both in the UK and the rest of the world, have been encouraged by the success of the LCCS, e.g. see The Observer (2004). The relevance of the Scheme to other urban areas has to be carefully considered. London has a large area that is highly congested, other urban areas may have a much smaller congested area, for example, Durham has road pricing for a single street. The effect of road pricing in a small area may be to divert much more traffic away to the surrounding road network. The proposed Edinburgh scheme, which is the only large area-wide scheme currently proposed in any detail, uses a double cordon system8. The inner cordon is 8

For further details of the scheme, see Transport Initiatives Edinburgh Ltd/The City of Edinburgh Council (2004). Although the proposal was supported by the conclusions of a public inquiry in

88

J. Peirson and R. Vickerman

set around the central area (much smaller than the LCCS area) and operates throughout the working day from 0700-1830, the outer cordon is set roughly along the line of the city’s Outer Ring Road (a much larger area than LCCS) and would operate in the morning peak only (0700-1000). A charge of £2 is incurred by any vehicle passing either cordon in the in-bound direction; the charge is imposed at the first operational cordon crossed. The reason for the double cordon is that a charge for the central area only would not have a significant impact on traffic entering the city during the peak and forecast increases in peak congestion are greatest at the edge of the built-up area. However, congestion remains high in the central area throughout the day and thus some deterrent to driving right into the centre is felt necessary at all times. The double cordon is predicted to achieve twice the reduction in congestion of a single central area cordon over the first five years of operation. By comparison with London, the Edinburgh scheme is a cordon scheme that does not charge vehicles moving entirely within the chargeable zone or those travelling out-bound from the city during the charging period. It is therefore much more focused on commuting traffic than the London scheme. The proposed technology is similar, and charging in one direction only simplifies operation of the cameras. As in London, the proceeds of the charging scheme are to be channelled into a package of public transport investments, although in Edinburgh there is an emphasis on capital investment. If congestion is caused by subsidised parking, it may be more efficient to charge the true cost of parking. This was an option considered by Transport for London. However, it should be noted that congestion is mainly caused by moving and not parked vehicles. Thus, the actual type of road pricing system adopted by other urban areas and the impact may be quite different from London. The important lessons from London are to recognise the importance of secure political backing, careful research of the feasible options, improvements in public transport, implementation of traffic management measures and use of the revenue to fund the transport system. These conclusions are similar to those suggested in ECMT (2004).

5.9 Conclusions The London Congestion Charging Scheme has shown that it is possible to introduce successfully an area-wide charging scheme for the Central Area of a large city with complex existing travel patterns, using relatively simple technology. What has surprised most observers is the impact that the scheme has had on private car traffic levels in the charging zone. Reductions in both levels of traffic entering the zone and in the levels of congestion experienced have been above the upper level of the prior forecasts. In this chapter, we have reviewed this evidence and

2004, it was significantly defeated in a local referendum in February 2005 and currently remains in abeyance.

The London Congestion Charging Scheme: The Evidence

89

presented some simple calculations which suggest that the price elasticities implied in this reduction are significantly higher than those claimed in other studies, and significantly higher than the received wisdom on own price elasticities for car travel. The higher estimate, of about -0.8, derives from two factors. One is the method of calculation of the relevant prices where we argue that it should be based on the marginal costs of driving and not the average costs of car use, since we do not believe that for most users the imposition of the charge at its current level will change their decision about car ownership. Secondly, the imposition of a charge of this type, which requires drivers to confront directly their decision to make a specific journey, or take a specific route, has a different behavioural impact from the general increase in the cost of car use observed in most previous studies which have looked at, for example, the increase in fuel prices which affect all journeys made. In deriving this estimate, we have also tried to examine in more detail, under some simple assumptions, what we think has happened to the suppressed car journeys. There has clearly been some significant mode switch to bus, although some of the increase in bus use has resulted from the improved bus service leading to switching from the Underground or from walking. There has been some diversion of route, although not to the extent that this is simply moving congestion to outside the charging zone. There must also be some reduction in travel in total; what remains to be seen is whether this reduction in travel has its most significant impact on economic activity in the charging zone, particularly in retailing, and whether this is an overall reduction in expenditure or a redistribution within the Greater London area. Despite the remaining uncertainties, it is clear that the LCCS has been more of a success than many anticipated; acceptance ratings in the business community remain high at about 60%; the 30% who believed in advance it would not be effective have reduced to 20% who currently do not support the scheme, see Transport for London (2003a, 2004). This success augurs well for the introduction of further schemes in the UK as more authorities see this as a real means of tackling congestion in a variety of situations, and moreover one which provides additional revenues which can go towards a wider range of transport improvements. The London scheme, and those likely to follow, are still however second best schemes which only go part way towards the theoretically desirable outcome of a full electronic road pricing scheme sensitive to real time information on the costs of driving in a congested area. Even introducing this partial scheme required considerable political courage from a maverick politician. How many more will have the courage to do this, let alone go further, remains an open question at this time.

90

J. Peirson and R. Vickerman

References Automobile Association (2000). Congestion Charging in London – The Views of the AA. London. Buchanan, C. (1963). Traffic in Towns: A study of the long term problems of traffic in urban areas. Reports of the Steering and Working Groups to the Minister of Transport. HMSO, London. Committee for Integrated Transport (2002). Public Attitudes to Transport in England. London. De Borger B., Proost, S. (eds.) (2001). Reforming Transport Pricing in the European Union. Edward Elgar, Cheltenham. Department of Transport (1994). Travel in London, London Area Transport Survey. HMSO, London. Department for Transport (2000). Ten Year Plan for Transport. The Stationery Office, London. Dix, M. (2002), The Central London Congestion Charging Scheme – From Conception to Implementation. Paper delivered at Conference on Implementing Reform on Transport Pricing Identifying Mode-specific Issues, Brussels. ECMT (2004). Managing Transport Demand Through User Charges: Experience to Date. OECD/ECMT Conference, London. Goodwin, P. (1992). A Review of New Demand Elasticities with Special Reference to Short and Long Run Effects of Price Changes. Journal of Transport Economics and Policy, 26: 155-169. Government Office for London (1995). The London Congestion Charging Research Programme. HMSO, London. Government Office for London (2000). Road Charging Options for London. (ROCOL), London. House of Commons Committee on Transport (2003). Urban Charging Schemes: First Report. HC390-I, Stationary Office, London. Livingstone, K. (1998). It Pays to Invest in Public Transport. Newspaper Plc 22 July 1998. Livingstone, K. (2000). Mayoral Manifesto. London Assembly (2004). Congestion Charging: A First Review. Greater London Authority, London. Observer, The (2004). Many cities look to emulate London congestion fee, 15 February 2004, London. Royal Institute of Chartered Surveyors (2004). RICS Research into the Impact of Congestion Charging on London Property. Royal Institute of Chartered Surveyors (2005). RICS Research into the Impact of Congestion Charging on London Property. http://www.rics.org/NR/rdonlyres/BE42797B-BC58-4AAE-8737813278CDB601/0/congestioncharging.pdf Santos, G., Shaffer, B. (2004). Preliminary Result of the London Congestion Charging Scheme. Public Works, Management and Policy, 9: 164-181.

The London Congestion Charging Scheme: The Evidence

91

Smeed, R J. (1964). Road Pricing: Economics and Technical Possibilities. Ministry of Transport, HMSO, London. Transport for London (2001). Transport Strategy. London. Transport for London (2003a). Impacts Monitoring First Annual Report. London. Transport for London (2003b). Congestion Charging: Three Months On. London. Transport for London (2003c). Congestion Charging 6 Months On. London. Transport for London (2003d). http://www.londontransport.co.uk/tfl/cc-ex/ Transport for London (2004). Update on Scheme Impacts. London. Transport Initiatives Edinburgh Ltd/the City of Edinburgh Council (2004). The Integrated Transport Initiative for Edinburgh and South East Scotland Proposed Congestion Charging Scheme: Statement of Case. City of Edinburgh Council.

6

The AKTA Road Pricing Experiment in Copenhagen

Otto Anker Nielsen and Majken Vildrik Sørensen Centre for Traffic and Transport, Technical University of Denmark

Abstract This chapter presents the AKTA road pricing experiment in Copenhagen and its main results. Conclusions are drawn on the behavioural impacts of the different pricing schemes and the accuracy of different survey and modelling techniques. AKTA followed 500 car users equipped with a GPS-based device in their cars. The participants' normal travel pattern was estimated using observations from a control period, after which pricing schemes (toll versus km-based) were implemented over an 8 to 12-week period. The participants earned the money they saved by changing behaviour compared to the control period. Surprisingly, it turned out that the participants’ behavioural changes were greater than expected based on prior surveys and modelling. Habits may have been expected to reduce changes, but the rather high amount of money involved (budget constraints) seems to be more important. Changes were analysed concerning the time of day, number of trips, average length of trips, costs of trips, etc. It appeared that km-based charging systems were more efficient than multi cordon-based.

Mai-Britt Herslund assisted in the work on the SP design and the interviews and Christian Würtz assisted in the work collecting and processing the GPS data. The AKTA project was primarily financed by EU’s 5th Framework Programme and the Municipality of Copenhagen with co-funding to the research part from the Technical University of Denmark (DTU). The principal contractor of AKTA was the City of Copenhagen, from which Poul Sulkjær is thanked. DTU, the Danish Road Directorate and the Ministry of Transport were assistant contractors. The AKTA SP was funded by the Danish Transport Council and the remaining work on this was finalised by funds from DTU and the Danish Transport Research Institute.

94

O.A. Nielsen and M.V. Sørensen

6.1 Introduction The chapter presents the design and some initial findings from the AKTA road pricing experiment in Copenhagen1. The main experiment encompassed 2x200 car drivers who were exposed to different road pricing schemes using a GPS-based technique where all trips are charged and logged. The participants were paid the difference between their expected road pricing if they did not change behaviour, and their actual behaviour. Since realistic and planned levels of road pricing were tested, the experiment is close to a real scheme. However, a third experimental round was applied, where the participants were paid money according to their measured driving pattern in a control period. They were then invoiced after the period of pricing. The amount of savings were between zero (if negative, payment was ignored) and 8,000 DKK (1,100 EUR) per participant for the maximum scheme. The data from the main experiment was supplemented by stated preference (SP) analyses, questionnaires, modelling exercises, focus group interviews and telephone interviews. Hence, the project provides a thorough knowledge of the impact of road pricing as well as an empirical basis for comparison of different survey and modelling techniques. In Section 6.2, the chapter describes the experimental design of AKTA and in section 6.3 the experience and problems of carrying out the experiment as well as some overall results. Section 6.4 presents the main results of the experiment, the behavioural changes of the participants. Section 6.5 discusses the findings and methodological issues raised by the experiment. Some of the sections of the paper are based on Nielsen and Jovicic (2003), while other includes further work on the AKTA data.

6.2 Design of the Experiment AKTA is the Danish part of the EU project PROGRESS, where road pricing is examined by various approaches in 8 European cities. The field trials of the AKTA project were carried out between autumn 2001 and spring 2003. The behaviour of the participants was followed by GPS with and without a road pricing system. AKTA is unique as each car was followed over a fairly long period, which provides more information on interpersonal variation than in most revealed preference (RP) datasets.

1

AKTA is a Danish abbreviation for Alternative Driving and Congestion Charging. AKTA was the Danish part study of the PROGRESS project (www.progress-project.org), which again was part of EU’s 5th Framework programme, “The Growth Programme on Sustainable Mobility and Intermodality” that supports several projects concerning pricing (http://www.transportpricing.net/). In PROGRESS, eight European cities assessed in different ways impacts of different urban pricing schemes. The cities were Bristol and Edinburgh (UK), Genoa and Rome (I), Helsinki (SF), Trondheim (N), Gothenburg (S) and Copenhagen (DK). AKTA ran for three and a half years with a total budget of about 13.5 million DKK (1.8 million EUR).

The AKTA Road Pricing Experiment in Copenhagen

95

The purpose of AKTA was to test whether road use taxes would change travel behaviour. The city of Copenhagen was equipped with virtual cordon rings and pricing zones2. The voluntary test drivers were equipped with a GPS-based vehicle positioning system, making it possible for them to read the virtual pricing systems on a display. The cars’ movements were logged in the system and a road price cost calculated for every trip. Each participant tested two different pricing schemes or one pricing scheme and one control period without pricing. To save costs, only 200 GPS units were used. The experiment was therefore conducted in two rounds, each with 200 participants, whereby a sample of 400 cars was obtained. The first round used two 8-week periods – one for each charge level (the first 200 participants). As this was found maybe to be too short, the second round used two 10-week periods (the next 200 participants). An additional third round with 100 participants was decided after this in order to further investigate the best charging scheme. This round consisted of a 10 to 12-week control period followed by 16 weeks of charging. At the end of each round, the test drivers were paid according to an estimate based on the difference in behaviour between the two periods. The control period was used to obtain information on participants’ usual driving patterns; time of travel, trip length, usual number of destinations, number of stops on a round trip, etc. In the control period, there was no pricing for road use and the display was turned off such that the participants were not expected to change behaviour at this time. The 300 participants from the second and third rounds participated in a SP experiment before the road pricing experiment.

6.2.1 Survey Setup The participants were selected in compliance with a full factorial design based on income group, commuting pattern (place of residence and work) and pricing schemes. All participants were members of one-car households. They all lived and/or worked within the road pricing area. No participants were allowed who neither lived nor worked within the pricing area (the geography of Copenhagen eliminates commuting through traffic). All participants had a daily need for transport. The participants completed two questionnaires before the experiment plus a telephone interview after; amongst other things, to test whether they had changed attitudes. For comparison, a telephone interview with 1,015 respondents living in the road pricing area was carried out in order to investigate the general awareness in the population and attitudes towards road pricing as well as to check for possible sample bias (i.e. to secure that the stratification of the participants in the experiments was representative of the population).

2

The multi cordon-based systems were applied in this project. It charged for the passing of one or more cordon lines. This is different from systems, where drivers are charged for permission to drive or stay within a zone.

96

O.A. Nielsen and M.V. Sørensen

6.2.2 Pricing Schemes Three different pricing schemes were set up: 1) A high km-based system with four different charging levels dependent on the zone with price rising towards the inner city (Fig. 6.1). The price was twice as high at peak hours compared to off-peak hours. The cost at peak hours varied from 1.00 DKK/km in the outer zones to 5.00 DKK in the city centre (1 EURO was approximately 7.5 DKK at the time the experiment was conducted). 2) A low km-based system designed with the same structure as 1), but only with charges in the peak hours and with half the cost here than in 1). 3) A multi-cordon system, with different charging passing each cordon. Four different cost levels were used ranging from 2 DKK for passing the outer cordons to 12 DKK to pass the city centre cordon in the peak hours. The charge was half in the non-peak hours. Fig. 6.1 shows the different cordons (22 in total) as arrows coloured according to the charge level.

Fig. 6.1. The different road pricing and toll schemes

The AKTA Road Pricing Experiment in Copenhagen

97

The GPS devices were pre-coded with the pricing level and they dynamically calculated the actual cost such that the participants could see the pricing level (zone), be informed of zone shift (cordon) and read the accumulated cost for the current trip. Position coordinates were logged each second and imported into a digital map (using GIS technology) after the field experiment had been completed and related (map-matched) to roads, junctions and origin/destinations for each trip.

6.3 Practical Issues Before general modelling results from the SP and RP experiments are presented, some particular issues related to this project are discussed. Recruiting of a sufficient number of participants turned out to be far more complicated than anticipated; a total of 25,000 people had to be contacted in order to get a sample of 500 one-car households distributed over the three rounds of experiments. Fortunately, during the experiment the drop-out rate was very low. Because the equipment was developed as prototypes especially for the AKTA experiment, the unit price was quite high. It was therefore decided to run the experiment in two rounds with 200 participants in each, whereby the 200 GPS units were used twice. The first round run over 8 weeks with one pricing level – or a control period with no pricing – and the 8 weeks with another pricing level – or a control period with no pricing. After this round, it was realised by the researchers that there was a need to explain the experiment better for the participants (refer to section 6.4.1 and Table 6.1). This, however, turned out not to help (refer to Table 6.2). It was also decided to add a control period of 2 weeks first and then extend the pricing level period to 10+10 weeks. The second round ran with this configuration on 200 new participants. In the first two rounds, the participants were paid according to their change of car use after the experiment was completed. After the data from the first two experimental rounds had been analysed, it was decided to launch a third round. Here only the high km-based charging was tested, since the two first rounds showed that this was the most promising pricing system. It was also decided to extend the period to 2x16 weeks to reduce the variance in the data due to natural variations of trips and that all participants should run with control first and then charging. Finally, it was decided to change the payment to the participants in the third round compared to the first two rounds. Now, they were paid before the charging period, where they received the full amount similar to the estimate charge level, if they did not change behaviour. After the charging period, they were then invoiced according to their car use. If they reduced their car use, this invoice would be lower than the a priori payment to them. Participants were subjected to three ordinary questionnaires and one combined RP and SP interviewer-assisted questionnaire. Only participants in the experimental

98

O.A. Nielsen and M.V. Sørensen

rounds 2 and 3 were subjected to the SP questionnaire and only participants in rounds 2 and 3 were guaranteed a period without charge (control period for reference).

6.3.1 Problems with the GPS Technology The GPS technology caused far more problems that anticipated. Each GPS-based observation is dependent on the number of satellites within “sight”, the quality of each signal (dependent on atmospheric conditions) and the driving course of the receiving unit. This caused the following problems: • Signals were lost (no coordinate observation) due to too few satellite signals. This happened often in “street valleys” where buildings shaded the signal and of course also in tunnels and parking garages, after which a delay was experienced before the GPS unit could find a signal again. About 90% of the trips lost signals to some extent. Most of the trips could be reconstructed unambiguously by analysing the log-files, but about 3% of the trips had fall-outs to an extent where the trip – or even trip chain – could not be recreated unambiguously and had to be estimated. • The coordinate accuracy was reduced due to too few satellites or atmospheric conditions. This happened far less frequently than total fallout of signals. When there were a sufficient number of satellite signals, the coordinates could be estimated relatively precisely. The coordinates’ accuracy and location were, however, altered systematically in a few situations when cars drove fast in a curve – e.g. on ramps to motorways. The routes could, however, in most cases be recreated unambiguously. • Segmentations in trips were sometimes recorded wrongly by the equipment: Some trips were segmented into sub-trips due to signal fallouts or significant bottlenecks on roads, whilst some trips with short intermediate trips were wrongly joined into one trip. This could be due to short time intervals for the trips (e.g. bringing children to kindergarten, where the engine was kept running), or due to unfortunate combinations of a short trip and fallout of satellite signals. The segmentation of trips was both problematic for the post analyses of trip patterns and behaviour, and for the map-matching programme whose algorithm works differently when a trip starts or ends compared to an ongoing trip. • Specific cars had significantly more fall-outs than others. This could be due to the place of installation of the equipment in the car, the construction of the car and other electronic equipment within the car. To reduce this problem, analyses of the log files could have been carried out after for example a 2-week period. However, it was estimated that the costs doing this were higher than the additional accuracy obtained by having a larger sample. The problems with the GPS equipment are described in more detail in Nielsen et al. (2003). Due to the significant amount of problems with the GPS observations, a mapmatching algorithm had to rebuild the routes (Nielsen and Jovicic, 2003). The post-

The AKTA Road Pricing Experiment in Copenhagen

99

survey and practical experience with the equipment revealed, however, some additional, more severe technical problems than those found in the log files, e.g. that 46 of the participants had experienced technical problems, the most frequent of which were: • 14% experienced that the unit stopped working. • 5% experienced that the car battery was discharged (total fallout). • 5% experienced that the unit did not show the pricing level – but worked nonetheless. • 5% on average had a non-functioning unit at any given time during the experiment, when all participants were contacted. • The remaining 17% had experienced other technical problems or the reporting did not specify the type of problem in detail. These problems seldom took a long time to solve and the participants were aware of these problems. Furthermore, neither the post-survey nor the focus group interviews indicated that the participants had discovered the number of problems with fallouts registered in the log files. It can therefore be concluded that this did not influence their driving pattern and decision-making in the main experiment. The analyses in AKTA were accordingly not influenced, as post experiment corrections and interpretations of the observations were made. If a unit, for example, did not register in a period, this should be defined as a technical fall-out, not as reduced trip making. It can, however, also be concluded that more work must address technology issues, if a GPS-based pricing system is to be implemented at full scale. In this respect, it is noted that the present algorithms for map-matching (linking GPS observations to the specific roads) and for recreation of routes (linking GPS points around a fall-out) use a sequence of points before and after the actual fall-out. To obtain the same real-time accuracy will demand significant algorithmic development. Nielsen and Jørgensen (2004) deal with this issue and present a new algorithm that solves the map-matching problem satisfactorily for the AKTA data.

6.3.2 Adding a Third Round After analysing the results from the first and second rounds – especially the responses from the focus group interviews – it was suspected that the design did not resemble real life under road pricing, as earning money is not the same as not paying money. In the first two rounds, the expected payments for the participants’ normal driving patterns were estimated. After completion of the field test, participants were reimbursed the difference between their normal behaviour and their actual behaviour during the road pricing experiment. Thus, participants were rewarded for (positively) changed behaviour though never facing the risk of additional payment. In this way, their incentive to change behaviour even further was small and the test was less realistic. To validate this, a third round was undertaken where all participants were paid according to their actual travel pattern in their control period. They were then told that a similar amount would be invoiced after the test if they did not change behaviour.

100

O.A. Nielsen and M.V. Sørensen

The charging period was longer for the third round (12 weeks rather than 8 or 10), since it was a matter of discussion whether the period was too short in the first two rounds, i.e. whether the participants could postpone trips to after the period or accept alternatives to car travel for a short period. The experimental design was also simplified and all participants were subject to a control period followed by a high km-based scheme (the most efficient of the schemes in the first two rounds). 85% of the participants earned money in the third round as compared with about 50% in the first two rounds. The amount of money paid was also higher. However, all followed the high km-based pricing level and for a longer period. The effect of receiving money for real before the charging experiment and having to pay some (or most) back after the charging was accordingly higher than the effect of "only" receiving money after the experiment. As the participants did change behaviour to a higher extent than in the first two rounds, the payment they received on average per day was higher than in the first two rounds. As the third round was longer, they also received payment for more days than in the first two rounds.

6.4 General Results During the experiment, information was collected several times from the participants. The initial screening to select participants was recorded, a questionnaire was filed by the participants at a pre-test information meeting, a stated preference questionnaire was applied while the meter was installed at the garage (2nd and 3rd round only), a post-field test questionnaire was used and finally, focus group interviews for some of the participants related to the undertaking of the test and behavioural issues was undertaken. In parallel to this, a general population survey was undertaken of 1,015 persons randomly drawn from the whole population in Copenhagen. It could hereby be controlled that the attitudes, travel behaviour and socioeconomic attributes of the 500 participants in the main AKTA experiments were representative of the population in the Copenhagen region.

6.4.1 Socioeconomic Variables for AKTA Participants The initial screening and the pre-test questionnaire mainly served to check whether the participants met the factorial design and as a source of socioeconomic information. The three rounds were undertaken at different times of the year over a period of 19 months. The following results have been corrected for seasonal variance where such could explain differences between the rounds. An effort was made to ensure that the participant was the household’s primary car user and that this person responded to all enquiries including questionnaires concerning the experiment. For 97% of the participants, the primary user did in fact participate in the surveys.

The AKTA Road Pricing Experiment in Copenhagen

101

Participants were defined as the primary car user of a one-car household. The choice of one-car households was made to prevent interchange of trips between cars, although this could result in a gender imbalance. Two-thirds of the participants were male. Distribution by age largely reproduced the distribution of age in the entire population. Geographically, half of the participants resided close to, but not in, the city centre of Copenhagen and one-third in the suburbs, and one-sixth in the rest of the region. Thus, most of the participants were affected by the road pricing schemes every day at their home-end of a commuting trip. The participants observed greater freedom in planning of the job in terms of flexibility in office hours, possibilities for working from home and frequency of working at home. More than half of the participants actually had the possibility of changing time of travel for that trip. For the households as many as 74% of the participants stated that they did not use the car as the main mode of commuting – this generally rising with the distance to their workplace. 77% of the car users used the car every day. The yearly demand for car usage was below 10,000 km for 27% and 10-20,000 km for more than half of the participants.

6.4.2 Experience and Attitudes Towards Road Pricing More than 8 out of 10 participants had practical experience with road pricing either as a fixed cost for e.g. a bridge or as a distance-dependent charge. The two tolled bridges in and to Denmark explain this relatively high figure given the limited attention road pricing has previously received in Denmark. Comparable figures for the whole population are not available. Furthermore, the attitudes towards the present tax system changed because of the experience of road pricing. Before the AKTA experiment, 18% of the participants thought that tax on registering cars was a good principle (54% thought it a bad principle) whilst after they had participated in the experiment, only 12% of the participants thought of this as a good principle, while 52% were against. Before the experiment, 69% of the participants were in favour of variable taxes on car use (road pricing) whilst 15% were against. Although the commuting trip is known to be determined by habit, 40% of the participants stated that they did to some extent change their habits. The participants were generally more likely to believe in an effect on other car users than on themselves. 53% believed other users would alter their driving pattern should a peak hour charge be introduced, whereas only 43% expected it to have an effect on their own driving. 42% of the participants expected that charges like those tested would make them drive less. If road pricing were to be established, 26% of the participants felt that the revenue should be used to improve the public transport system, followed by reduced fares in public transport (13%) and improvements in traffic safety (12%). Spending the revenue as indicated by the participants increased acceptability to 42%, although 40% clearly opposed the extra tax.

102

O.A. Nielsen and M.V. Sørensen

Prior to the field trial, 17% of the participants did expect to change their attitude towards road pricing, whilst 27% expected not to change their attitude. After the experiment, 31% of the respondents indicated that they changed their attitudes to some extent and half of the respondents did not change their attitude at all. 54% of the participants changed their attitude during the experiment, 29% becoming more positive and 25% becoming more negative. The participants were contacted after the experiment to answer some questions about their behaviour and the experiment in general. Table 6.1 shows the stated saving strategies distributed by the different pricing schemes in the experiment. Although the procedures in the experiment had been explained very thoroughly both orally and in writing, apparently many participants misunderstood the design. 23 participants stated for example that they changed behaviour in the control period due to the experiment, which is illogical since no charging was applied. This included even 3 participants who changed their stated behaviour in the control period only, i.e. they use as their stated saving strategy to reduce their trips in the free control period, but not in the charging period! It was hereby possible to control post-experiment behaviour changes (which turned out to be insignificant though). In the designs with no control period, one participant only changed behaviour in the low km charge period, but not in the high km period. A total of 24 participants (13%) had more or less misunderstood the experiment. These participants clearly did not behave in a cost minimising manner. Table 6.1. Stated saving strategies for the first round with 201 cars (183 answers). Number of participants and (per cent) Saving 2nd period only

25 (60%) 34 (63%) 6 (38%)

Saving 1st period only 2 (5%) 1 (2%) 0

Low km + high km Low km + toll High km + toll

9 (35%) 6 (35%) 13 (46%)

1 (4%) 0 0

5 (19%) 0 1 (4%)

Total

93 (51%)

4 (2%)

30 (16%)

Pricing levels

No saving

Control + high km Control + low km Control + toll

8 (19%) 12 (22%) 4 (25%)

Saving both periods 7 (17%) 7 (13%) 6 (38%)

Total

11 (42%) 11 (65%) 14 (50%) 56 (31%)

26 17 28

42 54 16

183

Note: Please notice that the two periods could be in both orders (e.g. high km before control or visa/versa), i.e. some participants run a control period followed by a control period. It was hereby possible to control for post-experiment behavioural changes (which turned out to be insignificant though). These have been joined in the table for simplicity. Shaded fields indicate illogical behaviour, bold fields a desired behavioural effect (see also Nielsen and Jovicic, 2003).

Of the participants who complied with the requirements of the experiment, some chose to change behaviour – others not; some participants did not believe that they had an alternative as the alternative was considered too inconvenient or they did not consider that the price was high enough to make them change behaviour (their willingness to pay and value of time were too high). In the experiments with two charge periods, some participants only changed behaviour in the period with the highest charge level.

The AKTA Road Pricing Experiment in Copenhagen

103

170 cars/households from the second round were interviewed, of which 91 tried to save money. Table 6.2 shows changes in behaviour distributed by different pricing strategies. A total of 21 participants had more or less misunderstood the experiment (12%). This was somewhat improved compared to the first round, but also disappointing, since extra care had been made to explain the experiment due to the experience of the first round. This confirms that a fraction of car users behave irrationally or not as utility maximisers even when a high level of information is available. Note that the figures only concern the participants’ stated strategies and therefore exclude (random) variation in trip patterns. Table 6.2. Saving strategies for the second round with 200 additional cars (172 answers) Saving 2nd period only 7 (21%) 6 (21%) 12 (44%)

Saving both periods 3 (9%) 9 (31%) 2 (8%)

Total

23 (69%) 13 (45%) 13 (48%)

Saving 1st period only 0 1 (3%) 0

Low km + high km Low km + toll High km + toll

11 (39%) 12 (40%) 7 (30%)

2 (7%) 2 (7%) 2 (9%)

2 (7%) 4 (13%) 1 (4%)

13 (46%) 12 (40%) 13 (57%)

28 30 23

Total

79 (40%)

7 (4%)

32 (19%)

52 (31%)

170

Pricing levels Control + high km Control + low km Control + toll

No saving

33 29 27

Note: Please notice that the two periods could be in both orders (e.g. high km before control or visa/versa), i.e. some participants run a control period followed by a control period. It was hereby possible to control whether there were any post-experiment behavioural changes (which turned out to be insignificant though). These have been joined in the table for simplicity. Shaded fields indicate illogical behaviour, bold fields a desired behavioural effect (see also Nielsen and Jovicic, 2003).

The two tables show that not all participants had understood the experiment or the general concept of the charging systems, since 12-13% used illogical saving strategies. About half (51% and 40%, respectively) of the participants stated that they did not change their travel behaviour due to the experiment, which may be due to high value of time, poor alternatives to their car trips or general preferences of the car.

6.4.3 The General Population Survey of Attitudes Towards Road Pricing In addition to the surveys of the participants in AKTA, an opinion poll was carried out with 1,015 randomly selected inhabitants in Copenhagen in autumn 2001. The respondents were geographically distributed over Greater Copenhagen with a quarter in the suburbs and the majority (73%) in the central city area. Half of the respondents were car owners; of these most were one-car owners (92% of the 50%) and ‘only’ 8% owned two or more cars. The yearly car use was below 10,000 km for 30% of the car owners and 10-25,000 for 53%. 14% of the car owners had a yearly need to use a car of more than 25,000 km. Half of the respondents were aware of road pricing – dominated by female respondents (60%, 40% for male) and dominated by car ownership (60%, 40% for

104

O.A. Nielsen and M.V. Sørensen

non-owners). Awareness increased with age. The rate of awareness also increased by income level, though some of this may be due to educational level. Understanding of road pricing was mainly described by toll for use of specific areas and/or particular hours of the day, specific roads and a cordon around the city. In the light of the fact that Denmark has two tolled major bridges and no other tolled road segments, the fact that awareness was spread over several types of road pricing schemes was reassuring. The existing tax system is structured with a lump sum tax when the car is acquired (only the first owner) followed by a yearly (lump sum) payment. This system was liked by 38% of the respondents and disliked by 43%. Females were slightly more in favour of the present taxation system than men and car owners disfavoured the system slightly more than non-car owners. A variable road pricing system was favoured by two-thirds (65%) of the respondents irrespective of gender, the positive view declining with age. Car owners were less in favour (58%) than non-owners (72%). The car owners’ assessment was unaffected by how much they used the car (mileage) as well as their income. Enthusiasm for variable road pricing was not fully maintained when an additional peak hour fee was suggested; the fee was not further quantified. Half of the respondents supported this idea, again dominated by non-car owners (60%) as compared with car owners (45%). More respondents with flexible working hours (45%) are against than respondents without flexible working hours (38%). As for the variable road pricing, enthusiasm declined with increasing income level. Belief in the effect of a peak hour fee on other car users was 40% though only 35% expected they might change their own behaviour – the latter covers over geographical differences. The figures were unaffected by whether or not the respondent had flexible working hours. Questioned on the effect of the peak hour kilometre rate on change of departure time, one in four respondents indicated the lowest rate (1 DKK/km) and one in three the highest rate (5+ DKK/km) without any variation by gender. The peak hour kilometre rate needed for change of mode was on average higher; 20% of the respondents intended to change at 1 DKK/km, 35% did not intend to switch mode before the rate was over 5 DKK/km (the highest option). This strongly suggests respondent bias as it appears that the respondents were deliberately trying to affect the results for political reasons. 53% of the respondents believed that a zone-based toll was good, car owners being more in opposition (45%) than non-car owners. There were only minor variations by gender. Of the car users, 57% believed that a zone-based system would reduce car use although only 46% believe they would reduce their own car usage. Again, there were only minor variations by gender. Even among opponents of road pricing, 39% regarded a zone based toll system as a good principle. This may be due to the clarity of the toll structure and the case of understanding the system. 28% of the respondents would alter their time of departure if a cordon fee of 4 DKK (lowest level) to the centre of the city were introduced, 35% stated that the cordon fee should be at least 12 DKK (highest rate). This was largely independent of

The AKTA Road Pricing Experiment in Copenhagen

105

gender. The pattern is similar for change of mode. Again, it seems that some respondents could not comprehend the question. The general opinion is that the revenue from road pricing should be used to improve public transport (extend services and reduce prices) whereas improvements of the road network was given lowest priority. Provided the revenue from road pricing is used as specified by the respondents, the majority (58%) were not willing to accept higher taxes for road use than today; males (65%) being more opposed than females (52%). The respondents largely recognised that variable road pricing would diminish traffic in city areas (62%), that road pricing would be fairer than the existing car taxation system (68%) and that the environment would benefit (57%). A fairly high fraction of the same respondents also found that the car user in general would have to pay more than today for car use (52%), that prices would become less transparent (58%) and that an additional level of bureaucracy would emerge (63%) if a road pricing system was introduced. More than 50% of the respondents believed that variable road pricing would reduce traffic in Copenhagen and reduce pollution related to traffic. The respondents were generally sceptical with regard to public spending of the revenue generated by road pricing as only one third believed that the revenue would be used to improve public transport. In general, more than half of the respondents were in favour of road pricing as an alternative to the existing system. Two out of three were supportive of variable road pricing and slightly more than half supportive of peak hour toll and zone-based tolls. Respondents living in the western and southern parts of Greater Copenhagen were more in favour of road pricing, than respondents living in other parts of Copenhagen. These findings were largely unaffected by income level and the distance to the nearest high service public transport. Women were slightly more supportive than men.

6.5 Behavioural Changes and AKTA: The Main Results Table 6.3 shows the main results of the experiment. Only results with a control period combined with a pricing scheme are shown. The km per day in the control period is lowest in the multi cordon-based system and highest in the low km-based system. This can perhaps be explained by the misinterpretation of the experiment by some of the participants (Table 6.1 and 6.2) or by personal variations (few persons within each segment). The payment estimate per day if the participants did not change behaviour could be estimated based on the control period. This estimate was higher in the multi cordon-based system than in the high km-based system. The behavioural responses (DKK per day, km per day) were greatest in the two high toll schemes, which could be expected, and the impacts of the low km-based tolls were so small (some of the responses even have perverse signs) that it must be concluded that no significant behavioural responses could be found, except perhaps a slight reduction of traffic in

106

O.A. Nielsen and M.V. Sørensen

the afternoon peak (e.g. postponing shopping trips to after the peak, where the toll is zero). The driven km per day decreased more in the high km-based system than in the multi cordon-based system, even though the absolute payment per day was about 30% lower in the km-based than the multi cordon-based system. This can be explained by the fact that it is easier to save by changing route and destination in the km-based system than in the multi cordon-based system. If, for example, an alternative destination is closer than the preferred, but is still on the other side of a cordon, then it is only possible to save in the km-based system. This result may, however, depend on the design of the cordon system. Bonsall et al. (1998) found for example a greater response to cordon tolls than a km-based system. The average paid DKK per km increased in the km-based system, even though the participants did reduce their travel. This means that they to a higher extent cancelled or changed trips with low payment per km, e.g. trips outside of rush hours, than expensive trips, e.g. typical commuting trips to the city centre in the rush hours. The interpretation must be that the commuting trips are difficult to change, since the location (work place) and time (fixed working hours) cannot be changed, whilst it is easier to change non-commuting trips to other destinations, or outside the afternoon rush hours, or to reduce the number of such trips (e.g. by shopping for larger quantities less often). This result is in line with the discussions during the focus group interviews. Table 6.3. Responses to different pricing schemes, main experiment (based on approximately 100,000 trips) Trips

High km-based toll Impact within 95%, 90%, 98% confidence intervals for each round Control

High

Low km-based toll No impact with 90% confidence interval

Change Control

Low

Multi cordon-based pricing No impact within 90% confidence intervals

Change Control Cordon Change

All period Morning peak Afternoon peak

4.3

4.0

-7.8%

4.2

4.2

0.1%

4.0

4.0

-0.5%

0.53

0.47

-12.0%

0.45

0.45

0.3%

0.51

0.48

-6.6%

0.76

0.72

-5.5%

0.76

0.73

-4.0%

0.70

0.71

0.7%

Weekday

2.0

1.8

-9.2%

1.9

2.0

1.5%

1.8

1.9

2.9%

Weekends

1.1

1.0

-5.2%

1.1

1.0

0.2%

1.0

0.9

-4.9%

DKK/km

0.75

0.78

4.0%

0.22

0.23

4.2%

1.2

1.1

-6.4%

DKK/day

26.2

24.4

-7.5%

9.4

9.6

2.4%

38.8

33.4

-16.2%

Km/day

38.2

34.0

-12.4%

46.5

45.2

-2.8%

34.8

31.5

-10.5%

The main reduction of the number of trips in the high km-based and multi cordonbased schemes was in the morning peak, which is somewhat surprising, since this must be modified commuting trips. The afternoon peak was relatively smaller changes, which is also a surprise since it could be assumed to be easier to postpone or

The AKTA Road Pricing Experiment in Copenhagen

107

modify shopping and leisure trips (time and destination are less constrained than commuting). The reason can be that afternoon trips are part of more complicated trip chains with less flexibility. The changes of behaviour on weekdays and at weekends are surprisingly high in the km-based system compared to the charging in the peak hours. This was not expected since the pricing was only 50% of the peak hours, and the alternative public transport service is less frequent and with fewer direct services outside peak hours. The response on the multi cordon-based system is more in line with what could be expected.

6.6 Discussion and Conclusions The results and findings of the experiment are discussed and summarised in the following.

6.6.1 GPS Technology The GPS experiment was perceived as realistic and it seems that the behavioural responses are realistic. A few participants acted illogically. This could also be the case in a real life implementation of road pricing. One could claim that the error term in discrete choice models also usually reflects this, since the choice set is reduced to a binary one. The GPS technology did not perform as well as anticipated. A significant amount of extra work had to be carried out in order to develop methods and software to process and repair the data. Since the participants had not experienced these problems, their behaviour was not affected. However, if a GPS-based system is to be used for a full-scale pricing scheme, further work needs to be carried out on methodological, software and technical issues.

6.6.2 The Main Experimental Design Theoretically, a factorial design is optimal since a small sample can be utilised better by this approach. However, it appeared that some of the participants could not understand the experimental design, not only those who tried different pricing schemes without any control period (which should be avoided in other experiments), but also those who ran a control period followed by one pricing scheme. Some even misunderstood the design in the second round, where the project leaders were aware of the problem and where further care had been taken to explain the experiment to the participants. By conducting a post-experiment survey by telephone, it was, however, possible to identify those participants who had misunderstood the design. It is recommended that participants be presented with their expected costs after the control period or, even better, to pay them the amount, as this makes the following period more realistic, since they have to pay back “real” money. To do this, data from

108

O.A. Nielsen and M.V. Sørensen

the GPS units have to be downloaded and processed in order to calculate the payment prior to the second round, which is a technical and organisational obstacle. The pricing periods in the experiment were too short; at least 12 weeks must be used for the control period and for each subsequent pricing level. This was compensated for in a third experimental round, where the initial results indicate a somewhat greater degree of change in behaviour.

6.6.3 Changes in Behaviour and Evaluation of Different Pricing Schemes It can be concluded that road pricing does indeed affect travel behaviour; it is not considered to be another fixed cost, but a marginal cost that drivers respond to. The high km-based pricing level clearly made an impression on the participants (between 1 and 5 DKK/km in the peak hours and half price in the off-peak hours). Even if they could not change behaviour, they had examined alternative travel options before rejecting them. The participants considered changing route, mode and “occasional” trips and some changed behaviour. The main changes were new routes and for “occasional” trips new destinations, time of day (to non-peak) and to some extent, fewer trips. Commuting trips are assumed to be difficult to change, e.g. shifting away from the peak hour, working at home (telecommuting) or using another mode (bicycle or public transport). However, it turned out that the participants nevertheless changed these as much – or even more – than other trips. The low km-based pricing level was in general not sufficiently high to change behaviour, although a few participants made some minor changes when it was easy to do so. The km-based schemes were in general considered more fair than the multi cordon-based ones. The participants had nevertheless greater difficulties to understand these than the multi cordon-based system. It is interesting to note that the fairer and economically more justified schemes are, the more difficulties are experienced in understanding them. Nevertheless, the km-based scheme turned nonetheless out to be the most efficient in terms of behavioural changes.

6.6.4 Attitudes Towards Road Pricing The participants' and interviewed persons’ attitudes to road pricing were less emotional than expected (especially considering the debate in the Danish press). Most participants did not consider surveillance as a problem (cars can be tracked by the logged coordinates). Neither was the possibility of controlling speed limits considered important. The participants disagreed on whether road pricing is fair or not, including whether society becomes more class-divided between people who can pay and those who cannot, and for people who cannot change behaviour (having fixed timetables or with children in school and day care). Only a few participants had very strong attitudes against road pricing.

The AKTA Road Pricing Experiment in Copenhagen

109

References Bonsall, P., Cho, H-J., Palmer, I., Thorpe, N. (1998). Experiments to determine differences in drivers' response to a variety of road user charging regimes. PTRC. September 1998. Nielsen, O.A., Jovicic, G. (2003). The AKTA road pricing experiment in Copenhagen. 10th International Conference on Travel Behaviour Research. Proceedings, session 3.2 Valuation/Pricing. Lucerne, Switzerland, August. Nielsen, O.A., Jørgensen, R.M. (2004). Map-matching algorithms for GPS data – Methodology and test on data from the AKTA road-pricing experiment in Copenhagen. Paper accepted for the Triennial Symposium on Transportation Analyses, TRISTAN V. Le Gosier, Guadeloupe, French West Indies, June 13-18, 2004. Nielsen, O.A., Kristensen, J.P., Würtz, C. (2003). Using GPS for road pricing – experiences from Copenhagen. Paper presented at the ITS world conference, Madrid. Presentation session PS117, paper No. 2131T.

7

Experience with Measuring Equity and Efficiency: A Case from Oslo

Farideh Ramjerdi, Knut Østmoe± and Harald Minken TØI, Institute of Transport Economics, the Norwegian Centre for Transport Research, Oslo, Norway

Abstract Road pricing has been discussed in the context of two objectives: improving resource allocation and financing the expansion of the capacity of the road network. Financing transport infrastructure by means of toll revenues in the traditional sense has a history that dates back almost 70 years in Norway. Since 1986, with the opening of cordon toll schemes in Bergen, Oslo and Trondheim, and more recently in Stavanger there has been a major shift in the location of tollfinanced projects from peripheries to urban areas. Meanwhile, the contribution from toll financing schemes to the total funds for transport infrastructure has increased by more than 35 percent and the scheduled projects for financing by tolls are almost complete. It is apparent that the Norwegian tolls have successfully achieved their purposes. The Norwegian tolls are approved for only a limited period of time, usually 15 years. There is now a growing interest in redefining the objectives of some of the Norwegian urban tolls from merely a financing scheme to a part of an integrated policy package that would address transport externalities. This chapter will discuss the performances of the Norwegian urban tolls and the issues related to their changing roles.

7.1 Introduction Issues of equity in transport have received extensive attention, especially more recently related to congestion pricing. This body of research has different focuses. Some address the distributions of economic gains and losses among different groups of users, usually by income, and suggest how to calculate these. Others have focused on alternative schemes for redistribution of the toll revenues in order to address equity concerns. Still others have focused on the identification of different stakeholders such as consumers, producers and operators, and subgroups among stakeholders that are affected differently by a policy. See Eliasson and

This study was partly supported through the SPECTRUM project under the EU 5th framework programme. Arild Vold has calculated the scenarios using RETRO. ±

After this chapter was written Knut Østmoe has died.

112

F. Ramjerdi, K. Østmoe and H. Minken

Lundberg (2002) for a survey. A few studies address the quantification of inequality formally. Among these are the EU-funded research projects AFFORD, MCICAM and PROSPECTS (see Fridstrøm, et al. 2000; Minken et al., 2002; MCICAM, 2003). The efficiency and equity of a toll scheme have most often been discussed while looking at the transport sector in isolation. There are, however, interactions between the transport sector and the rest of the economy, and there are distortions in the rest of the economy. Due to these distortions, the secondary effects of transport policies on the rest of the economy are relevant and should be evaluated. In a general equilibrium framework, these interactions are explicitly addressed. In a partial equilibrium framework, like an urban transport model or an integrated land use and transport model, interactions with the rest of the economy can at best be addressed implicitly, through the use of a so-called “marginal cost of public funds” (MCF). Roughly speaking, the marginal cost of public funds is the cost to society of raising a dollar’s worth of public revenue by distortionary taxation. As a rule, it is assumed that the distortionary tax that will have to be used is the income tax. However, different tax instruments, including the pricing instruments of transport, will have different MCFs. From an efficiency point of view, the instrument with the least MCF should be used. But efficiency is not our only concern. As Sandmo (2000) points out, a main reason for distortionary taxes is to address redistribution, otherwise uniform or arbitrary lump-sum taxes could have been levied. The redistributional impacts depend not only on which tax instrument is used but also on how revenue is used in the public sector or recycled to the households. Transport and integrated transport and land-use models are probably the most common planning tools. The level of detail varies among models with respect to geographical detail, presentation of the transport networks, alternative modes of travel, time periods (usually peak and off-peak) as well as the segmentation of the market by travel purposes. Behavioural responses with different time dimensions such as route choice, mode and destination choices and trip frequency are usually captured in transport models. Hence, the architecture of these models and the manner in which they are applied will determine the time horizon of the application of the models, usually from the very short to medium run. When disaggregated data are used for the estimation of urban and regional models, individual and household socio-economic characteristics can enter the model formulation as explanatory variables. Individual or household income, gender and age are among the socio-economic variables that are likely to influence behaviour. Consequently, it is possible to apply this class of models to evaluate the differences in response of different segments of a population to a transport policy. While partial equilibrium models must inevitably represent economy-wide distortions and distributional impacts in a coarse way, this level of detail in the representation of a few selected markets is a strong point with respect to equity analysis of policies. A general equilibrium approach, on the other hand, will usually lack important details in the transport and land-use markets. The main purpose of this paper is to illustrate some of the challenges that arise in addressing the efficiency and equity of a package of instruments with a more

Experience with Measuring Equity and Efficiency

113

traditional transport model system. There are two alternative approaches available to respond to equity considerations. One approach is to respond directly to the distributional concerns by adopting an explicit form of social welfare function and the choice of a desired inequality aversion parameter. An example of this type of model is TRENEN (see, for example, Proost and Van Dender, 2002). The second approach is to apply an inequality measure for the comparison of inequality of a given pair of distributions of a variable that changes as the result of a policy, such as income or accessibility. In the next section, we provide an overview of some equity measures and their properties. Section 7.3 presents the alternative packages of instruments for Oslo as a case study and an assessment of these packages. Section 7.4 focuses on the effect of the assumptions regarding MCF using a sensitivity analysis. Section 7.5 shows the performances of some of the equity measures in the case of Oslo. In Section 7.6, we summarise our conclusions.

7.2 Equity Measures The most central issue in the assessment of equity is related to how equity is defined. Equity can be defined along many dimensions such as justice, rights, treatment of equals, capability, opportunities, resources, wealth, primary goods, income, welfare, utility and so on (see Sen, 1982, 1992). Sen (1992, p12) states “that every normative theory of social arrangement that has at all stood the test of time seems to demand equality of something – something that is regarded as particularly important in that theory. The theories involved are diverse and frequently at war with each other, but they still seem to have that common feature.” Sen continues by suggesting that demanding equality in one space implies inequality in some other space. An important ethical issue is related to the equality of consideration. Sen (1992, p.18) suggests that “the need to defend one’s theories, judgements and claims to others who may be directly or indirectly involved, makes the equality of consideration at some level a hard requirement to avoid.” Moreover, the relative advantages and disadvantages that people have compared with each other, can be judged in terms of many different variables, e.g. their respective incomes, wealth, utilities, resources, liberties, rights, quality of life and so on. On human diversity, Sen (1992, p.20) states that “The plurality of variables on which we can possibly focus (the focal variables) to evaluate interpersonal inequality make it necessary to face, at a very elementary level, a hard decision regarding the perspective to be adopted. This problem of choice of the ‘evaluative space’ (that is, the selection of the relevant focal variables) is crucial to analysing inequality.” It is beyond the scope of this paper to provide an overview of the ways different social philosophies have defined equity and to compare them. What is relevant for our work is that different aspects of equity are important for different groups in society and it is important to provide measures for the evaluation of their concerns and to reflect their views.

114

F. Ramjerdi, K. Østmoe and H. Minken

In order to address equity, a unit of analysis and the variable along which equity is to be analysed have to be defined. In a social context, the unit of analysis can be an individual or a collective unit such as a nuclear family, women, elderly, disabled, a region, etc. The choice of the unit depends on the interpretation of the inequality measurement. In some contexts, it is natural to adopt an individual as the unit, for example when we are looking at exposure to pollutants. In other contexts, e.g. when we are examining the distribution of wealth or income, it might be more useful to adopt a collective unit such as a household. Furthermore, evaluation of inequalities along a certain dimension in terms of between and within groups such as between genders, regions, etc., is often required. Coherence and homogeneity are the important criteria in the selection of collective unit.

7.2.1 Properties of Equity Measures Different measures of inequality reflect different perceptions of inequality. The sets of weights that different views attach to transfers at various points in a distribution are different. This can result in contradictory rankings of a given pair of distributions (see Kolm, 1969; Atkinson, 1970; Sen, 1973). In this sense, inequality measures have both normative and descriptive content. These measures can be used to describe the differences in a population with respect to a given variable such as income, but they can also represent the manner in which these differences are measured. There are numerous axioms that place specific requirements on the properties of a measure of inequality. In the following, we summarise a number of these axioms (see Harrison and Seidl 1994; Myles, 2000). These axioms are used for the construction of the axiomatic measures of inequality. The symmetry or anonymity axiom requires the inequality measure for a given income distribution in a given population not to be affected by the order in which the individuals are labelled. In other words, it is not important who is rich and who is poor. This axiom seems very obvious. All the measures that are described in the following sections satisfy this axiom. The axiom of transfer or Pigou-Dalton principle says that a transfer of income from a rich person to a poor person should reduce the measured inequality as long as the income of the rich person stays higher than the poor person after the transfer. This view was originally expressed by Pigou (1954) and shared by Dalton (1920). The Pigou-Dalton principle is an important property that any acceptable measure of inequality should satisfy. The principle of population requires the inequality measure to be independent of the size of the population. The scale invariance axiom or relative inequality aversion axiom demands that the measured inequality should not change if all members of a population get the same proportional increase in incomes. Kolm (1976a, 1976b) regards this as a (politically) rightist view. The translation invariance axiom or absolute inequality aversion axiom requires that the measured inequality does not change by changing all incomes by the same

Experience with Measuring Equity and Efficiency

115

amount as long as the changes do not lead to a negative income. This is regarded as a (politically) left-wing view. The decomposability axiom requires that there should be a coherent relationship between inequality in the whole population and its constituent parts. The basic idea is that one should be able to define the inequality measure of the total population as a function of inequality within its constituent parts and inequality between subgroups.

7.2.2 Some Inequality Measures Inequality measures are often classified as statistical, welfare or axiomatic (see, for example, Myles, 2000 and Cowell, 1977). Statistical measures examine the distribution of any variable in a given population, such as income. Examples of these are range, variance, measure of variation, log variance, the Gini measure and Theil’s entropy measure. Welfare measures rely on welfare economics and incorporate equity concerns into a welfare function. Axiomatic measures are derived by addressing the properties that a satisfactory measure ought to have. These measures can be applied to the evaluation of inequality of any vector or distribution of observations, even to non-economic data such as the distribution of the ambient level of pollutants or accessibility over an area. The following measures are examined in this study. 1. Range, R, defined as R=Ymax –Ymin

(1)

2. Variance, V, defined as

V

1 n (Yi  Y ) 2 ¦ ni1

(2)

3. Coefficient of variation, c, defined as

c

V Y

(3)

4. Relative mean deviation, M, defined as

M

1 n Yi ¦ 1 ni1 Y

(4)

116

F. Ramjerdi, K. Østmoe and H. Minken

5. Logarithmic variance, v, defined as

v

1 n § Yi · ¦ ¨ log( ) ¸ n i 1© Y ¹

2

(5)

6. Variance of logarithms, vl, defined as

vl

1 n § Y · ¨¨ log( i ) ¸¸ ¦ n i 1© Ylog ¹

2

(6)

7. The Gini measure, G, defined as

G

n n 1 | Yi  Yj | ¦¦ 2n 2 .Y i 1 j 1

(7)

8. The Theil’s entropy measure, T, defined as

T

1 N Yi Y log( i ) ¦ N i1 Y Y

(8)

9. The Atkinson index, AH, defined as

AH

1 1H 1H

ª1 ªY º 1 « ¦« i » «¬ n i 1 ¬ Y ¼ n

º » »¼

(9)

10. Kolm’s measure of inequality, KD, defined as

KD

1 §1 N · log ¨ ¦ exp( D(Y  Yi )) ¸ D ©N i 1 ¹

(10)

In the above measures Y is a measure of welfare n is the number of observations on welfare

Y Ylog

is the mean level of welfare is the mean level of log of welfare

H and D in Atkinson and Kolm measures are parameters that address inequality aversion and H and D ! 0 .

Experience with Measuring Equity and Efficiency

117

The first 8 measures are classified as statistical measures, while the last 2 measures (Atkinson and Kolm) are welfare measures. The following table summarises some of the properties of these measures. The impacts of a package of instruments can be measured using non-economic data. An example of the application of the equity measures to non-economic data is related to the changes in the distribution of emission of pollutants over the area of study. It might even be feasible to evaluate the changes in terms of within and between segments of the population. The segments can be defined in terms of the socio-economic characteristics of the population or by locations in the study area. A decomposable measure is necessary for this purpose (see, for example, Myles, 2000 and Cowell, 1977). Table 7.1. A summary of the properties of inequality measures Measure Variance Coeff. of variation Relative mean deviation Logarithmic variance Variance of logarithms Gini Theil’s entropy Atkinson-Kolm Kolm

Definition Eq. (2) Eq. (3) Eq. (4) Eq. (5) Eq. (6) Eq. (7) Eq. (8) Eq. (9) Eq. (10)

Some important properties Scale Translation invariance invariance Yes No Yes Yes (weak) Yes No Yes Yes No No Yes No No Yes No Yes (weak) Yes No Yes Yes No Yes Yes No Yes No Yes Transfer

7.3 Evaluation of Alternative Packages of Instruments for Oslo The greater Oslo area has a population of about one million with an area of 5,305 km2. The population density is about 140 inhabitants/km2. Oslo city has a population of about 512,000. The Oslo toll ring was established in 1990 as a financing scheme. Originally, the toll revenue, supplemented by about equal funds from the central government, was to finance the “Oslo Package” (now referred to as “Oslo Package 1”), comprising some 50 new road projects. It is estimated that by 2007 the total contribution of the scheme to Oslo Package 1 will amount to NOK 9.1 billion (2002 NOK), approximately 15-20 per cent above the initial estimate. In 2000, the Parliament approved an increase in the toll fee for financing an investment package on public transport projects, referred to as “Oslo Package 2”. There is much debate and some interest in changing the direction of the scheme to a congestion pricing scheme from 2008. Amongst the different alternatives that have been evaluated for Oslo, there is a time differentiated toll scheme with the purpose of reducing car traffic during peak periods. Revenues would be allocated to public transport and to the extension and improvement of roads in the region. The Oslo scheme is likely to continue in some form or other after 2007.

118

F. Ramjerdi, K. Østmoe and H. Minken

For the purpose of this study, a reference scenario and some 12 alternative scenarios were constructed. The reference scenario comprises the trends in income, socio-economic and demographic changes, changes in land use and any planned and approved changes in the transport networks. Alternative scenarios are formulated by adding alternative packages of instruments to the reference scenario. The instruments include: x A time differentiated toll ring scheme x Fuel tax x Expansion of public transport services by increasing frequency of services x Selected reduction in speed limits on the car links with low capacity. A toll ring complemented by a distance-based pricing scheme performs closer to a “first-best pricing scheme” where ideally users are charged in relation to their social marginal costs. Since fuel tax approximates a distance-based pricing scheme and at the same time is a good instrument to address environmental externalities, especially related to CO2 emissions, it is an important instrument in an optimal package of instruments. These pricing instruments will reduce the demand for car use and increase the demand for public transport. To satisfy this additional demand, public transport frequency can be increased optimally. Improvements in public transport services can also address the adverse effects of pricing instruments on equity. Finally, among the instruments is a selection of speed limit reductions on car links with a low capacity. An instrument such as a reduction in the speed limit decreases the car traffic and congestion, and decreases other external costs associated with car use. Contrary to a pricing instrument, a reduction in the speed limit results in a decrease in government revenues. This instrument clearly illustrates the importance of the assumption about the value of the MCF in the calculation of the net benefits of an instrument. These instruments and their levels and the packages used in this study do not reflect precisely any of the current proposals for the future of the Oslo scheme. The main purpose is to illustrate some important issues in addressing efficiency and equity by using these packages of instruments as examples. The lessons should be valid for other package of instruments. Table 7.2 shows the description of alternative policy scenarios that are constructed with a single instrument or the combination of these instruments.

Experience with Measuring Equity and Efficiency

119

Table 7.2. Description of alternative policy scenarios Alt

Scenario description

0. Ref.

Reference scenario

1. All

All measures

2. S

Selective reductions in speed limit

3. PTF

Increase in public transport frequency by +5.8%

4. T

Toll: Peak periods: 2.5 today’s prices, Other periods: Today’s prices

5. F

Increase in fuel taxes by +50%

6. S/T

Change in speed limit + toll

7. S/PTF

Change in speed limit + increase public transport frequency

8. S/F

Change in speed limit + increase in fuel taxes by 50%

9. T/PTF

Toll + increase in public transport frequency

10. T/F

Toll + increase in fuel taxes by 50%

11. PTF/F

Increase in public transport frequency + increase in fuel taxes

12. PTF/F/T

Increase in public transport frequency + increase in fuel taxes + toll

A multi-modal transport model, RETRO, is used in this study (Ramjerdi and Rand, 1992; Vold, 2003). RETRO has the following sub-models: i) Disaggregate and aggregate licence-holding models ii) Disaggregate car ownership models iii) Disaggregate models for travel frequency and models for mode and destination choices iv) Segmentation model v) Network model. EMME/2 (a software package) is used for the network model. The number of zones is 438. In this case study, it is assumed that the land-use changes are exogenously defined. The alternative scenarios are evaluated according to an objective function that accounts for the net benefits of all the affected actors, users, non-users, producers and government, defined as

OF

1

¦ 1  r (CS t

t

 PC t  MCFt * GSt  Env t  J t g t )

tt*

where OF t* r CSt PSt

is the net benefit is the horizon year is a discount rate is the change in the consumer surplus in year t is the change in the producer surplus in year t

(11)

120

F. Ramjerdi, K. Østmoe and H. Minken

GSt is the change in the government surplus in year t MCFt is the marginal cost of public funds in year t Envt is the external costs defined as accident, noise and pollution costs and other external effects Jt is the shadow cost of CO2 emissions, reflecting the national CO2 target for year t, is the amount of CO2 emissions in year t, gt The rule-of-half is used for the calculation of the consumer surplus. The changes in the producer surplus (revenues net of costs) should be calculated for all the transport operators. Since toll and parking operators in this study are government agencies, they will be addressed under the government surplus. The public transport operators must earn a surplus after subsidy. Therefore, their surplus is also accounted for under the government surplus. The tax revenue associated with car use and car ownership will be included in the government surplus. Tables 7.3 and 7.4 show the unit values that have been adopted in this case study. They are based on recommended Norwegian values in urban areas (Eriksen et al., 1999). Table 7.3. Values of externalities (in Euro/vehicle kilometre) Mode Car (average) Public transport (average for bus, and light rail)

Emissions (other than CO2)

Noise

Accidents

CO2

0.025

0.017

0.027

0.011

0.304

0.170

0.061

0.066

Table 7.4. Value of travel time (in Euro/hour) Mode of travel

Car

In-vehicle time

Public transport

5.64

4.70

Wait and transfer time

-

5.64

Auxiliary time

-

5.64

Tables 7.5 and 7.6 show the changes in the net benefits of the 12 scenarios relative to the reference scenario in 2015. The marginal cost of public funds that applies to the changes in the government surplus is assumed to be 1.0. In the next section, a sensitivity analysis by using a variable MCF of 1.0 to 1.4 will be presented. An examination of these tables indicates that: x Scenarios 10 and 12 have similar scores with respect to their net benefits. In scenario 10, fuel tax is increased by 50% and a time-differentiated toll scheme (for scenario descriptions, see Table 7.2) is introduced. In scenario 12, in addition to the previous two instruments, the public transport frequency is increased by 5.8%. Scenario 12 should be more desirable based on equity considerations.

Experience with Measuring Equity and Efficiency

x

121

Scenarios that include an increase in the fuel tax of 50% have positive net benefits. This is explained by the big increase in the government surplus that compensates the decrease in the consumer surplus and the positive impacts due to the reductions in externalities and CO2 emissions. Scenarios that include a reduction in speed limit have negative net benefits. Unlike the case of a fuel tax, both the consumer surplus and the government surplus are negative in this case. The decrease in externalities and the reduction in CO2 emissions do not compensate for the loss in the consumer and the government surplus. Scenarios that include a toll increase have positive net benefits. This again is explained by the big increase in government surplus that compensates for the loss in the consumer surplus as well by the big reductions in externalities and CO2 emissions. An exception is Scenario 8 where the package includes a reduction of the speed limit. The net benefit in the scenarios in which the public transport frequency is increased is slightly negative. The cost associated with the increase in frequency is the explanation.

x

x

x

Table 7.5. Changes in the net benefit relative to the reference scenario (in million Euro/year) 1).All

2) .S

3) P

4).Toll

5) Fuel

6) S/T

-572.79

-115

10

-145.5

-343.4

-259.5

324.8

-11.6

0.4

-22.8

378.7

-34.1

Annual car taxes

-30.6

0.0

0.0

0.0

-30.6

0.0

Toll revenue (net)

157.3

0.0

0.0

171.5

0.0

170.5

-4.1

0.2

0.3

-1.0

-2.5

-0.8

28.6

4.5

1.6

2.8

17.0

7.6

Consumer surplus Government surplus Fuel tax

Parking revenue Public transport revenue Public transport investment Total Externalities (emission of pollutions, noise and accident)

-23.1

-

-24.2

-

-

-

452.9

-6.9

-22.0

150.6

362.6

143.2

44.0

6.0

-1.0

11.0

27.0

16.0

CO2

7.0

1.0

0.0

2.0

4.0

3.0

Total

-68.914

-114.9

-12.99

18.104

50.2

-97.334

122

F. Ramjerdi, K. Østmoe and H. Minken

Table 7.6. Changes in the net benefit relative to the reference scenario (in million Euro/year) 7) S/P

8) S/F

9) T/P

10) T/F

11) P/F

12) P/F/T

-104.8

-453.6

-134.8

-474.0

-330.7

-464.5

-12.0

362.5

-22.8

344.6

378.0

343.0

Annual car taxes

0.0

-30.6

0.0

-30.6

-30.6

-30.6

Toll revenue (net)

0.0

0.0

171.6

158.8

0.0

158.7

Parking revenue Public transport revenue

-0.2

-2.2

-1.0

-4.1

-2.6

-4.1

6.4

21.8

4.7

21.1

18.7

23.0

PT investment

-24.2

-

-24.2

-

-24.1

-19.4

-30.0

351.4

128.3

489.8

339.5

470.6

Consumer surplus Government surplus Fuel tax

Total Externalities (emission of pollutions, noise and accident)

5.0

32.0

10.0

32.0

26.0

38.0

CO2

1.0

5.0

2.0

5.0

4.0

6.0

Total

-128.8

-65.2

5.5

52.8

38.8

50.1

7.4 A Sensitivity Analysis of MCF Table 7.7 shows the effects of the MCF on the net benefits of the 12 alternative scenarios compared with the reference scenario. The net benefit of a scenario changes substantially with the assumption about the size of the MCF. This is due to the importance of the contribution of the government surplus to the net benefit (see Tables 7.5 and 7.6). Indeed, all the pricing policies get a higher rank with an increase in the MCF. Furthermore, the assumption about the value of the MCF affects ranking of the scenarios. As briefly discussed earlier, the MCFs of different pricing instruments are different and more importantly they depend on the type of public expenditure it is used for or how it is recycled. Parry and Bento (1999) show that the labour market should not be ignored when setting urban road prices. The effect of a tax on work travel is similar to distortionary taxes on the labour market. In an urban setting, especially during the peak periods, most travel is work-related. Therefore, asserting an arbitrary value for the MCF might be without justification. The evaluation of the efficiency as well as the equity implications of any of these scenarios is incomplete unless the use of the government surplus is specified. The recycling of the government surplus to inefficient projects, public transport or roads is not justifiable based on efficiency or equity considerations. This is why the separation of the revenues raised in the transport sector from recycling to the rest of the economy becomes problematic.

Experience with Measuring Equity and Efficiency

123

Table 7.7. Effect of MCF on the high level objective function Alt

Scenario description

MCP= 1.0

MCP= 1.2

MCP= 1.4

17. All

All measures

-69

22

112

18. S

Selective reductions in speed limit Increase in public transport frequency by +5.8% Toll: Peak periods: 2.5 today’s prices, other periods: today prices

-115

-116

-118

-13

-17

-22

18

48

78

19: PTF 20.T 21.F

Increase in fuel taxes by +50%

50

123

195

22.S/T

-97

-69

-40

-129

-135

-141

24.S/F

Change in speed limit + toll Change in speed limit + increase public transport frequency Change in speed limit + increase in fuel taxes by 50%

-65

5

75

25.T/PTF

Toll + increase in public transport frequency

23.S/PTF

26.T/F

Toll + increase in fuel taxes by 50% Increase in public transport frequency + 27.PTF/F increase in fuel taxes Increase in public transport frequency + 28.PTF/F/T increase in fuel taxes + toll

5

31

57

53

151

249

39

107

175

50

144

238

7.5 An Evaluation of the Equity Implications of an “Optimal” Package The incidence of the net efficiency gains of a transport policy might be different for different segments of a population or over a geographical area. We pointed out that for a correct calculation of the net efficiency gains a spatial general equilibrium model is necessary. Addressing the interactions of the transport market with the rest of the economy, especially with the labour market, is crucial for a correct calculation of the distribution of the net efficiency gains among a population or over a region. It is, however, possible to use different measures of inequality and accessibility in order to obtain some indication of the distribution of the incidence of the net benefits. Equity and accessibility measures only suggest the likely direction of impacts, and should be treated as such. The ex-post equity analysis provides some information on how to recycle revenues to address equity considerations. This section focuses on the geographical distribution of the welfare changes of an “optimal” package (in this case, Scenario 12) compared to the reference scenario. For this purpose, two different approaches will be used for measuring accessibility (see Geurs and Ritsema van Eck, 2001; Baradaran and Ramjerdi, 2002, for a review of accessibility measures). The gravity or opportunities approach defined by:

124

Gi

F. Ramjerdi, K. Østmoe and H. Minken

Wj

¦ f (c , E) jL

(12)

ij

where Wj

stands for the mass of opportunities available to i at location j

f(cij,E)

is the deterrence function = f (cij , E)

E cij

is assumed to equal to 0.35 (WHY 0.35??) is the generalised cost of travel by car between i and j.

e

Ecij

Three alternative accessibility measures are constructed using this approach as follows: G_Emp G_65+ G_20-65 G_W

in which Wj is equal to the total employment at j in which Wj is equal to the total population over 65 years of age at j in which Wj is equal to the total population of age 20-64 at j is which Wj is equal to the female population at j

“Logsum” measure is used defined as:

log sum in

Max U nj|i i, jL

1 ln ¦ exp(P(v nj  c ijn )) P j HL

(13)

where

log sum in is the measure of accessibility at location i for individual n

U nj|i

is the utility of travel to location j given the individual n is located at i

v nj

reflects attraction at j

cijn P

is the travel cost between i and j is a positive scale parameter that is estimated

Table 7.8 shows the differences between the above accessibility measures in Scenario 12 and the reference scenario. Fig. 7.1 shows the different areas in the Oslo region. As can be expected, all the accessibility measures are negative in all areas in the Oslo region. An increase in fuel tax and a time-differentiated toll ring will drastically change accessibility by car (G_Emp, G_W, G_65+ and G_20-65). Note that G_W, G_65+ and G_20-65 measures indicate the accessibility of a particular segment of the population to different locations in the Oslo region while G_Emp indicates accessibility to employment in different locations. All these measures have similar patterns. They all indicate that accessibility by car to Upper Groruddalen will decrease the most for all segments of the population. Accessibility to

Experience with Measuring Equity and Efficiency

125

employment (G_Emp) and accessibility of the population of age 20-65 (G_20-65) have similar patterns. Table 7.8. Changes in the accessibility measures of Scenario 12 relative to the reference scenario Employment G_Emp

Women G_W

Age over 65 G_65+

Age 20-65 G_20-65

Logsum

1. Oslo West

-1.11

-0.82

-0.29

-1.31

-5.10

2. Oslo, East

-2.19

-1.30

-0.50

-2.01

-5.68

3. Oslo, outer West

-7.15

-5.96

-2.06

-9.57

-5.74

4. Lower Grorurddalen

-4.79

-3.00

-1.15

-4.66

-5.49

5. Upper Groruddalen

-16.09

-18.85

-6.24

-29.98

-8.72

6. Østensjøbyen

-7.86

-12.37

-6.22

-18.06

-4.81

7. Oslo South

-1.16

-3.65

-1.54

-5.53

-9.13

8. West region

0.00

0.00

0.00

-0.01

-3.67

9. Romerike

-0.24

-0.42

-0.14

-0.64

-7.92

10. Follo

-5.03

-8.89

-2.75

-14.05

-4.89

Fig. 7.1. The greater Oslo area

126

F. Ramjerdi, K. Østmoe and H. Minken

A main problem with the gravity approach is that the scale is ordinal. A “logsum measure” closely compares with the changes in the consumer surplus. It also captures the effect of provision of the public transport services. This measure suggests that the benefits from the package in Scenario 12 are not evenly distributed and hence have potential adverse distributional effects. The significance of the observed variations in the geographical distributions of welfare (captured by the logsum measure) is evaluated by the equity measures defined earlier in Section 7.2. Table 7.9 shows a summary of some of these inequality measures applied to the geographical distributions of welfare over the 49 zones that constitute the Oslo region. While almost all measures are quite similar in size in both scenarios, they suggest that the geographical distribution of welfare is more even in the reference scenario than in Scenario 12. Table 7.9. Summary of some inequality measures in Scenario 12 and the reference scenario for the Oslo region (49 zones) 49 zones Mean

Scenario 12

Reference

498.35

504.89

Range Ymax –Ymin

360.67

361.56

Variance

5175.71

5072.69

Coefficient of variation

0.144

0.141

Relative mean deviation

0.1070

0.1118

Logarithmic variance

0.0059

0.0056

Variance of logarithms

5.1210

4.5333

Theil

0.2480

0.2366

Table 7.10 shows the summary of all the inequality measures (described in Section 7.2) applied to the geographical distributions of welfare over 10 zones that represent the Oslo region. A comparison of the measures in this table with the corresponding measures in Table 7.9 shows that the level of zonal aggregation affects the size of most measures. This is partly due to the approximations in aggregation (not properly weighted) as well as the properties of the measures. This table also suggests most measures are quite similar in size in both scenarios and that the geographical distribution of welfare is more even in the reference scenario than in Scenario 12. Table 7.10 also shows the sensitivity of the Atkinson and Kolm measures to the inequality aversion parameter. The Atkinson measure is more sensitive to the value of the inequality aversion parameter than the Kolm measure.

Experience with Measuring Equity and Efficiency

127

Table 7.10. Summary of inequality measures in Scenario 28 and the reference scenario for the Oslo region (10 zones) 10 zones

Scenario 28

Reference

Mean

519.09

525.29

Range, Ymax –Ymin

115.20

113.01

Variance

1714.08

1710.49

Coefficient of variation

0.0798

0.0787

Relative Mean Deviation

0.0703

0.0697

Logarithmic variance

0.0013

0.0013

Variance of logarithms

5.2007

5.2205

Theil

0.0014

0.0013

H = 0,0001

0.0000003

0.0000003

H = 0,001

0.0000033

0.0000032

Atkinson

H = 0,005

0.0000163

0.0000616

H = 0,01

0.0000326

0.0001260

Kolm D = 0,0001

0.0373

0.0372

D = 0,001

0.3765

0.3757

D = 0,005

1.9607

1.9563

D = 0,01

4.0774

4.0663

0.04199

0.04118

Gini

While the property of a measure provides information about its change with a translation, it is relevant to get some sense of the level of the change, if any. To get an understanding of the size of the change, the measures were calculated for both scenarios (Scenario 12 and the reference scenario) after a translation. The translation was performed by subtracting from welfare (logsums) 443 units. The aim was to avoid negative values for the welfare measure as the result of the translation and to obtain small values for welfare. Table 7.11 shows the summary of the results. A comparison of Tables 7.11 and 7.10 shows that the size of the measures, which are not translation invariant, changes significantly. These measures suggest that the geographical distribution of welfare is more inequitable in Scenario 12 than in the reference scenario once the translation is performed. While this exercise suggests that accessibility and equity measures can be applied to the evaluation of potential changes in the distribution of welfare caused by a package of instruments, one needs to apply them cautiously. Accessibility measures, other than a logsum measure, are ordinal and hence it is problematic to apply equity measures to examine changes in their distribution. The logsum measures in Table 7.8 suggest that the distribution of benefits of the package in Scenario 12 is potentially uneven over the Oslo area. The difference

128

F. Ramjerdi, K. Østmoe and H. Minken

between the different areas is as high as 210 Euro/year for an average traveller. Yet the size of the different equity measures (see Tables 7.9, 7.10 and 7.11) varies significantly as the result of the level of spatial disaggregation and a translation in the measure of welfare. Similarly, some of the measures are quite sensitive to the scale of the welfare measure. This illustrates that relating the equity objective to a predefined value on any of these measures is not a desirable approach. Once we have defined the unit for comparison and the distributional concern to be addressed, it will, however, often be possible to rank alternatives with respect to equity. Furthermore, it is difficult to make a judgement about the equity implications of a policy on the basis of a single measure and without a thorough examination of several measures and their implications. Table 7.11. Summary of inequality measures in Scenario 28 and the reference scenario for the Oslo region (10 zones) after a translation in welfare by 443 units 10 zones & Trans 443

Scenario 28

Reference

Mean

76.09

82.29

Range, Ymax –Ymin

115.20

113.01

Variance

1714.08

1710.49

Coefficient of variation

0.5441

0.5026

Relative Mean Deviation

0.4796

0.4451

Logarithmic variance

0.6310

0.1687

Variance of logarithms

2.5538

2.5326

Theil

0.9287

0.7264

Atkinson H = 0,0001

0.000021

0.000017

H = 0,001

0.000214

0.000167

H = 0,005

0.001072

0.000838

H = 0,01

0.002150

0.001679

D= 0,0001

0.0373

0.0372

D=0,001

0.3765

0.3757

D= 0,005

1.9607

1.9563

D= 0,01

4.0774

4.0663

0.2860

0.2626

Kolm

Gini

This exercise relies on a partial equilibrium transport model and ex-post evaluation of the equity implication of a package of instruments. Nonetheless, the lessons can be extended to a general equilibrium approach where an explicit form of social welfare function and an inequality aversion parameter is used to address equity concerns. Table 7.10 shows that Atkinson measures with aversion parameters of up to 0.001 favour the reference scenario for equity. With aversion parameters of larger than 0.001, Scenario 12 becomes the favoured scenario. This example illustrates that it is

Experience with Measuring Equity and Efficiency

129

important to explore the implications of the form of the social welfare function (for example, Atkinson versus Kolm) as well as the aversion parameter, perhaps in the form of a sensitivity analysis.

7.6 Some Conclusions Partial equilibrium models of transport or integrated transport and land-use models are the most commonly used planning tools for the evaluation of the impacts of transport policies with respect to efficiency and equity. The lack of spatial details in general equilibrium models limits their applications. The main aim of this chapter is to illustrate some important issues related to the evaluation of efficiency and equity using a partial equilibrium model of transport with examples from Oslo. The first issue is related to the assumption about the marginal cost of public funds (MCF) in the calculation of the net benefit of a package of instruments. As illustrated in this study in the case of Oslo, the net benefits and the ranking of different packages change with the assumptions about the size of the MCF. Pricing instruments such as a fuel tax and a congestion pricing scheme result in significant government surplus but are potentially inequitable. The evaluations of the efficiency as well as the equity implications of any of these instruments are incomplete unless the uses of the government surplus are specified. The use of the government surplus to finance non-efficient public transport or road investments will not address efficiency or equity objectives. Ideally, the use of the government surplus to pursue an overall objective of efficiency and equity should not be limited to the transport sector. Equity and accessibility measures can only provide information about the potential distribution of welfare among a population or over a geographical area. The size of the equity measures is quite sensitive to the level of spatial disaggregation and to the scale and translation in the measure of welfare. While it should in many cases be possible to pass judgment on which among a set of alternatives is the most equitable, relating the equity objective to a predefined value of any of these measures is not a desirable approach. Furthermore, it is difficult to make a judgement about the equity implication of a policy on the basis of a single measure and without a thorough examination of several measures. Accessibility measures, other than a logsum measure, are ordinal and hence it is problematic to apply equity measures to examine the changes in their distributions.

130

F. Ramjerdi, K. Østmoe and H. Minken

References Atkinson, A.B. (1970). On the measurement of inequality. Journal of Economic Theory, 2: 244263. Baradaran, S., Ramjerdi, F. (2002) Performance of Accessibility Measures in Europe. Journal of Transportation and Statistics, 4 (2/3). Cowell, F.A. (1977). Measuring inequality. Philip Allan, Oxford. Dalton, H. (1920). The Measurement of the inequality of incomes. The Economic Journal, 30: 348-361. Eliasson, J., Lundberg, M. (2002). Vägavgifter i tätorter. En kunskapsöversikt ur svensk perspektiv. Vägverket Publikation 136, Stockholm. Eriksen, K.S., Markussen, T.E., Putz, K. (1999). Marginale kostnader ved transportvirksomhet. TØI rapport 464, Transportøkonomisk institutt, Oslo. Fridstrøm, L., Minken, H., Moilanen, P., Shepherd, S., Vold, A. (2000). Economic and equity effects of marginal cost pricing in transport, AFFORD deliverable 2A. VATT Research Report No 71, Helsinki. Geurs, K.T., Ritsema van Eck, J.R. (2001). Accessibility Measures: Review and Applications. RIVM report 408505 006, National Institute of Public Health and the Environment, Bilthoven, the Netherlands. Harrison, E., Seidl, C. (1994). Acceptance of distributional axioms: Experimental findings, in: Eichhorn, W. (ed.), Models and measurement of welfare and inequality, Springer-Verlag: 67-94. Kolm, S.-C. (1969). The optimal production of social justice, in: Margolis, J., Guitton, H. (eds.), Public Economics, McMillan, London. Kolm, S.-C. (1976a). Unequal inequalities I. Journal of Economic Theory, 12: 416-442. Kolm, S.-C. (1976b). Unequal Inequalities II. Journal of Economic Theory, 13: 82-111. MC-ICAM (2003). Modelling and Cost Benefit Framework, MC-ICAM Deliverable 3 Version 2. Available at http://www.its.leeds.ac.uk/projects/mcicam/reports.html Minken, H., Shepherd, S., Jonsson, D., Järvi, T., May, T., Page, M., Pearman, A., Pfaffenbicler, P., Timms, P., Vold, A. (2002). Developing Sustainable Land Use and Transport Strategies, A methodological guidebook, Deliverable 14 of PROSPECTS. Myles, G.D. (2000). Public Economics. Cambridge University Press, Cambridge. Parry, I.W.H., Bento, A.M.R (1999). Revenue Recycling and the Welfare Effects of Road Pricing. Policy Research Working Paper 2253, the World Bank, Washington DC. Pigou, A.C. (1954). The Economics of Welfare, 4th Edition. Mcmillan, London. Proost S., van Dender, K. (2002). Methodology and structure of the urban model, in: De Borger B., Proost, S. (eds.), Reforming Transport Pricing in the European Union: A Modelling Approach, Edward Elgar: 65-92.

Experience with Measuring Equity and Efficiency

131

Ramjerdi, F., Rand, L. (1992). The national model system for private travel, Report on Phase II of model development. TØI-report 150, Institute of Transport Economics, Oslo. Sandmo, A. (2000). The public economics of the environment. The Lindahl Lectures, Oxford University Press. Sen, A.K. (1973). On Economic inequality. Oxford University Press, Oxford. Sen, A.K. (1982). Choice, welfare and measurement, Harvard University Press. Sen, A.K. (1992). Inequality reexamined. Oxford University Press, Oxford. Vold, A. (2003). Integrating land use in an established transport model for the Greater Oslo area – Development, estimation and verification. Working paper TØI/1518, Institute of Transport Economics, Oslo.

8

Transport Costs in a Multiregional Equilibrium Job Search Model

Morten Marott Larsen1, Ninette Pilegaard2 and Jos van Ommeren3 1

AKF, Danish Institute of Governmental Research Danish Transport Research Institute 3 VU University, Amsterdam, the Netherlands 2

Abstract In this chapter, we introduce a multiregional equilibrium job search model to analyse the economic effects of intraregional and interregional transport cost changes. The key assumption is that unemployed job seekers and firms with vacancies have to search for each other. The regional unemployment and vacancy equilibrium rates, as well as the wage levels, are endogenously determined. According to the model, decreases in interregional transport costs tend to reduce local and national unemployment and increase vacancies. Model simulations indicate that wages are less sensitive compared to producer prices and that both labour-market search effects and negative externalities have substantial impacts on the overall effect of changes in transport costs.

8.1 Introduction Transport cost reductions are thought to benefit the efficiency of the labour market (Lakshmanan at al., 2001). In this chapter, we analyse the effects of local transport cost changes on the labour market using a multiregional equilibrium labour-market model. The improved efficiency argument relies on two underlying ideas. First, more workers can be recruited within reasonable commuting distances (Lakshmanan at al., 2001). Second, workers can be recruited at lower cost, because workers need less compensation for commuting costs (van Ommeren and Rietveld, 2002). Both ideas imply that local unemployment will fall. We will also allow for endogenous labour-market search, which may lead to higher local unemployment because unemployed workers stop searching for a job in other regions if the intraregional transport costs are reduced sufficiently. The effect on the local vacancy rate is also intuitively ambiguous. The vacancy rate may fall because of the increased probability of finding employees, or it may increase, because the region has become more competitive and will attract more firms with vacancies. It is common to distinguish between different time horizons to analyse the effects of changes in transport costs. Here, we distinguish between the short (typically less than one year), medium (typically 5 to 10 years) and long run (typically more than 30 years) (Blanchard, 2000). This chapter focuses on the

134

M.M. Larsen, N. Pilegaard and J. van Ommeren

medium run. In the short run, reductions in transport costs reduce the costs associated with commuting and may therefore induce unemployed job seekers to accept job offers they otherwise would have rejected. As a result, the local unemployment rate and the local vacancy rate will fall (Lakshmanan at al., 2001). In the short run, labour supply (via migration) and wages will react to changes in the labour market. Whether employers immediately adjust to the number of new job openings due to improved labour-market conditions is not very clear. Presumably, all labour-market effects are local. In the long run, the effects of transport cost reductions are different. House prices, wages, productivity levels and regional levels of labour supply (because of migration) adjust to the reductions and the effect on local and national equilibrium unemployment rates is probably absent. Employees are better off when the increased level of wages exceeds the additional tax that has to be paid for the transport cost reductions. In the medium run, the economy will move to its new equilibrium unemployment rate. It is probable that during the adjustment process wages will react to transport cost reductions and changes in labour-market tightness. For example, wages may fall, because employees receive less compensation for the commuting costs. The main difference in relation to the short-run effects is that the fall in wages induces firms to create more job openings and that labour markets in adjacent regions will be affected. The main differences between medium and longrun effects are that the effects of changes in productivity levels and in labour supply due to migration can still be ignored in the medium run (the cumulative effect of interregional migration is quite small in the medium run). In this chapter, we introduce three related multiregional equilibrium job search models. The essential idea is that the total labour force of the regions is fixed and that the number of vacancies, the number of unemployed and wage levels are endogenously determined, given productivity levels. Due to search frictions in each region, unemployment and a positive number of vacancies exist at the same time. Job seekers search for jobs in their own region and also in other regions. The number of matches between vacancies and unemployed job seekers in a region depends on the regional levels of vacancies and unemployment. Regional labour markets are in equilibrium when the number of individuals who become unemployed in a region is equal to the number of individuals who find employment in this region. Wage levels are endogenously determined. Job seekers and employers bargain about the wage conditionally on the commuting costs. Employers decide to create a vacancy when the expected profit from a filled job exceeds the hiring costs. A main contribution is that we take into account the fact that local transport cost reductions will induce firms to create more vacancies as they will pay lower wages in equilibrium (this is similar to the idea that the size of the labour pool is enlarged). Intraregional transport costs may have ambiguous impact on unemployment because transport cost reductions also change the search intensity of the unemployed workers. Intraregional transport cost reductions increase the local vacancy rate, but decrease the number of vacancies in adjacent regions. Interregional transport cost reductions between regions tend to reduce unemployment and increase vacancies in both regions, and vacancies are more sensitive than

Transport Costs in a Multiregional Job Search Model

135

unemployment to transport cost changes if the endogenous reactions from workers' willingness to search are sufficiently small. An important issue is to what extent wages are sensitive to changes in transport costs. Our results suggest that wages are less sensitive to transport cost changes than to producer prices. Finally, we allow for other environmental externalities and importantly, loss of leisure time. Approximately half of the gains from the labour market generated by the transport cost reductions are reduced through emissions, accidents, noise and lost leisure time. The main results can be summarised as follows: 1. Intraregional transport cost reductions have ambiguous impacts on both local unemployment and unemployment in adjacent regions; 2. intraregional transport cost reductions increase the local vacancy rate, but reduce the number of vacancies in adjacent regions; 3. interregional transport cost reductions between regions tend to decrease unemployment and increase vacancies in both regions; 4. vacancies are less sensitive to transport cost changes than unemployment if the endogenous reactions from workers’ willingness to search are sufficiently small; 5. wages are less sensitive to transport cost changes than to producer prices; 6. both labour-market search effects and negative externalities related to costs have substantial impacts on the overall assessments of the consequences of changed transport costs. 1. and 4. are discussed in more detail throughout this chapter, because ambiguity and sensitivity depend on the assumptions made. Results 2., 3. and 5. are supported in the three related search models examined here, whereas 6. is only related to the case in section 8.4.2 because this is the only part of the study which deals with the negative externalities: emissions, accidents, and noise. Road pricing is considered to be a transport cost. With the introduction of road pricing transport costs increase, but we have also dealt with the opposite case: lower road pricing and lower transport costs. In simulations with the basic version of the model and in the applied case of the island of Zealand, there are no endogenous congestion costs which affect the transport costs, but congestion costs are part of the applied case of road pricing in Denmark. Section 8.2 includes a theoretical description of the basic labour-market search model. Then in section 8.3, the basic model is used to simulate in a three-region setting in which there are different levels of productivity. Section 8.4 deals with theoretical extensions of the basic model and section 8.4.1 applies the extended model to the analysis of the effects of road pricing in Denmark in a multi-regional version. Section 8.4.2 describes a two-region case with lower transport costs on the island of Zealand and section 8.5 concludes.

136

M.M. Larsen, N. Pilegaard and J. van Ommeren

8.2 The Basic Model 8.2.1 The Matching Model Individuals are assumed to be identical, an assumption that is relaxed later. Firms are identical, except for the level of productivity, which is regionally specific and exogenously determined. In the long run, the assumption of exogenous productivity levels may be hazardous, but in the medium run it is a reasonable assumption. Individuals are either unemployed or employed. The unemployed search for jobs, the employed do not search. 1 Here, we use the interpretation that a firm only consists of one job, which is either filled or unfilled. Firms do not relocate filled jobs to other regions (due to prohibitively high relocation costs), but may relocate unfilled jobs freely at no cost. Thus, employers are free to choose the location of the vacancy. In order to fill a job, firms advertise a vacancy. We assume that individuals do not move residence either within or between regions. This assumption is quite strong, but must be viewed as a simplifying assumption namely that a large number of people are not completely mobile due to residential moving costs. Later on, this assumption will be relaxed. It is presumed that there are n regions defined as a homogeneous environment. So, within the region all agents (workers and firms) face the same environment. The labour force of each region contains Li individuals. We let ui denote the unemployment rate – the number of unemployed as a proportion of the labour force in region i, and vj, the vacancy rate, the number of vacant jobs, as a proportion of the labour force in region j (i, j =1,..n). The unemployed seek vacancies for which they perceive a positive probability of a match. The unemployed persons are matched to vacancies by a matching function. The matching function Mj gives the number of matches at any one time as a function of the number of unemployed looking for jobs and the number of firms looking for workers in region j (all firms with vacancies). This function is assumed to be increasing in the number of unemployed and vacancies and has constant returns to scale (Pissarides, 2000). The matching function Mj is usually assumed to be a Cobb-Douglas function. An example is:

Mj

– (u L ) i

i

k ij

(v j L j )

1 k ij

,

i

where

¦ k ij  1 . So k

ij

is the elasticity of the number of new filled jobs in region

i

j with respect to the number of unemployed in region i. From a theoretical point of view, the assumption that kij is constant, and therefore does not depend on the 1

An extension would be to allow on-the-job search.

Transport Costs in a Multiregional Job Search Model

137

spatial labour-market conditions, is undesirable. However, empirical studies largely support the assumption that search intensity (of job seekers and firms) is relatively constant compared to temporal and spatial changes in the UV ratio (Russo, 1996; Burda and Profit, 1996). A model with varying search intensity is discussed later. Let qij denote the rate at which unemployed located in region i are hired in region j. Let T ij denote the (regional) labour-market tightness, the number of vacancies in region j relative to unemployment in region i. It follows that:

T ij

Ljv j Li u i

.

The share of unemployed job seekers from region i that finds employment in region j depends on the probability that an unemployed job seeker is matched to a vacancy in region j. The rate at which unemployed job seekers in region i find employment in region j is denoted as fij. Given the properties of the contact function, fij is increasing in T ij , but decreasing in T i ' j , i ' z i . This makes sense, if the number of vacancies in j increases relatively to the number of job seekers in i, the job seekers from this region are more likely to find employment in j. The rate at which unemployed job seekers in region i find employment is written as fi.

8.2.2 Job Destruction It is assumed that jobs are destroyed at rate O , which does not vary among regions.

8.2.3 Equilibrium Employment and Unemployment In the steady state equilibrium, the number of individuals in region i who become unemployed (O(1-ui)) must be equal to the number of individuals in region i who become employed (ui.fi). The equilibrium regional unemployment rate ui can then be written as follows:

ui

O O  fi

.

Similarly, it can be shown that the commuting share, eij, defined as the number of jobs filled by residents from region j in region i, divided by the labour force in region i, can be written as:

138

eij

M.M. Larsen, N. Pilegaard and J. van Ommeren

f ij

O  fi

.

8.2.4 Job Creation Suppose that a firm opens a vacancy. The value of the vacancy depends on the search costs and the expected income stream of finding a suitable worker, which depends on the difference between regional productivity level pj and the costs connected with the job. Let Jij denote the value of a job filled by a worker living in region i. For example, the value of a job in region j occupied by a worker living in region i depends on the discounted income stream and can be written as:

rJ ij

p j  wij  OJ ij or, similarly, J ij

p j  wij rO

where wij is the (bargained) wage paid by a firm in region j to a worker in region i. Given Jij and information on the regional contact rate, the expected income stream can be calculated. In equilibrium, it is assumed that all profit opportunities from new jobs are exploited, driving rents from vacant jobs to zero. So, essentially we presume free entry (and exit) of firms. This assumption implies that in equilibrium, regional labour-market tightness is such that the expected value from a (new) job in region j must be equal to the expected recruitment cost. So, the difference between regional productivity and expected wage level is equal to the expected capitalised recruitment cost.

8.2.5 Workers Here, we determine the lifetime utility of the unemployed and employed workers who anticipate (future) changes in their labour-market status. For unemployed workers, lifetime utility depends on the expected utility of not working (mainly unemployment benefit) which is presumed to be the same in each region, and the expected utility of getting work in a region j. For employed workers, lifetime utility can be defined as the difference between the wage and the travel costs (monetary and time costs) plus the expected loss of losing the job.

8.2.6 The Spatial Wage Equation A bargaining process between firms and preselected job applicants determines the wage rate. The wage derived from the Nash bargaining solution is the wage that

Transport Costs in a Multiregional Job Search Model

139

maximises the weighted product of the workers’ and firms’ net return from the job match (Pissarides, 2000). The parameter E may be interpreted as a relative measure of labour's bargaining strength (0 d E d 1). E can be interpreted as the worker's share of the total surplus. It can be demonstrated that:

wij

(1  E )(t ij  z )  E ¦ T ij ' qij ' J ij  Ep j , j'

Thus, the wage is an increasing function of the commuting costs tij and the value of being unemployed, z. Since 1-E < 1, this implies that the net wage (the wage minus the commuting costs) is a decreasing function of the commuting costs. Furthermore, the wage is an increasing function of the productivity level in region j. The penultimate term can be interpreted as the average search/hiring cost per unemployed. Workers are rewarded for the saving of (future) hiring costs that the firm enjoys when a job is formed. Given the wage rate, it is straightforward to calculate the expected wage level. The regional productivity level has a positive effect on the wage level in a region for three reasons. The first reason is that a higher productivity level increases the surplus of the match (the last term). The second reason is that the expected commuting costs are an increasing function of the regional productivity level. In other words, workers are recruited from a larger area, which increases the commuting costs. The third reason is that the average wage costs are higher, since the ratio of vacancies to unemployment is less favourable increasing the labourmarket bargaining position of the unemployed.

8.2.7 Reservation Commuting Costs We have seen above that the wage is an increasing and the net wage a decreasing function of the commuting costs. It will be convenient to make this relationship explicit and we will write the wage rate as wij(tij). This implies that job seekers and firms will only form a match when the commuting costs are less than a certain maximum, which we will label the reservation commuting costs Tij. The reservation commuting costs Tij are defined by the condition that Jij is equal to 0. This condition implies that:

p j  wij (Tij )

0,

and using the wage equation, the reservation commuting costs can be written as:

Tij

pj  z 

E 1 E

¦T ij ' qij ' J ij ' . j'

140

M.M. Larsen, N. Pilegaard and J. van Ommeren

So, the reservation commuting costs are equal to the productivity level minus the unemployment benefits and minus the average search/hiring costs. Clearly, if the commuting costs exceed Tij, workers from i will reject all matches in region j. In the numerical simulations, we will assume that the commuting costs are less than Tij, so firms and job seekers always form a match.

8.3 Simulations with the Fundamental Model We have assumed an economy consisting of three adjacent regions, which contain a workforce of equal size. These regions have different levels of productivity (high, medium and low) and are therefore labelled HP, MP and LP (for details, see van Ommeren at al., 2004). In the base run, we find that in more productive regions, the unemployment rate is lower and the vacancy rate is higher. The main explanation for this result is that a high productivity level induces more firms to open vacancies, which reduces local unemployment levels. The regional variation in the vacancies is higher than the regional variation in unemployment (which is in line with empirical observations for the UK and Finland). Wages are higher in more productive regions. We assume that the regional variation in search intensity does not differ between regions. Nevertheless, the commuting flows are dramatically different. In the HP regions, workers tend to work in their own region, whereas in the LP regions workers tend to work in other regions. As a result, employment is not evenly distributed over space. Nevertheless, in contrast to models which exclude search imperfections, the results clearly imply simultaneously inward and outward commuting (see also Molho, 2001). The results indicate that the average productivity level increases most by reducing transport costs of connections to the most productive regions, because in these regions, the most productive jobs will be created. Reductions in costs of connections to less productive regions are less effective. Reductions in transport costs induce not only additional interregional commuting flows, but perhaps surprisingly, induce intraregional commuting flows. The main explanation for this result is that interregional improvements induce firms to create additional vacancies in the regions which are better connected, and the local labour force benefits the most from additional job openings. We have investigated the effects of changes of various exogenous variables (the regional productivity level, search costs, discount rate, bargaining strength, E, utility of not working, destruction rate). One of the main results is that changes in regional productivity levels have a strong effect on commuting flows. This result may explain the finding of Cameron and Muellbauer (1998) and Heyma and van Koppenhagen (2001), which shows that temporal changes in regional commuting flows are much stronger than those in migration. Vacancies also react much more strongly than unemployment to changes in regional productivity levels. One intuitive explanation is that firms are perfectly mobile (they may open and close vacancies in any region), whereas workers are immobile, but search in all regions

Transport Costs in a Multiregional Job Search Model

141

in competition with other job seekers. Because workers are able to search in adjacent regions, the effect of a change in the local number of vacancies is partly compensated by a change in the opposite direction in the number of vacancies in the adjacent regions. Furthermore, an increase in regional productivity always reduces local unemployment and increases the local vacancy rate, but may have adverse unemployment and vacancy effects on other regions. Nevertheless, wages in all regions increase as a result of a regional productivity improvement. In contrast to the effects of changes in productivity, the effects of changes in search costs, discount rate, bargaining strength, the utility of not working and the destruction rate on the unemployment rates are of the same magnitude as the effects on the vacancy rates. The effects on the commuting flows are negligible. Compared to the main results presented in the introduction of this chapter, we find that intraregional transport cost reductions reduce local unemployment, but may increase unemployment in adjacent regions. The result for local unemployment is unambiguous, because unemployed workers are not allowed to choose between regions when searching for a job. This assumption is relaxed later, which causes the effects on local unemployment to be ambiguous. This is also a reason why vacancies are more sensitive to transport cost reductions than unemployment. The rest of the main results is supported in this simulation with the basic version.

8.4 Theoretical Extensions As described earlier, the equilibrium job search model has been formulated in two other versions as well. These versions have some extensions and differences in some of the assumptions. First of all, these two versions have been formulated as general equilibrium models whereas the first version focused partially on the labour market. This implies that the workers maximise explicitly formulated utility functions while firms maximise profit functions. An important aspect of the explicit utility function is that the workers derive utility of consuming commodities and leisure. The utility function is formulated as follows:

U k ,t i , j Ct i , j  Q k ˜ l t i , j The k’th consumer has in each period t consumption of commodities, C, and of leisure, l. It is assumed that the costs of commuting consist of both pecuniary and time costs. This implies that the commuting costs are two-dimensional and that commuting influences both the consumption of commodities and of leisure. Both depend on the residence region i and the workplace region j. The workers have an exogenous individual preference for consumption of commodities relative to consumption of leisure, Q k , and thus an individual value of time. This means that some workers value leisure time consumption higher than other workers. The implication of this assumption is that some workers will be more willing to accept

142

M.M. Larsen, N. Pilegaard and J. van Ommeren

jobs in neighbouring regions than other workers. We assume that Q k is uniformly distributed. Another important aspect is that firms treat workers identically no matter where the workers live. The wage is negotiated with a representative union member who only searches for a job in the place of residence and this wage is paid to all the workers in a firm. This implies that the firms are not willing to compensate a worker for his individual interregional commuting costs. However, the workers are partially compensated for the commuting costs of local transport through the wage negotiation process. Transport costs now affect the number of commuters between regions by the worker’s individual decisions with respect to how many regions he is willing to search for and accept a job in. This implies that the regional labour supply depends, among other things, on local transport costs. By assumption, the cost of firms of having a job occupied does not depend on interregional transport costs due to the wage negotiation. If transport costs are low, then more workers are willing to search for a job in a neighbouring region. However, other variables also affect labour-market search behaviour and thereby the labour supply, e.g. the wage differences between regions and the unemployment benefits compared to the wages. Furthermore, the probability of finding a job in the residential region affects the workers’ willingness to commute to neighbouring regions. If it is difficult to find a job in a residential region, it is more attractive to accept a job in a neighbouring region. When an unemployed worker is to determine his search strategy, i.e. what regions to look for jobs in he compares the expected present value of each strategy. By solving and comparing the expected present value expressions in steady state for the different search strategies, we can find the marginal values of Q that determine the search strategy for a given consumer. Then the size of the population groups for each type of search strategy can also be found. Let v*(i) be this marginal Q for consumers living in i, i.e. the consumer who is indifferent between searching only at home, i, and searching in the neighbouring region as well. This v*(i) is determined by the relation:

Q*

T (i ) ˜ q(i ) ~ C (i, j )  C (i )  ˜ (C (i, j )  C (i, i )) G  s (i ) T (i ) ˜ q(i ) e l u (i )  l e (i, j )  ˜ (l (i, i )  l e (i, j )) G  s (i )

where C(i,j) is commodity consumption for the worker who is employed in j, and

~ C (i ) is commodity consumption if he is unemployed. lu(i) and le(i,j) are leisure

consumption in the case of being unemployed or employed in j. G is the workers’ expected interest rate, i.e. the rate with which they discount future incomes is 1/(1+G). T(i)q(i) is the probability of an unemployed worker being employed in his residential region i.

Transport Costs in a Multiregional Job Search Model

143

In the formulation of other parts of the model, there are some smaller modifications. One is in the definition of the matching function, which is defined slightly differently:

Mi

( JC i ) ki ( JOi )1 ki

where JC is the number of unemployed workers who search for a job in region i. JO is the number of job openings (vacancies) in the region. Again, we use a CobbDouglas function, but in contrast to section 8.2.1 and the basic version, it is assumed here that the number of matches in a given region is a function of the total number of job candidates in a given region and the number of vacancies in that region. This is a relevant definition since it is assumed that firms in a region do not distinguish between workers from different regions. The description of transport is further developed. In the second version transport is made possible through a foreign commodity that must be purchased when commuting. The third version models transport sectors that consume labour as factor input. The idea that transport is costly is important when evaluating transport. Besides the mentioned extensions, the two versions are developed in two different directions. In the second version, the model has additionally been extended with congestion and more regions and it is calibrated using Danish data. Furthermore, this version is formulated as a small open economy where relations with foreign countries are specified. In the third version, the model is formulated with two regions and as a closed economy and it has been extended with endogenous location of households. It is defined such that there is a possibility of migration and it is described in a two-region setting with negative transport externalities. The negative transport externalities are defined as separable in the utility function and cover air pollution, noise and accidents, but not congestion. Since the second and third versions of the model are general equilibrium models, clearing of the output market is explicitly formulated. However, the commodity markets are described differently in the two versions. In the second version, the commodity markets are local. This implies that domestically-produced commodities can only be consumed in the region where they are produced. This is a consequence of the fact that there are no shopping trips or inter-regional trade in this version. In the third version of the model, there are two types of regional commodities, namely mobile and non-mobile commodities. The regional nonmobile commodity is only consumed locally. The mobile commodity is not the same regionally. Consequently, all regional types of mobile commodities are consumed in all regions. In both versions, there is no modelling of shopping transport. Shopping transport influences both congestion and final consumer prices. In Denmark, the amount of shopping-related travel roughly corresponds to around half of the total commuting kilometres. Furthermore, freight transport is not considered. Freight transport is expected to have some feedback effect on the level of congestion. Freight transport prices would also be affected if road pricing on trucks was implemented or if new infrastructures were built. In the third version of the model

144

M.M. Larsen, N. Pilegaard and J. van Ommeren

in this chapter, the mobile commodities are simply delivered without charge. Both shopping and freight transport would improve the applied cases, but we would not expect major changes in the results. It is a far more important matter how the commodity markets are modelled. The spill-over effects from local production have significant consequences for the results. The second version of the model includes foreign trade. There is an imperfect substitute for the domestic commodity, produced abroad. The consumers in each region demand this commodity, but since there might be regional price differences for the domestic commodity, the foreign commodity might be demanded in different shares. Domestically-produced goods are similarly exported (abroad). In the third version of the model, there is no foreign trade. In both versions, unemployment benefit is national. In the second version, it is modelled as a fixed level while in the third version it is modelled as a fixed share of wages. Furthermore, these two model versions have been calibrated to actual data sets to represent more realistic economies. Version two describes the Danish economy while version three only describes a Danish region. We now turn to the presentation of a policy experiment in each of the two case versions.

8.4.1 Road Pricing and Regional Distributional Effects The first policy experiment is performed in a multi-regional version of the model. The model is presented in Pilegaard (2003). This version has been used to illustrate the regional distributional effects of a very simplified road pricing system. In this version of the model, the parameters are chosen to replicate total aggregate commuting in Denmark in 1999. Data for production, employment, wages etc. are also chosen to replicate the 1999 figures at both national and regional levels. Data and calibration are documented in Pilegaard (2003) and Munksgaard and Pilegaard (2000). To find the distances and the time use between the different regions, we have derived data from a GIS model. The data and the project behind its generation are described in Leleur at al. (2002). The spatial formulation is an important feature of the model. Since households are distinguished according to their location of residence and job, the model can be used to analyse the geographical distributional effects of policy changes. Furthermore, with the spatial formulation it is possible to assign commuting transport to an artificial network connecting each of the regions and not just to quantify total transport. This is relevant for several practical and political reasons. The economy's geography is described by a nested tree structure. This version is developed from a basic version with three regions where each region is split into three regions to a total of nine regions as presented in the Fig. 8.1 and 8.2. In principle, the description of the economy at each level is the same. In practice, however, this is not completely possible. Overall, the principle in the nest structure is that the total activities of the three connected regions at the lower level add up to the activity in the upper region. This

Transport Costs in a Multiregional Job Search Model

145

means, for example, that total commuting inside and between the three regions at the lower level equals total commuting inside the corresponding upper region. The total commuting in the economy now equals the sum of commuting inside each of the three regions at the upper level and total commuting between these three regions. In the following, the upper level is denoted the regional level and the lower level is denoted the local level. The total economy is the national level. The three-region structure makes it possible to describe the regional effects of region-specific policy changes, for example the construction or improvement of infrastructure or special regional subsidies and we are able to describe the geographical distributional effects. With the nested structure, we can obtain a more detailed geography. This is also important for the geographical distributional effects. When analysing the effects of local road pricing, it is important to include the geographical asymmetry as well as a level of detail where commuting takes place.

23

21

22

31

12 32 33 13 11

Fig. 8.1. The geographical map of Denmark

Production and employment are assumed to take place in the regions at the local level only, i.e. in the nine regions. The production of each region at the regional level is therefore the sum of production in the three corresponding local regions; the same applies to employment. The population is defined in the local regions.

146

M.M. Larsen, N. Pilegaard and J. van Ommeren

There are only wage differences in the three upper regions. Wage bargaining takes place at the regional level and the result is therefore potential wage differences in the three different regions, but not between three corresponding regions at the local level. The chosen level of detail for the Danish economy almost mirrors the counties. The population in each of the local regions ranges from around 200,000 to 800,000. In the following, to make the distinction between the two levels easier to understand we denote the regions as local level counties. At the regional level they will be denoted regions.

National level

Regional level Region : I

11

2

1

12

13

21

22

3

23

31

32

33

Local level County : i

Fig. 8.2. The nested structure of the economy

Let index I denote the regions at the regional level, I=1,2,3 and let i denote the counties at the local level, i=11,12,13,21,22,23,31,32,33 where i=11,12,13 are the three counties that correspond to region I=1 etc. J and j defined similarly. The population is initially defined at the local level. In each of the counties the total population is n(i). The search strategy of unemployed workers is only defined at one nest level, either the local or the regional level. At the local level, the consumers’ problem and search strategy are straightforward as explained in the basic model. For the unemployed workers who search at the regional level, they only decide in which region they want to search, I, and not at what local level county, i, they want to search in. It is assumed that they will always search in the whole of their residential region, i.e. all three related counties. This implies that an unemployed worker in region I=1 who decides to search in neighbouring region I=2 and/or I=3 automatically searches in all three of the local counties i=21,22,23 and/or i=31,32,33, as well as in the residential counties i=11,12,13. The workers

Transport Costs in a Multiregional Job Search Model

147

are distinguished according to search only in neighbouring counties at the local level or search in neighbouring regions at the regional level. We say that they belong to the local population, respectively the regional population. The distinction is determined by the marginal preference parameter Q*(reg)(I) that is determined at the regional level in the same way as at the local level. As mentioned earlier, an unemployed worker who searches at the regional level does not distinguish between three related counties, but searches in all of them. The distances between counties at the local level are exogenously given by the infrastructure. At the regional level, the average interregional distance is exogenous while the intraregional distance is endogenous. It is calculated on the basis of the weighted average distance that a worker living in region I has to travel weighted according to population densities in each county and the different probabilities for a worker, searching in all three counties, of obtaining a job in each of them.

Part of populationsearching only in local level regions

Q

0

(

)

1

Part of population searching in regional level regions

Fig. 8.3. Endogenous labour-market search

We have chosen to treat workers searching in the regional neighbouring regions as a pool of identical workers. If the workers from each of the local counties were treated differently their marginal value of Q for searching in regional neighbouring regions would probably be different, but for practical and interpretational reasons we have chosen the pool formulation. Other variables at the national level are similarly determined according to a weighted average. In steady state, the demand for labour will, as in the basic model, be equal to the number of employed workers. Now, commuters from the regional level also have to be included. Since firms consider all job searchers to be identical they choose their workers randomly from the pool of job searchers. This implies that the regional commuters implicitly are allocated to firms in the local

148

M.M. Larsen, N. Pilegaard and J. van Ommeren

counties according to the relative probability of finding a job in each of the counties. This nesting structure is chosen for different reasons. First of all, one nest level, i.e. three regions, is not enough to generate sufficient commuting between the regions. Therefore, we need more detail in the geographical description. We could have chosen to expand the basic model with more regions, but this would in practice have been problematic. The problem increases dramatically in complexity when more regions are added. Many potential search strategies have to be compared. Therefore, we chose this nesting structure with which we can increase the level of geographical detail without a dramatic increase in complexity. Furthermore, detailed data sets can be dealt with using the chosen nesting structure . We also find this nesting structure with the distinction between local and regional search strategy convenient for purposes of interpretation. It seems reasonable to believe that workers who are willing to accept a job in a region far from home are also willing to accept a job in a region next to it if the remaining job characteristics are the same. But, of course, it need not necessarily be so. Furthermore, this setting has the advantage that it is possible to scale the problem up and down and the national perspective can be maintained at the same time as regional distribution is described. Finally, the nesting procedure could (with some adjustments) in principle be extended to more than two nests. Congestion is included in the model in a speed-flow formulation so that the more commuters, the longer the travel time. It is formulated so that the time required to commute a given kilometre on a given part of the network (local or interregional in the regions) increases with the number of commuters on that given part of the network. This formulation of congestion implies that it has a feedback effect on the decision of the workers. The more congested a given part of the network is, the longer it takes to commute this distance and the less attractive it is to commute the distance. This implies that the congestion effect reduces the effects of given changes in transport policies. The inverse speed-flow relationship is formulated as an exponential function, which is a commonly used functional form (Mayeres 2000, O’Mahony at al. 1997). Road Pricing We now turn to the analysis of the regional distributional effects of a simplified road pricing system. The road pricing system is introduced in the form that all kilometres driven inside the capital, region 31, are taxed while all others are not. The tax is set equal to 0.30 DKK2 per kilometre. This experiment is performed, since it is often debated whether road pricing should be used as an instrument to reduce congestion. Road pricing is introduced here very simply and therefore it is not strictly comparable to a real system. We simply introduce a kilometre tax in one location. We do not differentiate according to time of day and thereby we are unable to examine whether the road pricing can actually make commuters change their time of travelling, which is one of the main arguments for road pricing. Furthermore, we 2

1 DKK § 0.14 EUR.

Transport Costs in a Multiregional Job Search Model

149

neglect mode changes and other transport purposes. However, we are able to examine how the economy reacts to this geographically asymmetric shock and how the different regions will be affected. This is relevant since an important argument against road pricing is that it severely and negatively affects the residents in the city. We simply assume that the increase in government revenue by this tax is paid out to all consumers as lump sum payments. First, we take a look at the national effects and some of the results that at first merit attention. There is a slight national reduction in total production and employment (0.12% and 0.15%). At the regional level, the production and employment increase in region 1, but decrease in the remaining regions. Total transport in the economy decreases by 2.4% and the lump sum subsidy increases by 0.55%. This illustrates a trade-off between reducing transport and congestion and ensuring production and employment. The first impression is that county 31 is not especially hit by this policy; not more than the two other counties in region 3 – at least not when looking at employment and production. As expected, fewer workers want to commute to region 3. Actually, now only a small share of the workers in region 1 commutes to region 3 whereas no one from region 2 commutes to region 3. Unemployment decreases relatively more in county 21 (-21.8%) and increases relatively more in counties 22 and 23 (+12.2% and +9.3%). In county 31, there is only a small change (+0.17%), while it is larger in counties 32 and 33 (-3.8% and +8.1%). Wages and prices are affected as well, but the changes are small. They decrease in region 1 and increase in the remaining regions. The level of consumption for employed workers increases in all counties in region 1 (around 0.35%). Other groups also experience increases (from 0.12% to 0.28%) except for the groups living in county 21 and county 31 and the commuters from 32 and 33 to 31. When studying the distributional effects, we analyse the consequences of the policy change for three selected counties, each in a different region: county 31, county 21 and county 13. The three counties have different characteristics. County 31 is the capital region with many inhabitants and in this region there are initially more workplaces than residences. Furthermore, county 31 is in the part of the country with the highest wages. County 21 is a more sparsely populated county, but there are still many workplaces relative to residences. This county is in the part of the country with the lowest wages. In county 13, there are initially fewer workplaces in relation to residences than in the other two counties. With respect to the distances from these counties to the neighbouring counties, there are of course also differences. County 31 has the longest internal distance and the shortest distances to the neighbouring counties as well as the longest distances to the other regions. In county 31, all kilometres used for commuting are taxed. This implies that all commuting is taxed for commuters inside 31 while commuters to and from 31 are only taxed for part of the trip. As a result, a commuter pays the same total road

150

M.M. Larsen, N. Pilegaard and J. van Ommeren

pricing for commuting inside, to and from 31. Employed residents in county 31 have no way of escaping the tax while all other workers can escape it. The increase in commuting costs in county 31 implies that fewer regional workers commute to region 3 and that fewer from counties 32 and 33 search in 31. This reduces the expected duration of an unemployment period for the workers who only search in 31, because the expected duration of a vacancy in 31 increases. The increase in commuting costs in region 3 puts an upward pressure on wages, which increase by 0.7%. This increase in the production costs of firms in 31 reduces production and employment in the firms. Even though fewer of the workers in 31 now come from other locations, the resulting unemployment for the residents in 31 increases a little (0.17%).

production

unemployment : Increase : Decrease

Fig. 8.4. Effects of the introduction of road pricing

For the employed workers from 31, consumption is reduced, since the cost of commuting is increased. On the other hand, unemployed workers in 31 experience a slight increase in consumption (0.01%), which is a consequence of the increased lump sum transfer. This increase outweighs the increase in the consumer price. County 21 shows some interesting effects which appear dramatic, which will be discussed in the following. What are the special features of county 21 that drive these results? The first we notice is as commuting to region 3 gets more expensive fewer prefer to work in region 3. No one is now willing to commute to region 3 from region 2. The labour market in county 21 initially only had limited interaction with the remaining labour markets. Road pricing makes this interaction disappear and the labour market in county 21 is now totally isolated. County 21 is the only one with a totally isolated labour market. Initially there were some workers from county 21 who commuted to region 3 and there were some workers commuting between the counties in region 2. The existence of these commuters in region 2 is a

Transport Costs in a Multiregional Job Search Model

151

result of the modelling of the regional worker; a worker who is willing to search for a job in a neighbouring region also searches in the neighbouring counties at the local level. Since commuting from the neighbouring counties disappears, there are fewer job searchers in county 21. This decrease looks quite dramatic, 22.4%, but actually covers a small absolute number since the number of jobs relative to population in county 21 initially is high (0.987%) and the number of unemployed relative to jobs initially is low (0.018%), lower than in all other counties. These changes imply that the expected duration of a vacancy increases quite a lot (59.8%) while the expected duration of unemployment by only searching in 21 is reduced by 22.3%. This further reduces the incentives to search in more labour markets. This implies increased production costs for the firms and thereby reductions in production and employment. At the same time, the consumers in county 21 experience income gains of the road pricing by an increase in the lump sum transfer. In total, the employed and unemployed workers in 21 experience a slight reduction in consumption. The time spent commuting and the distances commuted inside county 21 are almost unchanged. However, we do observe a slight increase in the total commuted kilometres. Note that this happens despite the reduction in employment of firms and the absence of long distance commuters since some unemployed workers in 21 replace workers from the neighbouring counties. A general comment about county 13 is that the effects of the policy here seem to be quite similar to the effects in the neighbouring counties 11 and 12. However, it is interesting that the effects for the counties in region 1 are typically the opposite of the effects in the other regions. For example, production and employment in firms increase in counties 11, 12 and 13 whilst decreasing in the remaining regions. Why does region 1 behave differently? In the counties in region 1, unemployment increases at the same time as employment in firms increases. This happens as fewer workers now commute to region 3 and therefore there are more workers competing for the jobs in counties 11, 12 and 13 and unemployment increases. The effect on congestion inside county 13 is a small reduction and the number of commuters from region 1 to region 3 exhibits a small decrease. Commuting by the local population in 13 increases, but commuting per local employed worker decreases. This follows from the fact that more people from 13 are employed in the home region relative to the neighbouring counties. Commuters in county 13 receive the increased lump sum transfer from road pricing without participating in its financing. Therefore, consumption from households in 13 increases and the demand effect creates increased output and employment.

152

M.M. Larsen, N. Pilegaard and J. van Ommeren

Change in number of commuters, local level : Increase in commuters : Decrease in commuters

Fig. 8.5. Introduction of road pricing – local level

Change in number of commuters, regional level

: Increase in commuters : Decrease in commuters

Fig. 8.6. Introduction of road pricing – regional level

Transport Costs in a Multiregional Job Search Model

153

It is important to note that when increases in government revenue are used for lump sum transfers as in this case, the unemployed are favoured compared to the employed who have to pay the road pricing. Therefore, this road-pricing scenario in general reduces the incentives to search for jobs in more labour markets and thereby reduces the labour supply compared to the situation where the revenue is used for lowering income taxes on labour. Note that if the revenue were used for lowering these distortionary income taxes then the negative effect on total employment would probably be smaller or even positive. The aggregate welfare effects can be measured by the equivalent variation: Table 8.1. Equivalent variation, local and national, in 1000 DKK per capita County 11 12 13 National

EV(i) 0.8111 0.7921 0.7795 -0.2435

County 21 22 23

EV(i) 0.1748 0.1292 0.0766

County 31 32 33

EV(i) -1.3415 -1.1158 -0.2106

Not surprisingly, the consumers of the capital county experience the largest welfare loss while consumers from the local neighbouring counties similarly lose, since they generally commute regularly to the capital county and their firms suffer from reduced labour supply. All other counties experience welfare gains from the introduction of road pricing. This is mostly a consequence of the income effect that these counties experience, as road pricing is paid in the capital region only. As assumed, this indicates that the welfare effects of road pricing are geographically biased, whilst the underlying regional effects might be smaller than expected. The road-pricing scenario shows that geographically biased policies can be dealt with in this model and that national, regional and local effects can be found. By introducing road pricing, commuting is made more costly in some regions and the incentives to search for jobs there are reduced. The results of the policy scenarios discussed in this section are based on assumptions and the calibration of the model. In the next section, some of these will be discussed in more detail. Local Commodity Markets The assumption of local commodity markets is one of the more crucial assumptions in the model and has an important implication. When workers commute to neighbouring locations to earn high wages, they spend their wages in their residential county. This means that counties, where a large share of the population commutes to neighbouring regions, benefit from their local spending on commodities since increases in local demand imply higher production and employment. We could have used alternative assumptions, but the assumption of local good markets is chosen as the best simplification when shopping transport is neglected. The cost of this assumption is that inter-regional competition among firms is absent. The implications of this assumption for the results of the policy scenario are tested under the alternative assumption of one national market equilibrium. Results

154

M.M. Larsen, N. Pilegaard and J. van Ommeren

are reported in Pilegaard (2003). It is found that the trade-off between employment and commuting is maintained and the relative importance of the effects is comparable. In terms of local effects, there are important differences. The effect on total welfare, EV, actually changes for county 33. Furthermore, in this scenario the national effects are reduced under the assumption of one national commodity market. This is a consequence of the more efficient production across counties. Production and employment losses and the resulting welfare loss that follows the introduction of road pricing, are therefore smaller. The qualitative effects on consumption are also changed for some groups. This is a logical consequence of the fact that output price only reacts at national level and not at local level. To conclude, the qualitative effects of the trade-off between employment and commuting are robust in relation to the definition of the markets of domestically produced commodities as well as the quantitative results of the relative importance of these effects. Furthermore, it is a robust result that different regions are affected differently by the policies. On the other hand, the regional distribution of the effects especially with respect to consumption and overall welfare is sensitive to this assumption and the results therefore indicate that the demand effect matters. Results from the Danish Case with Road Pricing We have presented a model for commuting behaviour and interactions between the labour market and transport where geographical differences are present and well defined. We agree that the definition of geography is important when analysing transport policies and their effects on the economy in general. Road pricing in the capital region has been discussed and analysed in the model. The scenario illustrates the trade-off between, on the one hand production efficiency and employment, and on the other hand total commuting effort and congestion. This was also found in a more aggregate and simplified version of the model for other policy scenarios. However, in this geographically detailed model there are considerable differences in the regional effects of the policy. This is important to remember since regional effects can be of crucial importance to the political evaluation of different transport policies. As this model captures some of the regional effects, it provides a tool for improving the understanding of these effects at the same time as the overall effects are maintained. We have demonstrated a CGE model that captures regional effects. However, an even richer geographical description would improve and expand the possible conclusions to be drawn from the model – especially when analysing policies such as road pricing, since a realistic road pricing system would involve considerably more detail. A richer geographical description could for example be achieved in the model by adding an extra nest level or by reducing the size of the economy which the model represents.

8.4.2 The Case of Zealand This section describes a case where two regions are closely linked together by for example infrastructure or a decrease in a road-pricing tax between the regions.

Transport Costs in a Multiregional Job Search Model

155

Economic and environmental effects arise because workers change their commuting and residential search behaviour. The case examined is the island of Zealand, Denmark’s largest island. Zealand is close to being one commuting area as Andersen (2000) demonstrates, but it does not imply that the economy of Zealand is completely integrated. The counties of Storstrøm and Vestsjælland have relatively poor links to the metropolitan area of Copenhagen. This section analyses a hypothetical transport investment between two Danish regions, the metropolitan area of Copenhagen and the two counties of Storstrøm and Vestsjælland defined as the rural area in Fig. 8.7. Note that here the islands of Lolland and Falster and other smaller islands are defined as Zealand.

ŰŰ The metropolitan area of Copenhagen ŰŰ The rural area

Fig. 8.7. The two regions of Zealand

The definition of the two regions is derived from by Tonboe (2002). The model is documented in Larsen (2005), which includes a detailed description of all equations. This section presents the key components of the model. Only the labour force of the two regions is included in the model and the only factor input in production is labour. The workers in the labour force are able to move residence and choose where to search for a job. Only unemployed workers search for a job and they select one of the following strategies: The residential search strategy (RSS), the commuting search strategy (CSS), or the moving search strategy (MSS). Compared to previous model versions, this one includes the MSS strategy. MSS is typically chosen if the worker has little or no utility from his residential location. The main differences

156

M.M. Larsen, N. Pilegaard and J. van Ommeren

between RSS and MSS are that a commuter has to consider commuting and time costs whereas a mover has to be willing to live in both regions. In both regions, there is a commodity-producing sector which produces a regional commodity which is different from that produced in the other region. The commodity-producing sector is the dominate sector in both regions. Two regional sectors produce local housing using local labour. In the long run, the supply curve for local housing is horizontal which would not have been the case if land was included in the model as a factor input. Hereby, all short-run problems in the local housing market are ignored. Housing in the model plays an endogenous role in the search strategy of the workers. A region with higher housing costs will be less attractive than a region with lower housing costs. The price of housing is related to development in the local wage and not directly with the number of inhabitants in the region. In the short run, it would be natural to assume that an increasing population would result in increasing housing costs. The same effect is in the model, but the underlying theory differs. With an increasing population, the local labour needed for producing local housing increases which implies that less mobile goods are produced in that region. This has a tendency to increase the price of the regional mobile goods and thereby also the regional wage and the regional housing price. Thus, growing population will result in increasing housing costs, other things being equal. Labour is also needed to make commuting possible. There are four commuting flows: Inside the metropolitan area, from the metropolitan area to the rural area, inside the rural area, and from the rural area to the metropolitan area. Each commuting flow is represented by a transport sector, which uses labour as an input. In this sector, it is assumed that the labour of the transport sector locates in the place of residence of the commuters. The regional public sector in the model does not produce any goods. This sector determines the level of the regional taxes and the regional tax deductions. Furthermore, a national level of unemployment benefit and tax deductions for commuting exist. The public sector has a balanced budget because difference between total profit and total loss is transferred via lump sum transfers at a national level to every member of the labour force. When constructing the model, initial work with building the database is important. Data are available, but assumptions have to be made in the construction phase. In the following, this calibration process is described. The key values for the regional labour market are exogenous in the calibration process and are presented in Larsen (2005). About 900,000 employed workers live in the metropolitan area and around 250,000 employed workers live in the rural area. 39,000 workers commute from the rural area to the metropolitan area, but only 11,000 commute from the metropolitan area to the rural area. The unemployment rate is 9.3% in the rural area and 8.7% in the metropolitan area. In the calibrated steady state, Fig. 8.8 illustrates the choices of search strategy for the workers who have preferences for living in the metropolitan area. P is a uniformly distributed parameter, which describes the degree of preferences for living in a given area, whereas Q is a uniformly distributed parameter for leisure. The top right corner represents a worker with maximum preferences for both place of residence and leisure whereas the bottom left corner is a worker with no

Transport Costs in a Multiregional Job Search Model

157

preference for both place of residence and leisure. The three lines inside the rectangle illustrate the values of the marginal worker who is indifferent between the two strategies. It is assumed that the share of workers who prefer to live in the metropolitan area is equal to the number of workers living in the metropolitan area in 1996, namely 77.4% of the labour force. This assumption in the calibration is chosen because no information is available on workers’ region-specific utility.

Fig. 8.8. Search strategies of the workers who prefer to live in the metropolitan area

94.0% of the labour force with preferences for living in the metropolitan area chose the search strategy RSS while 3.0% chose CSS or MSS, respectively. The number of moves between the metropolitan area and the rural area has been relatively constant for many years, and the net moves have been from the metropolitan area to the rural area from at least 1976. In the calibrated steady state, the net moves are zero. In 1996, around 22,000 people moved between the two regions. The choice of moving is typically not a simple question and therefore it is not known how many of the 22,000 people who have choice of work as an important parameter. The calibrated model yields 4,400 moving workers per year3. When it is remembered that it is only the labour force that is moving in the model, the calibrated number of moving workers seems to be fair. The picture is quite different for the workers who prefer to live in the rural area as Fig. 8.9 illustrates. 65.3% of these workers choose only to search in the

3

Christensen et al. (1987) find that 9.8% of the Danish people in the questionnaire say that workrelated conditions are the primary reason for the last move, together with the wish to relocate. In a questionnaire survey in the Nordic countries, Nordisk Ministerråd (2002), 17.7% of the Danish respondents say that work relations are one of the motives for their last move. Only one out of six work-related moves was by unemployed workers who got a new job, which indicates that search on the job could be a useful model extension.

158

M.M. Larsen, N. Pilegaard and J. van Ommeren

rural area while 27.0% and 7.7% choose the search strategies CSS and MSS, respectively.

Fig. 8.9. Search strategies of the workers who prefer to live in the rural area

As discussed previously, willingness to search for a job is greater in the rural area. One out of four is willing to commute because their valuation of leisure is sufficiently low and some workers are also willing to move to the metropolitan area if the preferences for living in the rural area are sufficiently low. The large share of workers choosing CSS results in commuting, which creates externality costs. It is assumed that the workers do not consider externality costs when they choose their search strategy. The lump sum transfers from the public sector are the same in the two regions. The regional profit is shared equally among all workers. When regional incomes are compared, it is also necessary to compare the price levels in the two regions. Commodity prices are the same in the two regions, but there is a difference in the level of house prices. Regional housing prices are determined by the regional wage, so housing prices are higher in the metropolitan area. Negative Transport Externalities There are three kinds of negative externalities arising from commuting in this model version: air pollution, noise and accident costs. The Danish Road Directorate (2002) estimates the level of air pollution in cities to be twice as high as in rural areas. These estimates are used with the extension that 25% of air pollution affects both regions. It is also assumed that noise costs are high in the metropolitan area. Accidents are differentiated with respect to interregional commuters because

Transport Costs in a Multiregional Job Search Model

159

it is assumed that interregional commuters tend to use the safer motorway. The local and global externality costs of commuting per worker are nearly 1,000 DKK in the metropolitan area and around 750 DKK in the rural area. By multiplying by the total number of workers, the costs to the regions are approximately 1.2 billion DKK. Lower Transport Costs It is assumed in the model that there is no cost of implementing improvements in infrastructure or by changing the level of the road-pricing tax. The policy appraisal is therefore that the change in the transport system should be carried out only if the total gain is greater than the actual cost of the new infrastructure or the administration costs of changing the road pricing scheme. It is assumed that only the connections between the two regions are improved. The first improvement in the case of Zealand is that less transport effort is needed to commute between the regions. The second improvement is a reduction in the commuting time of interregional commuters. It is assumed that the initial cut in transport cost is around 1,000 DKK per commuter per year and time savings from the metropolitan area to the rural region are around 6.0 minutes per trip, and because of less congestion commuters from the rural area to the metropolitan area save 6.7 minutes per trip. With the maximum valuation of time, this represents around 3,000 DKK per commuter per year. An additive welfare function with the same weight for all workers can be established to give a better overview of the results. When interpreting the regional welfare function, it is important to remember that migration has an effect on the regional result because the welfare function defined adds together the utility of the inhabitants living in the region. It is a question whether many inhabitants are desirable or not. The welfare function is based on the regional expected discounted utility and inhabitants in a given region could have discounted utilities from both regions because they also discount the possibility that they may live in the other region in some of the future periods. When the welfare function is defined, the equivalent variation (EV) can be calculated. EV is a measure of how much the population is willing to pay for improvement in the infrastructure. Table 8.2. Regional valuations

EV

Metropolitan area 193.8

Million DKK Rural area -78.3

Total 115.6

The total gain of the additive social welfare function is 115.6 million DKK. It is only the metropolitan area that has a gain, but the regional valuation is influenced by migration. The total valuation can be divided into four parts as presented in Table 8.3. The total income, which can be consumed, is increasing. A small gain is obtained because the labour force lives in the regions which they prefer. This is partly because reductions in transport cost make it possible for the labour force to locate

160

M.M. Larsen, N. Pilegaard and J. van Ommeren

in the region preferred. However, the transport investment results in less leisure and increasing externality costs. Altogether, it is the result of increased employment, but the gain is reduced by less leisure time and air and noise pollution, together with more accidents on the roads because of increased commuting. Table 8.3. Total valuation Million DKK Total result measured in EV 321.3 7.8 -117.7 -95.8

Real income Utility of place of residence Leisure Externalities EV

The flow utilities are only temporary states for a worker, but the development in a given flow utility tells whether or not the temporary state would be more or less attractive for the worker after the investment has been carried out. As expected, the flow utilities are increasing for the commuters who are interregional commuters. Unemployed workers and workers in the metropolitan area have small gains in flow utility whereas the flow utility of working and living in the rural area is slightly decreasing. This indicates that migration is not the only reason why the total EV of the rural area is negative. Lower transport costs imply that more workers choose a CSS. Workers choose CSS more because it is now cheaper to commute to the other area and the workers also save transport time. First, it has the positive effect that unemployment is reduced, which implies increased production. Table 8.4 shows the effects on the labour market. Table 8.4. Labour-market effects

Employed workers Place of production, metropolitan area Place of production, rural area Total Unemployed workers Labour force Unemployment per cent Number of job openings Number of job searchers

Place of residence: metropolitan area

Place of residence: rural area

Total

- 10,986

+ 11,535

+ 549

+ 12,486 + 1,500 - 1,202 + 298 - 0.1% point +3 - 35

- 10,353 + 1,183 - 1,480 - 298 - 0.5% point +135 + 215

+ 2,133 + 2,683 - 2,683 0 - 0.2% point +138 + 180

Second, as Table 8.4 also shows, the labour force in the metropolitan area increases by 298 workers, who have migrated from the rural area. It corresponds to a reduction in the labour force in the rural area of 0.1%. Third, relatively more workers are willing to commute from the metropolitan area to the rural region than the other way around. One reason is that more workers have preferences for living

Transport Costs in a Multiregional Job Search Model

161

in the metropolitan area and other things being equal it indicates this development of interregional commuting. This is not the only reason, because valuation of time, regional prices, taxes, and lump sum transfers also influence willingness to commute. With this interregional commuting pattern, more workers in the rural area by place of production results in greater production in the rural area. This indicates that producer prices in the metropolitan area compared to producer prices in the rural area increase because of a relatively lower supply of the commodities from the metropolitan area. The changes in regional production prices are reflected in regional wages. The regional wage in the metropolitan area divided by the regional wage in the rural area increases by about 0.5%. The size of the unemployment benefit is a convex combination of the regional wages. Here, the unemployment benefit is decreasing compared to the wages in the metropolitan area, which is the main reason why the metropolitan wage is decreasing compared to the producer prices in the metropolitan area. The opposite is the case in the rural area where the rural wage increases compared to the rural producer prices. The profit rate defined by regional producer prices over regional wage is therefore increasing in the metropolitan area and decreasing in the rural area. An implication of this is that it is more attractive for the metropolitan firm to produce, which implies more job openings and longer job opening spells in the metropolitan area. In other words, labour-market tightness increases in the metropolitan area. On the other hand, it also implies shorter unemployment spells. However, these effects are small in the case investigated (less than one day). The commodity-producing sectors are not the only sectors affected which table 8.5 shows. Table 8.5. Sector effects

Commodities Housing Transport From metropolitan to metropolitan area From metropolitan to rural area From rural to metropolitan area From rural to rural area Total

Employed workers by place of work Metropolitan area Rural area Total - 200 1,457 1,257 13 83 96 1,330 - 135 871

549

726 - 132 2,133

2,683

The overall result in Table 8.5 is that more workers become employed. Roughly half of them produce commodities and housing, but because the transport sector is factor consuming, half of the new employed workers produce transport in which there are no utility gains. More workers commute long distances which is the reason why transport uses so many of the extra workers. Minor effects on total employment are due to the housing market. Regional housing prices follow the regional wage. Therefore, housing prices increase rela-

162

M.M. Larsen, N. Pilegaard and J. van Ommeren

tively more in the metropolitan area. The housing price effect is in the concrete case at a sufficiently high level to dominate the income and migration effect, which results in more housing in the metropolitan area. However, the housing sector is growing more in the rural region. If the model was reformulated with a short horizon fixed housing stock the housing prices would not be determined by regional wages and the housing prices would increase relatively in the metropolitan area alone because of increasing income and the labour force. Larger effects on the total employment are due to the changed commuting pattern. More labour is needed to transport the workers both because of increasing production, and also because of longer total commuting distances. Regional commuting transport prices are also determined by regional wages. This implies that the metropolitan area experiences increasing transport prices compared to the rural area. The transport price effects are minor compared to the overall change in transport costs. The interregional commuters still benefit from lower transport costs as compared to the intraregional commuters. The main effect was, as noted above, that more workers chose CSS. All the derived price effects described below also affect the choice of search strategy. Lower housing and transport prices in the rural area attract more workers On the other hand, an increasing wage in the metropolitan area makes more workers search for a job in the metropolitan area. The result of the choice of search strategy is presented in Fig. 8.10 and 8.11.

Fig. 8.10. Development in the search strategies of the workers who prefer to live in the metropolitan area

Transport Costs in a Multiregional Job Search Model

163

Fig. 8.11. Development in the search strategies of the workers who prefer to live in the rural area

As the figures indicate, the dominant effect in both regions is that the horizontal marginal line separating RSS and CSS moves to the right. This is because lower transport costs make it more favourable to commute. Almost 110% more workers choose the CSS strategy in the metropolitan area and the increase is 30% in the rural area. RSS decreases by 3% and 12% and MSS decreases by 6% and 3% in the metropolitan and rural areas, respectively. The increasing wage in the metropolitan area moderates the tendency for an increase in commuting, but the wage effect is not marked because the value of unemployment benefit is reduced in the metropolitan area. The higher wage in the metropolitan area also implies that fewer workers choose MSS even though housing prices moderate this tendency. In the rural area, the MSS strategy is influenced by two main factors. The higher wage in the metropolitan area positively influences MSS in the rural area, but more workers prefer RSS to MSS because of better transport conditions. To summarise, the main results are that more workers commute over longer distances and more workers are employed. In this experiment this version also gives the perhaps counterintuitive result that economic activity increases in the rural area compared to the metropolitan area as a consequence of the lower transport costs. It is mainly due to the calibration of the workers’ preferences for leisure and place of residence and, most importantly, the assumption that the present share of workers living in a given area corresponds to the number of workers who actually prefer to live in that area. Thus, around three out of four workers are potential commuters from the metropolitan area to the rural area and only one out of four is a potential commuter in the other direction. Furthermore,

164

M.M. Larsen, N. Pilegaard and J. van Ommeren

the assumption of the uniform distribution of leisure and location preferences is not empirically founded and more information on this is desirable. Wages pull the results regarding commuters in the opposite regional direction because the relative regional wage increases in the metropolitan area. This implies that workers with RSS earn money in the metropolitan area and this is the main reason why the flow utility is increasing in the metropolitan area for a worker with RSS, whereas it is decreasing in the rural area. This effect together with the fact that more workers move to the metropolitan area imply that the regional EV is increasing in the metropolitan area and decreasing in the rural area. When interpreting the results, one should keep in mind that general economic tendencies are not included in the model. For example, if the demand for high-tech commodities is continuing to expand it could benefit the metropolitan area because the metropolitan commodity is more high-tech intensive. This would move economic activity from the rural area to the metropolitan area and perhaps dominate the overall effects described in this section. In other words, the results here are only one part of future economic development on Zealand – but the results could have significant positive (and negative) influence.

8.5 Conclusion In this chapter, we have investigated the labour-market effects of transport cost reductions employing a multiregional equilibrium search model, which extends the seminal work of Pissarides (2000). One of the main advantages of this model is that firms react to changes in the labour market by changing the number of job openings. In contrast to the general practice in spatial interaction models assuming that the number of jobs is given in all zones, we choose a formulation where the number of jobs may change as a result of changes in transport costs. We have formulated the model in different versions with different assumptions of the workers’ behaviour which implies that the effects of transport cost changes on commuting patters are driven by slightly different mechanisms. We observe two important differences in the resulting effects of transport cost changes in the different versions of the model. First we find that endogenous labour-market search includes the possibility that lower transport costs could in fact increase unemployment because the search effort of some workers may decrease (main result 1)4. Therefore, fewer unemployed workers will search in the neighbouring region, which increases the expected duration of their unemployment period. This is a result of the fact that the labour supply of the population of workers to each region is endogenous. This is in contrast to the situation where labour supply in all regions only depends on unemployment as in the basic version of the model. However, we find that lower interregional transport costs between regions tend to reduce unemployment and increase vacancies in both of the affected regions. 4

See the main results in the introduction.

Transport Costs in a Multiregional Job Search Model

165

The extension of the endogenous determination of the willingness to search affects the sensitivity of transport costs on vacancies relative to unemployment (main result 4). The reason is that the changes in transport costs as the first order effect change the search strategy of the unemployed workers and thereby the supply of labour in a given region. This affects unemployment. On the other hand, the costs of the firms of having a worker employed might be less influenced by this change in transport costs. This implies that employment in firms is less sensitive to transport costs than with the former model formulation. This difference in effects is again a result of the specification of the workers’ search decision and the single wage negotiation in firms. In both versions of the model, we find that main results 2 and 3 are supported, i.e. we find that intraregional transport cost reductions increase the local vacancy rate, but reduce the number of vacancies in adjacent regions and that interregional transport cost reductions between regions tend to decrease unemployment and increase vacancies in adjacent regions. We also find support for main result 5 in both versions that wages are less sensitive to changes in transport costs than producer prices. Finally, we find that labour-market search effects and negative externalities have a substantial impact on the overall assessment of changes in transport costs (main result 6).

References Andersen, A.K. (2000). Commuting Areas in Denmark. Copenhagen, AKF Forlaget. Blanchard, O. (2000). Macroeconomics, second edition, London, Prentice Hall International. Burda, M.C., Profit, S.. (1996). Matching across space: Evidence on mobility in the Czech Republic. Labour Economics, Elsevier, 3(3): 255-278. Cameron, G., Muellbauer, J. (1998). The housing market and regional commuting and migration choices. CEPR Discussion paper, no. 1945. London, Centre for Economics Policy Research. Christensen, A.L.S., Christoffersen, H., Madsen-Østerbye, K., Smitt, D.A. (1987). Boligmarkedet i Danmark. Copenhagen, AKF Forlaget. Heyma, A., van Koppenhagen, P. (2001). Verkenning van individuele pendel- en migratiebeslissingen. Delft, TNO, RRO, 025. Lakshmanan, T.R., Nijkamp, P., Rietveld, P., Verhoef, E.T. (2001). Benefits and costs of transport. Classification, methodologies and policies. Papers in Regional Science, 80, 2: 139-164. Larsen, M.M. (2005). Essays in Regional and Transport Economics. PhD Thesis, Institute of Economics, University of Copenhagen. Leleur, S., Kronbak, J., Rehfeld, C., Pilegaard, N. (2002). Utilisation of GIS in the TEAM model. Working paper. Centre for Traffic and Transport Research, Technical University of Denmark. Mayeres, I. (2000). The Efficiency Effects of Transport Policies in the Presence of Externalities and Distotionary Taxes. Journal of Transport Economics and Policy, 34, 2: 233-260.

166

M.M. Larsen, N. Pilegaard and J. van Ommeren

Molho, I. (2001). Spatial search, migration and regional unemployment. Economica, 68, 269-283. Munksgaard, M.B., Pilegaard, N. (2000). Team-modellen – Dokumentation. Working paper, unpublished, Copenhagen. Nordisk Ministerråd. (2002). Nöjda så in i Norden? Nordisk Ministerråd, Nord 2002:6, Copenhagen.. O'Mahony, M., Kirwan, K.J., McGrath, S. (1997). Modeling the Internalization of External Costs of Transport. Transportation Research Record, 1576: 93-98. Pilegaard, N. (2003). Essays in transport and the economy. PhD Thesis, Institute of Economics, University of Copenhagen. Pissarides, C.A. (2000). Equilibrium Unemployment Theory. 2. edition. Massachusetts Institute of Technology. Russo, G. (1996). Firms' Recruitment Behaviour. Amsterdam, Thesis Publishers. The Danish Road Directorate. (2002). Trafikøkonomiske enhedspriser 2000. www.vd.dk Tonboe, J. (2002). Værdiernes Geografi. Chapter 13 in Territorial Dynamik. Aarhus Universitetsforlag. van Ommeren, J., Rietveld, P. (2002). A multiregional equilibrium search model for the labour market. Free University, Amsterdam. Research Memorandum. van Ommeren, J.N., Rietveld, P., Woudenberg, S. (2004). The effect of infrastructure improvements on regional labour markets: a multiregional equilibrium job search model, mimeo.

9

Evaluation of the Introduction of Road Pricing Using a Computable General Equilibrium Model

Knud J. Munk Department of Economics, University of Aarhus and Danish Transport Research Institute

Abstract The introduction of road pricing has important budgetary and income distribution consequences. In countries like Denmark, due to high marginal rates of taxation, raising government revenue and redistributing income is associated with substantial distortionary costs and administrative costs. This chapter argues that an evaluation of the introduction of road pricing needs to take into account not only the effects on congestion and on the environment, but also the effects on the government’s budget and the income distribution consequences. A stylised Computable General Equilibrium (CGE) model which represents the interaction of the consumption of transport and of traffic congestion with leisure is used to illustrate this point. In addition, model simulations show that the introduction of road pricing may increase environmental damage, may make it desirable to reduce transport infrastructure and potentially may be associated with a significant double dividend.

9.1 Introduction This chapter assesses the introduction of road pricing within a second best public economic framework using a CGE (Computable General Equilibrium) model. Changes in transport policies often involve very substantial changes in government revenue and have important effects on income distribution. Raising government revenue and redistributing income are associated with substantial costs, distortionary and as well as administrative. In evaluating the consequences of changes in transport policy it is therefore not only the direct effect on traffic and the related effects on pollution, road congestion and accidents that are relevant. The effects on government revenue and the consequences for income distribution may be equally important. The main objective of this chapter is to provide a quantitative illustration of this point. CGE models are increasingly being used for policy analysis. However, the validity of the conclusions drawn is often compromised by convenient, but unrealistic separability assumptions, for example that household consumption and public goods are separable from leisure. An important contribution of the present analysis is the demonstration that more realistic assumptions in this respect can be

168

K.J. Munk

implemented relatively easily in CGE models and are important for the evaluation of the consequences of changes in transport policies. Road pricing can be defined as a tax on road transport differentiated according to the associated external effects (congestion externalities and environmental externalities); road pricing has become a realistic policy option and after the administrative costs associated with its application have been reduced significantly by the application of new technologies. We capture these characteristics by representing road pricing as a tax on transport where the rate of taxation differs between households according to the external costs of their transportation and by assuming that prior to the reduction of the administrative costs associated with road pricing it was not desirable to differentiate the tax on transport, whereas after the reduction it is desirable to do so. The optimal rate of taxation depends on whether the transport is job related (either commuting or input to production) or leisure travel. To simplify the exposition we only consider "transport" non-job related private road travel (leisure travel). It is possible to extend the analysis to other types of transport this would considerably complicate the analysis without contributing to the stated objective of the paper. In section 9.2, the theoretical model on which the analysis is based is formulated; in section 9.3, the public economic insight in the taxation of transport is reviewed based on this theoretical model and a number of stylised facts. In section 9.4, the model is parameterised in the form of a CGE model and in section 9.5 the parameterised model is fully specified. In section 9.6, simulation results derived from the CGE model are presented and interpreted with respect to the consequences, the desirability and the optimal level of road pricing. Section 9.7 summarises and concludes. The notation and the equilibrium conditions are presented in the appendix.

9.2 The Theoretical Model The theoretical model adopted is essentially that of Sandmo (1975, 2000), who considers optimal taxation in an economy where one commodity is associated with a public good externality which can be abated by government expenditures (see also Mayeres and Proost 1997, 2001). We consider an economy with many heterogeneous households who supply only one homogeneous primary factor, labour, labelled 0, and demand two produced commodities, transport and non-transport, labelled 1 and 2 respectively. We represent the households’ preferences by utility functions,

u h =u h  xh , e , hH, where xh {  x0h , x1h ,x2h is the hth household's net demand vector, and e, free road capacity, a public good negatively related to the consumption of transport. The households face consumer prices

q {  q0 , q1h ,q2 , hH, and receive lump-sum income, I h , hH. We assume

Evaluation of the Introduction of Road Pricing

169

that prior to the introduction of road pricing, the price of transport is the same for h

all households, i.e. q1

q1 , but after that it is differentiated according to how

much the households’ consumption of transport contribute to the externalities. The production side of the economy is represented by constant returns to scale production sectors, each producing one output and using only labour as input. Producer prices are

p {  p0 , p1 , p2 .

The government needs revenue for transport infrastructure, G, and for other government consumption, B. The government’s expenditures are financed by h

h

commodity taxes, t { q  p , hH, and a uniform lump-sum tax, L. Thus, unearned income is for all households equal to the uniform lump-sum tax,1 i.e.

I h =-L, hH. Free road capacity depends negatively on the households’ consumption of transport and positively on G, according to e derivative,





e ^ x1h , h  H ` , G . The

we , indicates to what extent the hth household’s consumption of h wx1

transport congests the roads. Using the expenditure function approach (see Dixit and Munk 1977), extended to represent the government’s provision of a public good and external effects (see h

Munk 2000), the conditions for (t , L, G ) to be compatible with market equilibrium may be expressed as

E h (q h , e, u h )

 L , hH

(1)

2

¦ ¦t k 0

h h k k

x (q h , e, u h )  HL  G  B 0

(2)

hH

^

`

e e x1h  q h ,e,u h , h  H , G h

h

h

(3) h

h

h

where E (q , e, u ) is the expenditure function and xk (q , e, u ) , k(0,1,2) the compensated demand functions of the hth household. The social preferences of the government, which is assumed to be inequality



1

2

averse, are represented by a social welfare function, W u , u ,..., u

H

. The

government chooses tax rates and the transport infrastructure to maximise social welfare.

1

L is negative if it is interpreted as the fixed element in a progressive linear income tax schedule.

170

K.J. Munk

9.3 Application of the Theory of Public Economics to the Taxation of Transport and Investment in Transport Infrastructure The externalities associated with transport may be divided into environmental externalities and congestion externalities. Environmental externalities2 (air pollution, noise and accidents) may, as an approximation, be considered separable from consumption, i.e. changes in these externalities do not impact on the pattern of consumption. However, congestion externalities clearly cannot be considered separable from consumption. The consumption of transport clearly depends on the level of congestion giving rise to the so-called feedback effect, i.e. a tax on transport reduces the transport by less than the tax in isolation because the decrease in congestion increases the consumption of transport. The total effect on free road capacity of an increase in the tax on transport is therefore:

de dx1

we 1 wx1 1  e ' E1e

(4)

we is the effect on free road capacity of a marginal increase in road capacity and wx1 1 is the feed back effect which indicates to what extent an increase in free 1  e ' E1e road capacity will increase the consumption of transport. Under first best assumptions, i.e. when the government’s revenue requirement can be raised by costfree lump sum taxes, the optimal rate of tax on transport, the so-called Pigovian tax, is equal to

t1

 MVe

de dx1

(5)

where MVe is the marginal monetary evaluation of free road capacity. However, under the more realistic second best assumptions that the government’s revenue must be raised by distortionary taxation, the optimal tax on transport must not only reflect the external effects associated with transport, but also how the taxation affects the government’s two overall objectives, to raise 2

These are not represented in the theoretical model formulated in section 9.2, but have been added to the parameterised version of the model in section 9.4.

Evaluation of the Introduction of Road Pricing

171

government revenue and to redistribute income. Based on this insight and on the following stylised facts we review the conditions for optimal taxation of transport and for optimal provision of transport infrastructure as a background for the subsequent interpretation of the quantitative assessment of the introduction of road pricing. We assume x that transport requires a relatively large amount of time for its consumption and, partly for that reason, may be considered complementary to the nonmarket use of time (leisure); x that transport is associated with negative external effects in the form of congestion and environmental externalities (air pollution, noise, accidents); x that the share of leisure travel in the consumption is higher for urban households than for rural households and associated with larger congestion costs; x that urban households are better off than rural households; x that an increase in free road capacity will increase the supply of labour.

9.3.1 Taking into Account that Leisure Travel is Complementary to Leisure When lump sum taxation is not feasible, raising government revenue involves discouraging the supply of labour. This can be alleviated by taxing commodities which are complementary to leisure at a higher rate than commodities that are less complementary to leisure (Corlett and Hague 1953). In a one household economy when transport is not associated with externalities, the optimal ratio of the tax on transport relative to the tax on non-transport must satisfy:

t1

t2

OP  H11  H 22  H10 K O OP  H11  H 22  H 20 K O

(6) (7)

H ij (i,j 0,1,2) are compensated demand elasticities, P the net marginal social welfare of income, O the marginal social value of government funds and K

where

is a positive constant. We have that H i 0

D 0V i 0 , where D 0

share of leisure in the household’s full income and where

q 0 x0 is the q0Z 0  I

V i 0 V 0i

is the

elasticity of substitution between commodity i and leisure. By the assumptions made, transport is more complementary to leisure (non-market use of time) than other goods, i.e. H10  H 20 and the optimal tax on transport is therefore, even in the absence of externalities and distributional considerations, higher than on other commodities.

172

K.J. Munk

Based on the assumption that transport is complementary to non-market use of time, it would – if it were possible to tax transport at a different rate than nontransport at no supplementary administrative costs – from an efficiency point of view be desirable to tax transport at a relatively high rate.

9.3.2 Taking into Account that Transport is Associated with Externalities In a one household economy, the conditions for an optimal tax structure may be expressed as:

t1

t2

OP P wT de  H11  H 22  H10 K  §¨ MVe  ¸· O we ¹ dx1 ©O OP  H11  H 22  H 20 K O

(8) (9)

Taking into account that transport is associated with externalities involves that an extra tax will be levied on transport compared to when this is not the case (compare (8) with (6)). The extra tax, however, not only depends on MVe

de as dx1

O P

and on

under first best assumptions, but also on the cost of government funds, how a change in free road capacity influences the tax base,

wT . The larger the we

cost of government funds the smaller the importance of the externality for the optimal tax structure and the more an increase in free road capacity expands the tax base, the higher the tax.

9.3.3 Taking into Account that Leisure Travel is Predominately Consumed by Households with a Relatively High Income Increasing the tax on transport and decreasing the tax on other commodities will redistribute income from the well-off (the urban households) to the less well off households (rural households). The optimal tax on transport will therefore be higher than it would be based only on efficiency considerations. The fact that the decrease in congestion is more important for the urban than for the rural households pulls in the opposite direction.

Evaluation of the Introduction of Road Pricing

173

9.3.4 Reasons for Taxing Transport Higher than Other Goods Based on these stylised facts, there are thus four reasons why a policy-maker may want to tax transport at a higher rate than other commodities: x it is more complementary to leisure (non-market use of time) than other commodities; x it is consumed by high income households relatively more than by the low income households; x its consumption is associated with environmental and congestion externalities; x a decrease in the congestion externality may stimulate the labour supply and thus expand the tax base.

9.3.5 The Optimal Size of the Transport Infrastructure Under first best assumptions, the condition for the optimal provision of the transport infrastructure is:

MVG

MCG

(10)

where MVG is the marginal social evaluation of transport infrastructure and

MCG the marginal costs of its production. However, under second best conditions the condition is

O§ dT · ¨ MCG  ¸ dG ¹ P©

MVG

where

O P

the marginal costs of government funds and

(11)

dT is the effect of an dG

increase in transport infrastructure on the tax base. If an increase in transport infrastructure expands the tax base (via its effect on the free road capacity), then the optimal level of transport infrastructure will be larger than if this effect is not taken into account.

9.3.6 The Effect of the Introduction of Road Pricing on the Optimal Taxation of Transport and the Optimal Provision of Transport Infrastructure The optimal level of road pricing is essentially determined based on the same considerations as the optimal level of taxation of transport. However, some additional

174

K.J. Munk

observations are important as background for the interpretation of the simulation results. Disregarding administrative costs, the introduction of road pricing will always increase social welfare as it provides the government with an extra instrument. Road pricing will make the taxation of transport a more efficient instrument as it will make it possible to differentiate the tax on transport according to the level of congestion externalities. The tax on the consumption of transport for urban households will increase whereas the tax for rural households will decrease. As the taxation of transport has become a more efficient instrument, the introduction of road pricing will justify an increase in the average level of taxation of transport. The introduction of road pricing will thus result in a reduction in the level of congestion. Assuming that tax rates are chosen optimally both before and after the introduction of road pricing, the introduction of road pricing will (disregarding administrative costs) increase social welfare through three different channels: first, by the reduction in the congestion; second, by the redistribution of income from the relatively rich urban households to the rural households; and third by the stimulation of the labour supply due to the higher taxation of transport and the decreased congestion. This latter effect is the result of two opposite effects: on the one hand, the decrease in congestion will encourage the consumption of transport and thus discourage the supply of labour; on the other hand, it will diminish the amount of time required to achieve a given amount of transportation. We have assumed that the latter effect dominates. As it reduces the urban households’ demand for transport, the introduction of road pricing decreases the marginal social evaluation of transport infrastructure,

MVG . It also decreases the marginal costs of government funds,

O P

, as road

pricing increases tax efficiency, which justifies an increase in transport infrastructure; however on balance, the optimal level of transport infrastructure is likely to be smaller after the introduction of road pricing. It is not possible a priori to establish how the introduction of road pricing will influence the total demand of transport. As the effect on environmental externalities is related to the total consumption of transport, road pricing may increase the environmental externalities associated with road pricing although it will decrease congestion.

9.3.7 The Introduction of Road Pricing will be Associated with a Double Dividend A green tax reform is a change in the tax system that reduces negative external effects. We may therefore consider the introduction of road pricing as a green tax reform. A green tax reform is said to be associated with a double dividend if the overall benefits in terms of social welfare exceed the benefits due to the reduction in negative external effects (see Bovenberg 1999). We assume that prior to

Evaluation of the Introduction of Road Pricing

175

improvement of monitoring technology the administrative cost of differentiating the tax on transport according to the level of congestion would have been prohibitive. Only after the reduction in administrative costs due to the improvement of technology the introduction of road pricing becomes desirable. The introduction of road pricing will therefore, as argued above, be associated not only with a first dividend in the form of a reduction in the external effects, but also a second dividend due to income distributional and tax efficiency gains, even if the tax system was optimal before the change in administrative costs. If the tax system was not optimal, the scope for the introduction of road pricing to yield a substantial double dividend is naturally far greater. This contradicts the established wisdom that a green tax reform cannot be associated with a double dividend (see Bovenberg 19993). The contradiction may be explained by the fact that we take into account administrative costs and distributional considerations and that we do not impose separability between the externality and leisure (see Munk 2000).

9.4 The Parameterised Model 9.4.1 Specification of Functional Forms for Free Road Capacity and Environmental Externalities Free road capacity, e, is negatively linked to the congestion externalities and positively linked to government investments in transport infrastructure according to

e Z e  ¦ a1h C1h  x1h , c0h  G

(12)

hH

where

Ze

h

h



h

h

is the endowment of road capacity, a1 C1 x1 , c0

the congestion

th

externality generated by the h household (as a function of its consumption of the h

h

h

transport good, x1 and the time use for transportation, c0 , where a1 is the transport congestion coefficient associated with the hth household) and G the government's provision of transport infrastructure. In addition, in order to represent environmental damage (air pollution, noise, accidents) associated with the consumption of transport, we extend this framework to include environmental externalities specified as

3

Bovenberg (1999) concludes: "The overall message of this paper is disappointing for those who expect substantial non-environmental benefits from green tax reforms. The analysis shows that stringent conditions need to be met in order for an environmental tax reform to yield a double dividend".

176

K.J. Munk

¦x

e

h 1

(13)

hH

9.4.2 Specification of Household Preferences Many CGE models that have been used to evaluate changes in environmental policies impose separability between consumption and leisure and between consumption and the externality (see, for example, Goulder et al. 1999). For the analysis of transport policy issues to impose such assumptions is highly unrealistic and may result in misleading conclusions. We have therefore chosen to specify the households’ preferences using a utility function with explicit representation of the use of time, (the CES-UT) (see Munk 2002) extended with a public good to represent how the provision of free road capacity influences household behaviour and with an additive term Ee e to represent the effect on welfare of the environmental externalities, Fig. 9.1)

 

e , associated with the consumption of transport (see







u U C CT C1  x1 , c10 ;V C1 , e;V T , C2  x2 , c02 ;V C2 ;V D , c00 ;V L  Eee (14) Ci

Ci  xi , c0i ; V Ci expresses for the transport input composite (i=1) and for

the other good composite (i=2) the preference for the amount purchased of i

commodity i, xi , relative to the time used for its consumption, c0 . The elasticity of

V Ci indicates the degree of substitutability between the two. CT  e, C1 ; V T expresses for the transportation composite the preferences of

substitution

free road capacity , e , relative to elasticity of substitution two.

C

the transport input composite , C1 . The

T

V indicates the degree of substitutability between the

C  CT , C2 ; V D expresses the preference for the transportation

composite, CT , and the other good composite, C2 . The elasticity of substitution,

V D , indicates the degree of substitutability between the two. U  C , c00 ; V L expresses the preference for aggregate consumption, C , 0

relative to pure leisure, c0

Z0  ¦ c0i  x0 where Z 0 is the household’s time i 1,2

Evaluation of the Introduction of Road Pricing

endowment. The elasticity of substitution, ability between the two.

177

V L , indicates the degree of substitut-

Fig. 9.1. Nested structure of utility function with explicit representation of the use of time

By specifying the share of time used in the consumption of transport to be larger than for non-transport and the rate of substitutions, composites, Ci

 x , c ;V , i i 0

i

Ci

V Ci , i 1, 2 ,

within the

1, 2 , as smaller than the rate of substitution,

V D , between the composites, H10  H 20 (see Munk 2002). The CES-UT representation of household preferences thus allows transport to be more complementary to leisure than non-transport. It also makes it possible to represent that a decrease in congestion stimulates the labour supply (see Table 9.1).

9.4.3 Real Income and Social Welfare We assume an additively separable, symmetric social welfare function defined on the households’ real income

W

¦ w R h

hH

(15)

178

K.J. Munk

where

w  R is a concave function and R h the real income of the hth household

defined as

R h { Y h.0  E h  q 0 , v h (q, I h ; e, e ); e 0 , e 0  I h.0 where E

h

(16)

 q , u ; e , e is the expenditure function evaluated at the 0

h

0

0

bench-

mark vector of household prices, free road capacity and environmental externality,

 q , e , e , u 0

0

0

h

v h (q, I h ; e, e ) the utility function evaluated at the vector of

household prices, lump sum income, free road capacity and environmental h

externality (q, I ; e, e ) and Y

h .0

the hth household’s nominal income in the

benchmark situation. In the benchmark situation E

h

 q , u ; e , e 0

h

0

0

I h . The

real income in the benchmark situation is thus by design equal to the nominal income in the benchmark situation, Y

h .0

 I h . The changes in real income for

different households due to the implementation of a project are not affected by how the real income in the benchmark situation is defined. The real income changes only depend on the behaviour characteristics of the household, which in principle can be established objectively. Given the choice of the social welfare function, the definition of the benchmark level of the real income function has normative significance, however. It is, in other words, of importance to the change in social welfare associated with a project. The hth household’s marginal evaluation in h

monetary terms of the congestion externality is Ee evaluation of the environmental externality is Eeh {

wE h and its marginal we

wE h . we

The marginal social welfare of income is defined as

Eh {

ww . As rural wR h

households, indexed R, have lower income than urban households, indexed U,

E R > E U (see Table 9.2). 9.5 The Specification of the Parameterised Model The parameter values required for the model are share values derived from the benchmark dataset, substitution elasticities and the benchmark marginal evaluation of the environmental and congestion externalities. The benchmark dataset in the

Evaluation of the Introduction of Road Pricing

179

form of a Social Accounting Matrix (SAM) as well as the other parameter values are provided in Munk (2003). From the parameters, the share parameters derived from the benchmark dataset and substitution elasticities and other values of the extended CES-UT utility function of the model, the matrix of compensated price elasticities at the level of the three traded commodities, contingent on the level of the free road capacity, has been calculated as they are important for understanding the behaviour of the model (see Table 9.1). Notice that transport is more complementary to leisure than the non-transport good. Table 9.1. Compensated price elasticities and expansion elasticities for the reference set of parameters in the benchmark situation Urban household

Hij Quantity of

Price of

Quantity of

Transport

Non-transport

Labour

Road capacity

Transport

-0.31

0.40

-0.10

0.56

Non-transport

0.06

-0.25

0.19

-0.30

Labour

0.02

-0.30

0.28

0.01

Rural household

Hij Quantity of

Price of

Quantity of

Transport

Non-transport

Labour

Road capacity

Transport

-0.27

0.00

0.00

Non-transport

0.05

0.27 -0.20

0.15

0.00

Labour

0.00

-0.25

0.24

0.00

Note: The elasticities are contingent on the initial on the benchmark level of free road capacity.

For the normative analyses, the social welfare weights indicated in Table 9.2 have been used. Table 9.2. Social welfare weights Household types Social welfare weights, E

Urban

Rural

1.10

1.00

h

180

K.J. Munk

9.6 Presentation and Interpretation of Simulation Results The change from taxation of transport to road pricing involves a differentiation of the tax on transport according to where and when transport is consumed to reflect differences in the associated externalities. To capture this, we have represented the introduction of road pricing as the move from a situation where the consumption of transport by all household types is taxed at the same rate, to a situation where the tax on transport can be differentiated between urban households and rural households, the consumption of the first being associated with congestion externalities and that of the latter with none. We present three types of analysis. We first calculate the consequences of two different projects based on different assumptions about the use of the revenue generated from road pricing (Consequence analysis). We then evaluate these two projects based on a set of supplementary value judgements (Project evaluation) and finally calculate the optimal policies under three different sets of assumptions with respect to which policy instruments are available to the government (Optimality analysis).

9.6.1 Consequence Analysis The political debate about the introduction of road pricing suggests that the way the revenue generated from road pricing is used will affect the popular support for such a reform. This is, related to the fact that reforms that provide an approximate Pareto improvement are politically easier to implement than reforms that imply substantial redistributions between social groups. We therefore consider different assumptions with respect to how the tax revenue is used. We calculate the consequences of the introduction of road pricing in the form of two alternative projects, each with a closure corresponding to alternative assumptions about how the revenue generated by road pricing is used. In the first case, we assume that the revenue from road pricing is used to reduce other transport taxes such as taxes on car ownership and on petrol. This corresponds to the view that for the introduction of road pricing to be politically feasible, the revenue needs to be used to reduce other taxes on transport such as taxes on petrol and car ownership. Since the tax on transport is the sum of these taxes and the road pricing tax, the introduction of road pricing will in this case be represented in the model as the tax on the transport for urban households being increased and that for rural households being decreased. In the second case, the revenue from road pricing is used to reduce the rate of income tax. This corresponds to the view generally held by economists that such revenue should enter into the government’s budget with no strings attached. For both projects, the tax on the urban households’ consumption of transport is increased due to the introduction of road pricing from the initial level of 80% to 139%. Technically, in the case of Project 1, the tax on the rural households’ consumption of transport is endogenous and all other tax instruments, other than the tax on the urban households’ consumption of transport, are kept fixed. Conversely, in the case of Project 2, the income tax rate is endogenous and all other tax

Evaluation of the Introduction of Road Pricing

181

instruments, including the tax on the rural households’ consumption of transport, are kept fixed. Table 9.3a. Consequences of introduction of road pricing: Tax rates Benchmark

Project 1: Revenue from road pricing used to reduce other taxes on private road transport

Project 2: Revenue from road pricing used to reduce income tax

Household taxes Labour

60%

60%

59%

Transport in urban areas Transport in rural areas

80% 80%

122% 63%

139% 80%

Other goods

25%

25%

25%

Table 9.3a shows the tax changes associated with Project 1 and Project 2. Project 1 involves a decrease in other transport-related taxes by 17 percentage points such that the government revenue generated by taxes on transport remains more or less the same. Combined with the increase in the tax on transport due to the introduction of road pricing, the tax on transport thus increases from 80% to 122% for urban households and decreases from 80% to 63% for rural households. Project 2 involves the same increase in the tax on transport as Project 1, i.e. the rate of tax on transport for urban households increases from 80% to 139% and for rural households it is kept at 80%, but is not combined with reductions in other transport-related taxes. Instead, there is a decrease in the rate of tax on labour from 60% to 59%. Project 2 thus implies that the introduction of road pricing increases the rate of tax on the consumption of transport. Table 9.3b shows the consequences of the two projects in terms of changes in the consumption of transport, free road capacity and the evaluation of the externalities. It also indicates the effects on the supply of labour, on the real income of the rural and urban households and on the social welfare of society. Project 1 reduces the urban households’ consumption of transport and increases that of the rural households. As a result, the congestion externality is reduced (free road capacity increases). The decrease in consumption of transport for the urban households is relatively small compared to the increase for the rural household, because of the feedback effect, i.e. because the decrease in road congestion stimulates the consumption of transport. In aggregate, Project 1 results in an increase of the consumption of transport and thus increases the environmental externalities. The supply of labour increases for urban households, but it decreases for the rural households. This reflects the assumption built into the model that transport is complementary to non-market use of time. The net effect on the supply of labour is positive, but rather small. This contrasts with the results obtained for Project 2, where the revenue from road pricing is used, rather than to reduce other transport related taxes, to reduce the income tax rate. In this case, the project results in a significant increase in the

182

K.J. Munk

supply of labour and a significant increase in social welfare. The consumption of transport declines not only as a result of the increase in the price of the transport good, but also as a consequence of the decrease in the income tax rate, which increases the cost of the time needed for transportation. Table 9.3b. Consequences of introduction of road pricing: Changes in transport, externalities and real income Benchmark

Project 1: Revenue used to reduce other taxes on private road transport

Project 2: Revenue used to reduce income tax

Changes compared to benchmark Consumption of transport Consumption of transport by urban households Consumption of transport by rural household Total Transport infrastructure Free road capacity Marginal evaluation Externalities Evaluation of congestion externalities Evaluation of environmental externalities Total Labour Supply of labour by urban household Supply of labour by rural household Total Real income and social welfare Real income of urban households Real income of rural households Social welfare Double dividend

70

-2.0

-2.6

30

2.9

0.9

100

0.9

-1.7

80 1.01

1.9 -0.14

2.7 -0.17

2.0

2.2

-0.2

0.4

1.8

2.5

1084

10.1

13.1

566

-9.8

-3.4

1650

0.3

9.7

813

-10.4

-6.2

417

12.2

15.4

1311

3.0 1.1

10.8 8.3

Source: Own calculations based on stylised CGE model. Note: The double dividend is calculated as the change in social welfare minus the evaluation in the change in the externalities.

Project 1 expands the tax base much less than Project 2 due to the much smaller expansion of the supply of labour. Project 2 is thus associated with a considerable greater gain in efficiency (as measured by the increase in aggregate real income).

Evaluation of the Introduction of Road Pricing

183

The marginal evaluation is in the benchmark situation 1.01. This means that one DKK investment in transport infrastructure increases social welfare by 1.01 units (disregarding the costs of financing the investment). The introduction of road pricing reduces the marginal evaluation of transport infrastructure (from 1.01 to 0.86 in Project 1 and to 0.83 in Project 2). The projects thus decrease the value of investment in transport infrastructure, which reflects that the projects decrease the consumption of transport and thus increase the amount of free road capacity, limiting the need for extra road capacity.

9.6.2 Project Evaluation Project evaluation may involve comparing each of the two projects with the status quo or comparing one project with the other. Project 2 dominates Project 1 according to the Pareto criterion, since the real income of both households is higher under Project 2 than under Project 1, but based on the Pareto criterion neither project is comparable with the status quo as the urban households lose while the rural households gain in both cases. However, introducing supplementary value judgements also makes the projects comparable with the status quo. In this context, it should be emphasized that to assume that the government attaches the same social welfare weight to the income of all household types, i.e. that the government is inequality neutral, as done in traditional cost-benefit analysis, naturally also implies the adoption of supplementary value judgements. Although such value judgements are convenient from a computational point of view, they are often not very relevant, i.e. they do not correspond to the value judgements of the political decision-maker for whom the analysis is prepared. On the basis of the social welfare weights in Table 9.2, the introduction of road pricing by either project is desirable, but the increase in social welfare is far greater when the revenue generated from road pricing is used to reduce the tax on labour income than to reduce other transport-related taxes. The implementation of both projects reduces, as already mentioned, the social evaluation of free road capacity. This suggests that after the introduction of road pricing, a project to reduce the government’s expenditures on road infrastructure would be desirable even if it had not been so prior to the reform. Comparing the value of the change in the externalities (1.8 in the case of Project 1 and 2.5 in the case of Project 2) with the change in social welfare (3.0 and 10.8) shows that the introduction of road pricing is associated with a significant double dividend, i.e. that the total benefits of the introduction of road pricing exceed the benefits of the reduction of the externalities (by 1.1 in the case of Project 1 and by 8.3 in the case of Project 2). There are three reasons for this. The first reason is that in the benchmark situation, the tax on transport is smaller than its complementarity to leisure would justify, i.e. that a further increase in the tax on transport would increase social welfare even disregarding the effect on the externality. The second reason is that the decrease in congestion reduces the use of non-market use of time for transportation and thus stimulates the labour supply (although only marginally, see Table 9.3). Both effects produce an efficiency gain.

184

K.J. Munk

The third reason is that the introduction of road pricing redistributes income from the urban to the rural households, who have a higher marginal social welfare of income than the urban households.

9.6.3 Optimality Analysis In order to calculate the optimal tax structure (in terms of the income tax rate and in terms of taxes on transport and on other goods) and the optimal level of the public good (transport infrastructure), supplementary value judgements have to be specified as in the case of project evaluation. We assume that the government attaches a higher social welfare weight to the real income of the rural household than to the real income of the urban household (see Table 9.2) and on this basis we calculate the optimal policy for the government under three different sets of assumptions with respect to which policy instruments the government is effectively able to use. In the first case, we consider the situation before a dramatic reduction of the administrative costs associated with road pricing has made road pricing feasible. We represent prohibitive administrative costs by the assumption that the government is not able to differentiate the tax on transport between the urban and the rural households. Taking a long-term perspective, we assume that the government is able to adapt the transport infrastructure to the optimal tax policy. In the second and third cases, road pricing has become feasible. The government therefore in these cases is assumed (for simplicity at no costs) to be able to differentiate the tax on the consumption of transport between the rural and the urban households. In the second case, we adopt a short-term perspective assuming that it is not possible to adapt the road infrastructure after the reduction in the administrative costs, whereas in the third case we adopt a long-term perspective as in the first case and assume that such adjustment is possible. In all three cases, we assume that the value of the lump-sum transfers remains unchanged in terms of the price of labour and therefore that the tax on labour can be fixed as a matter of normalisation. The optimal solutions for the three sets of assumptions are provided in Table 9.4. In Table 9.4 part a., the values of the instrument variables are indicated, whereas part b. contains values of a number of goal variables and other endogenous variables. The optimal solution in all three cases involves taxes on transport that are much higher than in the benchmark situation and a level of transport infrastructure much lower than the benchmark level. This is a consequence of the choice of parameter values chosen for the parameterised model. Since these values have been chosen for illustrative reasons, the results should therefore not be interpreted to imply that the Danish transport taxes necessarily are too low and investments in transport infrastructure too high.

Evaluation of the Introduction of Road Pricing

185

Table 9.4a. Optimal solutions: Transport infrastructure and tax rates Benchmark

OPT1

OPT2

OPT3

50

31

31

21

60%

60%

60%

60%

Transport in urban areas Transport in rural areas

80% 80%

202% 202%

446% 118%

495% 112%

Other goods

25%

8%

9%

8%

Transport infrastructure Government provision of transport infrastructure Household taxes Labour

Table 9.4b. Optimal solutions: Changes in transport, externalities and real income

Consumption of transport Consumption of transport by urban households Consumption of transport by rural household Total Transport infrastructure Free road capacity Marginal evaluation Externalities Evaluation of environmental externalities Evaluation of congestion externalities Total Labour Supply of labour by urban households Supply of labour by rural households Total Real income and social welfare Real income of urban households Real income of rural households Social welfare Double dividend

Benchmark

OPT1 Changes compared to benchmark

OPT2 OPT3 Changes compared to OPT 1

70

-11.5

-4.2

-5.9

30

-12.0

8.8

9.7

100

-23.5

4.6

3.8

80 1.01

-26.9 0.2

4.4 -0.6

-3.7 -0.4

6.4

5.9

7.7

4.9

-1.0

-0.8

11.3

5.0

7.0

1084

26.7

31.5

37.9

566

10.3

-33.4

-38.8

1650

37.0

-1.9

-0.9

813 417 1311

-1.0 23.1 24.4 13.1

-41.4 44.5 7.5 2.5

-50.8 53.4 7.9 1.0

Source: Own calculations based on stylised CGE model. Note: The double dividend is calculated as the change in social welfare minus the evaluation in the change in the externalities.

186

K.J. Munk

Compared with the benchmark situation, the optimal solution in the first case, OPT1, involves a considerable reduction in the consumption of transport. The reduction is relatively smaller for urban than for rural households due to the feedback effect. The policy changes result in a reduction in congestion externalities valued at 6.4. This reduction stimulates the consumption of transport of the urban households, but does not affect the rural households. Both types of households benefit from the reduction in environmental externalities, but in the model this has no effect on behaviour. The increase in the taxation of transport increases the tax base directly by increasing the supply of labour and indirectly by decreasing congestion, which in turn also increases the supply of labour. The policy change decreases the real income of the urban households because they suffer disproportionately from the reduction in the government’s provision of infrastructure, but have to share the benefit of the reduction in other taxes with the rural households. The policy change results in a considerable increase in social welfare, in part explained by the benefits due to redistribution of income. The optimal solution in the second situation, OPT2, shows that the introduction of road pricing (the possibility of differentiating the tax on the consumption of transport for the rural and for the urban households) makes it possible to increase social welfare, not only compared with the benchmark situation, but also compared with the optimal situation prior to the introduction of road pricing. The introduction of road pricing leads to an increase in the consumption of transport by rural households that is greater than the reduction for urban households. This results in an increase in the environmental externalities, but a reduction in the congestion externalities which only depend on consumption of transport by the urban households. Compared with the optimal solution without road pricing, OPT1, the change in the social value of the reduction of the externalities is 5.0. The policy change leads to a reduction in the supply of labour. The increase in urban households’ labour supply as a consequence of the increase in the tax on transport is, due to the feedback effect, smaller for urban households than the corresponding decrease in the supply of labour for rural households. The use of an extra policy instrument results, as one would expect, in an increase in social welfare. Part of this increase is due to the reduction in the externality and part is due to the redistribution from urban to rural households. There is, however, compared to OPT1, no benefit due to an increase in the supply of labour – on the contrary. In the third case, the optimal solution, OPT3, the introduction of road pricing by reducing the demand for transport infrastructure in urban areas, where it is associated with congestion, reduces the optimal amount of transport infrastructure. The possibility of adjusting the transport infrastructure justifies a further increase in the tax on transport in urban areas compared with the optimal solution in the previous case, OPT2. This is because the reduction of the government provision of road infrastructure increases in the marginal evaluation of the externality. In recent years, the so-called double dividend issue has attracted considerable attention both among policy-makers and economists. The issue is whether replacing existing taxes with taxes on commodities causing environmental damage will increase social welfare, not only by internalising the negative external effects,

Evaluation of the Introduction of Road Pricing

187

but also by reducing the distortionary costs of the tax system as a whole. Based on the idea that the tax revenue obtained from environmental taxes could be used to reduce pre-existing distortionary taxes, the initial contributions to the analysis of the issue suggested that a green tax reform in general would be associated with a double dividend. However, it has subsequently become clear that the intuition behind the initial suggestion was flawed by not taking into account the distortionary effects of the environmental taxes. Now, it seems that the established wisdom is that a green tax reform is unlikely to generate a significant double dividend and, if previous policies have been economically rational, that a green tax reform cannot generate a double dividend at all. The present analysis indicates that, in a situation where the tax on transport is too low (according to the relevant social welfare function), a higher tax on transport increase the social welfare beyond the social value of the decrease in the externalities associated with transport, i.e. is associated with a double dividend. When the initial situation is not optimal (i.e. in the benchmark situation), adopting an optimal tax on transport (OPT1) or introducing road pricing at the optimal level justified by the reduction of the administrative costs (OPT2 or OPT2) associated with a substantial double dividend (by 13.1, 15,6 and 14.1, respectively). When the initial situation has been optimal given the administrative costs initially associated with introducing road pricing (OPT1), the double dividend associated with the introduction of road pricing is still positive, but far smaller; by 2.5 if the tax reform is not associated with an adjustment in the road infrastructure (OPT2) and by 1 when it is (OPT3). The theoretical model allows a double dividend to arise from three sources: first that the change in the tax rates increases the tax base (in general by increasing the labour supply), second that the increase in the public good (free road capacity) has the same effect and third that the tax changes have desirable income distributional effects (see Munk 2000). The main source of the double dividend associated with the increase of the tax on transport from the benchmark to the optimal level (OPT1) is the increase in the supply of labour. In the case of the introduction of road pricing leading from OPT1 to OPT2, the double dividend is explained by the increase of the free road capacity resulting in a greater supply of labour and the redistribution of income from the urban to the rural households. Finally, in the case of the introduction of road pricing leading from OPT1 to OPT3, the double dividend is explained only by the redistribution of income from the urban to the rural households. Sandmo (2000) has suggested that when a green tax reform would be combined with an adjustment of the level of abatement, this would be associated with a third dividend. Comparing OPT3 with OPT2 shows that reducing the amount of road infrastructure after the introduction of road pricing naturally has resulted in an increase of social welfare. In this sense, Sandmo’s observation is naturally correct, but notice that the larger increase in social welfare associated with OPT3 compared

188

K.J. Munk

with OPT2 (7.9 compared with 7.4), the double dividend contributes less (1.0 compared to 2.5).

9.7 Conclusion This chapter has provided a quantitative assessment of the introduction of road pricing within a second best framework, i.e. where raising government revenue and redistributing income is associated with administrative and distortionary costs. The analysis has illustrated that how the government budget is balanced has important implications, both from an efficiency and a distributional point of view for the assessment of whether it is desirable to introduce road pricing and at what level to set the road pricing charge. The analysis illustrates that the introduction of road pricing as expected will result in a decrease in congestion, but that depending on how the revenue is spent it may result in an increase in environmental externalities. The analysis also highlights that road pricing is likely to reduce the need for transport infrastructure. The planning of the introduction of road pricing and investment in road infrastructure should therefore be closely coordinated. The simulation results have been based on a CGE model representing in an innovative way the complementarity of transport and congestion with leisure, and that different households contribute differently to congestion and are differently affected by congestion. The analysis demonstrates that the use of CGE models constitutes a realistic alternative to traditional cost-benefit analyses based on partial equilibrium, which cannot account correctly for the complexities of second best assumptions. However, the model which has been adopted for the analysis ignores important aspects of the taxation of transport and of investment in transport infrastructure. In order to be used in practice, the model needs to be expanded to represent collective transport and as far as private transport is concerned not only leisure travel, but also commuting and the use of transport as intermediate input in production. Furthermore, before CGE models can become an operational tool for transport policy analysis a significant investment must be made in database construction and estimation of key model parameters.

References Bovenberg, L.A. (1999). Green tax reform and the double dividend: a updated reader's guide. International Tax and Public Finance, 2: 157-183. Corlett, W.J., Hague, D.C. (1953). Complementarity and the excess burden of taxation. Review of Economic Studies, 21: 21-30. Dixit, A, Munk, K.J.(1977). Welfare effect of tax and price changes. A correction. Journal of Public Economics, 8: 103-107.

Evaluation of the Introduction of Road Pricing

189

Goulder, L.H., Parry, I.W.H., Williams, R.C., Burtraw, D. (1999). The Cost-effectiveness of alternative instruments for environmental protection in a second-best setting. Journal of Public Economics, 72: 329-360. Mayeres, I., Proost, S. (1997). Optimal tax and public investment rules and the congestion type of externalities. Scandinavian Journal of Economics, 99(22): 261-279. Mayeres, I., Proost, S. (2001). Marginal tax reform, externalities and income distribution. Journal of Public Economics, 79: 343-363. Munk, K.J. (2000). Administrative costs and the double dividend. Working Paper, EPRU, University of Copenhagen. Munk, K.J. (2002). What determines the optimal commodity tax structure from an intuitive point of view?, Working Paper, EPRU, University of Copenhagen. Munk, K.J. (2003). Computable general equilibrium models and their use for transport policy analysis. Report 4, Danish Transport Research Institute Sandmo, A. (1975). Optimal taxation in the presence of externalities. Swedish Journal of Economics, 77: 86-98. Sandmo, A. (2000). Public economics and the environment. Oxford University Press, Oxford.

190

K.J. Munk

Appendix: The Theoretical Model 1 Notation Index sets Households: Produced commodities, sectors: Primary factors: Commodities

H C F = (0) FC = C‰F

Variables Quantities Production levels, outputs: Input demand for labour: Demand for produced commodities: Net demand for labour:

Yi

v

k 0

x

h i

x

h 0

Z Labour endowments: Public good non-separable (Free road capacity): e

iC kC

i  C, h  H

hH

h 0

hH

Endowment of public good (Road capacity):

Ze

Consumption of produced commodities: Consumption of leisure:

cih { x ih c0h { Z 0h  x 0h

Net demand vector:

x {  x , i  FC h

i  C, h  H

h i

hH h H

Government consumption (investment in transport infrastructure):

xG {  x iG , i  FC

Taxes

Rate of tax on labour:

ti t0

Lump sum tax:

L

Rate of tax on produced commodities:

iC

Prices Market prices for produced commodities: Market price for labour:

pi p0

iC

Evaluation of the Introduction of Road Pricing

iC

Household prices for labour:

q i { pi  t i q 0 { p0  t 0

Household price vector:

q {  q i , i  FC

Household lump sum income:

Ih

Household prices for produced commodities:

191

hH

2 Equilibrium conditions (1)

Profit maximisation

v i0

Y (2)

i

i

iC

i

Y (p )

i C

Utility maximisation

x ih  q,e,I h

iC, hH

x

h 0

x  q,e,I

hH

I

h

L

x ih

(3)

v i0 (p i )

h 0

h



hH

Material balance

Yi

¦x

h i

 x iG

iC

hH

0

¦v

j 0

jC

(4)

 ¦ x 0h  x G0 hH

Government budget constraint

¦¦t x  ¦t x i

h i

hH iC

(5)

0

h 0

hH

 HL  ¦ pi x iG  p 0 x G0 +¦ S i (pi ) iC

0

iC

Tax-price equations Household prices for produced commodities

qi

pi  t i

iC

Household prices for primary factors

q0 (6)

p0  t 0

Public good externality as function of the consumption of commodity 1

§ · e e ¨ Z e , ¦ x1h , x G ¸ hH © ¹

10 Efficiency and Equity Considerations in Road Pricing Harald Minken and Farideh Ramjerdi Institute of Transport Economics, Oslo, Norway

Abstract Equity considerations are particularly important for the appraisal of road pricing schemes. We discuss and classify the most relevant aspects of equity in this context, and set out inequality measures from economics that may be used as outcome indicators with respect to these equity objectives. Road pricing is inherently a second-best problem due to the link between work trips and the labour market, where distortionary taxation exists. Consequently, efficiency and distributional issues both need to be considered simultaneously. To design the road pricing scheme in this case, it is suggested to solve the constrained optimisation problem of maximising welfare in the transport system as computed from a transport model, subject to relevant equity indicators reaching their target levels.

10.1 Introduction Due to increases in household car ownership rates, demographic changes and changes in the geographical patterns of housing, work and leisure activities, urban road networks are becoming increasingly congested in cities all over the world. This entails not only time losses for private and business transport, but also severe noise and pollution problems and degradation of the quality of life in the city centre and surrounding neighbourhoods. For the last 40 years, since the work of Walters (1961), Mohring and Harwitz (1962), Vickrey (1963, 1968) and Strotz (1965), economists have advocated road pricing as a solution to these problems, but somehow the idea still seems difficult to get across to the public. Singapore and London are the only cities with a road pricing system, i.e. tolling with a view to relieving congestion. Norwegian cities have toll rings with financing as their main purpose (Ramjerdi et al. 2004). Reasons for the reluctance and widespread opposition to road pricing are surveyed in Eliasson and Lundberg (2003). Concerns about the distributional impacts are prominent on this list. Equity reasons for opposition to the Norwegian toll rings are analysed in Langmyhr (1997). While some of the popular arguments against road pricing can be dismissed out of hand, all of the equity concerns merit close attention – not just to facilitate implementation of a measure that can improve the efficiency of the transport system, but because equity objectives are important social objectives in their own

194

H. Minken and F. Ramjerdi

right. Furthermore, it might be argued that analyses of equity impacts are more important for road pricing decisions than for decisions about infrastructure construction. There are three reasons for this. First, infrastructure construction is an ongoing process where those who did not get their new road this year might be the winners next year. Indeed, there is evidence that such compensatory thinking is part of the informal decision criteria of Norwegian decision-makers (Fridstrøm and Elvik 1997). Road pricing is different in this respect. It is a permanent redesign of the whole transport system, with no chance of the losers ever getting compensated by a reverse pricing policy next year. Second, road pricing only achieves its efficiency objective through comparatively large transfers of money from individuals to the government. Model tests indicate that the revenue from first or second-best road pricing schemes in different cities is from one to two times the size of the net benefit of the scheme (Ramjerdi 1995, Eliasson and Lundberg 2003) or even more (Fridstrøm et al. 2000). As a rule, motorists as a group stand to lose before recycling of the revenue. Third, since government is the winner by far before the use or recycling of the revenue, the equity impacts depend very much on how the revenue is used. The equity impacts are not well-defined until the use of the revenue has been decided. But neither are the efficiency impacts: if the revenue is used to cut back distortionary taxation elsewhere in the economy, there might be an additional gain (a “double dividend”). In this case, the double dividend is perhaps the main element of the efficiency gains (Fridstrøm et al. 1999, Parry and Bento 2001). This form of recycling, however, precludes compensation to losers. On the other hand, compensation to losers may well mean that no double dividend can be reaped, unless of course it takes the form of investing in efficiency-enhancing projects targeted particularly at the losers (Fridstrøm et al. 1999). Thus both equity and efficiency considerations are particularly important when designing a road pricing scheme. Furthermore, there is a conflict between them. With two conflicting objectives we need at least two policy instruments to achieve targets with respect to the objectives. That is why the recycling scheme should be an integral part of the road pricing scheme. Considering that equity has many aspects and that there will also be environmental and traffic safety objectives, road pricing and recycling will probably need to be parts of an even broader plan, consisting, for instance, of infrastructure provision, traffic management and public transport fares and service levels, to provide sufficient degrees of freedom to attain the objectives. The purpose of this paper is to propose an approach for addressing the equity and efficiency considerations as a constrained optimisation problem (maximising efficiency, subject to equity constraints). To do this, we first discuss what the relevant aspects of equity are (section 10.2) and suggest indicators for their measurement (section 10.3). These sections follow Minken et al. (2003). In section 10.4, we briefly discuss the concept of economic efficiency before considering the requirements for the modelling system that we need in order to predict the equity and efficiency impacts of road pricing. Section 10.5 sketches the optimisation problem. Experience with this approach for the design and appraisal of road

Efficiency and Equity Considerations in Road Pricing

195

pricing schemes is still very scant, which is why we limit ourselves in section 10.6 to a few remarks on experience with solving similar problems. Section 10.7 concludes. The AFFORD project (Fridstrøm et al. 1999, Fridstrøm et al. 2000) used optimisation with a transport model to find optimal first and second-best road pricing and explored the equity implications of these policies. This paper is taking the further small step of suggesting the inclusion of equity considerations as constraints in the optimisation problem.

10.2 Aspects of Equity Equity, like the related concepts of justice, fairness and right, is not a simple concept. Different people have different concepts of equity. Which of the aspects of equity that seems important will very much depend on particular contexts and circumstances (Langmyhr 1997). A first distinction can be made between formal equality (treating all people equally) and outcome equality, which may imply unequal treatment. Social inclusion objectives like accessibility for the mobility impaired or for those without a car are based on the notion that the outcome of a strategy should be favourable for the disadvantaged in the transport system, or at least meet their basic needs with respect to accessibility. Outcome equality might also be required with respect to different disadvantaged geographical areas and income groups. Such aspects have proved to be very important in the opposition to road pricing (Langmyhr 1997). On the other hand, the principle of formal equality may be invoked to make all users pay the same and letting no one use the transport system for free while others must pay, and to demand that the revenue is recycled to those who paid the charges. It must be admitted that such interpretations of formal equality are virtually impossible to reconcile with road pricing. Other interpretations of the same principle, such as “polluter pays” or the notion that everybody should pay the full costs to society of their actions, are on the other hand perfectly compatible with road pricing. Thus, struggles over how to apply formal equity principles seem to lead to either outright rejection of road pricing (a typical view of motorists’ organisations) or full endorsement of it (typical of most environmentalist organisations, at least in Norway). There seems to be little room for trade-offs, and consequently – with two exceptions – our focus in the following will be on forms of the outcome equality principle. The first exception concerns the inequality of the geographical distribution of the net benefits from a scheme (after recycling). While it is incompatible with road pricing to require that everybody should pay the same, regardless of where and when they travel, and it is virtually impossible to recycle the revenue to exactly the same people who paid the tolls, it should not be impossible to design a strategy with approximately the same net benefits to every geographical area.

196

H. Minken and F. Ramjerdi

This principle can be applied to districts within the urban area, and this is particularly relevant if each of these districts is represented by its own local government with a say in how the scheme should be designed. There must be something in it for everybody. The principle of equal geographical distribution of benefits could, however, also be applied to the distribution between the urban area and the nation as a whole. In that case, the revenue should be recycled to the inhabitants of the city, and not be siphoned off to taxpayers or public service users elsewhere. This principle is indeed increasingly recognised as a requirement for the introduction of road pricing. It makes recycling in the form of cuts in distortionary taxes difficult, since local taxes may not be the best candidates for such cuts. Consequently, it often takes the form that the revenue should be used to improve local transport. This is, for instance, stated in the new Norwegian law on road pricing. Parry and Bento (2001) find that this requirement almost wipes out the whole double dividend. Daganzo (1995) recognises that to avoid a situation where the gains from the scheme are spread nationwide while the losses are borne by the inhabitants of the city, the inhabitants may prefer restrictions on car use to road pricing. While the welfare losses may be larger, the gains will be fully reaped inside the city. The other exception where a formal equity principle is invoked concerns car taxes. It is claimed to be unfair that motorists pay much more to the state than they get back, and that this situation is made even more inequitable by introducing road pricing. This argument does not seem to rule out road pricing completely, but requires that other car taxes be cut. Moving on to outcome equity principles, the idea is that there is some good which is distributed unevenly among individuals or groups even before a road pricing scheme or a transport strategy is introduced, and that one of the objectives of the scheme should be to redistribute this good or to compensate for its unequal distribution. In the literature on inequality measurement, this good is of course almost universally understood to be income. For our purposes, however, it could equally well be accessibility. If the concern is about unequal distribution of accessibility, we are now not asking if accessibility gains from the scheme are distributed evenly, but if the scheme contributes to reducing the pre-existing inequality. Instances include concerns for the accessibility of those without a car, the mobility impaired, the poor, women or those living in less accessible parts of the city area. Avoidance of inequitable distributions of accessibility is probably a very important objective of urban transport policy. It may lie behind statements to the effect that road pricing will hit the poor the hardest as well as requirements that public transport should first be improved before road pricing is introduced, so that there is a real choice of mode for everybody. If there is a real danger that accessibility for some will become intolerably low, leading to social exclusion, poverty indicators, incorporating a “poverty line”, might be more appropriate than the inequality measures of the next section. Suitable poverty indicators are developed in Sen (1982) and Foster et al. (1984).

Efficiency and Equity Considerations in Road Pricing

197

If the concern is about the unequal distribution of income, we will probably not be interested in the monetary impacts alone. Instead, we will recognise that time gains can compensate for monetary losses and vice versa. To do this, we can create the concept of generalised income, adding time gains and monetary gains to income by way of the appropriate value of time in each instance. Using the concept of generalised income, we can investigate if its distribution has become more equal or not after the introduction of road pricing and the accompanying revenue recycling. This can be done across income groups, household groups, gender groups, locations, etc. Probably the most relevant concept of income for such analyses is household income per consumption unit. A consumption unit is defined in the following way: each household member is assigned a weight, equal to 1 for the first adult person in the household, 0.7 for any additional adults and 0.5 for children up to 17. With small variations, these weights are in line with OECD recommendations for household consumer surveys. The number of consumption units in the household is given by the sum of the weights attached to all household members. The concept of generalised income and this definition of household income per consumption unit was used for equity analysis in the Fourth Framework project AFFORD (Fridstrøm et al. 2000). Other aspects of equity are also relevant for road pricing, among them fairness in the form that government should keep its promises, and procedural fairness (a transparent and democratic planning process). These are, however, not properties of a scheme or a strategy, but of the way in which it is implemented.

10.3 Indicators of Equity Indicators of income inequality will inevitably have a normative as well as a descriptive content. The normative content becomes clear if we consider the properties that we want such an indicator to have. Some of them will be fairly uncontroversial. But to arrive at a specific mathematical formulation, we will also have to make more controversial choices. In experiments where people are asked if they consider an income distribution to be more or less unequal than another, none of the properties usually wins unanimous support. This is why we should be aware of the normative choices we make when we choose a particular indicator. Suppose we have recorded the income of the individual members of a given population and ordered them according to income. We want to measure the inequality of this distribution of income. The first property that we want our measure to have is anonymity (or symmetry). It says that if two members of the population swap incomes, the measure should be unchanged. It does not matter who the rich and poor are. Women earning twice as much as men is equally bad as men earning twice as much as women. The next property is the Pigou-Dalton property (the transfer principle). It says that if you take an amount from a richer person and give it to a poorer person,

198

H. Minken and F. Ramjerdi

inequality should diminish as long as the poorer person is still poorer than the rich one after the transfer. These properties seem uncontroversial. The population principle is also perhaps uncontroversial. It says that if we replace each income earner by the same number of clones, the inequality measure should not change. The controversial properties, however, are mainly two. Scale invariance says that if you multiply each income by the same positive constant, inequality is unchanged. That is often felt to be a right-wing view. On the other hand, translation invariance says that if you add the same amount to each income, inequality is unchanged. This is often felt to be a left-wing view. A compromise between these principles – a centrist view – is possible, but probably mathematically cumbersome. The Gini coefficient is the most commonly used income inequality measure. It can be explained with reference to Fig. 10.1 below. On the horizontal axis, a population is ordered by income from the lowest to the highest. On the vertical axis, there is the cumulative share of total income. If everybody had the same income, any ten per cent of the population would have ten per cent of the income, and the straight line “Equality” would be produced. In reality, the twenty per cent with the lowest income has only about 3 per cent of total income, the forty per cent with the lowest income has only about 25%, etc. This is shown by the “Empirical distribution” curve. This curve is called a Lorentz curve. (In actual fact, the depicted Lorentz curve shows the income distribution of Norwegian taxpayers in 1995.) Obviously, the area between the two curves is an indicator of income inequality, ranging from 0 for perfectly equal distributions to 0.5 for distributions where one person earns all income. The Gini coefficient is twice this area to get a measure of inequality varying between 0 and 1. A Lorentz curve Share of income 1.0 0.9 0.8 0.7 0.6 0.5

Empirical distribution

0.4

Equality

0.3 0.2 0.1 0.0 0

0.2

0.4

0.6

0.8

1

Share of population

Fig. 10.1. Lorenz curve for the taxpayer population of Norway 1995

Efficiency and Equity Considerations in Road Pricing

199

For our purposes, probably the most useful formulation of the Gini coefficient is:

G

n n 1 ˜ ¦¦ | xi  x j | 2n 2 x i 1 j 1

(1)

Here we have assumed a population of n individuals with incomes x = (x1,x2,…,xn). The average income is x . Suppose, however, that there are instead n income groups with incomes x = (x1,x2,…,xn), ni members of group i, i = 1, …,n and ¦ini =N. Then

G

n n 1 ˜ ni n j | x i  x j | . ¦¦ 2N 2 x i 1 j 1

(2)

The Gini coefficient complies with the first three principles and scale invariance, and consequently does not exhibit translation invariance. The Gini coefficient is not additively decomposable. Additive decomposability means that if the population consists of groups, the inequality measure can be decomposed into a term showing inequality within groups and a term showing inequality between groups. This is obviously useful for our purposes. For instance, if our population belongs to different zones, it might be interesting to see to what extent the unequal distribution of benefits among income groups is due to the unequal spatial distribution. The class of additively decomposable inequality measures was characterised by Shorrocks (1980). It turns out that the members of this class that exhibit the properties of symmetry, the Pigou-Dalton transfer principle, the population principle and scale invariance are of the following form:

S c ( x) S 0 ( x) S1 ( x)

c n ª º 1 § xi · «¨ ¸  1» for c z 0 or 1 ¦ nc(c  1) i 1 ¬«© x ¹ »¼ n 1 x log ¦ ni1 xi

(3)

x 1 n xi log i ¦ ni1 x x

where x = (x1,x2,…,xn) > 0 is the distribution of income among the n members of the population, and x is the mean income. The constant c can take all real values. This class of functions Sc is called the class of generalised entropy measures. For some values of c, they behave rather oddly as measures of income inequality. For instance, for c > 1, the measure is very sensitive to transfers of income among the rich, while for c < 0, it is very sensitive to transfers of income among the poor. Furthermore, only S0 will have the property that when decomposed, the weights on

200

H. Minken and F. Ramjerdi

the within-group terms are constants and sum to 1. Thus, S0 seems a very good candidate for our inequality measure.1,2 Decomposition of S0 takes the form:

B  ¦ wg S0 (x g )

S0

g

ng 1 B¦ g n ng

ng

¦ log x i 1

where

B

xg g i

n xg 1 g B  ¦ ¦ log g xi g n i 1

(4)

1 x ng log ¦ n g xg

Here, the groups are indexed by g, the population in group g is ng and average income in group g is x g . B is the across-groups inequality measure, resulting from abstracting from all income differences inside groups. (ng/n) is the weight of the inequality inside group g in the total measure S0. All of the measures treated so far exhibit scale invariance. For political balance and technical reasons, we will also have a need for inequality measures displaying translation invariance. Of course, if we are not certain which of our inequality measures embody the norms and values of the decision-makers, there is a third option, namely to present the distributional impacts of a strategy in a raw form, for the decision-makers themselves to pass judgement on whether or not inequality has decreased. The Kolm measure (Kolm 1976) obeys the first three principles and translation invariance. It is

K a (x)

1 §1 n · log¨ ¦ expa  x  xi ¸ a ©n i 1 ¹

(5)

where a > 0 is a transfer sensitive parameter. The technical reason for applying (5) is that it allows some (or all) xi’s to be negative, whereas (3) does not. This is useful for measuring the distribution of benefits, many of which will be negative in road pricing. We need different indicators for different purposes. As explained, a Kolm measure is technically well-suited as an indicator of the inequality of the distribution of net benefits. A Shorrocks measure, perhaps S0, is technically suited 1

The weights on the within-group terms of S1 will also sum to 1, but will be functions of between-group inequality. On the other hand, S1 (and all measures with c > 0) has the property that there is an upper limit to inequality, given by log n in the case of S1. This allows for a normalisation of the measure and is obviously convenient for expressing targets. 2

The S0 and S1 measures are originally due to Theil (1967).

Efficiency and Equity Considerations in Road Pricing

201

for measuring inequality with respect to accessibility or generalised income. It has the additional advantage of permitting decomposition. Thus, for instance, we can measure inequality with respect to generalised income and decompose the income groups by location, giving a picture of how much of it depends on income alone and how much on the geographic distribution of income groups. Or alternatively, we could measure inequality with respect to accessibility across space and decompose by income group. Equity issues involving only two groups, like the sexes or those with and without a car, could of course be measured by simpler indicators. With respect to the objective to retain benefits inside the city, the indicator could be government revenue as a percentage of total net benefits in the strategy. It is well known that welfare functions of a particular form may be derived from many of the inequality measures (see, for instance, Myles 1995). So we might in fact use these welfare functions instead of a utilitarian welfare function and a separate inequality measure. There are several reasons for not doing so at the moment. First, it might very well be the case that we are interested in several aspects of equity and would like to use more than one indicator. Second, as long as our models permit us to compute welfare without having to specify welfare weights for each individual or group (i.e. a “representative consumer” exists), it seems natural to keep efficiency and equity considerations apart. And third, incorporating equity concerns into the welfare function does not make it easy for decision-makers to interpret or discuss the results of the appraisal.

10.4 Equity, Efficiency and Modelling Perfect road pricing is, by its nature, pricing that maximises welfare. Road pricing in real world applications must be defined relative to a certain setting. By this we mean that we must define the system that we are studying, and consequently the system’s outside environment. Pricing inside the system is not thought to affect behaviour outside the system. Furthermore, a setting defines the level of detail at which marginal costs are studied and defined, as well as the agents of the system and the dimensions of choice open to them. Both first-best and second-best pricing must be defined relative to such a setting. First-best road pricing in a particular setting means that the price of any action open to any agent considered in this setting is equal to its marginal social costs. Thus, we use the concept of first-best pricing even for a pricing scheme where not every conceivable dimension of choice is taken into account. It is sufficient for us that every dimension of choice that is modelled in this setting is taken into account. In first-best pricing, we should ideally be able to set separate charges on the use of every link in the network, to differentiate between periods according to the level of traffic, and to differentiate between user groups to the extent that they perceive costs differently or their actions impose different marginal external costs. Furthermore, if our setting is confined to the transport system, we need to assume that prices in the outside economy are right, or else price or cost changes in the

202

H. Minken and F. Ramjerdi

transport system will have welfare implications outside transport – impacts that are not captured by computing benefits in the transport system alone (Jara-Diaz 1986). Considering the close link between work trips and labour market supply and the distortionary effect of labour taxes, there is really no such thing as first-best urban road pricing, even if we would be able to charge on all links. To this can be added the inherent difficulties in taking into account the real diversity and heterogeneity of the travellers, such as their different values of time, in setting the link charges (Eliasson 2000). Second-best pricing, then, is the prices that maximise welfare subject to constraints on the free use of the charges defined in a particular setting. These constraints may be technological, institutional, legal or political. Second-best solutions are defined relative to the same setting as the first-best, so they involve the same types of costs, the same dimensions of choice, and the same exogenously given environment as the first-best. For some reason, though, the free use of some of the instruments available in the first-best solution is now barred. For second-best pricing, the same inherent problems of taking account of real heterogeneity exist, so even if we are able to derive optimal second-best charges in our models and simplified settings, the solution does not necessarily carry over to the real world. Thus, we should be modest in our claims and realise that real world experience is needed. The work of Parry and Bento (2001) clearly indicates that it is unwise to ignore labour market effects when setting urban road prices. Taxes on travelling to work will affect labour supply negatively, thus adding to the effects of the income tax (remember that most motorists will experience a welfare reduction from road pricing before recycling). If nothing can be done about the income tax, this is a typical second-best situation. If, on the other hand, the revenue is used to reduce the income tax, a large double-dividend is reaped in Parry and Bento’s simple model. Thus, efficiency depends on the efficiency of the tax system. The efficiency of the tax system is captured by the indicator called the marginal cost of funds (MCF). If, for instance, raising an additional euro of public funds by raising all taxes in the same proportion leads to goods and services worth 0.20 euro not being produced and consumed, the marginal cost of funds is 0.20. This parameter (which is difficult to estimate) cannot be taken as an exogenously given fact when we study strategies and schemes that in fact alter the tax system. So if revenue is used to cut distortionary taxes, the MCF has a lower value than if it is used otherwise. Another feature of our second-best situation is of course that in practice only a few of the links can be tolled. Sandmo (2000) states that “When lump-sum taxes are not available, issues of efficiency cannot be separated from those of redistribution and equity, and when one considers the tax system as a source of revenue for the public sector, Pigouvian taxes must be analysed jointly with other aspects of the tax system” (page 33). Thus, there is a need to include equity indicators in the problem of setting second-best road prices. Different kinds of models, reflecting different settings, have been used to study the second-best problem. What we have said indicates a broad setting, reaching even beyond transport and including the labour market. However, such

Efficiency and Equity Considerations in Road Pricing

203

models will typically not be able to reflect the network and the full set of transport choices very well. That is why we assume that a transport model with at least modal choice and destination choice is used. The interaction with the labour market must be reflected by an appropriate choice of the shadow cost of public funds, reflecting how the revenue is used.

10.5 Optimal Road Pricing Subject to Equity Constraints In the models we will be using, such as nested logit models, a representative consumer can be said to exist and therefore welfare is well-defined even without making stronger normative assumptions about how each individual is to count. The fact that we do really care more about the welfare of some groups than others, is taken care of by the equity indicators. The welfare function can be written

W

UB  PS  (1  O ) PVF  EC

(6)

where UB are user benefits, PS are producer surpluses, PVF are government surpluses (present value of finance), EC are external costs and O is the marginal cost of funds. These elements (except O) are computed from model output. All of them are functions of the policy variables (the link charges) through their impacts in the transport model. O is set to reflect the recycling scheme. The problem is to set the link charges so as to maximise W. The resulting levels of the instruments are the second-best road prices. To reflect equity concerns, targets will have to be set for the level of the equity indicators. Constraints of the form that the equity indicators should at least reach their target levels can then be included in the optimisation problem, which becomes maximisation of W subject to the constraints. This is how we propose to analyse road pricing. Throughout this paper, we assume that economic efficiency is measured by a utilitarian welfare function. In the case where only one equity indicator is used as a constraint and this indicator implies a certain other welfare function, it might well be asked if the optimal result of the constrained optimisation problem equals the result of maximising the unconstrained implied welfare function. The answer is negative: while the welfare function that incorporates the equity concerns always allows for a trade-off between utilitarian efficiency gains and less inequality, the constrained optimisation approach makes the maximum allowable inequality an absolute requirement.

204

H. Minken and F. Ramjerdi

10.6 Solving the Optimisation Problem Algorithms to solve this problem exist and have been proven to work, even if a global optimum is not guaranteed. The PROSPECTS Methodological Guidebook (Minken et al. 2003) outline three approaches, the AMOEBA algorithm (Nelder and Mead 1965), the regression-based method (Fowkes et al. 1995, 1998) and general unconstrained optimisation methods (Luenberger 1984). There are limits to the number of policy variables (the number of links on which a charge can be levied) that may be used without requiring too many runs of the transport model, so the problem to be solved will definitely be a second-best problem. Until now, the location of a single toll ring has usually been pre-specified, and the tolls are restricted to be the same on each tolled link. These restrictions may, however, easily be modified, so that different locations of the ring and combinations of many rings may be tested. Kilometre-based charges can also be tested. The particular problem at hand, where only link charges are used as instruments, has however not been tested, so for the moment we have no empirical results. The right shadow price of public funds to use in each particular instance has not been investigated. There have been some tests with constraints derived from setting a target for an equity indicator, but in this particular case it turned out that the constraint was not binding. Thus, we are confident that the optimisation problem can be solved, but experience in setting targets for the equity constraints, finding out how the solution changes with these targets and with the shadow price of public funds is still lacking. Consequently, experience with presenting results to decision-makers and the public and using this to discuss and agree upon a road pricing scheme is also lacking, which is of course essential. To gain experience, various forms of equity indicators should be tried as constraints and results compared with respect to acceptability and ease of interpretation.

10.7 Conclusions We have emphasised the importance of distributional impacts and equity considerations for the appraisal of road pricing schemes, charted and classified the most relevant aspects of equity in this context, and pointed out that inequality measures used in economics may be used as outcome indicators with respect to these equity objectives. We have pointed out that road pricing is inherently a second-best problem due to the link between work trips and the labour market, where distortionary taxation exists. Consequently, efficiency and distributional issues need to be considered simultaneously. To design the road pricing scheme in this case, it is suggested to solve the constrained optimisation problem of maximising welfare in the transport system as computed from a transport model, subject to relevant equity indicators reaching their target levels. We point out that the shadow price of public funds to be used in the objective function must be derived from a

Efficiency and Equity Considerations in Road Pricing

205

general equilibrium model, taking into account the particular recycling scheme in each case. Alternatively, a spatial computable general equilibrium model with a sufficiently developed transport network may be used. Algorithms exist to solve such problems, and some experience with their use has been forthcoming from the Fifth Framework project PROSPECTS (Minken et al. 2003). However, there is no experience yet with the variant of this general problem where the policy instruments are link charges only and the shadow price of public funds is modified to take account of the recycling scheme. Crucially, experience in presenting results to decision-makers and the public and using this to discuss and agree upon a road pricing scheme is also lacking. An alternative to using an equity target as a constraint when optimising to find the road charges might be to incorporate the equity issue in the welfare function, as indicated in Myles (1995). The two approaches are not identical. The choice between them should be based on whether or not decision-makers are willing to make the equity target an absolute requirement, and which of the two approaches they find to be more transparent and easy to interpret.

References Daganzo, C.F. (1995). A pareto optimum congestion reduction scheme. Transportation Research, 29B(2): 139-154. Eliasson, J. (2000) Transport and location analysis. TRITA-IP FR 00-79, KTH, Stockholm. Eliasson, J., Lundberg, M. (2003). Road pricing in urban areas. VV Publication 2002:136E, Swedish Road Administration. Foster J., Greer, J., Thorbecke, E. (1984). A Class of Decomposable Poverty Measures. Econometrica, 52(3): 761-766. Fowkes, A.S., Bonsall P.W., Bristow A.L., May, A.D. (1995). The optimisation of integrated urban transport strategies: tests based on Edinburgh. ITS Working Paper 425 Institute for Transport Studies, University of Leeds. Fowkes, A.S., Bristow, A.L., Bonsall, P.W., May, A.D. (1998). A short-cut method for strategy optimisation using strategic transport models. Transportation Research A, 32(2): 149-157. Fridstrøm, L., Elvik, R. (1997). The barely revealed preference behind road investment priorities. Public Choice, 92: 145-168. Fridstrøm, L., Minken, H., Vold, A. (1999). Vegprising i Oslo: virkninger for trafikantene. TØIrapport 463, TØI, Oslo. Fridstrøm L., Minken, H., Moilanen, P., Shepherd, S.P., Vold, A. (2000). Economic and equity effects of marginal cost pricing in transport. Case studies from three European cities. AFFORD Deliverable 2A. VATT Research Reports 71, VATT, Helsinki. Jara-Diaz, S. (1986). On the Relation between User Benefits and the Economic Effects of Transportation Activities. Journal of Regional Science, 26(2): 379-391.

206

H. Minken and F. Ramjerdi

Kolm, S.-C. (1976). Unequal Inequalities I. Journal of Economic Theory, 13: 82-111. Langmyhr, T. (1997). Managing equity. The case of road pricing. Transport Policy, 4(1): 25-39. Luenberger, D.G. (1984). Linear and Nonlinear Programming, Second edition. Addison-Wesley. Minken, H., Jonsson, D., Shepherd, S.P., Järvi, T., May, A.D., Page, M., Pearman, A., Pfaffenbichler, P., Timms, P., Vold, A. (2003). Developing Sustainable Urban Land Use and Transport Strategies: A Methodological Guidebook. PROSPECTS Deliverable 14, TØI Report 619, TØI, Oslo. Mohring, H., Harwitz, M. (1962). Highway Benefits. An Analytical Framework. Northwestern University Press, Evanston, Illinois. Myles, G.D. (1995). Public Economics. Cambridge University Press, Cambridge. Nelder, J.A., Mead, R. (1965). Computer Journal, 7: 308. Parry, I.W.H., Bento, A.M.R. (2001). Revenue Recycling and the Welfare Effects of Road Pricing. Scandinavian Journal of Economics, 103(4): 645-671. Ramjerdi, F. (1995). Road Pricing and Toll Financing with Examples from Oslo and Stockholm. PhD Thesis, Royal Institute of Technology, Department of Infrastructure and Planning, Stockholm. Ramjerdi, F., Minken, H., Østmoe, K. (2004). Norwegian Urban Tolls. In: Santos, G. (ed), Road Pricing: Theory and Evidence, Elsevier Science. Sandmo, A. (2000). The public economics of the environment. The Lindahl Lectures Oxford University Press. Sen, A.K. (1982). Choice, welfare and measurement. Harvard University Press. Shorrocks, A.F. (1980). The Class of Additively Decomposable Inequality Measures. Econometrica, 48: 613-625. Strotz, R.H. (1965). Urban Transportation Parables. In: Margolis, J. (ed), The Public Economy of Urban Communities. Resources for the Future, Washington DC. Theil, H. (1967). Economics and Information Theory. North Holland Publishing Company, Amsterdam. Vickrey, W.S. (1963). Pricing in Urban and Suburban Transport. American Economic Review, 53: 452-465. Vickrey, W.S. (1968). Congestion Charging and Welfare. Journal of Transport Economics and Policy, 2: 107-118. Walters, A.A. (1961). The Theory and Measurement of Private and Social Cost of Highway Congestion. Econometrica, 29(4): 676-99.

11 Modelling the Economy, Transport and Environment Triangle, with an Application to Dutch Maglev Projects Jan Oosterhaven and J. Paul Elhorst Department of Economics and Econometrics, University of Groningen, the Netherlands

Abstract This chapter discusses modelling the ETE triangle from the perspective of the welfare consequences of local and regional policy measures. It argues that the interaction between the economy and the transport system needs to be modelled using sectors, household types, transport modes and spatial zones. For urban agglomerations, using a land-use/transportation interaction type of model should be weighed against using a spatial computable general equilibrium type of model. For interregional applications, a spatial equilibrium approach is superior because it enables the incorporation of economies of scale, substitution between inputs and heterogeneity of outputs. Furthermore, it is argued that – at the local and regional level – environmental externalities may be modelled without taking account of feedback effects on the economy and the transport systems. The modelling philosophy is applied to four Dutch magnetic levitation rail proposals showing that the location and trajectory of new transport infrastructures have an important impact on the size and mix of its direct transport effects, indirect economic effects and external environmental effects.

11.1 Introduction Modelling the interactions between the economy, the transport system and the environment (the ETE triangle) involves a wide range of choices. We discuss these choices from the perspective of analysing the impact of policy measures on people’s welfare at the spatial scale of individual cities and regions. It is important to note that the limitation of sub-national policy measures minimises the necessity to model the impact of exogenous factors at the macro level, such as technological breakthroughs and the growth of the world economy, since sub-national policy models can be based on exogenously provided national or international economic scenarios. The limitation of studying only welfare effects also constrains the choices at hand, but less so. The reason is that although the population in a particular country may care about the state of the environment in the rest of the world, it probably does not care about the state of the transport system and the economy elsewhere. This implies that only the environmental impacts need to be

208

J. Oosterhaven and J.P. Elhorst

modelled globally to be relevant to local policy decisions.1 This is not a major complication, since the feedback effects of global environmental impacts of local policy measures back on the local economy, transport and environment are negligible. In this chapter, we first compare the interactions between the elements of the ETE triangle, which prove to be different in nature. Next, the impact of the economy and the transport system on the environment and its consequences for modelling at the local and regional level are discussed in more detail. Finally, the interactions between the economy and the transport system are discussed in more detail, along with the modelling implications of the aim to simulate the impacts of policy interventions on the economy and the transport system. As an application of the modelling approach, four competing proposals to install new magnetic levitation (Maglev) rail projects in the Netherlands are discussed.

11.2 The Different Nature of the Three Types of Interaction Within this chapter, decisions about the production and consumption of transport services and of other goods and services will be treated separately. Hence, our definition of the economy and of economic activities is exclusive to the transport system and transport services. At the local and regional level, possible transport policy measures relate to such instruments as traffic tolls, the construction of new transport infrastructure, the (subsidised) provision of public transport and parking space. In some countries, local authorities may have additional powers such as levying fuel taxes and other transport-related taxes, such as parking tariffs. Generally, these instruments either have an explicit spatial dimension (i.e. location) or they have spatially strongly different impacts on the supply and demand for transport services and thus also on the (spatial) functioning of the economic system in the city or region in question. In market economies, local and regional authorities generally have limited possibilities to intervene in the economic system directly. At the local and regional level, possible economic policy measures relate to such instruments as taxing different types of land use, installing zoning regulations and sometimes also providing for rental housing, office space and labour market matching. With the exception of labour market matching, these instruments all have an explicit spatial dimension and will thus also influence the demand for transport services in a spatially differentiated manner. Two conclusions follow from the above discussion and may be summarised as follows (see also Fig. 11.1):

1

The recent guideline for social cost-benefit analyses of transport infrastructure projects in the Netherlands (CPB/NEI, 2000), for instance, restricts the measurement of the economic impacts to the national economy but insists on including worldwide environmental impacts, despite the obvious inconsistency in spatial scale (cf. Oosterhaven, 1999).

Modelling the Economy, Transport and Environment Triangle

1. 2.

209

Modelling the interactions between the economy and the transport system can only be done sensibly when it is done spatially. The interaction between the economy (i.e. production, consumption) and the transport system, in principle, runs both ways.

Neglecting the interdependent and essentially spatial nature of the relationship between the economy and the transport system will result in biased local and regional policy recommendations. To discuss the relationship between the environment on the one hand, and the transport system and the economy on the other, a distinction must be made between global impacts and local/regional impacts. Modelling global environmental impacts of local and regional policy measures is relatively simple, since feedback effects at the local/regional level are empirically negligible. Modelling their local environmental impacts is relatively complicated, since the spatial dimension (i.e. location) of the local impacts is at the heart of any local policy model. Let us take emissions as an example. Emissions occur at specific locations. Since their diffusion is spatial, the subsequent impact on welfare may be quite different in densely and sparsely populated neighbourhoods. The question of whether the feedback effects of local environmental impacts on the local economy or the local transport system are relevant is difficult to answer. Heavily disturbed and polluted neighbourhoods may become less attractive for households and even for firms. Consequently, location decisions and the demand for transport services will be influenced. But just as the feedback effects of global environmental impacts tend to be small, so are the feedback effects of local environmental impacts on the local level. Hence, two additional conclusions can be drawn (see also Fig. 11.1): 3. The interactions between the economy and the transport system on the one hand, and the environment on the other, may be modelled as a one-sided dependency of the environment on the economy and the transport system. This implies that environmentally motivated policy interventions in both the economic and the transport system may be treated exogenously. 4. Global environmental impacts can be specified non-spatially as opposed to local and regional environmental impacts that need to be specified spatially.

Economy (spatial)

Transport (spatial)

Environment, local (spatial) and global (non-spatial)

Technological or behavioural relationships Fig. 11.1. The nature of the ETE triangle at the local level

Policy-induced relationships

210

J. Oosterhaven and J.P. Elhorst

11.3 Modelling Environmental Impacts Many environmental impacts of both transport and economic activities are externalities; i.e. actors do not take account of the environmental impacts when deciding on their production and consumption levels and mix. Environmental impacts of decisions about the production and consumption of transport services constitute an important subcategory of all externalities, called transport-upon-environment externalities. Another important category of externalities of transport production and consumption decisions relates to such externalities as traffic congestion and traffic accidents. The valuation of these transport-upontransport externalities is psychologically different from that of environmental externalities. The reason is that congestion and accidents are commonly viewed as risks that people voluntarily choose to be subject to when they decide to produce or consume transport services. By contrast, environmental externalities are commonly viewed as involuntary risks, since they affect people that do not produce or consume the transport services in question. When the transport system and the economy are treated as separate entities, as in this chapter, no significant externalities seem to exist between the two systems. At the local and regional level, most of the economy-upon-economy externalities tend to be pecuniary and they are probably not very significant. For example, firms capture only a part of the direct transport benefits of new infrastructure because competition forces them to pass some or all of these benefits on to their clients. Non-pecuniary benefits at the local and regional level are probably restricted to transfers of knowledge, especially through the local labour market. At the local and regional level, transport-upon-economy and economy-upon-transport externalities do not seem to be different in nature from the economy-upon-economy externalities discussed above. Comparable to the transport system, but with a different mix, pollution, noise and the contribution to global warming appear to be the most important externalities originating from the economic system, i.e. economy-upon-environment externalities. From an environmental policy viewpoint, it is helpful to distinguish between intermediate and final externalities. Intermediate externalities may be defined as non-environmental externalities between actors in the transport system and the economy, which lead to (further) location, consumption and production reactions. These externalities must be modelled if they are significant, such as the transportupon-transport and economy-upon-economy externalities discussed above. Final externalities may be defined as …-upon-environment externalities, which do not lead to further behavioural reactions of actors in the economy or the transport system. It is to be noted that this distinction is not necessarily related to the distinction between direct and indirect environmental externalities, as both are final externalities in the sense defined above. The importance of the direct/indirect distinction is apparent when life-cycle analyses (LCA) of the total environmental cost of transportation are considered. LCA is best done by combining direct process data

Modelling the Economy, Transport and Environment Triangle

211

with more aggregate input-output data (Wilting, 1996). In an application to Dutch freight transportation, the direct/indirect energy use in the production and maintenance of infrastructure and means of transportation appeared to be 82/18% of the total energy use in road freight, 55/45% in rail freight and 40/60% in inland water transportation (Bos, 2000). Although these percentages are different and sensitive to assumptions made, data used and, especially, effects included, the rank order of the direct impacts did not appear to be different from the rank order of the total impacts. Essentially the same conclusions apply to Dutch passenger transportation (Bouwman, 2000). However, rank orders do change in applications at a more detailed level, as in GMC/Argonne (2002, cited in Harrington and McConnell, 2003, p.250) that studies the use of different fuels. The recommendation to distinguish intermediate and final externalities, and to neglect intermediate ones when estimating environmental impacts, implies that some indirect (backward) environmental externalities may also be neglected. This will not influence the rank order of the externalities per transport mode, as the Dutch LCA research has shown for final externalities alone. Perhaps more important is the problem that all of these direct/indirect calculations neglect the spatial dimension. Although the environmental cost involved in these national LCAs relates to the global level, the estimation of these global environmental costs is generally based on data of (sub-)national cost structures.

11.4 Modelling Transport-Economy Interactions There is city-level information that shows that lower population densities may be associated with fewer car miles per capita travelled (Newnan and Kenworthy, 1989). At the micro level, however, Dunphy and Fischer (1996) found evidence of fewer car miles and more public transport miles per capita for people living in higher density communities. In addition, they found that demographic characteristics determined location choice and travel behaviour simultaneously. Kitamura et al. (1997), using travel diary data, found personal attitudes to driving, the environment and other factors to be more important in explaining travel behaviour than land-use variables. The extensive literature on the balance between the number of jobs and the number of houses, following Hamilton’s (1982) provocative article on the concept of ‘wasteful commuting’, comes to a more or less comparable conclusion. The extent of wasteful commuting diminishes when more explanatory variables are added, such as double-earner families, different spatial distributions of particular job qualifications and discrimination in the housing market (White, 1999). Reviewing the literature on the relationship between land-use and travel demand, Harrington and McDonnell (2003, p.214) came to the following conclusions: 1. Compared with other variables, most studies find either no or only small effects of land-use variables on travel measures, such as vehicle miles travelled or vehicle ownership.

212

2.

3.

J. Oosterhaven and J.P. Elhorst

Urban form tends to have more impact on vehicle ownership than on miles travelled, and the effect tends to be greater at very high densities (i.e. above 5000 people per square mile, Walls et al., 2002). There are differences in land-use impacts through the effect on travel (time and distance) cost, at least for non-work trips.

This means that, even with significant effects at very high densities, policies directed at changing existing land-use are probably less effective in affecting travel patterns than policies that influence (time and monetary) travel cost more directly. This last relationship also plays a central role in theoretical models of urban form. In the monocentric base model (Alonso, 1964), commuters trade off lot size (i.e. density) and transportation cost. Using this model, the long-term decline in real (time and distance related) travel cost provides a straightforward explanation of suburbanisation and decreasing urban densities (Pickrell, 1999). In contrast, high land prices (i.e. density), high cost of road construction, and traffic congestion in the inner cities result in high generalised private transportation costs, which together may open the way to public mass transit. It also offers an alternative explanation for the decentralisation of housing and jobs (Ingram and Liu, 1999). In addition to the relationship between urban form and travel behaviour, that between urban form and domestic and holiday behaviour may also be of (environmental) importance. After controlling for different household characteristics, Diepen (2000) found that travel behaviour, holiday behaviour and domestic energy use depended on the residential neighbourhood. In the inner cities of two Dutch towns, each with about 150,000 inhabitants, total energy use per comparable household appeared to be lower. Interestingly, she did not find total energy use to be lower in sustainably designed neighbourhoods with sustainably designed houses. The location of other city neighbourhoods in relation to the city centre and the number of competing destinations appeared to be much more important. This brief review indicates that spatially detailed models provide the only way to adequately model the interaction between the economy and the transport system at the local and regional levels, especially when one is interested in environmental externalities associated with local transport and land-use regulation. Below, we will further discuss and compare two broad classes of such models, namely land-use/transportation interaction (LUTI) models and spatial computable general equilibrium (SCGE) models. LUTI models consist of linked transport models and land-use or location models. They generally employ a system dynamics type of modelling and are primarily developed to predict future growth and to analyse policy scenarios for large urban conglomerations (for example, Lee et al., 1995). There is a whole series of such models for different conglomerations. In this respect, the LINE model stands out as it is a LUTI type of interregional model for the whole of Denmark (Madsen and Jensen-Butler, 2003). LUTI models have a long history of gradual development over many decades and are currently typically very disaggregated with numerous spatial zones, sectors, household types, transport motives, modes of transport, etc. (for overviews see DSC/ME&P, 1998; Wilson, 1998).

Modelling the Economy, Transport and Environment Triangle

213

SCGE models are typically comparative static equilibrium models of interregional trade and location based on microeconomic theory, using utility and production functions with substitution between inputs. Firms may operate with economies of scale in markets with monopolistic competition of the Dixit-Stiglitz (1977) type. Empirical applications of this last approach are to be found in Venables and Gasiorek (1996) and Bröcker (1999). Interesting theoretical simulations with a SCGE model with a land market are to be found in Fan et al. (1998). These models are part of the new economic geography school (Krugman, 1991, Fujita et al., 1999), and have been around for less than a decade. In other words, we are comparing a mature methodology, possibly at the end of its life cycle, with a new methodology that is still in its infancy. The practical feasibility of LUTI models is wide. In particular, the transport sub-models are known to be very adequate in estimating all kinds of transport price and quantity impacts of policy measures in the transport sector itself. Given the scientific uncertainty in relation to the location behaviour of firms and the decrease in the relative cost of freight transport over time, this does not hold true to the same degree for the impact of transport measures on the location of industrial activities. Since the relative time cost of passenger transport has been increasing over time, due to increased congestion and rising real incomes, the location of service activities can be explained much better. However, as the location of most service activities primarily follows that of people and industrial activities, the location choices of service providers mainly play a role at the intra-urban level. Consequently, the strength of LUTI models lies especially in estimating the impact on intra-urban location decisions rather than in estimating the interregional locational effect of transport measures. Finally, most LUTI models are not well equipped to translate the impacts of transport and infrastructure measures into estimates of consumer benefits, as is needed in a sound cost-benefit analysis (CBA underpinned by welfare theory). At best, consumer choices relating to transport and location decisions are modelled and estimated by means of a discrete random utility approach. In contrast, producer location decisions are seldom modelled by means of discrete profit maximising behaviour, and producer production and price decisions tend to be modelled by fixed ratios. As a consequence, most LUTI models provide reasonable estimates of direct transport user benefits, and reasonable estimates of consumer benefits in as far as the latter are based on discrete choice behaviour. The existing LUTI models, however, are not able to estimate transport benefits that are based either on continuous consumer choices or on both discrete and continuous producer choices. SCGE models, typically, are theoretically well-suited for this evaluation task (see Venables and Gasiorek, 1998). The SCGE modelling problem, at the moment, is not theoretical in nature but rather empirical and computational. Consistent estimation of all the necessary consumption and production substitution elasticities is problematic, if only because of the lack of adequate data and the lack of a tradition of estimating such elasticities at the regional level. Moreover, the calibration of these models such that they reproduce recent history and simultaneously provide plausible (that is, stable) projections is also problematic, especially because of the highly nonlinear character of the behavioural equations. Another problem is that SCGE models

214

J. Oosterhaven and J.P. Elhorst

are not easily understood, making it difficult for policymakers and other researchers to assess the validity of the results. Whether LUTI models can easily incorporate imperfect markets, and internal and external economies or diseconomies of scale, is doubtful. The strength of most LUTI models lies in their segmentation and detail, that is, they usually contain many different zones, transport modes, household types, firm types, and so on. The benefit of having such detail lies in the homogeneity of behaviour and the assumed stability of relations at that level of detail. However, this detail is achieved at the cost of mathematical and theoretical sophistication, which results in assumptions of perfect competition, fixed ratios, linear relations and the absence of economies of scale. The current, still young SCGE models have the opposite properties, namely a lack of detail and sound empirical foundation, but a sophisticated theoretical foundation and complex non-linear mathematics. The latter is precisely the reason why SCGE models are able to model economies and diseconomies of scale, external economies of spatial clusters of activity, continuous substitution between capital, labour, energy and material inputs in the case of firms, and between different consumption goods in the case of households. Moreover, monopolistic competition of the Dixit-Stiglitz type allows for heterogeneous products implying variety, and therefore allows for cross hauling of close substitutes between regions. Finally, SCGE models lead to a direct estimation of the welfare effects, in particular of the non-transport benefits of new infrastructure, which is absent in most LUTI models. The modelling requirements that follow from our discussion are summarised in Table 11.1. Whether a further piecemeal improvement of LUTI models is preferable to the implementation of a theoretically superior but as yet untested alternative is essentially a matter of preference and belief. DSC/ME&P (1998) confess to the piecemeal improvement strategy. We would like to advocate the more promising but also more risky development of empirically-based SCGE modelling, at least when interregional as opposed to intra-urban problems are examined, as is illustrated in the next section. Table 11.1. Requirements for modelling the ETE triangle at the local level Spatial economy

Transportation

Environment

– sectors – household types – land use by zones – type of markets

– modes – freight/passenger types – origin/destination zones

– local, by destination zones – global, no zones

11.5 An Application to Dutch Maglev Proposals 11.5.1 Introduction Since 2001, the Dutch government has been considering two magnetic levitation (Maglev) rail projects, each with two variants. (1) An inner ring or an outer ring connecting the four largest cities (Amsterdam, The Hague, Rotterdam and Utrecht)

Modelling the Economy, Transport and Environment Triangle

215

in the heavily urbanised economic core in the west of the Netherlands (the socalled Randstad region). (2) A direct connection between Schiphol Airport in the Randstad region and Groningen in the more peripheral, rural north, either running along the south-east or along the north-west shore of the IJsselmeer lake in the middle of the country (see Fig. 11.2, for the four trajectories). The primary aim of a fast rail ring within the Randstad is to improve its internal accessibility by public transport. In turn, this may reduce traffic congestion and therefore also improve its internal accessibility by car. Both may strengthen the Randstad’s competitive position in attracting internationally mobile economic activities. In addition, compared to other regions in the Netherlands, the need for space for new residential areas and industrial sites is much more pressing in the Randstad. With new fast rail links, it might be possible to direct the urbanisation process away from the remaining vulnerable agricultural and natural areas within the Randstad. The primary aim of a fast rail link between the Randstad and the north is to stimulate the lagging northern economy. With a fast rail link, people would not have to leave the north for jobs in the Randstad, instead they could commute to them. This would increase demand for locally-produced goods, which would in turn initiate a multiplier process leading to a higher level of regional production and employment. A fast rail link would also lower the prices of services both supplied and demanded by firms located in the north, possibly shifting the competitive balance in favour of locations in the north in spite of the ‘two-way road’ argument (SACTRA, 1999, p.16). Both effects are seen as a key to the further economic development of the northern Netherlands. The secondary objective of a fast rail link between the Randstad and the north is to relieve the Randstad’s capacity constraints in transport, land and labour markets, which result in losses of time, high transport costs, labour shortages, high housing prices and high cost of living. As these costs are partly external to private decision-makers, they do not fully deter the spatial concentration of people and economic activities, as such costs are not taken into account in private location decisions (Elhorst et al., 1999). Whether a fast rail connection to the north will produce the desired relief in the Randstad remains to be seen, as the flow of industrial activity away from the economic core so far has mainly been directed towards adjacent regions and not towards the periphery. The net present value of the investment costs of the inner and the outer Maglev ring in the Randstad are estimated at EUR 6,835 and EUR 9,088 billion, respectively, and that of the core-periphery Maglev along the north-west or the south-east of the IJsselmeer at EUR 7,500 and EUR 6,666 billion, respectively. Each estimate includes a mark-up for uncertainties and risk. It has been assumed that the construction of the rail infrastructure would take place in the period 2010-2015 and that its use would start after completion. Costs and benefits are calculated as net present values for 2010 (in prices of 2000), using a social discount rate of 4% over a 30-year-period (20102040). Every project is evaluated in comparison with a spatially detailed baseline scenario that is based on the moderate ‘European Coordination’ macro-economic scenario of the CPB (1997).

216

J. Oosterhaven and J.P. Elhorst

Fig. 11.2. Trajectories for the four proposed Maglev rail projects

The research concerning the four Maglev proposals was part of wider investigations considering different routes, different service levels (frequency, schedule,

Modelling the Economy, Transport and Environment Triangle

217

waiting time), different price levels as well as different rail systems, such as regular rail and high-speed rail services and their accompanying infrastructure investments. In addition to direct transport effects and direct external (environmental) effects, which have been studied in the usual fashion, these investigations stand out as they also consider indirect effects. The estimated direct transport cost and benefits include operating costs and revenues, and transport costs and time benefits for both people and freight. The estimated indirect economic benefits relate to the so-called forward programme or induced effects. These are defined as the consequences of the reduction in transport cost for production, market and location decisions of people and firms, and the subsequent interregional redistribution effects with respect to income and employment of the population at large. In view of the theoretical considerations in the first part, the second part of this chapter will discuss in particular the efforts to model the interaction between the economy and the transport system, as well as the welfare effects of congestion and environmental impacts. The welfare effects will be scaled by computing them as a percentage of the investment costs.

11.5.2 Modelling the Interaction Between the Economy and the Transport System One of the main benefits of new infrastructure is the time benefit for people. We are particularly interested in the time benefits by car that occur due to reduced congestion as people substitute public transport for car transport. These time benefits can be split into direct effects, as usually calculated under the assumption of a fixed spatial distribution of population and employment, and indirect effects that are due to the changes in these spatial distributions. The direct benefits of reduced congestion are taken from NEI (2001a, 2001b) and have been calculated using a more or less standard 4-stage transport model (called LMS), which explains and predicts commuting flows, provided that the marginal totals of the trip distribution matrices are given. For major transport improvements, such as the Maglev proposals considered here, this approach is unsatisfactory, as the spatial distribution of population and employment is not exogenous to changes in the transport system. To model these indirect economic effects, we additionally considered two relatively independent main indirect effects and two derived interaction effects, as shown in Fig. 11.3. The first main effect relates to housing migration of the working population. When travel times diminish, due to improvements in the transport network, people may increase the quality of their housing accommodation and living environment by increasing the length of their commuting journey without changing their commuting journey time. This principle has been used to develop a commuter location model that takes actual commuting behaviour as given and then projects where people will choose to live given the location of their jobs.

218

J. Oosterhaven and J.P. Elhorst

Commuter location model: housing migration

Labour supply

Travel time reductions

Monopolistic competition type of spatial CGE model

Commuter location model: labour migration

Labour demand

Migrants’ expenditures input-output model

Fig. 11.3. Modelling scheme for calculating indirect economic effects in the Maglev application

Actual commuting behaviour is approximated by a commuting time distribution matrix, which specifies commuter shares by mode (car, public transport and slow transport), by time class (25 classes of 5 minutes) and by type of municipality (four biggest cities, municipalities with a railway station and municipalities without a railway station). The matrix is based on 70,886 observations, while potential changes in modal shares have been modelled with the help of an almost ideal demand system (AIDS, see Elhorst and Oosterhaven, 2006, for further details). With this set-up, the commuter-location model transforms the spatial distribution of employment into a spatial distribution of working population, both across 548 municipalities, dependent on the willingness to commute, and on municipality-bymunicipality travel-time matrices for the three modes of transport during peak hours. In addition, the transformation has been made dependent on the relative attractiveness of each municipality as a residential area. Although the location choice of each individual is free, the entire population is constrained by the total housing supply in each municipality. For this reason, this variable is suitable to test the fit of the model. It appeared that – with this attractiveness variable – the working population living in the 12 NUTS-2 and the 40 NUTS-3 regions of the Netherlands could be predicted with an average error of 7%. The amount of available land, however, better approximates the spatial preferences of people, the majority of whom prefer larger lots in areas that are greener (Elhorst et al., 1999; VROM, 2000). For this reason, the amount of available land is used to simulate longer term residential changes, assuming that the housing market has time to adjust to the changes in the transport system and to follow these residential preferences. The second main effect relates to travel-cost induced employment changes. If the transport costs of inputs and outputs change differentially in different locations, the optimal location and production size of the firm is expected to change. New economic geography (NEG) theory has pointed out that imperfect competition and increasing returns to scale in transport-using sectors are reasons why traditional location approaches may produce inaccurate estimates. There are two different NEG types of models. In the footloose labour models pioneered by Krugman (see

Modelling the Economy, Transport and Environment Triangle

219

Fujita et al., 1999), locations close to markets pay higher real wages than locations away from the major markets. They consequently attract labour, which further enlarges the market and causes a further concentration of economic activity. The forces of concentration depend on the level of trade costs and the proportion of the population that is mobile in response to wage differences. In the vertically linked industries model developed by Venables (1996), the process of cumulative causation is not driven by footloose labour but by cost and demand linkages between industries. Firms in the downstream industry will have lower costs if they locate close to upstream firms, they save trade costs on their intermediate inputs, while market access considerations draw the upstream industry to locations with relatively many downstream firms. In this study, a Venables-type of NEG model, called RAEM, the Dutch acronym for a spatial computable general equilibrium model, has been developed and calibrated (see Knaap and Oosterhaven, 2000). In contrast to the footloose labour type of NEG model, the working population in RAEM is assumed to be immobile and therefore cannot cause agglomeration. The reason for abandoning the footloose labour model is that wages in the Netherlands do not differ much between regions because they are determined on a national sectoral scale by collective bargaining. Consequently, there is little incentive to migrate between regions in order to receive a higher wage. Instead, the wage level has been assumed equal throughout the country, while the commuter location model is used to model the location choice of the working population. In contrast to standard NEG models, it has been assumed that transport costs relate to both freight and passengers (personal business travel and shopping travel by the customers of the firm). Just as in standard NEG models, the transport cost mark-up on f.o.b. prices for freight depends on distance, but for passengers it is made dependent on travel time. Furthermore, it has been assumed that the travel times related to personal business travel and shopping travel are different from those in the commuter location model. Instead of the three modes of transport and their corresponding travel times during peak hours in the commuter location model, travel time in RAEM consists of the travel time by car and by public transport during off-peak hours weighted by their modal shares in business/ shopping travel. One of the basic problems of monopolistic competition models of the spatial economy is the estimation of their parameters such that they reproduce recent history and simultaneously provide plausible (in particular, stable) projections. This is a difficult issue because of the non-linear character of the behavioural equations. To reach maximum accuracy, 14 sectors and 548 municipalities have been identified. Cobb-Douglas production and consumption expenditure shares have been taken from the (coincidentally also) 14 bi-regional input-output tables of the twelve Dutch provinces and the greater Amsterdam and greater Rotterdam regions (RUG/CBS, 1999, Eding et al., 1999). The 14 crucial elasticities of substitution, which inter alia determine the extent of market competition, and the 4 distance parameters, have been estimated econometrically. This is done by minimising the sum of squared residuals of predicted and observed trade flows taken from the bi-regional input-output tables (i.e. 588 flows of exports to, imports from and intra-regional transactions of the 14 regions, for the 14 sectors studied). The R2 of regressing the observed (log) flows of

220

J. Oosterhaven and J.P. Elhorst

trade on the (log) flows predicted by the model (without intercept) is 0.51. Although there is room for improvement, 2 any change in the model would lead to a compensating change in the parameters then to be re-estimated, which may have not much influence on the outcomes of the model. The first derived effect is the subsequent reaction of the working population to the predicted change in production and labour demand, and is labelled labour migration. Note that the commuter location model predicts housing migration as a result of reduced travel times starting with a given level of employment in each municipality, whereas this run of the commuter location model measures labour migration as a result of changes in employment opportunities. Total migration is the sum of housing migration and labour migration. The second derived effect relates to consumption-induced employment changes caused by the total migration of workers. Due to a lack of data, this effect is not determined at the level of the 548 municipalities, as were the three previous effects, but rather at the level of the 40 Dutch NUTS-3 regions, using a 40x40 employment multiplier matrix of working migrants (see Oosterhaven, 2005, for details). This matrix is again based on the 14 bi-regional input-output tables. The total labour demand effect is the sum of the travel cost-induced and consumption-induced employment effects. The empirical results may be summarised as follows (see Oosterhaven and Elhorst, 2003, for details). The primary aim of the urban-conglomeration proposals is to strengthen the (international) competitive position of the heavily urbanised Randstad. The results in this respect indicate that, due to the redistribution of labour demand within the Netherlands, employment in the Randstad will increase by 2,400 jobs with the inner variant and by 2,750 jobs with the outer variant.3 When looking at other regions and at intra-regional changes within the Randstad, it has been found that the urban rail link would strengthen the process of suburbanisation. Within the four big agglomerations, the central municipalities of Amsterdam, Rotterdam, The Hague and Utrecht would experience a population decrease, whereas surrounding municipalities close to a Maglev station would experience a population increase. This suburbanisation process would also extend to the regions adjacent to the Randstad. These regions would experience a decrease in the number of jobs, whereas their populations would increase. By contrast, the more peripheral regions such as the north of the Netherlands would hardly benefit from a fast rail link within the Randstad, neither in terms of employment nor in terms of population. The primary aim of the core-periphery proposals is to stimulate the peripheral north. The results in this respect indicate that employment in the north would increase by 3,950 jobs with the south-east variant and by 8,050 jobs with the northwest variant. The working population would increase by 4,000 people in the southeast variant and by 9,400 people in the north-west variant. In sum, the north would 2 RAEM is currently being further developed together with TNO Inro (Delft) and the Free University Amsterdam. 3

From an international viewpoint, employment in the Randstad would further increase by about 1,300-1,420 jobs (BCI, 2001).

Modelling the Economy, Transport and Environment Triangle

221

indeed catch up. Furthermore, it can be concluded that the north-west variant would be about twice as effective in stimulating the north, than the south-east variant, whereas the core-periphery projects would be far more effective in creating jobs in the north than the intra-core projects. The secondary aim of the core-periphery proposals is to relieve the (land, traffic and labour) pressures in the Randstad. The results indicate that in the south-east variant 7,045 people would leave the Randstad, whereas in the north-west variant the working population in the Randstad would increase by 100 people. This implies that, in view of this secondary aim, the south-east variant would be far more effective than the north-west variant, whereas both intra-core projects would be counter-productive in this respect.

11.5.3 A Welfare Evaluation of External Effects Although the interregional redistribution of economic activities described above provides useful information for policy purposes, it falls short of a welfare assessment of the new infrastructure. Here, the results of a welfare evaluation of most of the external effects are given. They are recorded in Table 11.2. The first two lines present direct and indirect transport-upon-transport externalities. The last four lines represent final environmental externalities. The estimates of the direct benefits of reduced congestion are taken from NEI (2001a, 2001b). The outcomes are derived from a large 4-stage transport model of the Netherlands (LMS), which explains and predicts commuting flows, provided that the marginal totals of the trip distribution matrix are given. Remarkably, the coreperiphery variants do not lead to any reduction of direct congestion, although it was found that about 6% of the car commuters between the cities with a train station along the north-west and south-west variant would switch to public transport. For the Netherlands as a whole, it involves about 8,000 commuters per day. Although 6% is relatively high, the reduction mainly occurs outside the urban core where congestion is not a major problem. In contrast, the congestion effects found for urban-conglomeration proposals show that these fully meet the first objective of reducing congestion; with the inner ring, congestion cost declines by 36% of the investment costs. The percentage found for the inner ring is greater than that for the outer ring for two reasons. First, the inner ring is cheaper and shorter (140 km against 170 km) as it stops at the edges of Rotterdam and Utrecht. Second, the inner ring attracts more passengers who leave their cars because of the relatively large time benefits, whereas the outer ring – connecting city centres – attracts more new passengers and passengers who already use other forms of public transport. The determination of the indirect benefits of reduced congestion is based on Elhorst et al. (1999). When the number of people and jobs decrease in the urban core and increase in the peripheral regions, so will the amount of traffic. When the amount of traffic decreases in the urban core and increases in the peripheral region, absolute congestion costs decrease considerably in the core region and increase only slightly in the peripheral regions. From Table 11.2, it is clear that all projects lead to a more

222

J. Oosterhaven and J.P. Elhorst

balanced spread of traffic over the country but that this effect is much larger for the core-periphery projects, especially for the south-east variant. This is because in this variant many more people leave the urban core. Table 11.2. Welfare impacts of the external effects in percentage of investment costs

Direct congestion benefits

Randstad Maglev, inner ring 36.0

Randstad Maglev, outer ring 16.5

SchipholGroningen, north-west 0

SchipholGroningen, south-east 0

Indirect congestion benefits

0.3

0.2

0.9

3.1

CO2 and NOX emission effects

1.6

0.4

-2.5

-2.5

Net open landscape benefits

0.4

0.3

1.1

3.8

Noise pollution cost

-0.7

-9.9

-5.1

-6.7

Qualitative estimate of remaining non-valued environmental effects

-/+

-/+

--

--

The final environmental externalities of a new rail link relate to the natural environment, carbon dioxide and nitrogen oxide emissions, noise, external safety, and land-use effects. Three different effects must be considered. First, the construction and service of a new rail link causes direct environmental cost through increased emissions and negative landscape impacts. Second, by contrast, the substitution of new public transport for old car transport causes environmental benefits through reduced emissions. Third, the relocation of employment and population may indirectly cause environmental costs or benefits in different regions, which may lead to a net national welfare effect. Although many of these effects are quantified, only the open landscape impacts, the CO2 and NOx emissions and the noise impacts could be valued in monetary units. The welfare effects of changes in CO2 and NOx emissions are taken from NEI (2001a, 2001b, see also Wee et al., 2003). Remarkably, they are positive for the urban-conglomeration projects and negative for the core-periphery projects. The explanation is that the urban-conglomeration projects, especially the inner variant, encourage far more people to leave their cars than the core-periphery projects, as discussed above. In the case of the intra-core projects, this more than compensates for the high energy use of the Maglev system. Net open landscape benefits occur because of the relocation of people and jobs. The restrictive spatial planning in the Netherlands aims at preserving open agricultural and natural landscapes. These are most scarce in the urbanised western part of the country and relatively abundant in the least urbanised northern part of the Netherlands, which leads to considerable price differences for technically comparable housing (Creusen, 1999). Especially in the core-periphery variants, the relocation of housing and jobs results in less pressure on open landscapes in the Randstad and more pressure in the north, which leads to a net increase in welfare. Housing price differentials are used to estimate the relative valuation of open landscapes in the different parts of the country.

Modelling the Economy, Transport and Environment Triangle

223

Second, public investment costs for developing new residential areas and construction costs in densely populated regions tend to be higher than in sparsely populated regions. In addition to this, they also tend to be higher in urbanised parts within these types of regions than in non-urbanised parts (Sievers and Keers, 1992). The total estimate of the benefits due to imperfections in the housing market is derived from Elhorst et al. (1999). The results regarding the open landscape benefits are comparable to those concerning the indirect benefits of reduced congestion because they are based on the same interregional relocation figures reported in section 11.5.2. The welfare impact of noise pollution depends on factors such as the driving speed of the trains, the number of train passengers and the number of people living along the new rail tracks. It also depends on whether the new railway line is combined with existing road or railway infrastructures. In the latter case, annoyance hardly increases and may even remain unchanged if the noise emission of the Maglev falls (more than 3dB) below that of road traffic (for example, if the Maglev reduces speed when approaching a stop). The annoyance effect of the inner variant of the urban-conglomeration project is relatively small (Gotink, 2003) as it is almost completely combined with the existing road infrastructure. By contrast, the annoyance effect of the outer variant of the urban-conglomeration project is ten times higher. This is striking because it is almost completely combined with the existing railway infrastructure. Apparently, a combination with continuous noiseproducing road infrastructure is much less disturbing than a combination with infrequent noise-producing railway infrastructure. The annoyance effect of noise pollution in the core-periphery projects lies midway between the two urbanconglomeration projects because these projects are partly combined with existing road and partly with existing railway infrastructure. First attempts to evaluate the direct external effects on external safety and the natural environment indicate that these effects are relatively small. The ‘-/+’ sign in the urban-conglomeration projects is used to indicate that the non-valued environmental effects are negative with respect to the built and natural environment and are positive with respect to safety, such that the overall effect is uncertain.

11.6 Conclusion This chapter has discussed modelling the ETE triangle from the perspective of the welfare consequences of local policy measures. It has been argued that economytransportation interaction needs to be modelled using sectors, household types, transport modes and spatial zones. For urban agglomerations, using a land-use/ transportation interaction type of modelling should be evaluated against using a spatial computable general equilibrium type of model. For interregional applications, a spatial equilibrium approach is superior because it enables the incorporation of economies of scale, substitution between inputs and heterogeneity of outputs. Furthermore, we conclude that  at the local and regional level  environ-

224

J. Oosterhaven and J.P. Elhorst

mental externalities may be modelled without taking account of feedback effects on the economy and the transport system. The modelling philosophy is applied to four Dutch magnetic levitation rail proposals. When concentrating on the welfare effects of congestion changes and the environmental impacts, the greatest strength of the urban-conglomeration projects is their reduction of congestion, whose valuation is up to 36% of the investment costs when connecting the edges of cities. As this type of connection also attracts more passengers who leave their cars, a net decrease in carbon dioxide and nitrogen oxide emissions is possible. Noise pollution can be kept within bounds if the new railway line is combined with the existing road infrastructure. The greatest strength of the core-periphery projects is their reduction of congestion and pressure on open landscapes in the urban core due to the differential growth of employment and population from relatively overcrowded to relatively rural regions, whose combined valuation accounts for up to 7% of the investments costs if its route is carefully chosen.

References Alonso, W. (1964). Location and Land Use: Toward a General Theory of Land Rent. Harvard University Press, Cambridge, Mass. BCI (2001). Strategische internationale effectenstudie Rondje Randstad. Buck Consultants International, Nijmegen Bos, S. (2000). Direction Indirect: The Indirect Energy Requirements and Emissions from Freight Transport. University of Groningen, Ph-D, Groningen. Bouwman, M. (2000). Tracking Transport Systems: An Environmental Perspective on Passenger Transport Modes. University of Groningen, Ph-D, Groningen. Bröcker, J. (1999). Trans European Effects of “Trans-European Networks”: Results from a Spatial CGE Analysis. Technological University, Dresden. CPB (1997). Economie en fysieke omgeving, beleidsopgaven en oplossingsrichtingen 1995-2020. Central Planning Bureau, The Hague. CPB/NEI (2000). Evaluatie van infrastructuur projecten, Leidraad voor kosten-batenanalyse, Central Planning Bureau/Netherlands Economic Institute, The Hague. Creusen, H. (1999). Housing Construction: Between Competition and Regulation. Central Planning Bureau, Report 1999/2, 20-23, The Hague. Diepen, A. van (2000). Households and their Spatial-Energetic Practices: Searching for Sustainable Urban Forms. University of Groningen, Geographical Studies 266, Groningen. Dixit, A.K., Stiglitz, H. (1977). Monopolistic Competition and Optimum Product Diversity. American Economic Review, 67: 297-308. DSC/ME&P (1998). Review of Land Use/Transportation Interaction Models. Report to SACTRA, DETR, UK.

Modelling the Economy, Transport and Environment Triangle

225

Dunphy, R.T., Fischer, K. (1996). Transportation, Congestion and Density: New Insights. Transportation Research Record, 1522: 89-96. Eding, G.J., Oosterhaven, J., de Vet, B., Nijmeijer, H. (1999). Constructing Regional Supply and Use Tables: Dutch Experiences. In: Hewings, G.J.D., Sonis, M., Madden, M., Kimura, Y. (eds.), Understanding and Interpreting Economic Structure, Springer Verlag., Berlin: 237-63.. Elhorst, J.P., Oosterhaven, J., Sijtsma, F., Stelder, D. (1999). Welfare Effects of Spatial Deconcentration: A Scenario for the Netherlands. Tijdschrift voor Economische en Sociale Geografie, 90: 17-31. Elhorst, J.P., Oosterhaven, J. (2006). Forecasting the Impact of Transport Improvements on Commuting and Residential Choice. Journal of Geographical Systems, 8: 39-59. Fan, W., Treyz, F., Treyz, G. (1998). Towards the Development of a Computable Geographic Equilibrium Model. Paper to the North American RSAI Meetings, Santa Fe. Fujita, M., Krugman, P.R., Venables, A.J. (1999). The Spatial Economy, Cities, Regions and International Trade. MIT Press, Cambridge, MA. Gotink, A. (2003). Monetaire waardering van geluidsoverlast veroorzaakt door de magneetzweefbaan. MA-thesis, University of Groningen. Hamilton, B. (1982). Wasteful commuting. Journal of Political Economy, 90: 1035-51. Harrington, W., McConnell, V. (2003). Motor Vehicles and the Environment. In: H. Folmer and T. Tietenberg (eds.), The International Yearbook of Environmental and Resource Economics. Edward Elgar, Cheltenham: 190-268. Ingram, G.K., Liu, Z. (1999). Determinants of Motorization and Road Provision. In: GomezIbanez, J.A., Tye, J.A., Winston, C. (eds.), Essays in Transportation Economics and Policy. The Brookings Institute, Washington DC, ch. 10. Kitamura, R., Mokhtarian, P., Laidet, L. (1997). A Micro-Analysis of Land Use and Travel in Five Neighbourhoods in the San Francisco Bay Area. Transportation, 24: 125-58. Knaap T., Oosterhaven, J. (2000). The Welfare Effects of New Infrastructure: An Economic Geography Approach to Evaluating a New Dutch Railway Link. Paper to the 47th North American RSAI meetings, Chicago. Krugman, P. (1991). Geography and Trade. MIT Press, London. Lee, S.Y., Haghani, A.E., Byan, J.H. (1995). Simultaneous Determination of Land Use and Travel Demand with Congestion: A System Dynamics Modeling Approach. TRB Paper No. 950716, Washington D.C. Madsen, B., Jensen-Butler, C. (2003). Theoretical and operational issues in sub-regional modelling, illustrated through the development and application of the LINE. AKF Forlaget, Copenhagen. NEI (2001a). Verdiepte KKBA van een snelle verbinding met het Noorden. In: BCI (ed.), Verdieping maatschappelijke kosten-batenanalyse Zuiderzeelijn. Buck Consultants International, Den Haag.

226

J. Oosterhaven and J.P. Elhorst

NEI (2001b). KKBA van een snelle verbinding tussen de vier grote steden: ‘Rondje Randstad’. Rotterdam. Newnan, P.W.G., Kenworthy, J.R. (1989). Cities and Auto Dependence: A Source Book. Gower, Aldershot, UK. Oosterhaven, J. (1999). De evaluatie van grote infrastructuur: over MER’s, EER’s en MKBA’s. Openbare Uitgaven, 31: 60-70. Oosterhaven, J. (2005). Spatial Interpolation and Disaggregation of Multipliers. Geographical Analysis 37(1): 69-84. Oosterhaven, J., Elhorst, J.P. (2003). Indirect Economic Benefits of Transport Infrastructure Investments. In: Dullaert, W., Jourquin, B.A.M., Polak, J.B. (eds), Across the Border. Building upon a quarter century of transport research in the Benelux. De Boeck, Antwerpen: 143-62. Pickrell, D. (1999). Transportation and Land Use. In Gomez-Ibanez, J.A., Tye, W.B., Winston, C. (eds.), Essays in Transportation Economics and Policy. The Brookings Institute, Washington DC. RUG/CBS (1999). Regionale Samenhang in Nederland, Bi-regionale Input-output Tabellen en Aanbod- en Gebruiktabellen voor de 12 Provincies en de Twee Mainport Regio’s. Stichting Ruimtelijke Economie, University of Groningen, REG-publicatie 20, Groningen. SACTRA (1999). Transport and the Economy. Standing Advisory Committee on Trunk Road Assessment, DETR, TSO, London. Sievers, A., Keers, G.P. (1992). Investeringskosten nieuwe woningbouwlokaties overige stadsgewesten 1995-2015. Rigo Research en Advies b.v., Amsterdam. Venables A.J. (1996). Equilibrium Locations of Vertically Linked Industries. International Economic Review, 37: 341-359. Venables, A.J., Gasiorek, M. (1996). Evaluating Regional Infrastructure: A Computable Equilibrium Approach. (Mimeo) Report to the European Union. Venables, A. J., Gasiorek, M. (1998). The Welfare Implications of Transport Improvements in the Presence of Market Failure. (Mimeo) Report to SACTRA, DETR, UK. VROM (2000). Nota Wonen; Mensen, Wensen, Wonen (Summary). Ministry of Spatial Affairs, The Hague. Walls, M., Krupnick, A., Harrington, W. (2002). Population Density, Transit Availability, and Vehicle Travel: Results from a Nested Logit Model of Vehicle Use. Paper to the AEA Meetings, Atlanta. Wee, B. van, Van den Brink, R., Nijland, H. (2003). Environmental impacts of high-speed rail links in cost-benefit analyses: a case study of the Dutch Zuiderzee line. Transportation Research Part D, 8: 299-314. White, M.J. (1999). Urban Areas with Decentralized Employment: Theory and Empirical Work. In: E.S. Mills and P. Cheshire (eds.) Handbook of Regional and Urban Economics, Volume 3. Elsevier, Amsterdam: 1375-1412.

Modelling the Economy, Transport and Environment Triangle

227

Wilson, A.G. (1998). Land-use/Transport Interaction Models, Past and Future. Journal of Transport Economics and Policy, 32: 3-26. Wilting, H. (1996). An Energy Perspective on Economic Activities. University of Groningen, PhD, Groningen.

12 A Systems Approach to Modelling the Regional Economic Effects of Road Pricing Bjarne Madsen1, Chris Jensen-Butler2±, Jacob Kronbak3, Steen Leleur4 1

The Centre for Regional and Tourism Research, Denmark Department of Economics, University of St. Andrews, Scotland 3 Department of Maritime Research and Innovation, University of Southern Denmark 4 Centre for Traffic and Transport, Technical University of Denmark 2

Abstract In Denmark, there has been a substantial debate in recent years concerning the consequences of introducing road pricing. This chapter examines some of the regional economic consequences of the full implementation of a GIS-based road pricing system for all roads in Denmark. A Danish model system (MERGE) consisting of an Interregional General Equilibrium model, LINE, a national transport model (LTM model) and an environmental sub-model (TIC-MAP) are presented. LINE is used to make an initial analysis of the primary regional economic effects. In the first step, price changes as a result of road pricing are calculated. In the second step, changes in regional competitiveness affecting demand are calculated. In the final step, revenue from road pricing is re-entered in order to ensure institutional balance and the effects on production, income and employment are calculated.

12.1 Introduction As road traffic grows rapidly in Europe, the problems which it creates become increasingly serious. Congestion is a major problem in many European countries, largely, though certainly not exclusively, an urban phenomenon. The costs of congestion are considerable, perhaps in the magnitude of 2% of GDP (GomezIbañez 1997, Mayeres & Van Dender 2001). Environmental damage, both local and global, arising from transport-related emissions also represents a substantial and increasing cost, with potentially extremely costly though uncertain consequences. Both problems involve externalities and raise issues concerning the extent to which external costs can and should be internalised (Rothengatter 2000). A further problem is the cost of transport infrastructure provision, which is considerable, whilst its net benefits are difficult to calculate in the presence of externalities (Jansson 2000a). These issues are of course linked, as taxation of transport can both reduce some of the negative externalities, creating a welfare improvement, ±

After this chapter was written Chris Jensen-Butler has died.

230

Bjarne Madsen et al.

and it can also provide revenue for the construction of more transport infrastructure or a subsidy for public transport. Furthermore, revenues from taxation on transport can be used to reduce taxation on earned income, giving rise to a possible double dividend effect (Pearce 1991). Parry & Bento (2001) have recently claimed that recycling tax revenues into transport sector subsidy is markedly less efficient than using the tax revenues to reduce income tax. There is growing interest in road pricing as a policy instrument directed at the solution of congestion problems, seen as one of a set of alternative measures (Button 1998, De Borger et al. 2001a) as well as being a measure which simultaneously addresses other problems. Road pricing can also be used to internalise environmental damage costs (Johansson-Stenman & Sterner 1998) and there are a number of important issues linking congestion pricing to road and transport infrastructure investment (McDonald 1995, Hau 1998). Road pricing raises a range of questions, both theoretical and practical. The welfare implications of road pricing and the potential distortionary effects of taxation are the subject of discussion (see, for example, Arnott et al. 1994, Jansson 2000b, Hau 1998). There are four components of welfare in the transport market: consumer and producer surplus, tax revenues and external costs. Unlike many other markets, time costs must be taken into account when dealing with consumer and producer surpluses. In addition, different surpluses and costs fall typically on different groups. Furthermore, different transport sub-markets are interdependent. For example, reduced congestion through road pricing will increase consumer surplus for users of road-based public transport. Alternatively, improvement of rapid transit and metro systems will confer benefits on car users, through reduced congestion. The welfare gains of road pricing (Proost & Van Dender 1998) seem to be superior to most other forms of regulation of congestion. However, efficiency issues are inextricably linked to the difficult problem of evaluation of social marginal costs. Practical problems are related to the technology to be employed in managing the system, particularly at the user interface, and to the problem of traffic diversion (McDonald 1995). These problems are, in turn, closely related to the recurrent theme of political and social acceptability (Jones 1998, Rietveld & Verhoef 1998, Larsen & Ostmoe 2001). The European Union and many individual countries are moving towards road pricing as a key element in future transport policy (De Borger et al. 2001b). In Germany, road pricing for lorries on motorways using a GPS-based system was planned to be implemented in 2003, but it was delayed due to start-up problems of a technical nature. It is to be expected that a number of European countries will follow and that it will be expanded to cover all types of roads and possibly in the longer term, all types of vehicles. The European PROGRESS project has established experiments and demonstrations in road pricing systems in eight different European cities, including Copenhagen (Nielsen 2003, Herslund 2003 on the Danish AKTA sub-project). This increased political interest in road pricing is fuelling research interest in the field. A number of studies have attempted to assess the effects of road pricing on congestion, the environment, tax revenues and welfare in concrete geographical contexts (see De Borger & Proost 2001 for a number of case studies). The studies are usually ex ante, as only a few projects

A Systems Approach to Modelling Regional Economic Effects

231

have actually been implemented and these are limited in extent and impact. Amongst others, Proost & Van Dender (2001b) have developed an urban model based upon a transport market equilibrium derived from the maximisation of a social welfare function which includes the four components named above, and which permits assessment of the effects of different policies on transport supply and demand. Most of these studies treat transport as an independent commodity, while in reality it is derived demand. The spatial configuration of the economy and the development of spatial patterns of economic activity play a substantial role in the pattern and growth of transport activity. This means that road pricing will have effects both on the level and distribution of economic activity and, in turn, patterns of regional economic development will affect both the need for, and revenues from, road pricing. There are both efficiency and distributional issues when examining the relationship between road pricing and regional economic growth. The relationship between transport infrastructure and regional economic development has been examined in a number of studies (see, for example, Rietveld & Nijkamp 2000, Jensen-Butler & Madsen 1999). Equity issues, directly related to the distribution of economic activity and population in space, have been the subject of some studies (for example, Richardson & Bae 1998). However, there have been few attempts (Madsen & Jensen-Butler 2001 2002c) to include road pricing within a regional or interregional economic model and even fewer in the context of local economic modelling. Eliasson & Mattsson (2001) present one of the few attempts to relate road pricing to location and transport flows, using a simulation model for a city. They conclude that impacts on location seem to be modest, compared with impacts on traffic. In an empirical analysis using an interregional economic model and Danish data, Jensen-Butler & Madsen (1996) find evidence to suggest that environmental gains from distance-related taxation of transport far outweigh income loss. The present paper is a contribution to the study of the regional economic effects of the introduction of road pricing, using Denmark as an empirical example. The two-stage approach adopted is to model the effects of road pricing on regional economies. The paper covers only the first stage, where a local economic model is used to forecast changes in commodity prices and household income, together with the effects on demand, as a consequence of the introduction of road pricing. At this stage of the modelling process, it is assumed that the underlying behavioural relationships are constant. The second stage, which is not dealt with in the present paper, is to incorporate changes in behavioural relations, transforming the modelling approach in the first stage to a Computable Interregional General Equilibrium model, permitting a more complete analysis of changes in prices on economic activity in a regional system.

232

Bjarne Madsen et al.

12.2 A Systems Approach to Regional and Sub-Regional Economic Modelling At different levels, international, national and regional, complex problems of analysis and planning of the relationships between specific sectors and the broader economy are emerging. One example of this complexity is the relationship between the transport sector, the broader economy and the environment. Another example is the relationship between agriculture, the economy and the environment. There are two fundamental issues to be examined when considering the nature of the relationship between a specific sector and the broader economy: i) data and accounting principles and ii) modelling approaches.

12.2.1 Data and Accounting Principles Statistics and accounting aim to provide a picture of the complex system. At this level, either the complex system can be represented through a system of national/ regional accounts together with satellite accounts for specific sectors, or it can be represented by sector-specific statistics which have been extended to include the broader economy. Principles for setting up national/regional accounts together with satellite accounts for specific sectors are provided by the United Nations (1993) and the European Union (Eurostat 1996). Similar activities are to be found at the sectoral level. The main advantage of the economic system approach is uniformity in statistical framework and, therefore, they are comparable and consistent with constraints. However, the basic units of accounting are value, rather than physical units, such as transport volumes and quantities of emissions. Description of physical units, where this appears, is highly aggregated. The main advantage of sector-specific approach is the level of detail with respect to physical components, whereas the value-based information in these accounts is more ad hoc, less consistent and highly aggregated. At the regional and sub-regional levels, the statistical material available is much less developed than is the case for the national level. The statistics relating to the broader economy are based upon local social accounting matrices (SAMs) or local national accounts. Satellite accounts do not exist at this spatial level, except for tourism, where they are being developed at the present time. Sector-based statistics exist in the form of detailed accounts for agriculture and in the form of ad hoc data bases on transport flows in the case of the transport sector.

12.2.2 Modelling Principles In terms of modelling, both in relation to the broader economy and individual sectors, several trends seem to be emerging. First, there is a growing interest in the interaction between the broader economy and specific sectors seen from the perspective of the broader economy. Second, modelling the interactions between

A Systems Approach to Modelling Regional Economic Effects

233

specific sectors, such as agriculture and transport, and the broader economy, seen from the sectoral perspective, is developing. Broadly speaking, there are therefore two alternative modelling approaches. Economic system modelling involves modelling the economy as the core question, whilst sectors are treated as additions, usually as a front end or back end sub-model, and sometimes as an integrated element in the core model. In Denmark, examples are provided by models such as ADAM (Dam 1995) and AAGE (Frandsen et al. 1995) and in relation to transport see Fosgerau & Kveiborg (2003). Sector-specific modelling on the other hand has as a point of departure, a partial or free-standing model for the sector in question. This model is then usually extended to capture interactions with the broader economy. In Denmark, examples are provided by (Stryg 1992, Hansen 2001 and Madsen 1999). At the regional and sub-regional levels, the same two issues arise but in an even more complex form. Stand-Alone Approaches In relation to modelling activity, a number of analyses of the broader economy have been undertaken. Starting with straightforward single region Keynesian inputoutput models (Groes 1982, Madsen 1992a), interregional input-output models were developed (Holm 1984, Madsen 1992b) followed by interregional General Equilibrium models (Madsen et al. 2001). Regional sector-specific models have been mainly concentrated in the agricultural and transport sectors. For agriculture, the approaches adopted have included input-output models (Pedersen 1986), linear programming models (Stryg et al. 1991) and econometric models (Jensen et al. 2001, Hansen 2001). For the transport sector, a number of models have been developed (for an overview see: Madsen 1999). There is a national transport model as well as transport models which have been developed to analyse specific infrastructure investment projects, for example, the Femer Belt Link (Trafikministeriet 1999, Jensen-Butler & Madsen 1999) and the Øresund link (Øresundskonsortiet 1997) and upgrading or extending of existing rail networks, for example the Ringsted model (Nielsen 1998). A traditional approach and fundamentally stand-alone approach involves the identification of changes in transport flows and changes in direct and indirect (time) costs for travellers. This information is then used to undertake a cost-benefit analysis of the changes in the regulation of the transport system, involving an evaluation of changes in direct costs, time savings and changes in other costs, such as accidents and environmental costs. Normally, the diffusion of changes in the costs of the transport sector to commodity prices and incomes in other sectors is not dealt with. This means that the traditional cost-benefit approach cannot be used for analysis of distributional questions in relation to regions, factors, sectors and household types. In addition, as behavioural reactions from producers and consumers are not included explicitly in the cost-benefit approach, determination of the value of time for different categories of traveller is made exogenously rather than endogenously.

234

Bjarne Madsen et al.

Different Approaches to Model Integration Treating the general economy and the specific sector as two independent components is theoretically less than satisfactory, as development of the individual sector is partially dependent upon the general economic development and vice-versa. Therefore, there is a need to develop a more integrated approach to the problem. Four avenues in the development of integrated models at the regional and subregional levels are identifiable. ƒ Extension of sector-specific models ƒ Extension of regional and sub-regional models ƒ Loosely coupled models (framework approach) ƒ Fully integrated models The first can be illustrated by the extension of models specific to the transport sector with regional and macroeconomic elements. An important representative of this approach is use of the established growth factor model in forecasting total traffic flows. For passenger transport, attempts have been made to develop strategic models which include more developed modelling of interaction between the transport system and the broader regional economy including migration and commuting flows (Husted & Christensen 2001). The second avenue is represented by the extension of regional and subregional models of economic activity to include specific sectors, again typically transport. In the case of the Great Belt Link (Jensen-Butler & Madsen 1996), the analysis included not only the effects of the establishment of the Great Belt link on regional economic activity (employment and GDP by region), but also an analysis of the consequences for structural change in interregional trade patterns. In the case of the Femer Belt link (Jensen-Butler & Madsen 1999), a simple transport model was incorporated into the analysis to estimate the shifts in traffic flows by transport corridor in order to estimate the redistribution of regional economic activity. The third avenue is based upon a different modelling strategy, where economic models and sector models are linked together in a loosely coupled framework, consisting of separate and independent models linked together inside a general framework. Loose coupling implies that the models are independent but that output from one model constitutes a data input to another. This strategy has developed most strongly in the agricultural sector. In an initial attempt to evaluate the regional economic consequences of restructuring the European Union’s agricultural policy in 1991 (the McSharry proposal), the results emerging from an agricultural sector model (Stryg et al. 1992), consisting of changes in gross value added (GVA), other taxes linked to production, employment together with estimates of impacts on the food industry, all by region, were taken as inputs to the single region model EMIL (Madsen 1991a). This contrasts with an integrated model, where the different sub-models are solved simultaneously, constituting a fourth avenue. Here, one single model user has an overview of the entire system.

A Systems Approach to Modelling Regional Economic Effects

235

12.2.3 Optimal Model Structure The optimal model structure depends on both the nature of the object of analysis and the analytical capacity of the model users. If the focus is on a specific sector, then the choice of a specific sector model (a partial model) should be made. Here, it is assumed that the key parameters required for the operation of the model are supplied unambiguously (for example, the inputs of a growth factor model to a transport model). Alternatively, if it is assumed that there are strong interactions between the specific sector and the general economy, then it is necessary to integrate the models. The two basic forms of integration can, as mentioned above, be described as loosely coupled models or fully integrated models. In the case of transport and agriculture, loose coupling is more appropriate here, whilst in the case of the public sector and of tourism an integrated approach is more relevant. This is because the transport and agriculture sectors are complex, requiring a special modelling approach, which makes it necessary to build up specific models for each of the two sectors reflecting the complexity of the technical relations involved. Analysis of tourism and the public sector involves lower levels of technical specification, being more directly based on the conventional theory and models of consumption and production. An important requirement for the loose coupling of models is that they rest on the same or a very similar theoretical foundation. In addition, there has to be consistency in exogenously given assumptions and parameters. Also, practical considerations related to the fact that single research groups only cover specific areas and model sets promotes the use of loose coupling.

12.2.4 A Loosely Coupled Model for Transport and Agriculture On the basis of the above discussion, a combination of a fully integrated and a loosely coupled modelling system has been developed and used in a number of studies of the interaction between developments in specific sectors and regional economies in Denmark, also including interactions between international, national and local levels. The model system is presented in Fig. 12.1. The horizontal dimension shows in the centre general (equilibrium) models for the entire economy whilst to the right and left at this level the specific sectoral models (for agriculture and transport) are shown, which are loosely coupled together with the general models. Tourism and the public sector are, however, fully integrated models to be found in the centre of the figure. The vertical dimension shows the different levels of spatial resolution from the international level to the sub-municipality level. Each box represents an independent model and the arrows between the boxes represent flows of data (results) between the models. As can be seen from the figure, the total model system has been developed on a top-down principle, starting with a model of the international economy, GTAP, (Bach et al. 2000) to estimate the equilibrium values of central variables in the international economy, for example gross output, disposable income, aggregate

236

Bjarne Madsen et al.

demand, imports and exports. On this basis, the model for the national economy, AAGE, (Frandsen et al. 1995) determines equilibrium values for the Danish economy, including national values for the agricultural sector. The links between the economic model and the agricultural model and between the international, national and local level models are documented in Hasler et al. (2002). The link between the economic model and the transport model is documented in Kronbak & Leleur (2003) and in this paper.

Fig. 12.1. Linking Danish models of different types and levels

A Systems Approach to Modelling Regional Economic Effects

237

The economic model, LINE, is linked to the models for the transport sector by an interaction model, MERGE (Kronbak 2002). MERGE not only links the models in the general economic sector with the models in the transport sector, but also links the models within the transport sector together. The models used in the transport sector are the environmental impact model TicMap (Wass-Nielsen & Hviid Steen 2001), the transport model TSM and the assessment model COSIMA (Leleur 2002).

12.3 The Linking Procedures The linking procedures are described in a bottom-up way where first the linking of models within the transport sector is dealt with and second the linking of the transport and the regional economic models. By giving some detailed information about the individual model elements, the potential of comprehensive modelling on the basis of the suggested systems approach – also for other types of examination than road pricing – is more easily apprehended.

12.3.1 Linking the Models Within the Transport Sector The models within the transport sector are connected using a geographical information system as linkage. This GIS-based linkage has been named MERGE (Model for Exchanging Regionalised Geographic Entities). MERGE The keyword in MERGE is model integration. Within the transport sector, MERGE has to link the transport, environmental impact and assessment models together into a decision-making tool by making procedures for transferring input and output data between the models. Besides that, MERGE also has to make the linkage between the models in the transport sector and the interregional economic model (LINE). For each type of model within the transport sector, an existing model has been chosen for the decision-making tool: ƒ The Transport Sketch Model, TSM, developed at the Centre for Traffic and Transport, Technical University of Denmark. ƒ The environmental and economic impact assessment tool TicMap, developed at the Centre for Traffic and Transport, Technical University of Denmark. ƒ The Composite Model for Assessment (COSIMA) developed at the Centre for Traffic and Transport, Technical University of Denmark. The integration of these models has addressed some questions that are not only relevant for the specific models in question but it also has a more general application. This means that some of the procedures developed in MERGE are quite universal for loosely coupled model integration. As for most loosely coupled models,

238

Bjarne Madsen et al.

the starting point for MERGE has been a more general approach to model integration (including data transfer) and from there on to focus on the specific models. This approach has had the advantage that if/when other (or better) models become available, they can be utilised with less effort than if MERGE was specifically designed for the models mentioned in a fully integrated model. This can be referred to as model modularity and the principle is that other models can be “snapped” on to MERGE and used in coherence with the total modelling framework. Besides the modularity of MERGE, another important factor in the loosely coupled model integration has been consistency. One of the objectives of MERGE has been to provide data exchange between the transport, impact and assessment models (and also to the regional economic model). All these types of models rely to a certain extent on spatially distributed data but not necessarily on data with the same spatial distribution or level of aggregation. When integrating the models into a decision support system, it is important to ensure a common consistent basis so that results and conditions are identical or at least consistent to some extent. It is especially important to be able to reproduce results and datasets. An example is that if a model requires 50 zones, MERGE has to be able to generate this number of zones, preferably from any base dataset, under a number of different conditions, e.g. equal number of inhabitants within the zone; equal area of the zones, etc. At the same time, it has to be possible to keep track of where data originated from and to give some estimates of the accuracy of not only the original data, but also of the generated data. This is commonly known as metadata (or data on data). Generation of new datasets from an existing dataset is where GIS has proven to be a very powerful tool. Results and input data all have some kind of spatial attributes, e.g. population data can be on a municipal or a parish level. It is not necessary that all the integrated models actually use the spatial reference, but the spatial reference can be used in MERGE to generate and exchange datasets. Transport Sketch Model (TSM) The TSM is a more or less traditional 4-step traffic model. Some of the advantages of this model are that it is already fully integrated in a GIS (ArcInfo) and since it is conceptually simple, it runs fast – even on large networks. Although national and regional individual road transport are at the focus of the model, it also provides an assessment of transport by rail, bus and ferry in Denmark. Traffic Impact and Cost Mapping (TicMap) The TicMap model (Wass-Nielsen & Hviid Steen 2001) is a tool for traffic impact calculation. The basis of TicMap is four GIS-based impact models for: 1. Accidents 2. Noise 3. Emission 4. Severance and perceived risk Each of these impact models is based on Danish impact assessment models and can be run individually from within the geographical information system (MapInfo).

A Systems Approach to Modelling Regional Economic Effects

239

The results from the impact models are used to make an assessment of the impacts on each road segment in the network. For a closer description of the impact models, see for example Wass-Nielsen & Hviid Steen (2001), Clausen et al. (1991) or Leleur (2000). Composite Model for Assessment (COSIMA) The COSIMA model has been worked out to provide a more comprehensive assessment of transport initiatives than made possible by applying a conventional cost-benefit analysis (CBA). Thereby, COSIMA deals with a mix of CBA effects and non-CBA effects. Typically, the non-CBA effects – when seen from a modelling viewpoint – are more difficult to handle compared to the CBA effects where handbook approaches (pricing and procedures) are available for many transport planning problems. In brief, one can refer to the CBA effects as effects where pricing manuals and procedures exist and to the non-CBA effects as multicriteria analysis (MCA) effects as this type of analysis, stemming from operations research, becomes relevant for the extension of the conventional CBA. The idea of COSIMA can be described briefly by the following seven steps as formulated for the assessment of a number of alternative by-pass projects for a Danish town currently in need of relieving the through traffic (Leleur 2001). 1.

2.

3.

4.

5.

The first task is to determine the CBA effects being relevant for the concrete appraisal study. In the by-pass example, following the Danish Road Directorate’s standard model, the effects are: travelling time, vehicle operating costs, accidents, maintenance costs, noise, air pollution and severance & perceived risk. The investment enters the analysis denominated as the construction costs. The next task is to determine the MCA effects of relevance. These may be measured either in some type of quantitative unit, an example could be changes in strategic mobility see Kronbak (1998), or by judgement using a +5, .., 0, .., -5 scale. Three possible MCA effects of relevance for many Danish road projects are: network accessibility, urban planning and landscape. With CBA and MCA effects laid down, the so-called “anchoring” part of the model formulation can take place. This means determining the importance of the MCA effects against the CBA effects and in-between each other. Several MCA techniques are relevant here: direct weights, pairwise comparison, swing weights, etc., see Leleur (2000). Criteria importance is denominated by weights on the individual criteria adding up to 100%. At this stage, a base case scenario is modelled and presented to the decisionmakers together with the assumed interesting assessment questions; these concern issues that may have a principal influence on the decisions to be made from the study. The involvement of the decision-makers may lead to revision of both the kind of MCA impacts included and their weights in the base case scenario. Part of this exchange with decision-makers is also to formulate suitable additional scenarios. Afterwards, COSIMA is run for all the scenarios and the assessment questions are scrutinised and related to possible sources of uncertainty. This “determin-

240

6.

7.

Bjarne Madsen et al.

istic run” of the model and the identification of “varying levels” of information give intermediate results which are examined in the following model step as concerns their “feasibility risks” which indicate that a result in the deterministic run may be associated with such uncertainty that precaution is needed. Next, the so-called “stochastic run” of COSIMA is undertaken. This in fact is a Monte Carlo simulation where parameters and data have been replaced with suitable probability distributions that can represent the actual information level. In the by-pass example used as background for this overview of steps in COSIMA, use is made of both distributions arrived at empirically by data fitting and distributions set on the basis of reasoning. One could refer to those as “objective” and “subjective” probability assessments. At this stage, the assessment questions are addressed on the basis of the model results and the assumptions in the background and a second exchange with the decision-makers is carried out. With the layout of the COSIMA model as a transparent tool box, two principal possibilities are now available. The study may simply end here if the decision-makers are confident about the model outcome, or the decision-makers may want to feed back into the process and re-address some of the previous model settings to shed light on some issues.

It should be noted that COSIMA is more or less tailored dependent on the concrete application. As should appear from the overview of methodological steps above, features for applying both scenarios and risk examinations are available. When incorporating COSIMA in the MERGE model software, the way the other three model categories are set for the actual planning problem will influence the possibilities for the assessment analyses to be carried out in the COSIMA module.

12.3.2 Linking the Transport and Regional Economic Models The role of transport costs and the transport system are integrated into LINE, though the integration is still incomplete. In LINE, supply and demand for transport are modelled in a full and consistent manner, while transport in physical terms has not yet been modelled. Transport costs and transport system changes feed into the present version of LINE in an ad hoc way and the effects of changes in the local and regional economy on traffic flows are not yet modelled explicitly. In order to illustrate the division of labour between transport models and regional economic models in a fully developed modelling framework, an idealised set of relationships between LINE and a standard sequential transport model can be seen in Fig. 12.2. The linked model is simultaneous as there are linkages in both directions. Spatial interaction forms the link from the transport model (transport costs) to the regional economic model (to the cost-price circle) and from the regional economic model (interregional interaction, such as trade or shopping) to the transport model (to the real circle).

Trip frequency [Trips/person]

Commuting Pattern [trips]

Modal split [rail/road/ air/sea]

Commuting flows [means of transport]

Trip frequency [Trips/person]

Trade pattern [ton]

Modal split [rail/road/ air/sea]

Trade flows [tonkm/means of transport]

Modal split

Assignment

Trade pattern (origin-destination)

Trip frequency

Commuting [persons]

Interregional and international trade [EUR]

Trip types (generation)

Shopping flows [means of transport]

Modal split [rail/road/ air/sea]

Shopping pattern [trips]

Trip frequency [Trips/EUR]

Shopping [EUR]

Tourist trip flow [means of transport]

Modal split [rail/road/ air/sea]

Tourist trip pattern [trips]

Trip frequency [Trips/EUR]

Tourist trips [EUR]

Transport Cost [EUR]

Demand for the transport commodity

A Systems Approach to Modelling Regional Economic Effects

Fig. 12.2 An idealised integrated model of transport and regional economic change

241

242

Bjarne Madsen et al.

Compared with a traditional free-standing regional economic model, determination of transport costs takes place in the transport model, which replaces exogenous estimates of transport costs. In relation to a traditional free-standing sequential transport model (see, for example, Wilson et al. 1969), the economic model has replaced the trip generation, attraction and distribution steps by an economic model for interaction (trade, commuting, etc.) and a model for trip frequency. Transport Inside LINE A number of features of the treatment of transport inside LINE are important. First, the model distinguishes between mobile and immobile commodities. Mobile commodities are transportable commodities whilst for immobile commodities, place of demand is by definition the same as place of production, including various forms of private and public service, for example, hairdressing and hospitals. For immobile commodities, the relation between demand and supply of commodities is direct as there is no interregional trade and, therefore, there are no transport costs. In the case of hairdressing, the problem of transport costs is related to shopping trips. This is also the case for a number of components of consumption, for example, services related to real estate. Second, different price concepts are used. Commercial margins and net commodity taxes enter into the full model and in relation to the transport commodity, which means that price depends on the level of net commodity taxes (for example, fuel taxation and road pricing). Third, in a detailed version of LINE, the transport sector can be subdivided into different transport sub-sectors, each having different productivity and employment levels.

12.4 LINE: the Full Model, a Graphical Presentation Here, a brief graphical presentation of LINE is made. The full model and its equations are described in detail in Madsen et al. (2001a). The data used in the model, together with the interregional SAM, are described in Madsen & JensenButler (2002b) and Madsen et al. (2001b). LINE is based upon two interrelated circles: a real Keynesian circuit and a dual cost-price circuit. Fig. 12.3 shows the general model structure, based upon the real circuit employed in LINE. The horizontal dimension is spatial: place of work (denoted R), place of residence (T) and place of demand (S). Production activity is related to place of work. Factor rewards and income to institutions are related to place of residence and demand for commodities is assigned to place of demand. The vertical dimension is more detailed and follows with its five-fold division the general structure of a SAM model. Production is related to activities; factor incomes are related to i) activities by sector ii) factors of production with labour by sex, age and education and iii) institutions: households; iv) demand for commodities is related to wants (aggregates of commodities or components of final demand and intermediate consumption); v) commodities, irrespective of use.

A Systems Approach to Modelling Regional Economic Effects

Place of production (R) Activities (Sectors) (E)

Gross output Interm. consumption GVA GDP at factor prices Earned income (RE)

Factors of Production (education, gender, age) (G)

Earned income Employment

(RG) Institutions (households, firms, public sector) (H)

Place of residence (T)

243

Place of demand (S)

Earned income Employment Unemployment Taxes and transfers Disposable income (TG) Earned income Taxes and transfers Disposable income (TH)

Demand components (W)

Local private consumption Residential consumption Tourist expenditure

(TW)

Intermediate consumption Local private consumption Tourist expenditure Public consumption Investments (SW)

Commodities (V)

Local production Exports to other municipalities Exports abroad (RV)

Local demand Imports from other municipalities Imports from abroad (SV)

Constant prices Current prices

Fig. 12.3. Simplified version of LINE: the real circuit

The real circuit corresponds to a straightforward Keynesian model and moves clockwise in Fig. 12.3. Starting in cell RE in the upper left corner, production generates factor incomes in basic prices, including the part of income used to pay commuting costs. This factor income is transformed from sectors (RE) to sex, age and educational groups (RG). Factor income is then transformed from place of

244

Bjarne Madsen et al.

production (RG) to place of residence (TG) through a commuting model. Employment follows the same path from sectors (RE) to sex, age and educational groups (RG) and further from place of production (RG) to place of residence (TG). Employment and unemployment are determined at place of residence (TG). Disposable income is calculated in a sub-model where taxes are deducted and transfer and other incomes are added. Disposable income is distributed from factors (TG) to households (TH). This is the basis for determination of private consumption in market prices, by place of residence (TW). Private consumption is assigned to place of demand (SW) using a shopping model. Private consumption, together with intermediate consumption, public consumption and investments constitute the total local demand for commodities (SV) in basic prices through a use matrix. In this transformation from market prices to basic prices (from SW to SV), commodity taxes and trade margins are subtracted. Local demand is met by imports from other regions and abroad in addition to local production. Through a trade model, exports to other regions and production for the region itself are determined (from SV to RV). Adding export abroad, gross output by commodity is determined. Through a reverse make matrix, the cycle returns to production by sector (from RV to RE). The stylised version of the model with the real circle illustrated, as well as the price concepts used, is shown in Fig. 12.3, where the price level of real circle variables (constant/current) is shown. Again using the stylised version of the model shown in Fig. 12.3, the anticlockwise cost/price circuit shown in Fig. 12.4 corresponds to the dual problem. In cell RE, sector basic prices (current prices) are determined by costs (intermediate consumption, value added and indirect taxes, net in relation to production). Through a make matrix, sector prices by sector are transformed into sector prices by commodity (from RE to RV). These are then transformed from place of production to place of demand (RV to SV) and further into market prices through inclusion of retailing and wholesaling costs and indirect taxes (from SV to SW). This transformation takes place using a reverse use matrix. Finally, private consumption is transformed from place of demand to place of residence in market prices (from SW to TW). Fig. 12.3 shows the structure of LINE in more detail.

A Systems Approach to Modelling Regional Economic Effects

Place of production (R) Activities (Sectors) (E)

Gross output Interm. consumption GVA GDP at factor prices Earned income (RE)

Factors of Production (education, gender, age) (G)

Earned income Employment

(RG)

Place of residence (T)

245

Place of demand (S) Intermediate consumption (SE)

Earned income Employment Unemployment Taxes and transfers Disposable income (TG)

Institutions (households, firms, public sector) (H)

Earned income Taxes and transfers Disposable income

Demand components (W)

Local private consumption Residential consumption Tourist expenditure

(TH)

(TW) Commodities (V)

Local production Exports to other municipalities Exports abroad

Intermediate consumption Local private consumption Tourist expenditure Public consumption Investments (SW) Local demand Imports from other municipalities Imports from abroad

(RV)

(SV)

Basic prices (exclusive transport costs) Market prices Basic prices (inclusive transport costs)

Fig. 12.4. Simplified version of LINE: the cost-price circle

12.4.1 The Dimensions of LINE In the standard version of LINE, the dimensions of the axes are normally the following: Sectors: 12 sectors aggregated from the 133 sectors of the national accounts. Factors: 7 age, 2 sex and 5 education groups. Households: 4 types, based upon household composition Needs: For private consumption and governmental individual consumption 13 components, aggregated from the 72 components in the detailed national accounts. For

246

Bjarne Madsen et al.

governmental consumption, 8 groups. For gross fixed capital formation, 10 components. Commodities: 20 commodities, aggregated from 131 commodities in the national accounts. Regions: 277 municipalities (the lowest level of spatial disaggregation). Regions are defined either as place of production, place of residence or as place of demand. It is possible to aggregate the (277) municipalities into any regional unit. Standard regions are the (16) counties and (45) labour-market districts. In the version of the model used here, 23 sectors are used, 27 commodities and an aggregation of factors and households has been made, so that they do not enter the model. It uses 16 counties (regions). org öteb To G

G

Bornholm Nordjyllands

Viborg Århus

F

Ringkøbing

Frederiksborg Vejle

E

Copenhagen Vestsjællands Roskilde

Ribe

Copenhagen M Frederiksberg M

To Malmø

A

B Fyns Storstrøms

D

To G erma ny

Sønderjyllands

C

Fig. 12.5. Danish regions (counties and two municipalities with county status, Copenhagen M and Frederiksberg M). Three fixed links: A: Oresund, to Malmo, Sweden, B: Great Belt, C: Femer Belt to Germany. Four ferry routes: D: Spodsbjerg-Taars, E: Odden-Aarhus, F: Odden-Ebeltoft, G: Frederikshavn-Gothenburg (Sweden). Other very local and international ferry routes are not shown

A Systems Approach to Modelling Regional Economic Effects

247

12.5 Road Pricing and Modelling its Impacts The design of road pricing systems relates to general issues concerning transport policy and technical constraints and possibilities, in concrete social and cultural contexts.

12.5.1 Road Pricing Five basic dimensions underlie all road pricing systems. 1) Whether the tariff is kilometre-dependent, time dependent or flat rate 2) Whether the tariff depends on the type of road to be used (motorway, main road, secondary road, etc.) 3) Whether the tariff depends upon type of area in which the road is used (for example, urban, rural or simply the national territory) 4) Whether the tariff depends on time of day (rush hour, daytime, evening, night) 5) Whether the tariff depends upon behaviour (changes in speed, for example) For example, an urban cordon is a flat-rate charge for entering a specific type of area (as in the case of London). A motorway toll is typically kilometre and road type dependent (as in the case of the proposed German motorway lorry charges). Choice of system depends partly upon the dimensions chosen and partly upon technological, administrative and political alternatives available. In Denmark, a pilot project (AKTA, see Nielsen 2000, 2003, 2004, Herslund 2003, Herslund et al. 2001) is at present being developed involving both kilometre and area dependent tariffs. In the future, more complex systems, which also take account of the road hierarchy and time of day, may be developed. Technically, it is based on GPS technology, permitting precise identification of the location of the vehicle and thereby its road use related in turn to toll level for the road. This level depends on road status in a road hierarchy, time of day and level of urbanisation. Cars are fitted with receivers which function as meters registering both current expenditure as the road is used, and cumulated expenditure for a given period. The meter is read at periodic intervals and the user is charged.

12.6 Results from the Danish Road Pricing Toll Study Application of LINE permits analysis of the consequences for the regional economy of any change in the transport system which affects the costs of transport. The change in transport costs examined in this case, the introduction of road pricing, has been described above. The analysis begins in the interaction components of the cost-price circle shown in Fig. 12.4. Starting with trade in commodities (RV to SV), the prices of commodities decline. Given the point at which the analysis commences, the presentation

248

Bjarne Madsen et al.

follows the two circles, where first the effects on commodity prices are derived, followed by the real effects on demand, production, employment and income. For other cases, the starting point depends on the nature of the initial shock.

12.6.1 Changes in Transport Costs Changes in transport costs are calculated using a) an interregional satellite account for transport used to determine levels of transport costs and b) exogenously given interregional transport costs, based on a digital road map, Vejnet DK, used to calculate changes in transport costs. The data in the interregional satellite accounts are estimated in four steps: a1) Taking the national make and use tables, national transport activity is determined by i) transport mode ii) subdivided by transport costs related to intermediate consumption (by sector) and to private consumption (by component) and iii) by external (transport firm based) and internal (own transport, within a non-transport producing firm or a household) costs. Six different modes of transport activity are used in the interregional satellite accounts, four for passenger transport and two for goods. Passenger transport is divided into car, rail, aeroplane and other and freight is divided into lorry and rail. a2) In the second step, national transport activity related to passenger transport is subdivided (using data from the National Travel Survey) by trip purpose: i) commuting ii) shopping, iii) tourism iv) business travel and v) recreation. a3) National transport activity is then divided by origin and destination using data on intra and inter-regional trade (freight and business trip transport activity) and interregional shopping, tourism and commuting (personal trip transport activity) a4) Regional transport activity is then corrected (using regionalised National Travel Survey data) to ensure that the data reflect regional transport activity by mode. Changes in interregional transport activity for car transportation are estimated using the digital road map Vejnet DK. Transport costs in Vejnet DK are based upon both time and distance where the generalised cost has been calculated as Time costs + Distance costs. Also included are costs (tickets, tolls) of travelling by ferry and using fixed links. In addition, costs are calculated both with and without road pricing. The calculations are based on assumptions shown in Table 12.1.

A Systems Approach to Modelling Regional Economic Effects

249

Table 12.1. Maximum speeds, distance and time costs, road pricing tariffs (DKK) (DKK 7.50 = approx. 1 Euro = approx. 1$US) Motorway Non-urban highway Urban Distance cost per kilometre Time cost per hour Road pricing – urban Road pricing – rural

Car Lorry 110 km/h 80 km/h 80 km/h 70 km/h 50 km/h or local restrictions taken Max 50 km/h or local restrictions from VejnetDK if under 50 km/h 1.82 DKK 2.60 DKK 0.75 0.60 0.30

2.78 DKK. -

The estimation of level of transport activity by region, mode, purpose and by type of consumption described above reflects a basic assumption that data on transport activity obtained from National (transport satellite) Accounts used in a top-down procedure are superior to data on transport activity obtained from different statistical sources used in a bottom-up approach. In this paper, only results based upon road pricing for private cars is presented. Table 12.2 shows the consequences of introducing road pricing. It is assumed that all ferry routes and fixed links will continue with unchanged ticket prices. The table shows that transport costs in general increase from 2% to 13%, outside the main cities of Copenhagen and Aarhus least in the interregional links where use of rural roads is important and/or where a significant part of the journey uses ferries. In Copenhagen and Aarhus, transport costs decline as road pricing results in a reduction of congestion, and thus, transport costs.

Table 12.2. Changes in total transport costs (%) for transport between Danish regions after road pricing for cars Cop FrC RoC VsC SsC BhC FyC SjC (%) Cop. C & Frb M Copenhagen & Frederiksberg M 0.91 0.96 1.02 1.01 1.06 1.07 1.00 1.05 1.07 Copenhagen C 0.96 1.13 1.13 1.13 1.13 1.12 1.00 1.08 1.09 Frederiksborg C 1.02 1.13 1.10 1.12 1.11 1.12 1.00 1.08 1.09 Roskilde C 1.01 1.13 1.12 1.08 1.13 1.11 1.00 1.07 1.09 Vestsjællands C 1.06 1.13 1.11 1.13 1.12 1.11 1.01 1.05 1.08 Storstrøms C 1.07 1.12 1.12 1.11 1.11 1.13 1.02 1.07 1.09 Bornholms C 1.00 1.00 1.00 1.00 1.01 1.02 1.09 1.02 1.03 Fyns C 1.05 1.08 1.08 1.07 1.05 1.07 1.02 1.12 1.12 Sønderjyllands C 1.07 1.09 1.09 1.09 1.08 1.09 1.03 1.12 1.15 Ribe C 1.07 1.09 1.09 1.09 1.08 1.09 1.03 1.12 1.12 Vejle C 1.07 1.09 1.08 1.08 1.07 1.08 1.03 1.12 1.12 Ringkøbing C 1.05 1.05 1.04 1.09 1.08 1.09 1.03 1.12 1.11 Århus C 1.01 1.02 1.02 1.02 1.00 1.03 1.01 1.10 1.10 Viborg C 1.04 1.05 1.05 1.05 1.04 1.06 1.02 1.11 1.11 Nordjyllands C 1.04 1.05 1.05 1.05 1.04 1.05 1.02 1.12 1.12 VjC

1.07 1.09 1.08 1.08 1.07 1.08 1.03 1.12 1.12 1.12 1.12 1.11 1.07 1.11 1.11

RbC

1.07 1.09 1.09 1.09 1.08 1.09 1.03 1.12 1.12 1.12 1.12 1.11 1.10 1.11 1.11

1.05 1.05 1.04 1.09 1.08 1.09 1.03 1.12 1.11 1.11 1.11 1.11 1.11 1.11 1.11

RkC

1.01 1.02 1.02 1.02 1.00 1.03 1.01 1.10 1.10 1.10 1.07 1.11 0.97 1.10 1.09

ÅrC

1.04 1.05 1.05 1.05 1.04 1.06 1.02 1.11 1.11 1.11 1.11 1.11 1.10 1.11 1.09

ViC

1.04 1.05 1.05 1.05 1.04 1.05 1.02 1.12 1.12 1.11 1.11 1.11 1.09 1.09 1.12

NjC

250 Bjarne Madsen et al.

A Systems Approach to Modelling Regional Economic Effects

251

12.6.2 Changes in Commodity Prices and Disposable Incomes Table 12.3, column 1, shows that prices for commodities produced domestically at the national level increase by 0.10%. However, prices fall in Greater Copenhagen, whilst they increase with increasing distance away from Copenhagen, except for Bornholm, which benefits from transport via Copenhagen. Likewise, price increases in Aarhus, the other main urban centre, are lower than elsewhere in the periphery. Import prices (column 2) are not really affected by road pricing, for imports on their way to the Danish market. This means that prices for the total demand (column 3) rise by a similar, though slightly smaller percentage and the regional distribution of price rises has the same pattern as for demand for domestic products. Column 4 shows that prices for intermediate consumption are similarly affected by the reduction in transport costs, related to the fact that only service commodities enter into the calculations. These changes affect prices of gross output (column 5) though their effect is smaller as intermediate consumption is a part of gross output. Column 6 shows that export prices relative to foreign prices increase by 0.12%, again least in Greater Copenhagen and most in the industrial and peripheral regions of Jutland. The regional pattern is similar to that in column 1 and for the same reasons. Column 7 shows private consumption by place of demand, and exhibits a similar structure to the previous columns in the table. However, when private consumption by place of residence is examined (column 8), significant changes appear because road pricing affects the cost of trips for shopping, trips to visit family and friends, cultural and recreational visits, as well as tourist trips. There are now substantial price increases in the outer areas of Greater Copenhagen and Aarhus. This is because of longer trips and higher road price tariffs.

0.10

Whole country

0.17

Viborg County

0.17

0.08

Aarhus County

0.06

0.20

Ringkøbing County

Outside the regions

0.25

Nordjyllands County

0.26

0.04

Bornholms County

Vejle County

0.21

Storstrøms County

Ribe County

0.18

Vestsjællands County

0.19

0.09

Roskilde County

0.24

0.12

Frederiksborg County

Sønderjyllands County

-0.04

Copenhagen County

Fyns County

-0.10

Copenhagen & Fr. berg Municip.

(1) Demand: domestic production (SV)

0.00

-0.01

-0.01

-0.01

0.00

-0.02

-0.02

-0.02

-0.01

-0.02

-0.01

-0.02

-0.02

-0.02

-0.03

-0.01

0.00

(2) Foreign import (RV)

0.08

0.04

0.14

0.14

0.07

0.16

0.20

0.21

0.20

0.16

0.03

0.18

0.15

0.07

0.09

-0.04

-0.08

(3) Demand (SV)

0.09

0.05

0.13

0.13

0.08

0.15

0.19

0.20

0.19

0.15

0.04

0.16

0.14

0.08

0.09

-0.02

-0.04

(4) Intermediate consumption (RE)

Table 12.3. Price changes for demand and supply by type, by geographical origin. Percentage changes

0.14

0.07

0.22

0.23

0.12

0.29

0.32

0.34

0.29

0.23

0.07

0.24

0.26

0.12

0.14

0.00

-0.07

(5) Gross output (RE)

0.12

0.01

0.11

0.11

0.16

-0.01

0.03

-0.04

0.10

0.08

0.16

0.05

-0.01

0.14

0.10

0.39

0.13

(6) Foreign export (RV)

0.07

-

0.11

0.10

0.05

0.11

0.14

0.16

0.15

0.12

0.03

0.12

0.11

0.06

0.07

-0.01

-0.04

(7) Private consumption Place of market place (SW)

0.44

-

0.62

0.59

-0.03

0.63

0.74

0.74

0.83

0.65

0.38

0.69

0.64

0.48

0.55

0.46

-0.17

(8) Private consumption Place of residence (TH)

252 Bjarne Madsen et al.

A Systems Approach to Modelling Regional Economic Effects

253

12.6.3 Changes in Demand, Production and Income In general, road pricing means that real private consumption declines markedly, as consumer prices increase. On the other hand, export declines less because road pricing for cars only affects export prices marginally. Total demand and total production are therefore affected to a lesser degree than is the case for private consumption. Table 12.4, column 1 shows the consequences for disposable income (in current prices) of increases in transport costs which affect commuting. As most commuting is towards Copenhagen, then the principal increases are to be found outside Copenhagen and Aarhus. The decrease in disposable income registered for Aarhus arises because transport costs in Aarhus decrease, as a result of reductions in congestion and the shorter commuting distances in Aarhus, as compared to Copenhagen. Column 2 shows how real disposable income changes if price reductions on commodities are also included. The overall reduction in real disposable income is almost 1%. The effect is almost identical to that for private consumption by place of residence (column 3). Here again, Copenhagen and Frederiksberg municipalities and Aarhus have small absolute gains, while all other regions have losses which become greater with increasing distance from Copenhagen. Column 4 shows the increase in private consumption by place of demand, reflecting the more even spread of the effects noted in column 3 through shopping trips. Column 5 shows the real effect on total demand, which is more than for private consumption. The impacts are negative for all regions, though these negative effects are lowest in the cities. This pattern is the same for changes in real production, intermediate consumption and Gross Value Added (see columns 9 and 10). Column 8 shows the effects on exports of both the price increases for exports noted in Table 12.4 and the regional commodity export composition, as different commodities have different price elasticities. Correspondingly, import from abroad is calculated as a function of changes in domestic prices, which rise, and import prices from abroad, which in general are unchanged (see column 6 in Table 12.4). LINE uses the national import and export price elasticities used in Statistics Denmark’s national macro-economic model, ADAM (Dam 1995).

12.6.4 Changes in Employment and Income Table 12.5 column 1 shows that employment declines by 9,764, (by place of production).

Table 12.4. Consequences for demand, production and income. Percentage changes. Disposable Real disPrivate Private Demand income conposable con(SV) (current sumption income sumption prices) Place of (TH) Place of (TH) residence demand (TW) (SW) (1) (2) (3) (4) (5) Copenhagen & Frederiksberg M -0.05 0.17 0.17 -0.06 -0.05 Copenenhagen C -0.21 -0.71 -0.71 -0.68 -0.23 Frederiksborg C -0.28 -1.24 -1.24 -1.20 -0.54 Roskilde C -0.27 -1.03 -1.03 -0.99 -0.50 Vestsjællands C -0.26 -1.40 -1.40 -1.37 -0.64 Storstrøms C -0.27 -1.50 -1.51 -1.47 -0.70 Bornholms C -0.20 -0.95 -0.96 -0.97 -0.41 Fyns C -0.25 -1.41 -1.41 -1.40 -0.62 Sønderjyllands C -0.34 -2.02 -2.03 -1.98 -0.77 Ribe C -0.26 -1.56 -1.57 -1.56 -0.67 Vejle C -0.28 -1.43 -1.43 -1.41 -0.61 Ringkøbing C -0.25 -1.48 -1.49 -1.48 -0.59 Aarhus C 0.01 0.05 0.05 -0.01 -0.08 Viborg C -0.27 -1.50 -1.51 -1.49 -0.59 Nordjyllands C -0.26 -1.41 -1.42 -1.39 -0.57 Outside the regions -0.04 Whole country -0.20 -0.98 -0.98 -0.98 -0.41 (7) -0.05 -0.25 -0.52 -0.49 -0.65 -0.70 -0.41 -0.63 -0.77 -0.69 -0.63 -0.62 -0.10 -0.61 -0.58 -0.05 -0.42

-0.03 -0.17 -0.56 -0.50 -0.60 -0.65 -0.36 -0.55 -0.61 -0.54 -0.50 -0.46 -0.03 -0.46 -0.46 -0.01 -0.33

Demand Domestic production (SV)

(6)

Foreign imports (SV)

-0.01 -0.15 -0.34 -0.35 -0.57 -0.48 -0.18 -0.43 -0.45 -0.60 -0.59 -0.50 -0.26 -0.42 -0.39 0.00 -0.33

(8)

Foreign export (RV)

-0.08 -0.23 -0.48 -0.46 -0.66 -0.66 -0.37 -0.59 -0.72 -0.71 -0.66 -0.63 -0.17 -0.58 -0.56 -0.16 -0.41

(9)

Gross output (RE)

-0.07 -0.25 -0.49 -0.47 -0.64 -0.69 -0.41 -0.61 -0.75 -0.67 -0.61 -0.59 -0.11 -0.58 -0.56 -0.13 -0.40

(10)

GDP at factor prices (RG)

254 Bjarne Madsen et al.

Copenhagen & Fr.berg M Copenhagen C Frederiksborg C Roskilde C Vestsjællands C Storstrøms C Bornholms C Fyns C Sønderjyllands C Ribe C Vejle C Ringkøbing C Aarhus C Viborg C Nordjyllands C Outside the regions Whole country

Pct -0.08 -0.21 -0.41 -0.39 -0.52 -0.55 -0.32 -0.50 -0.60 -0.56 -0.52 -0.51 -0.11 -0.47 -0.46 0.00 -0.35

Number -291 -787 -648 -376 -675 -598 -64 -1125 -755 -674 -954 -757 -347 -579 -1134 0 -9764

Pct -0.14 -0.19 -0.32 -0.30 -0.46 -0.50 -0.31 -0.49 -0.59 -0.55 -0.50 -0.50 -0.14 -0.46 -0.45 -0.35

Number -434 -604 -638 -389 -686 -623 -63 -1146 -753 -646 -916 -735 -449 -562 -1120 -9764

Table 12.5. Impacts on employment and disposable income (2) (1) Employment Employment at place of at place of residence production (TG) (RG)

-0.13 -0.39 -0.52 -0.48 -0.57 -0.61 -0.44 -0.56 -0.71 -0.58 -0.59 -0.56 -0.03 -0.57 -0.56 -0.43

(3) Earned Income (TG)

-0.09 -0.36 -0.48 -0.48 -0.52 -0.54 -0.41 -0.50 -0.65 -0.50 -0.53 -0.48 0.00 -0.52 -0.50 -0.38

0.33

(5) Taxes (TG)

0.13 0.18 0.39 0.43 0.39 0.33 0.17 0.33 0.49 0.55 0.52 0.70 0.13 0.46 0.39

(4) Income transfers (TG)

-0.05 -0.21 -0.28 -0.27 -0.26 -0.27 -0.20 -0.25 -0.34 -0.26 -0.28 -0.25 0.01 -0.27 -0.26 -0.20

(6) Disposable income (TG)

A Systems Approach to Modelling Regional Economic Effects 255

256

Bjarne Madsen et al.

The distribution of these employment decreases closely follows the distribution of increases in Gross Output from Table 12.4. Fyn and North Jutland have the biggest declines in absolute terms whilst Greater Copenhagen and Aarhus have, in relative terms, the smallest declines. By place of residence, differences are somewhat reduced. Column 3 shows the consequences for earned income (wages and salaries plus surplus on self employment) that decreases markedly in the areas which suffer from higher commuting costs. These losses are modified by decreases in tax payments, (column 5) because of the income effect. Changes in transfer incomes (column 4) further modify the pattern of income change. Increases in transfer incomes are, other things being equal, greatest in the regions where employment declines faster. Column 6 shows the net result of these changes on disposable income. Greater Copenhagen and Aarhus are almost neutral and declines are related to increasing distance from Copenhagen.

12.6.5 Recycling the Revenue from Road Pricing: a Balanced Budget The revenue arising from road pricing can be used for different purposes. The allocation of these revenues by activity will have different economic effects as well as different effects in terms of regional distribution. The following alternatives are examined: 1. Reduction of income tax: high rate taxation 2. Reduction of income tax: low rate taxation 3. Reduction of commodity taxes: VAT 4. Reduction of commodity taxes on fuel Table 12.6a shows the consequences of reducing tax income by the state, whilst Table 12.6b shows the combined effects of road pricing and reductions in taxation. Table 12.6a shows that disposable income for households increases most if income taxes are reduced and least if commodity taxes are reduced. This can be explained by the fact that reductions in commodity taxes in part are transferred abroad via exports and to other components of final demand. There are regional differences in the different tax reduction strategies. Reduction of income taxes benefits Greater Copenhagen marginally more than the rest of the country. This is because the marginal tax rate in Greater Copenhagen lies above the national average because of the progression in the tax system. Alternatively, reduction of commodity taxes benefits the peripheral areas in Jutland because the marginal propensity to consume is higher in these areas. Regionally, reduction of income taxes at the lower bound provides the Greater Copenhagen area (except for the municipalities of Copenhagen and Frederiksberg) with a slightly greater increase in disposable income than reduction of upper bound taxes. This is probably related to higher levels of housing consumption, which in turn explains higher relative deductions from gross income, for tax purposes, as interest can only be deducted at the rate corresponding to the lower bound.

Table 12.6a. Consequences for demand, production and income. Effects arising from reductions in taxation, percentage changes. Real disposable income at place of residence (TG) Employment at place of production (RG) State income tax Commodity taxes State income tax Commodity taxes Lower Transport Lower Transport bound Top bound General commodity bound Top bound General commodity (1) (2) (3) (4) (5) (6) (7) (8) Copenhagen & Fr.berg M 1.19 1.14 0.67 0.67 0.30 0.29 0.17 0.19 Københavns C 1.25 1.17 0.67 0.67 0.34 0.33 0.17 0.23 1.22 1.18 Frederiksborg C 0.72 0.72 0.38 0.37 0.22 0.24 Roskilde C 1.18 1.17 0.72 0.72 0.43 0.43 0.21 0.27 Vestsjællands C 1.08 1.12 0.69 0.69 0.38 0.39 0.11 0.29 Storstrøms C 1.07 1.11 0.74 0.74 0.40 0.41 0.21 0.31 1.03 1.09 0.74 0.74 Bornholms C 0.35 0.36 0.02 0.34 Fyns C 1.09 1.12 0.69 0.69 0.36 0.37 0.14 0.29 Sønderjyllands C 1.06 1.08 0.75 0.75 0.31 0.32 0.16 0.35 Ribe C 1.07 1.09 0.70 0.70 0.33 0.34 0.04 0.32 1.09 1.11 0.72 Vejle C 0.72 0.33 0.34 0.08 0.29 Ringkøbing C 1.05 1.08 0.71 0.71 0.31 0.32 0.05 0.33 Aarhus C 1.13 1.14 0.74 0.74 0.36 0.37 0.17 0.28 Viborg C 1.06 1.10 0.74 0.74 0.32 0.33 0.08 0.32 1.06 1.11 0.74 Nordjyllands C 0.74 0.33 0.35 0.08 0.31 Outside the regions 0.00 0.00 0.00 0.00 Whole country 1.13 1.13 0.71 0.70 0.34 0.35 0.14 0.27

A Systems Approach to Modelling Regional Economic Effects 257

Copenhagen & Fr.berg M Københavns C Frederiksborg C Roskilde C Vestsjællands C Storstrøms C Bornholms C Fyns C Sønderjyllands C Ribe C Vejle C Ringkøbing C Aarhus C Viborg C Nordjyllands C Outside the regions Whole country

1.36 0.54 -0.02 0.15 -0.32 -0.44 0.08 -0.33 -0.96 -0.49 -0.33 -0.43 1.17 -0.44 -0.35 0.15

1.31 0.46 -0.06 0.14 -0.28 -0.40 0.14 -0.29 -0.93 -0.47 -0.31 -0.40 1.19 -0.41 -0.30 0.15

0.84 -0.05 -0.52 -0.31 -0.71 -0.76 -0.21 -0.72 -1.26 -0.86 -0.71 -0.78 0.78 -0.76 -0.67 -0.27

0.84 -0.07 -0.62 -0.38 -0.69 -0.75 -0.13 -0.69 -1.22 -0.83 -0.70 -0.74 0.73 -0.71 -0.64 0.28

Real disposable income at place of residence (TG) State income tax Commodity taxes Lower Transport Top bound General bound commodity (1) (2) (3) (4) 0.22 0.13 -0.03 0.04 -0.14 -0.15 0.03 -0.14 -0.29 -0.23 -0.18 -0.20 0.26 -0.15 -0.13 0.00 -0.01

0.21 0.12 -0.04 0.04 -0.13 -0.14 0.05 -0.13 -0.28 -0.22 -0.18 -0.19 0.27 -0.14 -0.11 0.00 -0.01 0.09 -0.04 -0.19 -0.18 -0.41 -0.34 -0.30 -0.36 -0.45 -0.52 -0.44 -0.46 0.06 -0.40 -0.38 0.00 -0.22

0.11 0.02 -0.17 -0.11 -0.23 -0.23 0.02 -0.21 -0.25 -0.25 -0.23 -0.18 0.17 -0.16 -0.15 0.00 -0.08

Employment at place of production (RG) State income tax Commodity taxes Lower Transport Top bound General bound commodity (5) (6) (7) (8)

Table 12.6b. Consequences for demand, production and income. Net effects (gross effects minus effects arising from tax reductions), percentage changes

258 Bjarne Madsen et al.

A Systems Approach to Modelling Regional Economic Effects

259

Table 12.6b shows the net effect of road pricing and tax recycling. The principal result is that combined road pricing and tax reduction gives a small increase in disposable income. This can be explained by the fact that even though there is a balanced budget, the expenditure on road pricing by the individual household is tax deductible. The other result is that recycling via commodity taxes results in a decrease in disposable income. This can again be explained by the fact that reductions in commodity taxes do not benefit domestic private consumption to the full extent of the reduction, as discussed above. Regionally, the net effects are similar to the gross effects because the regional differences in tax reductions are very limited.

12.7 Limitations of the Model and Future Development Strategies A central problem faced by input-output or more general demand-side approaches to regional economic analysis is the influence of supply side conditions on production, this being a primary concern of CGE approaches. Another set of problems relates to some of the central concerns of contemporary urban and regional analysis: the existence of imperfect competition; externalities; product variety and growth in productivity all of which in a modelling context usually involve non-linear functional forms. In its present form, LINE builds on linear relationships. This raises the question of the suitability of LINE to deal with such issues. In relation to the issue of supply-side conditions, as noted above, development of appropriate links between the real circle and the cost/price circle is the way forward. This road pricing study illustrates the first step, where links have been established between relevant prices and disposable income, foreign exports and imports. In addition, the public sector’s budget and finance problems have been considered. A future strategy involves development of similar links for example, in relation to labour participation and productivity, or relating changes in commodity prices to the commodity composition of demand. Adjustments to changes in trade balances can also be included. The basis for this relation is the establishment of equilibrium in both commodity and factor markets and a steady-state equilibrium for institutional balances. Even though this modelling strategy is similar to mainstream CGE approaches, the development of LINE nevertheless deviates from such CGE approaches to regional and interregional modelling. Both approaches are consistent with established economic theory and ensure equilibria with respect to markets and institutional balances. However, the mathematical complexity involved in the typical CGE approaches, which are often based upon non-linear functional forms, is attained at the expense of detail in the treatment of the regional and local economy. Thus, mainstream approaches, despite theoretical sophistication and complex non-linear mathematics, face severe problems related to lack of detail and a sound empirical foundation. They also frequently involve serious problems of calibration and solution.

260

Bjarne Madsen et al.

One way of dealing with these reservations is to set up a model development strategy which involves the stepwise inclusion of behavioural components requiring non-linear functional forms. As described above, LINE can be developed in different ways. One important direction is a further development of the links between the real circle and the cost-price circle in order to model productivity, wage determination and labour force participation rates in the labour market and substitution and income effects in relation to commodity markets (private consumption, export, etc.). Another important direction is to model in greater detail the consequences of a change in the institutional balances, for example, changes in household savings and consumption behaviour. These development strategies will transform LINE into a more advanced CGE model, as these essentially non-linear relationships are included in a step-bystep manner. It might appear that the two modelling approaches converge towards a similar solution, this being an ideal CGE model. However, in reality the model development strategy employed in LINE, basically involving a linear system, represents a fundamentally different strategy to that utilised by mainstream CGE modelling, where there are often non-linearities in the central model equations which are solved simultaneously. This places substantial constraints on the number of different actors, markets and level of detail which can be incorporated in the models. In LINE, the complexity of the regional and sub-regional economic systems, including the possibility of choosing an appropriate model configuration, is in focus, whereas the mathematical complexity of selected behavioural components will be introduced in a gradual manner.

12.8 Conclusion The requirements faced by model-builders today are very diverse, determined by the great variety of problems faced by regions and local authorities and which are reflected in the variety of studies which they generate. The ideal modern modelling tool must be able to respond to this diversity of demands placed upon it. The aim of the construction of LINE described above is to ensure a high degree of flexibility, achieved in the basic structure which involves an interregional SAM. On this basis, it is possible to ensure flexibility in the following areas: 1.

2.

The model can be structured as either a sub-regional model, a regional model, or a combined model, as it includes both commodity markets and tourism, usually treated as regional phenomena, and labour markets, commuting and shopping which are typical sub-regional phenomena. The overall model includes different types of agent which enter into a SAM, activities, factors, institutions, needs, commodities, which means that there is a degree of flexibility involving the option of including either all accounts or a choice of accounts, for example, a model covering activities or commodities (a local production environment model) or a model covering institutions and needs (an extended shopping model).

A Systems Approach to Modelling Regional Economic Effects

3.

4.

261

Use of three basic types of location. Place of production, place of residence and place of demand enter into the overall model. Choice of combination of type of locality permits the addressing of specific problems, such as a commuting model involving place of production and place of residence and a trade model involving place of production and place of demand. The overall model involves a real circle and a cost-price circle. This permits development of a fixed price model by using the real circle alone or a general equilibrium model based upon links between the real circle and the cost-price circle.

Furthermore, given the above-mentioned more strategic modelling choices, even more flexibility can be introduced by choice of aggregation of categories of activities, factors, institutions, etc. The paper describes LINE and its use in a number of studies. The configuration of the model and the aggregation of the SAM accounts used in the analysis of the regional economic effects of road pricing are presented together with a summary of empirical results from the analysis. The magnitudes and directions of change are all satisfactory and theoretically defensible. The modelling system set up and demonstrated on an application case concerning the introduction of road pricing may also be relevant for other cases. Thus, it may be highly relevant to model the consequences of large new transport infrastructure. Here, it would be of major interest to address regional economic effects together with other effects relating to changes in transport efficiency and quality. The latter are normally the main focus of such studies but planners and decision-makers could also benefit from the new possibilities in modelling set out with the suggested systems approach in this type of examination.

References Arnott, R., de Palma, A., Lindsey, R. (1994). The welfare effects of congestion tolls with heterogeneous commuters. Journal of Transport Economics and Policy, 28: 139-61. Bach, C.F., Frandsen, S.E., Jensen H.G. (2000). Agricultural and economic – wide effects of European enlargement: Modelling the Common Agricultural Policy. Journal of Agricultural Economics, 51, 2. Button, K. (1998). Road pricing and the alternatives for controlling traffic congestion, in: Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment, Edward Elgar, Cheltenham: 113-135 Clausen, F., Wätjen, W., Leleur, S. (1991). The Danish Highway Priority Model, EURET, Concerted Action 1.1. The European Union 1991. Dam, P.U. (ed.) (1995). ADAM – En model for dansk økonomi. Marts 1995. Statistics Denmark. De Borger, B., Peirson, J., Vickerman, R. (2001a). An overview of policy instruments, in: De Borger, B., Proost, S. (eds.), Reforming transport pricing in the European Union, Edward Elgar, Cheltenham: 37-50.

262

Bjarne Madsen et al.

De Borger, B., Peirson, J., Vickerman, R. (2001b). Policy reform packages, in: De Borger, B., Proost, S. (eds.), Reforming transport pricing in the European Union, Edward Elgar, Cheltenham: 51-61. De Borger, B., Proost, S. (eds.) (2001). Reforming transport pricing in the European Union. Edward Elgar, Cheltenham: 37-50. Eliasson, J., Mattson, L.-G. (2001). Transport and location effects of road pricing: A simulation approach. Journal of Transport Economics and Policy, 35, 3: 417-456. Eurostat (1996). European system of accounts ESA 1995. European Commission, Luxembourg. Fosgerau, M., Kveiborg, O. (2003). A review of some critical assumptions in the relationship between economic activity and freight transport. Paper presented at the TRIP Conference on the economic and environmental consequences of regulating traffic, Hillerød, Denmark, 2-3 February 2003. Frandsen, S.E., Hansen, J.V., Trier, P. (1995). GESMEC – En generel ligevægtsmodel for Danmark, Dokumentation og anvendelse. Det Økonomiske Råds Sekretariat, Copenhagen. Gomez-Ibañez, J.A. (1997). Estimating whether transport users pay their way: The state of the art, in: Greene, D.I., Jones, D.W., Delucchi, M.A. (eds.), The full costs and benefits of transportation, Springer Verlag, Berlin: 149-172. Groes, N. (1982). En regionalmodel for Sønderjylland. Working Paper 13, Institute for Border Region Research, Aabenraa. Hansen, L.G. (2001). Modelling the Effects of Complex Regulatory Constraints – the Case of Danish Nitrogen Regulation. SØM-publication no. 45, AKF Forlaget, Copenhagen. Hasler, B., Jensen, J.D., Madsen, B., Andersen, M., Huusom, H., Jacobsen, L.B. (2002). Scenarios for Rural Areas Development – an Integrated Modelling Approach. AKF Forlaget, Copenhagen. Hau, T.D. (1998). Congestion pricing and road investment, in Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment, Edward Elgar, Cheltenham: 39-78. Herslund M.B. (2003). Road pricing: user reactions with focus on consequences for traffic and congestion. Paper presented at the TRIP Conference on the economic and environmental consequences of regulating traffic, Hillerød, Denmark, 2-3 February 2003. Herslund, M.B., Ildensborg-Hansen, J., Jørgensen, L., Kildebogaard, J. (2001). FORTRIN programmet: et variabelt kørselsafgiftssystem – hovedrapport. Danmarks Tekniske Universitet, Center for Trafik og Transport, Rapport 2001-1, Lyngby. Husted Rich, J., Christensen, L. (2001). Altrans: Adfærdsmodel for persontrafik. Faglig Rapport fra Danmarks Miljøundersøgelser, 348, Copenhagen. Jansson, J.O. (2000a). Transport infrastructure: the investment problem, in: Polak, J.B., Heertje, A. (eds.), Analytical transport economics. Edward Elgar, Cheltenham: 141-171. Jansson, J.O. (2000b). Transport infrastructure: the problem of optimum use, in: Polak, J.B., Heertje, A. (eds.), Analytical transport economics. Edward Elgar, Cheltenham: 172-209.

A Systems Approach to Modelling Regional Economic Effects

263

Jensen, J.D., Andersen, M., Kristensen, K. (2001). ESMERALDA – A regional econometric sector model for Danish agriculture. Danish Institute of Agricultural and Fisheries Economics, Copenhagen. Jensen-Butler, C., Madsen, B. (1996). Modelling the Regional Effects of the Danish Great Belt Link. Papers in Regional Science, 75: 1-21. Jensen-Butler, C., Madsen, B.. (1999). An eclectic methodology for assessment of the regional economic effects of the Femern Belt link between Scandinavia and Germany. Regional Studies, 38, 8: 751-768. Johansson-Stenman, O., Sterner, T. (1998). What is the scope for environmental road pricing? In: Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment, Edward Elgar, Cheltenham: 150-170. Jones, P. (1998). Urban road pricing: public acceptability and barriers to implementation, in: Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment, Edward Elgar, Cheltenham: 263-284. Kronbak, J. (1998). Trafikplanlægning og GIS-baserede konsekvensberegninger (Traffic planning and GIS-based impact analysis). Ph.D. dissertation, Department of Planning, Technical University of Denmark. Kronbak, J. (2002). Data exchange between transport models, regional economic models and impact models, specifically designed to facilitate environmental impact evaluation. Paper presented at the Aalborg Traffic Days, Aalborg, Denmark, 26-27 August 2002. Kronbak, J., Leleur, S. (2003). The economic and environmental consequences of regulating traffic. Paper presented at the TRIP Conference on the economic and environmental consequences of regulating traffic, Hillerød, Denmark, 2-3 February 2003. Larsen, O.I., Ostmoe, K. (2001). The experience of urban toll cordons in Norway: Lessons for the future. Journal of Transport Economics and Policy, 35, 3: 457-472. Leleur, S. (2000). Road Infrastructure Planning – A Decision-Oriented Approach. Polyteknisk Forlag, Denmark, ISBN 87-502-0824-1. Leleur, S. (2001). Transport Infrastructure Planning: Modelling of Socio-Economic Feasibility Risks. Proceedings of the EUROSIM Congress, Delft, 2001. Madsen, B. (1992a). EMIL-modellen. AKF Forlaget, Copenhagen. Madsen, B. (1992b). AIDA-modellen. AKF Forlaget, Copenhagen. Madsen, B. (1999). Øresundsforbindelsens trafikale og mijlømæssige betydning. AKF Forlaget, Copenhagen. Madsen, B., Jensen-Butler, C. (2001). Modelling the local economic impacts of road pricing. SØM publication no. 46, AKF Forlaget, Copenhagen. Madsen, B., Jensen-Butler, C., Dam, P.U. (2001a). The LINE-model. AKF Forlaget, Copenhagen. Madsen, B., Jensen-Butler, C., Dam, P.U. (2001b). A Social Accounting Matrix for Danish Municipalities, SAM-K. AKF Forlaget, Copenhagen.

264

Bjarne Madsen et al.

Madsen, B., Jensen-Butler, C. (2002a). A model-based decomposition of income growth in Danish regions 1980-99, in: Norstrand, R., Andersen A.K., Indkomster, flytninger og uddannelse. AKF Forlaget, Copenhagen. Madsen, B., Jensen-Butler, C. (2002b). Regional economic modelling in Denmark: Construction of an interregional SAM with data at high levels of disaggregation, subject to national constraints, in: Hewings, G.J.D., Sonis, M., Boyce, D. (eds.), Trade, networks and hierarchies, Springer, Berlin: 445-456. Madsen, B., Jensen-Butler, C. (2002c). Theoretical and operational issues in sub-regional economic modelling, illustrated through the development and application of the LINE model. Paper presented at the 14th International Conference on Input-Output Techniques, Montreal, October 2002. Mayeres, I., van Dender, K. (2001). The external costs of transport, in: De Borger, B., Proost, S. (eds.), Reforming transport pricing in the European Union. Edward Elgar, Cheltenham: 135-169. McDonald, J. (1995). Urban highway congestion: An analysis of second-best tolls. Transportation, 22: 353-369. Nielsen, J.K. (2000). Road pricing – en prototype. ISP skriftserie nr. 253, University of Aalborg. Nielsen, O.A. (1998). En ny model for passagerers rutevalg i kollektiv trafik. Trafikdage, Aalborg Universitet, 1998: 137-156. Nielsen, O.A. (2003). The AKTA road pricing experiment in Copenhagen. Paper presented at the TRIP Conference on the economic and environmental consequences of regulating traffic, Hillerød, Denmark, 2-3 February 2003. Pearce, D. (1991). The role of carbon taxes in adjusting to global warming. Economic Journal, 101: 938-948. Pedersen, O.G. (1986). Landbruget i amterne. Jordbrugsøkonomiske Institut, Frederiksberg.

En

input-output

analyse.

Statens

Proost, S., van Dender, K. (1998). Variabilization of car taxes and externalities, in: Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment. Edward Elgar, Cheltenham: 136-149. Richardson, H.W., Bae, C.-H. C. (1998). The equity impacts of road congestion pricing, in: Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment. Edward Elgar, Cheltenham: 247-262. Rietveld, P., Nijkamp, P. (2000). Transport infrastructure and regional development. In: J.B. Polak and A. Heertje. Analytical transport economics. Edward Elgar, Cheltenham: 208-232. Rietveld, P., Verhoef, E. (1998). Social feasibility of policies to reduce externalities in transport, in: Button, K., Verhoef, E.T. (eds.), Road pricing, traffic congestion and the environment. Edward Elgar, Cheltenham: 285-308. Rothengatter, W. (2000). The external costs of transport. In: J.B. Polak and A. Heertje. Analytical transport economics. Edward Elgar, Cheltenham: 79-116.

A Systems Approach to Modelling Regional Economic Effects

265

Stryg, P.E., K.Å. Hornshøj, M.H. Knudsenand, Andersen, F. (1991). Fremtidsperspektiver for dansk landbrug. Papers from the Department of Economics, Royal Danish Agricultural University, Study no. 28. Stryg, P.E., Madsen, B., Olsen, P., Groes, N. (1992). Forslag, forlig og fremtid – regionaløkonomiske regnestykker for dansk landbrug i EF. AKF, Copenhagen. Telle, M.H., Hanghøj, J. (2002). Udenlandske investeringer i hovedstadregionen. Copenhagen Economics/Copenhagen Capacity. Trafikministeriet (1999). FemerBælt-forbindelsen. Department of Transport, Copenhagen. United Nations (1993). System of National Accounts 1993. Washington DC. Wass-Nielsen, M., Hviid, Steen, C. (2001). Development of a MapInfo tool for mapping and analysis of traffic impacts (in Danish). Centre for Traffic and Transport, Technical University of Denmark. Wilson, A.G., Hawkins, C., Hill, G.J., Wagon, J. (1969). Calibrating and testing the SELNEC transport model. Regional Studies, 2: 337-350. Øresundskonsortiet (1997). Øresund Traffic model: User manual – system developed by Comvin. Øresundskonsortium, Copenhagen.

13 External Effects and Road Charging Jeppe Rich and Otto Anker Nielsen Centre for Traffic and Transport, Technical University of Denmark

Abstract Negative external costs as a result of road traffic are an important issue in the process of developing new infrastructure and much effort has been concerned with how to measure, valuate and internalise the various external effects in order to establish a more efficient transport system. A flexible and increasingly popular instrument for internalisation is road charging, in order to internalise negative external effects and force a behavioural reaction towards a more efficient transport system. The aim of the present chapter is to analyse how, and to what extent, different road charging systems impact travel demand and derived external effects. The analysis is based on a recent study from Copenhagen and experience from the implemented toll-ring systems in Stockholm and London.

13.1 Introduction The mechanism behind road charging is that increased costs will cause car drivers who have the lowest willingness to pay to cancel or reduce their trip activity. As a result, there will be a reduction in demand, which then indirectly will reduce the negative external effects to the extent these effects are connected to transport demand. One problem in this respect, however, is that the different external effects are related to demand in different ways. For some effects, such as CO2, there will be an almost proportional relationship. For example, if mileage increases by 10%, then CO2 will increase by approximately the same per cent. For other effects, such as traffic noise, a more complex relationship exists. In the literature, there have been various studies on specific urban road charging experiments. However, very few studies have been comparable with respect to how the external effects are likely to change as a result of charging. One reason for this might arise from the common viewpoint that since cities are different in many respects, a quantitative comparison tends to be meaningless. We believe this is not the case. Comparisons between models and measurement methods are a completely integrated part of standard model building and validation. This exchange of ideas and best practice has forced most models to conform to certain standards and, in the case of models for external effects, relatively simple ones. As a result, it makes good sense to evaluate the behavioural reaction to road charging in different cities, to clarify whether people from different cities correspond to an expected response pattern, which in turn makes

268

J. Rich and O.A. Nielsen

them suitable for the standard traffic model apparatus. And if not, what could be the reason for this disparity? The best source for information is from the monitoring programmes in London and Stockholm (the latter during the trial period in the first half of 2006) found on the respective websites1. The philosophy of this chapter is to carefully relate the results of the two studies in order to verify similarities and differences as well as the causes of these. Moreover, we relate these observations to a recent Danish road-charging analysis described in Rich and Nielsen (2006) and Wrang et al. (2006). The London charging system is described in the reports from the monitoring programme (Transport for London, 2006) as well as in Santos and Fraser (2006) and Knorr and Eisenkopf (2006). See also Ison and Rye (2006) for a contribution with a more policy-oriented analysis of London, Cambridge and Singapore. The Stockholm trial has been fully reported in Stockholm Stad (2006). The chapter is organised in five sections. In section 13.2, we consider the nature of the various external effects. Section 13.3 is concerned with demand effects due to road charging in London, Stockholm and Copenhagen. The effects on derived external effects are analysed in section 13.4 and a summary and concluding remarks are offered in section 13.5.

13.2 External Effects: Background Usually, the external costs of traffic have been divided into four cost categories (Newberry, 1990): congestion; road damage costs; accident externalities and environmental costs. Environmental costs can be further sub-divided into traffic noise, climate effects and local emissions. Clearly, the importance of the different types of effect varies with the type of region under consideration. For instance, congestion is the dominating negative externality when considering larger cities, whereas the contribution when looking at countries as a whole is smaller. In a study by the Danish Environmental Protection Agency (DEPA, 2002), total external traffic costs for Denmark were estimated at 4.4 billion EUR. According to the study, more than 50% of all external effects were due to accidents, whereas climate effects and local emissions accounted for approximately 30% and noise for 8%. Congestion was estimated to account for less than 6%, which, however, seems to conflict with more recent studies2. In a congestion study from 2004 (Nielsen and Landex, 2004), congestion costs in the Copenhagen region alone were estimated at 0.8 billion EUR. However, although there are great uncertainties related to the measurement of externalities, international studies seem to indicate that among the various effects, accidents are the most important. Edlin and Mandic (2006) estimated the Pigovian tax for internalisation of accident 1 2

www.tfl.gov.uk and www.stockholmsforsoket.se

The calculation of congestion costs was based on a German speed-flow relationship from 1997, which makes it somewhat uncertain compared to more recent studies.

External Effects and Road Charging

269

externalities to $66 billion in California, which at present is more than the state taxes collected. The direct measurable costs due to accidents in EU-15 are estimated at 45 billion EUR by the European commission (EC White Paper, 2001). However, according to the Road Safety Action Programme (EC, 2003), indirect costs of accidents may be as high as 160 billion EUR. Given the enormous magnitude of traffic externalities and the fact that congestion increases exponentially as a function of demand 3 , more focus has been placed on how to internalise these costs. Road pricing is often mentioned as the instrument to achieve this, but not all types of externalities are perfectly dealt with in a road-charging system. Starting with congestion, road charging appears to be the optimum method of internalisation because a price mechanism replaces the present queuing mechanism which is allocatively inefficient. The fact that people change their driving behaviour (reduce demand for transport) in accordance with road pricing has been documented in the Danish AKTA experiment (Nielsen, 2004 and Chapter 6 in this book) as well as in numerous real-world charging systems in cities like London (Transport for London, 2006), Stockholm (Stockholm Stad, 2006) and Singapore (Keong, 2002). With respect to climate effects, road pricing is not the optimal instrument for internalisation of costs. Taxes on fuel, for instance, charge emissions in a more direct way and are simple to design. For local emissions, however, road pricing is the best alternative. Externalities related to accidents also seem to be well-suited to road charging, basically because there is a well-understood relationship between the number of accidents and the traffic on a given type of road (Hemdorff and Greibe, 2001). Clearly, there are great variations in the distribution of the severity of accidents for different types of road, which translate to different valuations of external costs for different types of road. However, this can be tackled by proper spatial price discrimination. An interesting finding is due to Parry (2003), who indicates that a uniform mileage tax can achieve only ¾ of the welfare gains compared to a tax differentiated on the driver’s age. For example, further price discrimination based on driver characteristics may be needed if the marginal costs due to accidents are to be internalised. Traffic noise has often been mentioned as a good example of an externality that might be dealt with by road pricing. This is due to the fact that noise pollution appears exactly where the transport activity takes place and depends on the nature of the area where the pollution occurs. However, recent experiments seem to indicate that the noise externality is much more complicated and that there is a complex relationship between traffic demand and noise when looked at from the perspective of an overall charging system. On the basis of empirical evidence from London, Stockholm and Copenhagen, it has been shown that some reduction in traffic noise can be expected as a reaction to road charging. However, it depends significantly on the road-charging system and the specific location. Even for roads with traffic reduction, noise may increase as a result of increased speed. Moreover, road charging may produce detouring along a toll ring, which can produce unexpected negative noise effects locally. A factor that might reduce the amount of 3

The BPR formula (Bureau of Public Roads, 1964) is the most commonly used speed-flow relationship.

270

J. Rich and O.A. Nielsen

traffic noise is if traffic can be redistributed to roads with better coverage of noise shields.

13.3 Demand Effects from Road Charging Road charging has become a popular instrument for the internalisation of external effects. This is due to the fact that the transport demand can be regulated in a flexible manner in the space-time domain. It is clear, however, that charging systems can be very different both in design as well as in their effect on transport.

13.3.1 Charging and Demand: Experience from London, Stockholm and Copenhagen Three toll rings from London, Stockholm and Copenhagen are analysed in the following. Although London is the only active system, the two other studies provide good insight. The Stockholm trial was a completely realistic full-scale installation running for the first six months of 2006 (Stockholms stad, 2006), whereas the Copenhagen study was a model simulation based on a large scale GPS experiment with volunteer 500 drivers (Nielsen, 2004; Rich and Nielsen 2006; Wrang et al. 2006). Table 13.1 below illustrates the charging levels for the three systems. Table 13.1. Charging levels for three toll-ring systems (measured in 2005 EUR) Day All London Stockholm Copenhagen

Peak

Night or weekend Near-peak

Off-peak

8 4.3 8

3.3

2.3

0 0

4

0

In addition to the charging levels shown in Table 13.1, the London and Stockholm schemes operate with a maximum daily payment of 8 and 6.7 EUR, respectively. For the Danish experiment, no maximum was used, and lorries were charged triple the price of cars in order to account for higher marginal costs. Obviously, the first thing to consider when comparing different systems is the charging level. As seen in Table 13.1, there are differences in the charging design, which makes it somewhat difficult to compare the three charging levels directly. However, it seems that the charging levels for London and Copenhagen are somewhat similar; whereas the level for Stockholm is significantly lower (4050%). If we look at how the charging scheme affected the transport demand in London and Stockholm, there is an almost similar response.

External Effects and Road Charging

271

Table 13.2. Demand effect for London and Stockholm City London Stockholm

Crossings 22% 22%

Mileage within charging area 17% 14%

If we compare Tables 13.1 and 13.2, it can be seen that even though the charging level in Stockholm is only half the price of London, the response level is similar. This indicates significantly higher demand elasticity in Stockholm. There could be several explanations for this. One explanation could be due to the difference in the level of congestion. Normally, we would assume that the more congestion in the pre-charge situation, the greater the effect of charging. This is true in terms of reduced congestion; however, in terms of reduced demand, it may be the other way round. Since the speed-flow relationship becomes steep, when approaching the capacity limit a marginal increase in car mileage will result in an increase far more than proportional to the mileage. Clearly, this is also true when reductions in mileage are made. In other words, if London is in a more critical state of congestion than Stockholm, the marginal effect of one charging unit will be higher in London, as illustrated in Fig. 13.1. Speed

Stockholm - after Stockholm - before London - after London - before

Flow

Flow

Flow

Fig. 13.1. The nonlinear speed-flow relationship

So, if drivers in a high congestion regime pay double charge compared to a low congestion regime, the corresponding reduction in congestion could outweigh costs. We would expect London to be more heavily congested than Stockholm, but existing observations do not provide unambiguous support for this. In fact, congestion is measured in different ways and with different degrees of detail, which makes a direct comparison difficult. Another reason could be due to the shape of the demand function. Certainly, it seems to indicate strict convexity in the sense that the marginal effect on demand

272

J. Rich and O.A. Nielsen

decreases with the cost. The fact that there is a large initial response could be due to a pool of people with a low value-of-time (students, pensioners, low-skilled workers), who are very sensitive to even small changes. However, for higher charging levels the marginal effect declines rapidly. This has also been found in the Danish AKTA study (Chapter 18 in this book). It conforms well to empirical findings, which indicate that value-of-time and income are log-normal distributed with long tail. A similar pattern has been found in practice in that the effect of the charge increase from 5 to 8£ in London in July 2005 (a 60% increase) “only” resulted in an additional 6% decline within the charging zone. It should be noted that other arguments could add to the differences, e.g. difference in terms of purchasing power parity, cost of registration of a car, as well as the state and structure of the public transport system. Moreover, it should be noted that Stockholm experienced a fuel increase of approximately 9% during April 2005, which is estimated to have reduced traffic across the toll ring by 3%. The experiment from Copenhagen, although it is not based on a real road pricing scheme, emerges with slightly lower results than the London and Stockholm schemes, with a 10% reduction in the central part of the city and a 5.4% reduction for the greater Copenhagen region. This points to a more general problem with standard traffic models, in that they tend to underestimate the effects of road charging. Presumably, part of the explanation for this is that the models are short-term based and seem to underestimate how quickly people respond to changes, as seen in Stockholm as well as in London.

13.4 Internalisation of External Effects Through Road Charging When politicians argue in favour of road charging, the most prominent argument is that a road-charging system will reduce external effects. For example, the city will become quieter, safer and less polluted and congestion will be history. The recent monitoring from London and Stockholm seems to indicate a more complicated picture, however.

13.4.1 Road Safety Accidents, one of the most important external effects when looked at from an economic perspective, were reduced significantly in all three cities. For London, the reduction was 18.8% between 2002 and 2004 within the charging zone, which appears to correspond well to the reduction in traffic. However, for the rest of London there has been an accident reduction of 15.7% due to broader road-safety initiatives. Assuming this trend to have continued, then within the charging zone and on the ring-road, casualties are only reduced by 4.3% as a result of charging. This amounts to approximately 100 casualties. For Stockholm, the observation period has been shorter and the results more uncertain. It is estimated that accidents have declined by 5-10% within the charging area as a result of charging. The

External Effects and Road Charging

273

experience from London suggests that traffic safety might be increased more efficiently by other means than road charging. In the Copenhagen model, the overall reduction in accidents has been 4% corresponding to a 5.4% reduction in car mileage. In the charging zone, the estimated reduction in accidents was 6-8% compared to an 8-10% reduction in traffic. This seems to correspond reasonably well with Stockholm. In fact, the Copenhagen case is interesting, because what partly causes the reduction is a change of the travel pattern from city streets to bypass roads, especially the ring motorway. The fact that cars make detours around the toll-ring boundary, on the bypass motorways, imposes a reduction in the number of accidents4. This was seen very clearly in a parallel experiment (Rich and Nielsen, 2006) of a small toll ring. In this experiment, car mileage increased by 0.2% compared to the base scenario, but with a decrease in accidents of 2.4%. In other words, the level of traffic demand matters, but so does the distribution on road types. Accidents (in London) involving cars have fallen the most compared to accidents involving two-wheelers, pedal cycles and pedestrians. However, there seems to be no evidence that accidents are more or less severe according to experience from London.

13.4.2 Traffic Noise Noise is an external effect, which is often used as an argument for road charging. It has been mentioned alongside other external effects in the Danish as well as the Swedish public debate. However, neither London nor Stockholm can identify significant changes in noise levels as a consequence of road charging, although there seems to be a slight overall reduction of 0-3dB. The problem with noise is a somewhat diffuse relationship to the level of transport. Less traffic may not result in less noise. What is often more important is the travel speed, which tends to increase due to road charging. In other words, road charging as a means to reduce traffic noise is not the ideal instrument. In the Copenhagen study, however, there is an estimated overall 2% reduction in noise exposure measured in terms of a noise annoyance index referred to as SBT (Danish Road Directorate, 1998). Compared to the overall traffic reduction in mileage of 5.4%, this seems a fairly large improvement compared with the findings of London and Stockholm. There are two explanations for this. Firstly, with more observations we believe London and Stockholm would be able to statistically verify a small change in noise level caused by road traffic. With a high price, even small changes of 1dB or less would be important in the valuation process. A 1dB reduction in the A-weighted continuous sound pressure (a 24-hour sound-pressure indicator) at the 60dB level will change the SBT by 10-20%. The second reason why noise reductions are found to be important in Copenhagen is due to the detouring pattern caused by the toll ring. Cars are redistributed from roads in the 4

Dickerson et al. (2000) report significant differences in the accident density across road types. The Danish standard is described in Hemdorff and Greibe (2001).

274

J. Rich and O.A. Nielsen

inner city to bypass motorways, which are all equipped with noise shields. Hence, in this redistribution the same positive effect observed for accidents occurs. As for accidents, this is evidenced in the small toll-ring experiment, where increasing mileage and decreasing noise exposure are experienced at the same time.

13.4.3 Emissions CO2 emission is closely related to travel demand. In London and Stockholm, a 16% and 13% reduction was observed, which is slightly less than the reduction in demand. However, CO2 also depends on the speed and thereby the congestion level (Beevers and Carslaw, 2005). In London, half of the reduction in CO2 is estimated to result from reduced congestion. For NOx, there has been an overall reduction of 13% in central London in 2003 and the reduction was 18% in 2005. However, recently (Transport for London, 2006) it is estimated that a significant part of this reduction is due to a general emission abatement programme and the fact that the car fleet is gradually becoming more environmentally friendly. An estimate of the true charging effect on NOx emissions is therefore only 8% and 7% for PM10 in 2003. In Stockholm, there has been a reduction in NOx by 8.5%, which is comparable to the results in London. For PM10 it is slightly higher in Stockholm with a 13% reduction. For Copenhagen, the reduction in CO2 and local emissions (NOx, PM10 and PM5) is estimated to between 1 and 3% for the whole Copenhagen region. This seems to correspond with the findings of the two other studies considering the lower demand effect in Copenhagen.

13.5 Summary and Conclusions The detailed monitoring of the road charging systems in Stockholm and London combined with model exercises for Copenhagen have been used to gather insight into how demand and external effects are affected by toll rings. This chapter considered the way demand has been affected. By comparing London and Stockholm, it is evident that the response pattern for Stockholm is more elastic per charging unit. This indicates all else being equal that there is a relatively stronger reaction to the first units of charging and a less significant reaction to a higher charging level. The relatively weak demand response when the London charge was raised from 5 to 8£ confirms this. The finding is in line with other previous studies and seems to indicate a relative large pool of “inexpensive” trips, which are more sensitive to charging. It was also found that the congestion level is likely to affect the demand and could be part of the explanation. The more congestion, the more relief that results from a marginal reduction in traffic. Hence, if London has more congestion relative to Stockholm, a smaller demand response in London is needed to bring about the same reduction in generalised travel costs.

External Effects and Road Charging

275

The impact on external effects indicates that accidents are reduced, but less than the reduction in traffic. In London, factors other than the charging were responsible for most of the reduction. This seems to indicate that other road-safety initiatives may be more important or that there is a positive self-perpetuating cycle between road charging and safety initiatives. Based on findings from London and Stockholm, traffic noise does not seem to change significantly. Clearly, this highlights the fact that traffic noise levels are often more sensitive to speed than to traffic load. In other words, if charging reduces traffic and speed therefore increases, the noise pressure may increase. However, in the case of London and Stockholm, the insignificance of noise could alone be due to a measurement problem. The Copenhagen model highlighted another issue related to noise as well as accidents, namely, that a toll ring may produce detouring around the charging boundary, which in turn will affect the way external effects are distributed. The fact that traffic was routed from city streets to bypass roads with better safety and a higher proportion of noise shields resulted in a positive effect. Emissions were reduced significantly in all studies, although less than the decrease in traffic. For London, approximately fifty per cent of the reduction was due to structural changes caused by new technology.

References Beevers, S.D., Carslaw, D.C. (2005). The impact of congestion charging on vehicle emissions in London. Atmospheric Environment, 39: 1-5. Bureau of Public Roads (1964). Traffic Assignment Manual, US Department of Commerce US, Urban Planning Division, Washington D.C. Danish Road Directorate (1998). Road Traffic and Noise – A textbook. Danish Road Directorate. Rapport no. 146. Accessible from www.vd.dk. Dickerson A., Peirson J., Vickerman, R. (2000). Road Accidents and Traffic Flow: An Econometric Investigation. Economica, 67: 101-121. DEPA (2002). Valuation of the external effects of transport. Environmental Project No. 734, Danish Environmental Protection Agency. EC (2003). Road Safety: Road Safety Action Programme (2003-2010). EC Report COM(2003) no. 311. EC White Paper (2001). European transport policy for 2010: Time to decide. EC Report COM(2001) no. 370. Edlin, A., Mandic, P.K. (2006). The Accident Externality from Driving. Journal of Political Economy, 114(5): 931-955. Hemdorff, S., Greibe, P. (2001). Håndbog i Trafiksikkerhedsberegninger. Danish Road Directorate, Rapport no. 220.

276

J. Rich and O.A. Nielsen

Ison, S, Rye, T. (2006). Implementing Road User Charge: The lessons learnt from Hong Kong, Cambridge and Central London. Transport Reviews, 25(4): 451-465. Keong, C.K. (2002). Road Pricing – Singapore’s Experience. Paper presented at the 3rd Seminar of the Imprint-Europe, 23-24 October 2002, Brussels. Knorr, A., Eisenkopf, A. (2006). Road-pricing in Practice – the London Congestion Charge. Paper Presented at the Transport Conference, Banff, Alberta, Canada, 2006. Newberry, D. (1990). Pricing and Congestion: Economic Principles Relevant to Road Pricing. Oxford Review of Economic Policy, 6(2): 22-38. Nielsen, O.A. (2004). Behavioural responses to pricing schemes: Description of the Danish AKTA experiment. Journal of Intelligent Transportation Systems, 8(4): 233-251. Nielsen, O.A., Landex, A. (2004). The Congestion Project – Modeling Congestion, CTT Report no. 2004-3 (In Danish). CTT/DTU, Accessible from www.ctt.dtu.dk. Parry, I.W.H (2003). Comparing Alternative Policies to Reduce Traffic Accidents. Discussion Paper 03-07, RFF, Washington. Rich, J.H., Nielsen, O.A. (2006). Road-charging in Copenhagen – Effects related to traffic. Danish Environmental Assessment Institute, ISBN: 87-7992-043-8 (In Danish). Santos, G., Fraser G. (2006). Road pricing: Lessons learned from London. Economic Policy, 21(46): 263-310. Stockholms stad (2006). Facts and Results from the Stockholm Trial. Final version, December 2006 (In Swedish). Accessible from www.stockholmsforsoket.se. Transport for London (2006). Central London Congestion Charging – Impact Monitoring, Fourth Annual Report. Accessible from www.tfl.gov.uk. Wrang K., Nielsen, U., Kohl, M. (2006). Road Charging in Copenhagen – A Cost-benefit analysis. Danish Environmental Assessment Institute, ISBN:87-7992-043-8 (In Danish).

14 Assessing the Impacts of Traffic Air Pollution on Human Exposure and Health Ole Hertel1, Steen Solvang Jensen1, Martin Hvidberg1, Matthias Ketzel1, Ruwim Berkowicz1, Finn Palmgren1, Peter Wåhlin1, Marianne Glasius1, Steffen Loft2, Peter Vinzents2, Ole Raaschou-Nielsen3, Mette Sørensen3 and Helle Bak3 1

National Environmental Research Institute, Roskilde, Denmark Institute of Public Health, University of Copenhagen, Panum Institute, Denmark 3 Institute of Cancer Epidemiology, Danish Cancer Society, Denmark 2

Abstract It is well known that exposure to air pollution can be linked to adverse health effects in the population. This has been demonstrated in various epidemiological studies. The associations determined between pollution exposure and health effects rely on the quality of the exposure assessment. Proper assessment of human exposure is therefore crucial for a correct determination of the association between the pollution load of the population and the negative health outcomes. Focus in this chapter is on assessment of the impact of traffic generated air pollution, which is the major source of human exposure in many countries including Denmark. Some of the methodologies for assessing exposure to traffic-induced air pollution are outlined and examples from Danish exposure studies are presented. To model air pollution concentrations in streets within the Danish exposure studies, a number of locally-developed models have been applied, including the regional scale model DEHM, the Urban background model UBM and the street pollution model OSPM. These models are briefly described here. However, one of the major difficulties in exposure modelling is to obtain proper input data for the calculations. In the first Danish exposure studies, these data were obtained manually or through use of questionnaires sent to local authorities. In recent years, a GIS-based tool, AirGIS, has been applied. AirGIS takes advantage of information from the unique Danish register databases that are available together with digital maps for building images and road network. Currently, AirGIS is being applied to exposure assessment in a number of Danish epidemiological studies.

14.1 Introduction Human exposure to ambient air pollution has been associated with severe health effects. The most classical example is the London smog (smoke and fog) episode in 1952, during which the mortality rate in the city increased dramatically (Wilkins, 1954). During this episode, the concentrations of particulate matter and sulphur dioxide in London were respectively 56 and 7 times above the normal level

278

O. Hertel et al.

at that time and the episode caused premature death of an estimated 12,000 people. However, ambient air pollution may cause adverse health effects even under much less extreme conditions. Studies of long-term exposure to air pollution (especially particles) suggest an increased risk of chronic respiratory illness and of developing various types of cancer, whereas higher prevalence of bronchitis, acute cardiovascular decease, asthma and other symptoms have been associated with shortterm exposure to increased air pollution concentrations during episodes. The population is generally most at risk in urban areas where many people spend a significant part of their time and where pollution levels often are high due to the high density of pollution sources and to the poor dispersion conditions because of the presence of obstacles such as buildings. Over the last decades, emissions from traffic have become the major local source of air pollution in many larger European cities. This development is the result of legislation concerning emissions from power plants and industries, together with a steady growth in road traffic. Furthermore, these releases take place in the streets with poor dispersion conditions and close to the population. Traffic exhaust fumes contain pollutants such as nitrogen oxides (NOX defined as the sum of nitrogen monoxide (NO) and nitrogen dioxide (NO2)), volatile organic compounds (VOCs), carbon monoxide (CO) and particulate matter (PM). In recent years, an increasing focus has been on determining the possible association between adverse health effects in the population and various traffic-related air pollutants. A significant part of this focus has been on particulate matter (Künzli et al., 2000).

14.2 Air Pollution in Urban Areas The urban environment constitutes the areas where the population is most at risk of negative health outcomes in relation to air pollution. Therefore, it is useful to describe the pollution conditions in urban areas. Air pollution in an urban environment is the result of local emissions as well as contributions from pollution transport from both nearby and remote sources outside the city. The distribution between contributions from different source types and source areas to a given site varies between pollutants. Furthermore, it varies in time as a function of variations in releases as well as meteorological parameters governing the dispersion conditions. Industries, power plants and other sources with releases from tall chimneys seldom contribute to enhanced concentrations at ground level in urban areas. The reason is that pollution from high sources is usually transported some distance before it is dispersed down to ground level. As a rule of thumb, the width of the pollutant plume grows as a result of the dispersion as one tenth of the transport distance. This means that for a 200m tall chimney, the pollution will reach the ground at a distance of about 2 km downwind of the source. Pollution from tall sources therefore contributes primarily to regional pollution.

Assessing the Impacts of Traffic Air Pollution

279

Vehicle transport, local domestic heating and small-scale industrial firms have low release heights, for which the pollution is not diluted as efficiently as for releases from tall sources. Furthermore, these releases take place close to the population and therefore contribute significantly to human air pollution exposures. The emissions from traffic follow a fixed pattern through out the day and the week. However, due to variations in meteorological conditions, this is not the case for the resulting pollution levels. In general, the dilution increases with wind speed, especially in urban areas where the highest concentrations appear at low wind speeds (below 2 m/s).

14.2.1 Urban Background and Street Pollution The urban background pollution is the result of regional pollution and contributions from the city itself. The size of the city domain and the emission density determine the urban area’s contribution to the pollution level. Air pollution in an urban street is the result of several contributions and to a great extent it is governed by the physical conditions around the street. The special airflow in streets and around buildings may result in very different concentrations at different locations in the street (Fig. 14.1). When the wind blows perpendicularly to the street, the pollution concentrations may be up to 10 times higher on the leeward side compared to the windward side.

wind

Background pollution

Recirculated pollution

Leeward

Direct emission

Windward

Source: Berkowicz (1998). Fig. 14.1. Schematic illustration of the flow and dispersion inside a street canyon. In this case wind above roof level is blowing perpendicularly to the street. Inside the street canyon, a vortex is created and the wind direction at street level is opposite to the wind above roof level. Pronounced differences in air pollution concentrations on the two pavements are the result of these flows.

280

O. Hertel et al.

Street canyons have generally higher pollution concentrations than more open street sections with similar traffic densities. Open streets are windier due to the lack of buildings which provide a shield for the wind. In the street canyon, pollution is circulated instead of being transported away from the street. The dispersion due to traffic-induced turbulence has been found to be very important for pollution levels at low wind speeds. The higher the driving speed, the quicker the pollution is mixed with “clean air” of the surroundings. This also means that the relationship between traffic intensity and air quality is non-linear.

14.3 Air Pollution Exposure Assessment Human exposure refers to an individual’s contact with a pollutant concentration and may be determined by application of direct as well as indirect methods (Zartarian et al., 1997). Direct methods are measurements made by personal portable monitors or measurements of biological markers. The personal monitors are carried by a study subject during a campaign period. Biomarkers may be measurements of pollutant concentrations or their degradation products in, for example, blood samples or urine, but they can also be measurements of parameters for the effect of pollution exposure, for example, in terms of oxidative damage of DNA determined in blood samples. In the application of indirect methods, exposure is determined by combining information about pollution concentrations with information about time spent in the specific environments. Information about pollutant concentrations can be obtained through measurements, application of transport-chemistry models or using combinations of measurements and models. In the use of the indirect method, it is common to apply the concept of microenvironments. A microenvironment is defined as a three-dimensional space where the pollution concentration at a specific time is spatially homogeneous or has wellknown statistical properties. Examples of microenvironments are the interior of a car, inside a house or urban, suburban and rural areas. This concept is particularly useful in the assessment of exposure for larger cohorts where general time-activity patterns may be obtained. Time-activity patterns are used to describe a person’s movement between the different microenvironments in time and the associated activity.

14.4 Direct Methods for Exposure Assessment Personal monitoring is a direct method for exposure assessment and is especially useful in connection with studies of smaller cohorts during shorter campaign periods. The reason is that such measurements are highly resource-demanding and therefore usually impossible to carry out for larger cohorts. However, data from

Assessing the Impacts of Traffic Air Pollution

281

such studies are crucial in the evaluation of the reliability of exposure models and also in testing the usefulness of the commonly used application of measurements from fixed site monitoring stations as a proxy of personal exposure. The exposure monitors may be divided into two groups of instruments: 1. 2.

Integrated samplers that over a specific period of time collect pollutants which afterwards are analysed in the laboratory. Continuous monitors that use a self-contained analytical system to measure pollutant concentrations on location.

Both types of instrument may be based on passive as well as active monitors. The active monitors use a pump with a portable power source in order to move sampled air through a collector or sensor, whereas the passive monitors use diffusion to bring the air pollutant into contact with the sensor or collector.

14.5 Indirect Methods for Exposure Assessment Indirect methods for exposure assessment seek to estimate personal exposure by combining information about concentrations at fixed sites with information about time spent in specific environments. A crude indirect method for exposure assessment is the categorical classification of the population, based on indicators such as type of residence (e.g. rural, urban or industrial), job classification (occupational exposure), presence of indoor sources like gas stoves, environmental tobacco smoking, etc. This method is, however, considered inadequate for application in air pollution epidemiology. The use of fixed site measurements is a more widely used method for obtaining an indicator for population exposure, based on routine monitoring. However, information about concentrations in the environments where people spend time may be obtained from monitoring as well as application of models.

14.5.1 Monitoring Networks For indirect estimation of exposure, monitoring networks with continuous sampling of ambient air pollution serve as important sources of information concerning pollution concentrations and their trends, mainly at fixed sites. Furthermore, data from monitoring networks are crucial in connection with the development and validation of air pollution models. However, monitoring networks are usually established with the aims of: x evaluating actual pollution levels against air quality standards, x providing warnings in connection with pollution episodes, x estimating contributions from various sources, x following trends in pollution concentrations, x in some cases also for conducting process studies.

282

O. Hertel et al.

Monitoring stations represent conditions at the local site, which may be considerably different from nearby sites. Careful consideration is therefore needed when these data are used in connection with exposure assessments. Urban monitoring networks usually consist of street stations as well as urban background stations. Urban background pollution is typically measured at roof level or in parks and backyards, where local sources are not allowed to dominate the measurements. Pollutant levels at street stations are dominated by emissions from traffic, but may also to some extent be influenced by emissions from local industry. Typical compounds included in the monitoring are nitrogen monoxide (NO), nitrogen oxides (NOx), carbon monoxide (CO), sulphur dioxide (SO2), soot and PM10 or TSP (total suspended particulate matter). We will return to these compounds in the following sections. Easily measurable indicator components may allow estimation of concentrations of other compounds for which monitoring is less straightforward or expensive. Therefore, even though a compound is not harmful in itself, it may turn out to be a useful compound in the monitoring programme, as an indicator for other more harmful compounds. An example is the use of black soot as an indicator of ultrafine particles.

14.5.2 Application of Models for Exposure Assessment Models for specific use in exposure assessment include transport-chemistry models that are based on the microenvironment approach and statistical models that relate exposure to selected and easily determined parameters. Stochastic components may, furthermore, be added to the physical models in order to reflect the variability, for example, in pollutant concentrations in a given microenvironment. In the following, we will focus on the physical models, since this type of model has been applied in the examples shown later in this chapter. The wide spectrum of atmospheric phenomena governing air pollution concentrations takes place on various temporal (from seconds to month and years) as well as spatial scales (from a few meters to thousands of km). When modelling human exposure, we are mainly interested in obtaining a good description of the pollution phenomena in urban areas. However, pollution concentrations in urban areas may arise from local sources as well as remote sources. Therefore, models for long-range transport, pollution dispersion within the urban area, as well as for single streets may be of interest in this context.

14.5.3 Modelling Long Range Transport of Pollution Air pollution is a transboundary problem where pollutants may be transported over thousands of kilometres. A wide range of different types of models (Kallos, 1998) is currently available for describing these pollution phenomena. Such models may have different mathematical expressions for describing the governing physical and

Assessing the Impacts of Traffic Air Pollution

283

chemical processes (emission, dispersion, transport, transformation and dry and wet deposition), but they all require proper input data for meteorology, emissions and land use for the selected model domain (see Fig. 14.2).

Source: Hertel et al. (2002). Fig. 14.2. Annual emissions of nitrogen oxides (Tonnes NO2/grid) in Europe in 1999 on a 16.67km x 16.67km grid net. Note the strong impacts of ship traffic and the impact of the road network in central Europe.

Integrated model systems, where a hierarchy of models is combined, are being developed at many institutes these years. The THOR system (Tilmes et al., 2002) is an example of such an integrated model system, which has been applied to the modelling of long-range transported pollution concentrations in the exposure assessments presented in the next sections. Fig. 14.3 shows ground level ozone concentrations modelled with the THOR system.

284

O. Hertel et al.

Fig. 14.3. Modelled ground level ozone concentrations (ppb) in Europe from the THOR integrated model system

14.5.4 Modelling Pollution in the Urban Background Several models have been developed for describing pollution levels in the urban background (see references in Mestayer (1998)). The dispersion conditions are strongly affected by the city structure, which may generate movements at scales as large as the whole city area itself. Within the Danish exposure studies, measurements from urban background stations have been applied whenever possible. However, where observations have not been available, a simple model for the urban background pollution (UBM; Berkowicz, 2000a) has been applied (Fig. 14.4). In UBM, the urban area is subdivided into a grid net of a resolution of 2km x 2km. Contributions from individual area sources are integrated along wind direction paths. It is assumed that the dispersion increases linearly with distance from the source. Horizontal dispersion is accounted for by averaging the calculated concentrations over a certain wind direction dependent sector, centred on the average wind direction.

Assessing the Impacts of Traffic Air Pollution

285

NO2 (μg/m3) 10 - 12 12 - 15 15 - 17 17 - 21 21 - 26 26 - 33

Fig. 14.4. Calculations of nitrogen dioxide using the simple UBM

14.5.5 Modelling Urban Street Pollution Since a significant part of the exposure of the population is caused by emissions from traffic in urban streets, street pollution models may serve as important tools in exposure assessment. Examples of these models are STREET, CPBM, OSPM and CAR (see references in Berkowicz, 1998). Here, we will concentrate on the OSPM model (Berkowicz, 2000b), which has been applied to exposure assessment in a number of studies discussed in the following sections. In the OSPM model, concentrations of exhaust gases are calculated hour by hour, combining a plume model for the direct impact of vehicle-emitted pollutants with a box model for computation of the additional impact due to pollutants recirculated within the street by the vortex flow (Fig. 14.1). OSPM makes use of a simplified parameterisation of flow and dispersion conditions in a street canyon, but has been found to reproduce the observed hourly mean concentrations of traffic pollutants in urban streets well.

14.5.6 Modelling Nitrogen Oxide Chemistry in Street and Urban Background Nitrogen oxides from traffic are mainly emitted as nitrogen monoxide (NO). The retention time is relatively short (few minutes) and only very fast reactions take place inside the street canyon. The distribution between the harmless NO and the airway irritant nitrogen dioxide (NO2) is therefore to a great extent determined by the fast reaction between NO and ozone (O3) and the similarly fast photo dissociation of NO2 back to NO and O3 (Equation 1).

286

O. Hertel et al.

NO  O3 o NO2  O2 NO2  hQ o NO  O

(1)

O  O2 o O3 The concentration of NO2 is thus strongly determined by the long range transported O3. Model calculations are in good agreement with observations, even with this simplification of the nitrogen oxide chemistry to include only three reactions (Palmgren et al., 1996).

14.5.7 Determination of Emissions and Emission Factors Emission data on an annual basis for Europe can be obtained from the European Monitoring and Evaluation Programme (EMEP) (www.emep.int). Temporal variations on an hourly, weekly and monthly basis are estimated based on information about the various source types. However, when focusing on modelling pollution from traffic in a specific street, such data are insufficient. Here, information concerning the traffic intensity and emission factors for the current car fleet is needed. Emission factors may be obtained from the COPERT traffic emission model, which is available at (http://vergina.eng.auth.gr/mech/lat/copert/copert.htm). Alternatively, an emission model can be built directly into the dispersion model for traffic pollution, as in the case of the windows-based version of the OSPM (http://ospm.dmu.dk). Often, emission factors for different vehicle categories are not known or the evolution in these factors may not be well-described, e.g. for particle number emissions or re-suspension of road dust. In these cases, inverse modelling is a strong tool. It involves a backward calculation procedure, using street pollution models, air quality measurements from fixed-site street and urban background monitors, meteorological data and traffic counts, to estimate the emission factors for the current car fleet (Palmgren et al., 1999). The emission factors obtained can be used for calculations in other streets for which monitoring data are not available. For particle numbers, no common emission database is yet available, but inverse modelling has been applied to estimate the average fleet emission factors for Danish conditions (Ketzel et al., 2003).

14.6 Particle Pollution Ambient air contains a complex mixture of particles with different sizes and chemical compositions. Size is crucial for effects, since it governs the particles atmospheric processes as well as their deposition in the human respiratory system. Particles typically appear in three rather distinct size classes (or modes) usually

Assessing the Impacts of Traffic Air Pollution

287

termed ultrafine (diameter: 0.01-0.1Pm or 10-100nm), fine (diameter: 0.l-2.5Pm) and coarse (diameter>2.5Pm) particles (Fig. 14.5).

Number Mass

Condensates & other primary particles (mainly traffic)

coarse Secondary particles (wear of tire & road long range transport

construction work natural sources

Soot (traffic)

0.001

0.01

0.1

Dust from breaks (traffic)

1

10

100 μm

Nano particles Ultrafine particles PM2.5 (fine particles) PM10

Fig. 14.5. Size distribution of particles in urban air shown in both mass and number concentrations. The horizontal axis is the particle diameter in Pm. The continuous line shows mass distribution dominated by the coarse and secondary particles. The broken line is the number distribution, dominated by ultrafine particles. Note that one particle with a diameter of 10Pm has the same weight as 1 billion particles with a diameter of 0.01Pm

Only the particle mass concentration of particles 0.1 μm in diameter. However, recent health studies relate both acute and long-term health effects to ultrafine particles (particle sizes < 0.1 μm) (Wichmann and Peters, 2000). Although ultrafine particles represent a minor contribution to mass concentrations, they constitute most of the particles in terms of number concentration.

14.6.1 Source of Traffic Particles and Modelling In busy streets, a significant fraction of particle pollution originates from traffic. The direct emissions from car exhaust contain particles formed inside the motor as well as in the air just after the exhaust pipe. These particles are mostly in the ultrafine particle fraction. However, traffic also contributes to mechanically formed particles in the fine and especially the coarse fraction. The particles in the coarse fraction are produced from wear of tyres and road material as well as resuspended dust. Particles from brakes contribute similar amounts to the fine and the coarse fractions.

288

O. Hertel et al.

For particle mass (PM10), long-range transport is the dominant source for regional background levels. Less than 10% of the urban PM10 originates from local urban sources. For particle numbers, as well as NOx, greater differences between rural, urban and kerbside levels are observed, indicating a large contribution from local traffic sources (Fig. 14.6). Concentration relative to urban background (=100)

600 kerbside urban background near-city location rural background

500 400 300 200 100 80 60 40 20 0 ToN

ToV

PM10

NOx

Source: Adopted from Ketzel et al. (2004). Fig. 14.6. Comparison of average concentrations of total particle numbers (ToN), particle mass (PM10) and NOx at rural, near-city, urban and kerbside stations relative to urban background levels in the Copenhagen area. The concentration bars are stacked so that only the additional contributions are marked with the pattern shown in the legend. Note that the scale of the vertical axis changes at 100.

A striking feature of urban particles is the high correlation (R>0.9) between concentrations of NOx and total particle numbers, indicating that both compounds originate from the same (traffic) source. They are emitted in a similar ratio particle number/NOx by the different traffic categories, i.e. high NOx emitters (diesel vehicles especially heavy-duty vehicles) are also high particle emitters. OSPM calculations reproduce the observed particle numbers (Fig. 14.7) well when treating particles as inert tracers (disregarding transformation and loss processes). The particle emission factors depend on ambient temperature with higher emissions at lower temperatures, which is accounted for in the simulations.

Assessing the Impacts of Traffic Air Pollution

289

80000 70000

ToN_mod ToNdiff_obs

60000 50000 40000 30000 20000 10000 0 -10000 18-06-2001

25-06-2001

02-07-2001

09-07-2001

Fig. 14.7. Time series of OSPM predictions (grey) and measurements (black) (difference between street and urban background) for total particle numbers ToN [cm-3] for a three-week period in June/July 2001. The average hourly emissions are determined by inverse modelling, treated separately for weekdays, Saturdays and Sundays

14.6.2 Other Particle Sources Long-range transport contributes significantly to the fine fraction particles and leads to the main part of particulate sulphate and ammonium and a large part of particulate nitrate. These secondary particles are formed from anthropogenic sulphur dioxide, ammonia and nitrogen oxides emissions. Another part of the particulate nitrate appears in the coarse fraction, which also contains contributions from sea spray and resuspended dust (including road dust) that has a relatively large mass and quickly deposits by gravitational settling. Coarse particles, therefore, have a shorter lifetime in the atmosphere compared with fine particles. Combustion in wood stoves is a source of particle pollution and it contributes to about 90% of the total particle emission related to domestic heating. Road traffic and the use of wood stoves are the biggest Danish sources of particle exposure of the population, due to low release height and because the emissions take place where people live. The particles emitted from wood stove combustion are soot particles with high contents of polycyclic hydrocarbons (PAH). People spend a significant part of their time indoors and exposure to air pollution in the home is therefore important for their overall exposure. A Danish study in an uninhabited apartment in central Copenhagen revealed that particle pollution inside the apartment to some extent was linked to the activity level of the neighbouring apartments (Schneider et al., 2004).

290

O. Hertel et al.

14.7 Examples of Exposure Assessment in Danish Studies A number of exposure assessment studies have been carried out in connection with the Danish National Environmental Research Programme (Hertel et al., 2001). The results from these studies are presented in the following.

14.7.1 Exposure of Bus Drivers and Postmen in Copenhagen Some occupations are associated with high exposure to ambient air pollution. Persons in such occupations are, thus, at higher risk of pollution-related adverse health effects compared with the rest of the population, especially persons who stay in the occupation for long periods of their life. Examples are bus drivers and postmen. It has been found that Copenhagen bus drivers have a higher frequency of several forms of cancers when compared to reference groups (Soll-Johanning et al., 1998). Monitoring exposure of workers in their working environment is demanding in terms of resources and therefore there is a need for comprehensive modelling tools for estimation of pollution exposure in epidemiological studies. This was the background for testing exposure assessment based on the OSPM against campaign measurements1. NO2 measurements were made using passive diffusion samplers inside a bus, outside a bus, on a postman and on a postman’s bicycle during a three-week campaign period in Copenhagen, Denmark. Sampling was performed for entire working days of the bus driver and for the total mail route of the postman. Information concerning the configuration of the streets (street orientation, street width, height of the buildings along the street, etc.) was obtained manually on location and information about traffic was obtained from the local municipality. Calculations with the OSPM were then undertaken for 22 selected street sections along the bus route and the postman’s mail route. Applying information from bus schedules and diaries, the average exposures were obtained for the bus and the postman’s bicycle. The measurements showed that NO2 concentrations inside and outside the bus were almost identical. Similarly, the NO2 concentrations on the postman and the bicycle were similar. The OSPM reproduced the measurements well (Fig. 14.8 and 14.9), and the study showed that such a model might serve as a useful tool for exposure assessment in occupational epidemiology studies of bus drivers and postmen.

1

Campaign measurements refer to monitoring during short intensive periods.

Assessing the Impacts of Traffic Air Pollution

291

Mixing ratio NO2 (ppb)

70 60 50 40

modelled measured inside measured outside measured Jagtvej

30 20 10 0 28-aug

31-aug

03-sep

06-sep

09-sep

12-sep

15-sep

Source: Hertel at al. (1996). Fig. 14.8. Average nitrogen dioxide mixing ratios over a whole working day of the Copenhagen bus driver (8 hours) for the campaign period 28 August-17 September 1995. The figure shows measurements for the entire working day of the bus driver obtained inside and outside the bus and results obtained from OSPM calculations representing the average concentrations along the bus route. Average concentrations for the working hours at the monitoring station Jagtvej (one of the streets on the bus route) are also shown. Arrows on the time axis indicate Sundays.

Mixing ratio NO2 (ppb)

70 60 50

modelled measured on postman measured on bicycle

40 30 20 10 0 28-aug

31-aug

03-sep

06-sep

09-sep

12-sep

15-sep

Source: Hertel at al. (1996). Fig. 14.9. Average nitrogen dioxide mixing ratios on the mail route of the Copenhagen postman (typically between 2 and 4 hours) for the campaign period 28 August – 17 September 1995. The figure shows measurements sampled for the time spent on the mail route of the postman and on his bicycle and the results obtained from OSPM calculations representing the average concentrations along the mail route. Arrows indicate Sundays (when no mail is delivered).

292

O. Hertel et al.

14.7.2 Exposure of Danish Children to Traffic Air Pollution The possible link between exposure to traffic-related air pollution and childhood cancer was investigated in the Children cancer study (Raaschou-Nielsen et al., 2001). Exposure at the front door address was taken as an indicator of the personal exposure of the child. Input data for the OSPM calculations (street configuration and traffic data) were obtained from a questionnaire sent to the local authorities. The information in the questionnaire was digitised and a programme was developed for interpretation and subsequent generation of input files for the model. The system was evaluated against a series of NO2 and benzene measurements obtained from 200 addresses. A reasonably good agreement was obtained (Fig. 14.10) and the system was subsequently applied to nearly 20,000 addresses of Danish children in the period 1960 to 1991. It has been estimated that the time spent by the local authorities on filling out the questionnaires of all these addresses corresponds to about 3 man-years. This illustrates clearly that such an approach to exposure assessment is very costly.

80 Calculated NO2 (ppb)

70 60 50 40 30 20 10 0 0

10 20 30 40 50 60 70 80 Observed NO2 (ppb)

Fig. 14.10. Comparison of about 1200 observed and calculated monthly mean NO2 mixing ratios at the address (front door) of 200 Danish children (100 in Copenhagen and 100 in rural areas). Data are identical to those presented in Raaschou-Nielsen et al. (2002a)

14.7.3 Personal Monitoring of Air Pollution Exposure in Copenhagen No models perform better than the quality of the available input and validation data. A field study of exposure to air pollution in Copenhagen was carried out during four seasons in both 1999 and 2000 (Sørensen et al., 2003).

Assessing the Impacts of Traffic Air Pollution

293

Personal exposure to PM2.5 (and black smoke analysed from the filters), NO2 and BTX (Benzene, Toluene and Xylenes) was measured for 50 students living and studying in central Copenhagen. Measuring campaigns of 5 weeks’ duration took place in autumn, winter, spring and summer. Measurements were conducted over 2-day periods for each study subject. Urban background concentrations were measured simultaneously at rooftop level in central Copenhagen. For 30 of the students, pollutant concentrations were measured simultaneously at the front door of the residence building and inside the bedroom. Morning blood samples were collected at the end of each 2-day campaign and 24-hour urine samples were collected on the second day of the campaign. Blood and urine samples were analysed for various biomarkers e.g. for DNA damage and the biomarker results were compared to air pollution exposure levels measured. Detailed information about the sampling equipment and methods of analysis is provided in Sørensen et al. (2003). The results showed that outdoor pollution concentrations are not always a good predictor of personal exposure. For NO2, the relationship between personal exposure and the front door concentration was dependent upon the season, with a stronger association in the warm season compared to the cold season, and for PM2.5 and black carbon the same tendency was observed. Time exposed to burning candles had a significant impact on the personal exposure levels of PM2.5, black carbon and NO2. Time exposed to environmental tobacco smoking (or passive smoking) had a significant impact on personal exposure levels of PM2.5, but did not affect the exposure levels of NO2 and black carbon. These findings imply that personal exposure to PM2.5, black carbon and NO2 depends on many factors in addition to outdoor levels and that information on the season or outdoor temperature and residence exposure improves the accuracy of personal exposure estimation. The analysis of the association between biomarkers and personal exposure suggested that even moderate exposure to concentrations of PM can induce oxidative DNA damage and that personal PM2.5 is more important in this aspect than is ambient PM2.5 background concentration.

14.7.4 Using GIS in Street Pollution Modelling for Exposure Assessment Geographical Information Systems (GIS) are used for a variety of purposes including exposure assessments. An example of a GIS-based system is the Danish AirGIS 2 (Jensen et al., 2001) that is now applied to exposure assessment in a number of Danish epidemiological studies.

2

See also: http://www.dmu.dk/International/Air/Models/AirGIS/.

294

O. Hertel et al.

of

ca rri a

ge

wa

y

or ie nt at io St re et oin t

W in d

se ct

or

D to ista ro nc ad e f ce rom nt re fac lin ad e e Pe rp en dic ula r Re ce pt or p

th

n

W id

Address point Road centre line Building

Fig. 14.11. Illustration of the 2½ dimensional model applied for generation of street configuration data in AirGIS. The calculation procedure is built on digital maps and data from Danish national registers. The parameters that are determined include general building height in the area, building height in wind sectors, width and orientation of the street and distances to street intersections AirGIS is based on the OSPM, available technical and cadastral maps (buildings,

roads, address points, property limits) and available Danish national administrative databases on buildings, property limit, and populations. The digital maps and registers applied depend on the purpose of application and the approach taken to generate the required data. The basic requirements for calculating air quality levels at a particular address are the availability of digital maps of streets with traffic data, buildings with building heights and addresses. Data on building heights may be obtained in different ways: (a) object heights above sea level of buildings minus terrain heights (b) heights obtained from the Building and Dwelling Registry using property limits to geo-code buildings (c) heights obtained from Digital Elevation Models. For population exposure assessment, population data can be linked to an individual address using the Central Population Register. Determining street configuration data from manual observations or from questionnaires filled in by local authorities is costly, especially when a large number of addresses is studied, for example in connection with a large cohort in an epidemiological study. This is clearly illustrated by exposure assessments presented in the two previous sections. Application of a GIS based system is a strong tool for obtaining such data using fewer resources. In the AirGIS system, a so-called 2½ dimensional landscape model determines these data. This is done by interpretation of digital maps and information from the available databases, which leads to generation of street configuration data for the OSPM (Fig. 14.11). Obtaining information about traffic on the road network is one of the main problems in generating the required input data for model calculations in AirGIS. The administration of the Danish road network is subdivided into three levels of

Assessing the Impacts of Traffic Air Pollution

295

government: municipalities, counties and state. Data for the county and state road network can be obtained relatively easily, whereas many municipalities do not have easily accessible databases for traffic. This constitutes a problem since the municipalities administer the major part of the total road network. For the Greater Copenhagen area, traffic information has been obtained from the CopenhagenRingsted Railway project. In this project, most of the necessary data were collected, though for an aggregated network and not a GIS network. To make the data useful in a GIS network, a link was established to transfer traffic data from the aggregated network to the GIS network. The AirGIS system allows for assessment of exposure at thousands of addresses, which would be impossible using measurements and highly resource demanding using the previously applied methods (manually or using questionnaires) for obtaining street configuration and traffic data (see the two previous sections). A system for tracking study subjects using a Global Positioning System unit (GPS) has been developed (Fig. 14.12). This system is based on GPS built into a mobile phone and transfers of position data using SMS, as well as routes are handled in the AirGIS (Fig. 14.13).

Fig. 14.12. Illustration of tracking an individual along a route in AirGIS

296

O. Hertel et al.

#

#

# #

# # # # # # # # ## # #

# # # # # #

# #

Centre #

# #

#

# # # # #

Radius

Po sit ion ing

# #

##

#

#

# 6

# # # # # #

#

# # # # ## ### # ## ### ## ### ### # #### # ### #### # ## # ## ### # # # ## # # ## # # # # ## # # ## ### ## ### # # ### ## # ### # # # ##### # ## ## ## # # # ## ## # ## # # ## ## # # ## # ## # # # # ### # ##### ## # ## ## # ## ##### ##### ## # # #### # # # ### # ## ## # # ## # # # # # ## ## # ## ## # ## # ## # ##### ### # ## # ## ## ##### # ## # # ## ## # ## # ## ## ## ## ### # ## ### ### ## # #### ####### ###### # ## # # #### # ### ## ### # # ####

# 0

# 7

Tracking program

# 9

# 5

PC and GSM modem

GPS

ArcView script for visualisation

Fig. 14.13. Example on results from the system for tracking people using a Global Positioning System (GPS). The GPS is built into a mobile telephone and SMSs are used for transferring positions of a study subject every 20 seconds to a computer. A tracking programme collects the GPS data. A dynamic area-tracking feature ensures that SMSs are only sent when the subject is moving (settings may be changed on-line by telephone). The movement of a study subject can be traced on digital maps in real time

The AirGIS system has been validated by careful comparison of model results to observations from fixed site monitoring stations as well as to personal exposure measurements from the field study referred to in the previous section. These validations reveal that the 2½ dimensional landscape model provides high quality street configuration data and that the total model system is capable of reproducing fairly well the observed concentrations and personal exposures (Jensen et al., 2004).

14.8 Assessment of Health Effects of Air Pollution Exposure Most of the assessments of health effects of air pollution exposure presented in the literature are based on data from fixed site monitoring or simple methodologies are applied for estimation of the exposure of the population and extrapolations are the obtained for exposure/dose-effect relationships from the literature. On the initiative of the WHO Europe, the national agencies from Austria, France and Switzerland have assessed the public health costs of total air pollution and traffic-related air pollution (Künzli et al., 2000). The results showed that in Switzerland, which has a population similar to Denmark, a reduction in the population-weighted PM10 concentrations from the present 21.4 Pg/m3 to 7.5 Pg/m3

Assessing the Impacts of Traffic Air Pollution

297

would result in a substantial reduction in mortality and morbidity, with up to 3,300 premature deaths per year. The methodology of this study has been applied to Denmark (Raaschou-Nielsen et al., 2002b). However, a slightly different method for determining PM10 is used in the Danish study. Here, the urban background concentrations in Copenhagen were calculated based on traffic data and calculated NOx emissions were derived on the basis of the traffic data, application of empirically-determined relationships between NOx and PM10 emissions from traffic, meteorological data and measurements of NOx in rural areas. The concentrations from Copenhagen were then scaled down for smaller cities and towns based on the city size. Using this methodology, the average population PM10 exposure was estimated to be 22 Pg/m3. Approximately one-third could be related to natural sources. The man-made level of PM10 was estimated to cause annually approximately 3,400 deaths, 3,700 hospital admissions, 3,300 cases of chronic bronchitis, 12,000 cases of acute bronchitis, 150,000 asthma attacks and close to 1.8 million restricted activity days. This type of assessment is based on transferring exposure-effect relationships from studies abroad. Currently, exposure assessment is being added to a number of Danish ongoing and historical epidemiological studies. When the exposure assessments are ready, these data will be analysed with respect to collected health data from various cohorts.

14.9 Perspectives Tools such as the AirGIS have only been available for exposure assessment for a short time. Ongoing studies where AirGIS is applied will be concluded and published in the near future. The results from these studies will indicate how well the methodologies behind these tools work in practice. In these applications, air pollution exposure at the front door of the address is in most cases used as a proxy for personal exposure. Currently, calculations are being performed for about 20,000 addresses. In later project, similar AirGIS calculations will be performed for more than 200,000 addresses.

References Berkowicz, R. (1998). Street Scale Models. In: Fenger, J., Hertel, O., Palmgren, F. (eds.). Urban Air Pollution – European Aspects. Kluwer Academic Publishers, the Netherlands: 223-251. Berkowicz, R. (2000a). A simple model for urban background pollution. Environmental Monitoring and Assessment, 65(1-2): 259-267. Berkowicz, R. (2000b). OSPM – A parameterised street pollution model. Environmental Monitoring and Assessment, 65(1-2): 323-331.

298

O. Hertel et al.

Hertel, O., Wilhardt, P., Berkowicz, R., Skov, H. (1996). Exposure of bus drivers and postmen to air pollution from traffic in their working environment. Proceedings of NOSA/NORSAC Symposium, Elsinore, Denmark, 15-17 Nov., P. Hummelshøj (ed.), Risø-R-934 (EN): 20-22. Hertel, O., De Leeuw, F.A.A.M., Raaschou-Nielsen, O., Jensen, S.S., Gee, D., Herbarth, O., Pryor, S., Palmgren, F., Olsen, E. (2001). Human exposure to outdoor air pollution (IUPAC technical report). Pure and Applied Chemistry, 73: 933-958. Hertel, O., Ambelas Skjøth, C., Frohn, L.M., Vignati, E., Frydendall, J., de Leeuw, G., Swarz, S., Reis, S. (2002). Assessment of the atmospheric nitrogen and sulphur inputs into the North Sea using a Lagrangian model. Physics and Chemistry of the Earth, 27(35): 1507-1515. Jensen, S.S., Berkowicz, R., Hansen, H.S., Hertel, O. (2001). A Danish decision-support GIS tool for management of urban air quality and human exposures. Transportation Research Part D: Transport and Environment, 6: 229-241. Jensen, S.S., Hvidberg, H., Brocas, M., Larsen, M.R.B., Berkowicz, R., Hertel, O., Sørensen, M., Loft, S. (2004). Personal Exposure and Health Related to Traffic Air Pollution (In Danish: Personlig eksponering og sundhed for trafikskabt luftforurening). Nordic Transport Conference, Aalborg, Denmark. Kallos, G. (1998). Regional/mesoscale models. In: Fenger, J., Hertel, O., Palmgren, F. (eds.). Urban Air Pollution – European Aspects. Kluwer Academic Publishers. – Environmental Pollution 1: 177-196. Ketzel M., Wåhlin, P., Berkowicz, R., Palmgren, F. (2003). Particle and trace gas emission factors under urban driving conditions in Copenhagen based on street and roof level observations. Atmospheric Environment, 37: 2735-2749. Ketzel, M., Wåhlin, P., Kristensson, A., Swietlicki, E., Berkowicz, R., Nielsen, O. J., Palmgren, F. (2004). Particle size distribution and particle mass measurements at urban, near-city and rural level in the Copenhagen area and Southern Sweden. Atmospheric Chemistry & Physics, 4: 281292. Künzli, N., Kaiser, R., Medina, S., Studnicka, M., Chanel, O., Filliger, P., Herry, M., Horak, F., Puybonnieu´x-Texier, V., Quenel, P., Schneider, J., Seethaler, R., Vergnaud, J.C., Sommer, H. (2000). Public-health impact of outdoor and traffic-related air pollution: a European assessment, Lancet, 356: 795-801. Mestayer, P.G. (1998). Urban scale models. In: Fenger, J., Hertel, O. and Palmgren, F. (eds.). Urban Air Pollution – European Aspects. Kluwer Academic Publishers, the Netherlands: 223251. Palmgren, F., Berkowicz, R., Hertel, O., Vignati, E. (1996). Effects of reduction of NOx on the NO2 levels in urban streets. Science of the Total Environment, 190: 409-415. Palmgren, F., Berkowicz, R., Ziv, A., Hertel, O. (1999). Actual car fleet emissions estimated from urban air quality measurements and street pollution model. Science of the Total Environment, 235: 101-109. Raaschou-Nielsen, O., Hertel, O., Thomsen, B.L., Olsen, J.H. (2001). Air pollution from traffic at the residence of children with cancer. American Journal of Epidemiology, 153: 433-443.

Assessing the Impacts of Traffic Air Pollution

299

Raaschou-Nielsen O., Hertel O., Vignati E., Berkowicz R., Jensen S.S., Larsen V. B., Lohse, C., Olsen J.H. (2002a). An air pollution model for use in epidemiological studies: Evaluation with measured levels of nitrogen dioxide and benzene. Journal of Exposure Analysis and Environmental Epidemiology, 10: 4-14. Raaschou-Nielsen, O., Palmgren, F., Jensen, S.S., Wåhlin, P., Berkowicz, R., Hertel, O., Vrang, M.-L., Loft, S. (2002b). Health effects of particular air pollution in Denmark – an attempt to quantify (In Danish: Helbredseffekter af partikulær luftforurening i Danmark – et forsøg på kvantificering). Ugeskrift for Læger, 164(34): 3959-3963. Schneider, T., Jensen, K.A., Clausen, P.A., Afshari, A., Gunnarsen, L., Wåhlin, P., Glasius, M., Palmgren, F., Nielsen, O.J., Fogh, C.L. (2004). Prediction of outdoor particle penetration and deposition in an uninhabited Copenhagen apartment for the size range 0.5-4 μm. Atmospheric Environment, 38: 6349-6359. Soll-Johanning H., Bach E., Olsen J.H., Tuchsen, F. (1998). Cancer incidence in urban bus drivers and tramway employees. A retrospective cohort study. Occupational and Environmental Medicine, 55: 594-598. Sørensen, M., Autrup, H., Møller, P., Hertel, O., Jensen, S.S., Vinzents, P., Knudsen, L.E., Loft, S. (2003). Linking exposure to environmental pollutants with biological effects. Mutation Research/Reviews in Mutation Research, 544: 255-271. Tilmes, S., Brandt, J., Flatøy, F., Bergström, R., Flemming, J., Langner, J., Christensen, J.H., Frohn, L.M., Hov, Ø., Jacobsen, I., Reimer, E., Stern, R., Zimmermann, J. (2002). Comparison of Five Eulerian Air Pollution Forecasting Systems for the Summer of 1999 Using the German Ozone Monitoring Data. Journal of Atmospheric Chemistry, 42: 91-121. Wichmann, H.E., Peters, A. (2000). Epidemiological evidence of the effects of ultrafine particle exposure. Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences, 358(1775): 2751-2768. Wilkins, E.T. (1954). Air pollution aspects of the London fog of December 1952. Quarterly Journal of the Royal Meteorological Society, 80: 267-271. Zartarian, V.G., Ott, W.R., Duan, N. (1997). A quantitative definition of exposure and related concepts. Journal of Exposure Analysis and Environmental Epidemiology, 7: 411-437.

15 Car Use Habits: An Obstacle to the Use of Public Transportation? Berit Møller and John Thøgersen Aarhus School of Business, Department of Marketing and Statistics, University of Aarhus, Denmark

Abstract It is often claimed that many drivers use their private car rather habitually. The claim gains credibility from the fact that travelling to many everyday destinations fulfils all the prerequisites for habit formation: it is recurring, performed under stable circumstances and produces rewarding consequences. Since the decision is made quite automatically and only one choice alternative is considered (the habitually chosen one), behaviour guided by habit is difficult to change. The implications of car use habits for converting drivers to commuters using public transportation is analysed based on a survey undertaken in the Copenhagen area. The study reveals that a relatively low percentage of drivers (10-20%) consider commuting by public transportation in the near future, which is hardly a surprise. A hierarchical analysis, where reported use of public transportation is regressed on intentions to do so, car use habit, and the interaction between the two, confirms the theory-derived hypothesis that car use habits act as an obstacle to the transformation of intentions to commute by public transportation into action.

15.1 Introduction Even though the average fuel consumption per car kilometre fell by 11% during the 1990s, CO2 emissions from road transport in Denmark increased by 16%, the reason being that the increase in kilometres driven (29%) exceeded the fall in fuel consumption (Danmarks Statistik, 2002). This situation is not unique for Denmark, on the contrary. Worldwide, rapid growth in car driving is a source of concern both because of its local and global environmental impacts (Mackenzie, 1997; OECD, 1996; Trafikministeriet, 1995, 1999). Solving the traffic-related environmental problems is not easy, however. A wide range of regulations, system changes, and adaptations are needed, which can only be provided through the concerted effort of many parties, including car producers, providers of public transport services, public authorities, and users of transport products and services (Mackenzie, 1997; OECD, 1996; Sperling and Shaheen, 1995; Vlek et al., 1993; Vlek and Steg, 1996). Final consumers, who in their private lives and for private purposes demand transport solutions, play one of the key roles. Person transport constitutes most of total

302

B. Møller and J. Thøgersen

transport, and private cars are being increasingly and dominantly used (Danmarks Statistik, 2002; Mackenzie, 1997). Private consumers of transport products and services are also the majority of voters in most countries. Hence, in a democratic society, whatever solution one proposes to relieve the environmental and societal burdens of transport, it stands or falls with the acceptance by these individuals, either as buyers of transport products and services or as voters (Schlag and Teubel, 1997). In this chapter, we focus on individuals in their former capacity. Since most person transport is made by individuals as private consumers, and since their dual capacity as consumers and voters makes it virtually impossible effectively to force people to change their transport behaviour, much more insight is needed into how people can be persuaded to accept more environmentally-friendly transport solutions (Steg and Vlek, 1997; Tertoolen, Kreveld, and Verstraten, 1998). In this book, the focus is on road pricing as a means of persuasion. In recent years, Dutch (Aarts, Verplanken, and Knippenberg, 1998; Verplanken, Aarts, Knippenberg, and Knippenberg, 1994; Verplanken, Aarts, Knippenberg, and Moonen, 1998) and Swedish (Gärling, Boe, and Fujii, 2001) researchers in particular, have made substantial advances in demonstrating the importance of habits for the choice of mode of transportation. Consistent with these findings, a recent Danish study found that past behaviour is a better predictor of the use of public transportation than attitude towards the use of public transportation, public transportation’s ability to cover the individual’s transport needs, and perceived social pressure to use public transportation (Thøgersen, 2006). If the travel mode is generally chosen habitually, this has a profound influence on the effectiveness of different means of persuasion (Aarts, Verplanken, and Knippenberg, 1997; Assael, 1987; Ronis, Yates, and Kirscht, 1989; Thøgersen and Ölander, 2006). In particular, it means that drivers are unlikely to search for – or even contemplate – new information before choosing a travel mode. Hence, in order to be noticed at all, persuasive information needs to be obtrusive and to be perceived as personally relevant (Dahlstrand and Biel, 1997; Hoyer and MacInnis, 1997). Besides usually being obtrusive themselves, economic instruments of regulation, such as road pricing, have the advantage of catching media attention, which increases the chances of regulation being noticed by those being targeted. However, if the regulatory instrument obtains its personal relevance by threatening individuals’ economic security and freedom, as fees and taxes usually do, there is a risk that it produces a psychological reactance (Brehm and Brehm, 1981; Jacobsson, Fujii, and Gärling, 2000); a motivational state directed toward re-establishing the threatened free behaviour. Hence, although regulation based on strong economic incentives is likely to persuade at least some drivers to change behaviour, they may in addition produce an increased desire for the targeted behaviour (driving) (Tertoolen et al., 1998; Van Vugt et al., 1996). The risk of resistance, as well as political action against the regulation, is increased if the regulation is perceived as unfair because it hits low income groups harder than high income groups (Gardner and Stern, 1996; McClelland and Canter, 1981). There is no doubt that commuting to work by car has the potential of becoming habitual: it is frequently and extensively performed, in stable surround-

Car-Use Habits: An Obstacle to the Use of Public Transportation?

303

ings, and usually the destination is reached in a timely and comfortable manner, i.e. the experience is perceived as rewarding. These are the three prerequisites for habit formation identified, for instance, by Ouellette and Wood (1998). Conditions such as these facilitate a learning process, which allows the individual to reduce the amount of cognitive effort in decision-making and to perform the behaviour with an increasing degree of automaticity (Thøgersen and Ölander, 2006). Following Verplanken and Aarts (1999), we define a habit as a learned sequence of acts that has become an automatic response to specific cues, and is functional in obtaining certain goals or end-states. The instigation of habitual behaviour is volitional and intentional, but subsequent acts may be unintentional (Ouellette and Wood, 1998; Verplanken and Aarts, 1999). Ouellette and Wood note that many established behavioural routines in daily life, such as car use, have both volitional (planning to go somewhere) and automatic elements (taking the car, driving). A habit, which was functional in obtaining some goal(s) at the time when it was formed, may lose its functionality if the goal(s) change at a later point in time (Verplanken and Aarts, 1999). In such cases, the habit may become counterintentional. Counter-intentional habits are particularly prevalent when the behaviour is based on short-term, hedonistic motives at the expense of long-term goals (Verplanken and Faess, 1999). The impact of habits that counteract attitudes and intentions has been studied for healthy dieting behaviour (Verplanken and Faess, 1999), but not previously with travel mode choice. Studies of the habit-intention-behaviour relationships regarding travel mode choice have confirmed Triandis’s theory (Triandis, 1977; Landis et al., 1978) that behavioural intentions and habits interact in determining behaviour, implying that the stated intention is a good predictor of behaviour only under conditions of weak habits while intention is a bad predictor of behaviour when habits are strong (Aarts et al., 1997; Gärling et al., 2001; Verplanken et al., 1994, 1998). For instance, Verplanken et al. (1994) found that the correlation between the attitude towards a specific travel mode option and travel mode choice (for shopping trips to either of two cities located approximately 5 miles away and where a realistic public transport option existed) was significantly weaker for individuals with strong rather than weak habits. This study is inspired by the work of Verplanken and his associates (Verplanken et al., 1994; Verplanken et al., 1998). However, the approach here is slightly different. We investigate whether a habit of using one travel mode (a private car) is an obstacle to choosing another (public transport). Our approach is an attempt to investigate the influence of counter-intentional habits (Verplanken and Aarts, 1999) on travel mode choice, where car use habits are perceived as counter-intentional with respect to an intention to commute by public transport. Specifically, we test the following hypothesis: x H1: For commuters with a strong car use habit, there is a weak correlation between intention and behaviour in relation to the use of public transportation and for commuters with a weak car use habit the correlation is strong.

304

B. Møller and J. Thøgersen

15.2 Method This study is based on a survey of drivers in the Copenhagen area, the data having been collected in October 2002. Only 37.6% of the households in the Copenhagen area have a car (Danmarks Statistik, 2001). Hence, rather than aiming for a representative picture of the total population, this study focuses on the segment that is most likely to have developed a habit of commuting by private car. Respondents were screened according to the following criteria: The household has a car at its disposition and the respondent possesses a driver’s licence, is in work or studies, has not held a travel pass for public transport in the Copenhagen area in the last year, and does not need a car to perform his or her job. If more than one person in the household fulfilled the criteria, the next birthday method was used to select the respondent. Of those meeting the screening criteria, 1071 agreed to participate, resulting in a response rate of 75%. Listwise deletion is used in the case of missing values with the result that the sample size is reduced by 27 to 73 respondents in the following analyses.

15.3 Data Behaviour was measured as a frequency on a scale from 0 to 10 using the item: On how many of the last 10 journeys have you used public transportation for the trip between home and work/educational institution? 1 Although low, the mean (0.57) differs significantly from zero (t = 9.832, p < 0.001). A frequency analysis shows that 85% of the respondents had not commuted by public transport on any journey out of the last ten, 8% had used public transport on one or two journeys and 7% had used public transport more than twice in the last ten commuting journeys. The intention to commute by public transport was measured using two items on 11-point scales, following the form suggested by Ajzen and Fishbein (1980) (see Table 15.1).

1

Note that the measure of behavioural frequency is made at the same time as the measure of behavioural intention. Hence, as always in cross-sectional analyses, the measures are not registered in their assumed temporal sequence. This, of course, means that we cannot be sure of the direction of causality between the measured variables. However, previous studies have found travel mode choices to be quite stable over time (Bamberg and Lüdemann, 1996) and even over a period of one year (Thøgersen, 2006), which indicates that using behaviour in the recent past as an indicator for behaviour in the near future is defensible.

Car-Use Habits: An Obstacle to the Use of Public Transportation?

305

Table 15.1. Items and descriptive statistics for intention (n = 1052)

Mean

St. Dev.

On how many of the next 10 journeys do you plan to use public transport for the trip to work/educational institution? (0: zero times to 10: ten times)

0.44

1.68

If you consider the next month, how often do you think you will use public transport for the trip to work/educational institution? (0: never to 10: every time) SCALE intention

0.54 0.49

1.70 1.62

Cronbach’s Alpha

0.92

The intention scale is constructed by averaging the scores on the two items. Cronbach’s Alpha indicates excellent reliability (Cronbach, 1987). The scale mean is significantly different from zero (t = 9.820, p < 0.001). A frequency analysis shows that 89% of the respondents did not plan to commute by public transport one single time out of the next ten and 81% did not expect to commute by public transport at all during the following month. Table 15.2. Items and descriptive statistics for the response frequency habit measure (RF) I now mention some travel mode situations. Please say as fast as possible for each of them the transport mode that springs to mind. There are no wrong answers, but it is important that you answer as quickly as you can, and state the transport mode that first comes to mind. Which mode of transport would you choose if you are …

% choosing car

n

picking someone up at the central station?

77.4

1065

visiting someone at the other end of the capital area?

84.8

1064

going to the beach with friends for a day?

76.4

1063

going shopping in the centre of Copenhagen?

33.8

1063

going to the nearest woods for a walk?

70.6

1067

going to sports or other recreational activities?

55.3

1054

going shopping in the nearest supermarket?

43.6

1070

going to the nearest post office?

30.5

1069

going to visit someone in Jutland?

79.0

1059

going to visit a sight outside the Copenhagen area?

95.1

1068

Scale RF (mean = 6.48, St. Dev. = 2.12)

1016

Two habit measures were employed. The response frequency (RF) measure developed by Verplanken and his associates (1994) attempts to capture mental representations of habitual activities. It is based on the assumption that habitual responses are guided by mental representations of past behaviour, i.e. scripts (Aarts et al., 1997). When a habit appears in different contexts (for example, habitual car

306

B. Møller and J. Thøgersen

use in shopping, recreational and/or work-related contexts), each of the contexts may evoke its own specific scripted sequence of acts (Verplanken and Aarts, 1999). If the goal and the behavioural response are the same across different situations, a general habit may evolve that is capable of being triggered by specific cues in different situations. The RF measure is designed to capture such a general habit, rather than the habit of choosing a particular travel mode in a particular situation. In order to reflect scripts in the semantic memory, the RF measure should elicit automatic (as opposed to strategic) responses. Automatic responses are facilitated by imposing time pressure and by asking about the first travel mode that comes to mind when confronted with the choice situation (Aarts et al., 1997; Verplanken and Aarts, 1999). The instrument contains 10 trips of short, medium or long distance. The number of times that “car” is mentioned serves as a measure of the respondent’s general habit of choosing the car. Hence, the RF measure can vary from 0 to 10. Items are presented in Table 15.2. Table 15.3. Items and descriptive statistics for the of SRHI-measure, n = 1064 Mean

St. Dev.

I often use the car as a means of transport between home and work/educational institution

3.86

1.67

I automatically use the car as a means of transport between home and work/educational institution

3.57

1.78

It would be difficult not to use the car as a means of transport between home and work/educational institution

3.07

1.76

To use the car as a means of transport between home and work/educational institution belongs to my routines

3.66

1.74

I would find it hard not to use the car as a means of transport between home and work/educational institution

3.20

1.76

To use the car as a means of transport between home and work/educational institution is something that is typically me

3.35

1.73

To use the car as a means of transport between home and work/educational institution is something I have done for a long time. SCALE: habit (SRHI)

3.68 3.09

1.75 1.78

Kronbach’s Alpha

0.95

Note: Answers are given on a 5-point scale with the categories: totally disagree (1), partly disagree (2), neither agree nor disagree (3), partly agree (4), totally agree (5).

Car-Use Habits: An Obstacle to the Use of Public Transportation?

307

The other habit measure, the self-reported habit index (SRHI), has recently been developed by Verplanken and Orbell (2003). It is designed to capture the features of automaticity assumed to be most descriptive of a habit: uncontrollability, efficiency and lack of awareness, in addition to a history of repetition and the degree of reflecting identity or personal style. Verplanken and Orbell had 12 items in their original measuring instrument. Here, the number of items was reduced after a pre-test to 7. The items are shown in Table 15.3. The SRHI scale is constructed by averaging the scores on the seven items. Cronbach’s Alpha indicates excellent reliability.

15.4 Results The hypothesis is tested by means of hierarchical regression analysis. In a hierarchical regression analysis, the independent variables are entered in the order that one expects them to influence the dependent variable. In this case, it is assumed that the intention to use public transport is the immediate antecedent of the behaviour, if it is not blocked by car use habits. Hence, when predicting commuting by public transport, the intention to use public transport is entered first, followed by a measure of car use habits. In this second step, a possible additive effect of habit on behaviour is captured. Finally, the hypothesised moderating effect of habit on the intentionbehaviour relationship is captured by including the interaction term (intention x habit) into the equation. Hence, the analysis involves three steps: 1. Behaviour = f(Intention) 2. Behaviour = f(Intention, Habit) 3. Behaviour = f(Intention, Habit, Intention x Habit) If the variable entered has a significant effect on the dependent variable not 2 accounted for by the variable(s) previously entered, the F-value of the change in R from one step to the next is statistically significant. In order to avoid multicollinearity, intention and habit measures must be centred (scored as deviations from the scale mean) before conducting the analysis (Aiken and West, 1991; Cronbach, 1987). First, we present the results of the analysis using the SRHI measure of habit (Table 15.4). Table 15.4. Hierarchical regression: The moderating effect of habit (SRHI) on the intentionbehaviour-relationship (n = 1044) 2

1: B = f(I) 2: B = f(I, H) 3: B = f(I, H, I*H)

Adj. R 0.50 0.50 0.53

2

' R

'F

d.f.

0.000 0.028

0.346 60.772

1; 1041 1; 1040

Sig. ' F 0.000 0.580 0.000

Note: H = car habit (SRHI measure), B = behaviour: use of public transport, I = intention to use public transport. Estimated parameters for equation 3: Behaviour = 0.58I – 0.04H – 0.21(I*H); parameters for I and I*H are significant at p < .001.

308

B. Møller and J. Thøgersen

Table 15.4 shows that the model predicts behaviour satisfactorily (53% of the variation is explained). The standardised E-coefficients reveal, as expected, a significant and positive direct effect of intention on behaviour. Also as expected, the direct effect of car use habit on commuting by public transport (behaviour) is negative, but it is not significant at the .05 level. As predicted, there is a significant interaction between car use habit and intention. The coefficient of the interaction term shows how much the effect of one of the variables depends on the level of the other. In the presence of an interaction effect, the coefficient of a centred variable represents its effect at the mean of the interacting variable (Aiken and West, 1991). At one standard deviation above (below) the mean of the interacting variable, the coefficient of the variable in focus is equal to its value at the mean of the interacting variable plus (minus) the coefficient of the interaction term. Hence, when the car use habit is strong (one standard deviation above its mean value), the effect of intention on behaviour is weak, .37 (.58 + (-.21)) and when the car use habit is weak (one standard deviation below its mean value) the effect of intention on behaviour is strong, .79 (.58 - (-.21)), consistent with the hypothesis that car use habits act as an obstacle to transforming intentions to commute by public transport into action. A similar hierarchical regression analysis was made with the RF measure representing habit. The results are shown in Table 15.5. Again, the model predicts behaviour satisfactorily (50% of the variation is explained), although not as well as above. The reason for this difference is that the effect of the interaction term is weaker when the habit is represented by the RF measure rather than the SHRI measure. However, the interaction term still has a (marginally) significant negative effect in this case. Hence, the conclusions of the analysis remain substantively the same; that intention to commute by public transport is more likely to be transformed into action under conditions of weak rather than under conditions of strong car use habits. Table 15.5. Hierarchical regression: The moderating effect of habit (RF) on the intention-behaviourrelation (n = 998) 2

1: B = f(I) 2: B = f(I, H) 3: B = f(I, H, I*H)

Adj. R 0.499 0.498 0.500

2

' R

'F

d.f.

Sig. ' F

0.000 0.002

0.064 3.425

1; 995 1; 994

0.800 0.065

Note: H = car habit (RF measure), B = behaviour: use of public transport, I = intention to use public transport. Estimated parameters for equation 3: Behaviour = 0.68I – 0.01H – 0.05(I*H); parameters for I and I*H are significant at p < .001 and p < .07 respectively.

15.5 Summary and Implications Based on reasoning, suggesting that it is likely that car users develop a habit of commuting by car and on previous research suggesting that travel mode choices are strongly influenced by habits, this study sets out to investigate whether a habit of

Car-Use Habits: An Obstacle to the Use of Public Transportation?

309

driving may be an obstacle to transforming an intention to commute by public transport into action. In this perspective, a car use habit may be conceived as counterintentional. The study is based on a survey of car users in the Copenhagen area. Behaviour is operationalised as behaviour frequency. Behavioural intention is operationalised as a plan for the near future. Habits are operationalised as either a general, cross-situational script linking transport goals to a travel mode (the response frequency measure) or as a more specific self-descriptive measure for commuting by car that captures features of automaticity, a history of repetition, and the degree to which commuting by car reflects identity or personal style (the self-reported habit index). Two results of the analyses are particularly important: 1. Descriptive statistics show that only 10-20% of the interviewed car users have any intention of commuting by public transport in the near future and that few have done so in the recent past, which should come as no surprise. Many drivers commute by car because there are no alternatives that fulfil their specific needs and probably an even larger number do so because they perceive the car as the best among the available alternatives. Some of them may be wrong, they may be ill-informed about the options available. However, our focus is not on the ill-informed, but on the minority of drivers who have formed a conscious intention to commute by public transport. A car use habit seems to be an obstacle to transforming such an intention into action. 2. Hierarchical regression analysis shows that intentions and behaviour regarding commuting by public transportation are strongly correlated. However, as predicted by the proposition that car use habits form an obstacle to transforming an intention to commute by public transport into action, the intentionbehaviour correlation depends on the strength of the habit of driving by car. When the habit is weak, the intention-behaviour correlation is strong, and when the habit is strong, the intention-behaviour correlation is weak(er). This is the result regardless of which of the two habit measures included is used, but the strength of the moderating effect varies between the two measures. The SRHI measure is a considerably stronger moderator than the RF measure. We cannot be sure why the two habit measures are not equally strong moderators of the commuting by public transport intention-behaviour relationship. Evidence presented in Table 15.2 suggests that the most likely reason is that few drivers have developed a cross-situational habit of driving by car that is as general as suggested by the RF measure. Table 15.2 shows that the percentage of drivers choosing the car as their mode of transportation to the ten destinations varies considerably (from 30.5 to 95.1%). Hence, it seems likely that the RF instrument contains situations not covered by most of the drivers’ car habits. This could explain why the RF measure is a weaker moderator than the SRHI measure, which focuses specifically on car use for commuting2 (remembering the discussion about the correspondence principle in attitude theory, Ajzen and Fishbein, 1980). If this is the explanation, 2

In fact, commuting is not even included among the situations covered by the RF measure.

310

B. Møller and J. Thøgersen

the RF measure should become a stronger moderator of the intention-behaviour relationship if items diverging from the general pattern were removed. In the calculations reported in Table 15.6, we removed the three situations where fewer than 50% responded that they would use the car. This leads to an increase in the numerical value of the regression coefficient of the interaction term .05 to .08. Also, the inclusion of the interaction term now leads to a highly significant increase in R2. This obviously lends support to our proposed explanation for the weakness of the RF measure. Table 15.6. Hierarchical regression: The moderating effect of habit (revised RF scale) on the intention-behaviour-relation (n = 998) 2

1: B = f(I) 2: B = f(I, H) 3: B = f(I, H, I*H)

Adj. R 0.499 0.499 0.503

2

' R

'F

d.f.

Sig. ' F

0.000 0.005

0.588 9.116

1; 995 1; 994

0.444 0.003

Note: H = car habit (revised RF measure), B = behaviour: use of public transport, I = intention to use public transport. Estimated parameters for equation 3: Behaviour = 0.67I – 0.01H – 0.08(I*H); parameters for I and I*H are significant at p < .001 and p < .003 respectively.

Previous research and reasoning about the nature of decision-making regarding habitual behaviour suggests that information about opportunities to use and the benefits of public transport has limited chances of breaking a habit of commuting by car (Verplanken et al., 1998). Instead, we suggest that there is a need to explore the possibilities for implementing structural changes that are both strong enough to persuade or entice car users at least to try available public transport options and be acceptable for car users (voters) as well as for politicians. Road pricing could be a means to this end, although preliminary results indicate that it is not easy to gain public acceptance (Jacobsson et al., 2000). If it is possible to create an economic incentive, which is perceived as a carrot rather than as a stick, acceptance is more easily obtained (Barde and Opschoor, 1994; Gardner and Stern, 1996). In our current research, we test the effectiveness of the opportunity to use public transport free of charge for one month (an instrument that also has been tested in Japan with promising results, see Fujii and Kitamura, 2003), but there are other possibilities that should be pursued.

References Aarts, H., Verplanken, B., Knippenberg, A. v. (1997). Habit and information use in travel mode choices. Acta Psychologica, 96: 1-14. Aarts, H., Verplanken, B., Knippenberg, A. v. (1998). Predicting behavior from actions in the past: Repeated decision making or a matter of habit. Journal of Applied Social Psychology, 28: 13551374.

Car-Use Habits: An Obstacle to the Use of Public Transportation?

311

Aiken, L.S., West, S.G. (1991). Multiple regression: Testing and interpreting interactions. Sage Publications, Newbury Park. Ajzen, I., Fishbein, M. (1980). Understanding attitudes and predicting social behavior. PrenticeHall, Englewood Cliffs. Assael, H. (1987). Consumer behavior and marketing action. Kent Publishing Co, Boston. Bamberg, S., Lüdemann, C. (1996). Eine Überprüfung der Theorie des geplanten Verhaltens in zwei Wahlsituationen: Rad vs. Pkw und Container vs. Hausmüll. Zeitschrift für Sozialpsychologie, 27: 3246. Barde, J.-P., Opschoor, J.B. (1994). From stick to carrot in the environment. The OECD Observer, 186: 23-27. Brehm, S.S., Brehm, J.W. (1981). Psychological reactance: A theory of freedom and control. Academic Press, San Diego. Cronbach, L. (1987). Statistical tests for moderator variables: flaws in analyses recently proposed. Psychological Bulletin, 102: 414-417. Dahlstrand, U., Biel, A. (1997). Pro-environmental habits: Propensity levels in behavioral change. Journal of Applied Social Psychology, 27: 588-602. Danmarks Statistik. (2001). Familiernes bilrådighed 2001. Copenhagen: Danmarks Statistik. Danmarks Statistik. (2002). CO2-udledning fra vejtransport er steget (Nyt fra Danmarks Statistik: Trafik og miljø 2002 No. 477). Danmarks Statistik, Copenhagen. Fujii, S., Kitamura, R. (2003). What does a one-month free bus ticket do to habitual drivers? An experimental analysis of habit and attitude change. Transportation, 30: 81-95. Gardner, G.T., Stern, P.C. (1996). Environmental problems and human behavior. Allyn and Bacon, Boston. Gärling, T., Boe, O., Fujii, S. (2001). Empirical tests of a model of determinants of script based driving choice. Transportation Research F, 4: 89-102. Hoyer, W.D., MacInnis, D.J. (1997). Consumer behavior. Houghton Mifflin, Boston. Jacobsson, C., Fujii, S., Gärling, T. (2000). Determinants of private car users' acceptance of road pricing. Transport Policy, 7: 153-158. Landis, D., Triandis, H.C., Adamopoulos, J. (1978). Habit and behavioral intentions as predictors of social behavior. Journal of Social Psychology, 106: 227-237. Mackenzie, J.J. (1997). Driving the road to sustainable ground transportation in: Dower, R., Ditz, D., Faeth, P., Johnson, N., Kozloff, K., Mackenzie, J.J. (eds.), Frontiers of sustainability, Island Press, D.C. Washington: 121-190. McClelland, L., Canter, R.J. (1981). Psychological research on energy conservation: Context, approaches, methods in: Baum, A., Singer, J.E. (eds.), Advances in environmental psychology, Vol. 3. Energy conservation: Psychological perspectives, Lawrence Erlbaum, Hillsdale: 1-26.

312

B. Møller and J. Thøgersen

OECD. (1996). Towards sustainable transportation. The Vancouver conference. Paris: OECD. Ouellette, J.A., Wood, W. (1998). Habit and intention in everyday life: The multiple processes by which past behavior predicts future behavior. Psychological Bulletin, 124: 54-74. Ronis, D.L., Yates, J.F., Kirscht, J.P. (1989). Attitudes, decisions, and habits as determinants of repeated behavior in: Pratkanis, A.R., Breckler, S.J., Greenwald, A.G. (eds.), Attitude structure and function, Lawrence Erlbaum, Hillsdale, N.J.: 213-239. Schlag, B., Teubel, U. (1997). Public acceptability of transport pricing. IATSS research, 21(2): 134142. Sperling, D., Shaheen, S.A. (eds.) (1995). Transportation and energy: Strategies for a sustainable transportation system. American Council for an Energy-Efficient Economy, Washington, D. C. and Berkley, California. Steg, L., Vlek, C. (1997). The role of problem awareness in willingness-to-change car use and in evaluating relevant policy measures. In: Rothengatter, T., Vaya, E.C. (eds.), Traffic and transport psychology. Theory and application, Pergamon, New York: 1 ed. 465-475; chap. 448. Tertoolen, G., Kreveld, D.V., Verstraten, B. (1998). Psychological resistance against attempts to reduce private car use. Transportation Research-A, 32: 171-181. Thøgersen, J. (2006). Understanding repetitive travel mode choices in a stable context: A panel study approach. Transportation Research Part A: Policy and Practice, 40, 621-638, Thøgersen, J., Ölander, F. (2006). The dynamic interaction of personal norms and environmentfriendly buying behavior: A panel study. Journal of Applied Social Psychology, 36: 1758-1780 Trafikministeriet (1995). Transportsektorens miljøbelastning (rapport). Copenhagen: Trafikministeriet. Trafikministeriet (1999). Begrænsning af transportsektorens CO2 udslip. Trafikministeriet, Copenhagen. Triandis, H.C. (1977). Interpersonal behavior. Books/Cole, Monterey. Van Vugt, M., Van Lange, P.A.M., Meertens, R.M., Joireman, J.A. (1996). How a structural solution to a real-world social dilemma failed: A field experiment on the first carpool lane in Europe. Social Psychology Quarterly, 59: 364-374. Verplanken, B., Aarts, H. (1999). Habit, attitude, and planned behaviour: Is habit an empty construct or an interesting case of goal-directed automaticity? European Review of Social Psychology, 10: 101-134. Verplanken, B., Aarts, H., Knippenberg, A. v., Knippenberg, C. v. (1994). Attitude versus general habit: Antecedents of travel mode choice. Journal of Applied Social Psychology, 24: 285-300. Verplanken, B., Aarts, H., Knippenberg, A. v., Moonen, A. (1998). Habit versus planned behavior: A field experiment. British Journal of Social Psychology, 37: 111-128. Verplanken, B., Faess, S. (1999). Good intentions, bad habits, and effects on forming implementation intentions on healthy eating. European Journal of Social Psychology, 29: 591-604.

Car-Use Habits: An Obstacle to the Use of Public Transportation?

313

Verplanken, B., Orbell, S. (2003). Reflections on past behaviour: A self-report index of habit strength. Journal of Applied Social Psychology, 33: 1313-1330. Vlek, C., Hendrickx, L., Steg, L. (1993). A social dilemma analysis of motorised transport problems and six general strategies for social behaviour change. In: European conference of ministers of transport. ECMT Publications, Paris: 210-225. Vlek, C., Steg, L. (1996). Societal reasons, conditions and policy strategies for reducing the use of motor vehicles: a behavioural-science perspective and some empirical data. In: OECD (ed.). Towards sustainable transportation. OECD, Paris: 2-9.

16 Road Pricing in Denmark – User Attitudes and User Reactions Mai-Britt Herslund Centre for Traffic and Transport, Technical University of Denmark

Abstract Road pricing is often considered an appropriate way in which traffic can be better regulated. At the Technical University of Denmark, two research programmes on this topic have been carried out. The pre-study FORTRIN (a feasibility study) has been followed by a real life test, namely the AKTA Road Pricing Experiment in Copenhagen. Here, the city was equipped with virtual cordon rings and pricing zones and 500 test drivers had a GPS-based device installed in their cars testing different road-pricing schemes. This paper describes the technology as well as attitudes and reactions of road users in both studies.

16.1 Introduction The FORTRIN programme defined and specified a variable road-pricing system based on car type, number of kilometres driven, time and place. Furthermore, a prototype based on GPS technology was developed. Different tariff structures were analysed statistically and optimal structures pointed out. User attitudes and user reactions were studied using a tariff scenario based on traffic planning considerations. And traffic effects of the road-pricing system were analysed by means of model predictions. The FORTRIN programme was finished in May 2001 (Herslund et al., 2001; Herslund, 2001). The FORTRIN programme has now been followed by a large-scale field test, namely the AKTA Road Pricing Experiment in Greater Copenhagen. Here, car drivers were exposed to different road pricing schemes using a GPS-based technique by which trips are charged and logged (Herslund, 2003; Nielsen and Jovicic, 2003; Nielsen et al., 2003; Herslund. 2004). AKTA is the Danish part of the PROGRESS project (www.progressproject.org), which again is part of EU’s 5th Framework programme, “The Growth Programme on Sustainable Mobility and Intermodality”, which supports several projects concerning pricing (http://www.transport-pricing.net/). In PROGRESS, eight European cities assess in different ways impacts of different urban pricing schemes. The cities are Bristol and Edinburgh (UK), Genoa and Rome (I), Helsinki (SF), Trondheim (N), Gothenburg (S) and Copenhagen (DK). In the following, results from FORTRIN and AKTA will be presented focusing on user attitudes and user reactions towards road pricing. The methods

316

M.B. Herslund

used in the studies are different kinds of interviews, questionnaires, plus analyses of diaries and log-data.

16.2 User Studies of the FORTRIN Project The FORTRIN research programme on traffic informatics investigated the possibilities of better regulating traffic by means of a GPS-based road pricing system focused on car type, number of kilometres driven, time and place. Users' attitudes and reactions towards such a system were investigated by the following methods: x Focus-group discussions x Analyses by questionnaire x Calculation and modelling of specific trips entered in diary logs.

16.2.1 Pricing Strategy A Goal orientation Scenario was used in which the tariff structure was designed to reduce the traffic load in areas and at times of day where and when the driving has the most adverse consequences. The scheme operated with three road classes (primary traffic roads, secondary traffic roads and local roads) as well as three zones (rural area, city area and dense city area). This equalled a price structure with nine levels. Furthermore, a peak hour charge was added to the city and dense city zones (during the period 07.00-09.00 and 15.00-17.00 hours). The kilometre rates of the scenario varied from DKK 0.20 (0.03 EUR) to DKK 1.45 (0.20 EUR) per km. The scenario assumed that the Danish registration tax would remain while the annual weight duty/the green owner’s duty would be abolished (on average 2,900 DKK ~ 392 EUR per year, dependent on vehicle type). It should be mentioned here that the existing tax system in Denmark is structured with a lump sum tax when the car is acquired (only first owner) followed by an annual tax. Table 16.1 shows what car users in Denmark have to pay for a typical family car like a Ford Focus and the table also shows the annual tax and the price of one litre of petrol. Table 16.1. Price of new car plus annual tax and petrol price in Denmark Purchase Net price incl. duty Registration tax VAT (25%) Sales price

DKK 99.154 142.476 60.408 302.038

EUR 13.339 19.167 8.126 40.632

Annual tax Petrol price (1 litre)

2.900 8

392 1

Road Pricing in Denmark – User Attitudes and User Reactions

317

16.2.2 Test Population The selection criterion for participants of the FORTRIN study was that they owned a private car and had certain transport needs. The population was divided into six groups, based on the desire to study variations as regards attitude and choice of mode of transport that were dependent on where the respondents lived. The six groups were organised in pairs so that: x Two groups lived in a city area with ample access to public transport (Copenhagen, 0.6 million inhabitants, and Århus, 0.2 million inhabitants), x Two groups lived in a provincial towns (Næstved, 30,000 inhabitants and Herning, 35,000 inhabitants) x Two groups lived in a rural area with poor public transport coverage (Lolland and West Jutland). The test population matched the national average overall with respect to distribution by age and marital status.

16.2.3 Focus-Group Meetings A focus group can be defined as a group interview, centred on a specific topic and facilitated by a moderator (interviewer/facilitator). During the focus-group discussion, the interaction that occurs within the group generates qualitative data. The method is often used when new topics are to be studied – especially due to its opportunity for individual expression (Morgan, 1988; Kitzinger, 1995; Kvale, 1997). The focus-group meetings of FORTRIN took place ultimo year 2000. A question outline ensured that the focus of the group discussion was mildly controlled and that all the groups touched upon the same subjects, namely attitudes and possible reactions towards road pricing in their area. After the discussion, each participant was to weight some characteristic statements from the discussion. This weighting indicates the focus of the group in relation to the subject discussed. The material revealed both differences and similarities among the groups. Among the similarities can be mentioned that the principles behind road pricing were understandable enough for everyone and that most respondents considered the price differentiation between city and rural area as reasonable, whereas the same agreement on introducing a peak hour surcharge was not available. The price differentiation by road type was difficult to understand and accept for most participants. The uncertainty felt concerning the GPS technology was not generally as great as was the uncertainty connected with the possibility for the politicians to turn road pricing into a tax spiral. The difference between the six groups was most obvious when comparing the statements, which each respondent had weighted after the discussions. The three statements from each group, which were given the highest priority, appear in Table 16.2.

318

M.B. Herslund

Table 16.2. Statements in order of priority at the focus-group discussions (Nielsen and Herslund, 2002)

Lolllland

land West Jutla

Næstved

ing Hernin

Århus

Copenhagen

City

Type of area

Statement with 1st priority

Statement with 2nd priority

Statement with 3rd priority

Metropolitan area. Significant congestion in peak hours. Efficient buses and urban rail for some journeys Urban, with little congestion. Mainly bus as public transport

Variable road pricing will turn into an extra tax

Tolls are preferable to variable charges

Variable road user charges will redistribute traffic – spread traffic more in time

The price becomes confusing

The system of variable road pricing will not induce any change in behaviour Families with small children are hit hard

It is a new tax element

Provincial town. No congestion. Poor bus system

Road pricing will be an extra tax mechanism for the Government

Town. No congestion. Poor bus system. Some commuters to Copenhagen

Cheap and efficient transport will change people's transport habits

The number of cars will not be reduced by introducing variable road user charges

Variable charges should be viewed in addition to the present ones in order to shift traffic

Rural area. No congestion. Slow inefficient rural buses

The principle of paying for using a car is fair

It is commendable to pay for using the car rather than a fixed charge

Environmental considerations require action

Rural area. No congestion. Slow inefficient rural buses

The principle of paying for using a car is fair

It is commendable to pay for using the car rather than a fixed charge

Environmental considerations require action

Fixed road user charges are preferable to variable ones

All the weighted statements from the focus-group discussions can be roughly divided into three main categories (one category for statements on tax, one category dealing with behavioural statements and a third category dealing with all other statements). Fig. 16.1 illustrates the share of statements within each of the three categories.

Road Pricing in Denmark – User Attitudes and User Reactions

319

100% 80% 60% 40% 20% 0% Århus

København

Herning

Næstved

Vestjylland

Lolland

Fig. 16.1. Shares of statements within each category by city (black = statements on tax, grey = statements on behaviour and white = all other statements)

This analysis clearly illustrates that the tax/duty dimension generally had been focused on the least during the group discussions, while the group "other" (dealing, for instance, with security/surveillance/fairness/clarity as well as management) had been focused on the most. The difference in focus during the discussions revealed that the groups from the rural areas were most interested in fairness and environmental problems in relation to variable road pricing, while the groups from the big cities were most interested in the tax dimension of the duty system. The respondents from Århus appeared overall to be the group which was most negative towards variable road pricing, which is also seen in Table 16.2.

16.2.4 The Study by Questionnaires All respondents in the focus groups also answered questionnaires. Again, it appeared that price differentiation by zone and time was assumed to be the most powerful element in the Goal orientation Scenario. The greatest effect would be expected from price differentiation by zone, where it becomes expensive to drive in dense city areas with a lot of traffic and cheap to drive in the open countryside with little traffic. However, there was a tendency to expect that an extra charge on private driving would influence others more than oneself (Table 16.3). This finding is consistent with the general finding that people perceive themselves to be less able, but more motivated than the general population to act in a pro-social way, i.e. that others’ pro-social behaviour is primarily restricted by lack of motivation, but own behaviour is primarily restricted by lack of ability (Herslund et al., 2001; Herslund, 2004). Table 16.3. Expectation of others and own behaviour in relation to price differentiation by zone

Yes No Indifferent

Will taxes on driving in big city areas reduce private driving in the big cities? 47% 36% 17%

Yes No Indifferent

Will taxes on driving in big city areas make you drive less in the big cities? 34% 50% 16%

320

M.B. Herslund

As regards the effect of a peak hour surcharge, Table 16.4 shows the same tendency, other drivers were expected to be more influenced by the pricing scheme than the one being asked. Table 16.4. Expectation of others and own behaviour in respect of peak hour charging

Yes No Indifferent

Will a high duty in rush hours make drivers move the trips to other times of the day? 44% 48% 8%

Will a high duty in rush hours make you drive at other times of the day? Yes No Indifferent

36% 59% 5%

16.2.5 Trip Logs Thirty-one of the participants registered 113 trips in private logs with up to nine trips a day. Each trip had a specific starting and end point. Thus, transport to and from work was to be logged as two trips. The trips were mainly made outside peak hours (57%) and mainly with destinations such as home (44%), workplace (24%), or leisure activity (11%). The individual trips were then priced in relation to the tariffs of the Goal orientation Scenario and the logs were returned to the individual respondent with an indication of the price of each trip, total price for all trips of the day and price if the trip was made at some other time of the day (in the peak hour or outside the peak hour). The respondents then filled in a comprehensive questionnaire about each trip in relation to the price level experienced, prioritisation of trips and the possibility of shifting the trip in time. The answers show that if the road pricing of the trips was as calculated then: x Expenses for private driving would increase x 11% of the trips would be shifted to cheaper routes x 2% of the trips would be shifted in time x 4% of the trips would be cancelled x Road pricing would have only little impact on private driving to and from workplace.

16.2.6 Conclusions of the FORTRIN Study Five main conclusions have been drawn from the FORTRIN study: x Road pricing was regarded as being more fair than the present tax system in Denmark, where the fixed charges are up to 180% registration tax and 25% VAT when you buy a new car. On top of that, you pay an annual tax of an average of 2,900 DKK (~ 392 EUR) – varying by car type and energy use (see Table 16.1)

Road Pricing in Denmark – User Attitudes and User Reactions

x x x x

321

The GPS system provoked no great fear of surveillance. The greatest fear of the participants was that politicians will turn variable road pricing into a tax spiral There seemed to be differences in attitudes towards road pricing depending on where the respondents reside – or rather depending on the experienced congestion There was the greatest effect on the driving pattern when the price differentiation was based on zones and time. But road pricing was believed to influence other drivers more than oneself Participants found that the tariffs of the road-pricing scenario would increase expenses, but still it had only little impact on private driving in the short run, because the price level was too low to significantly change behaviour.

16.3 AKTA – the Road Pricing Study of Copenhagen As opposed to the FORTRIN pre-study, AKTA made a real life test of whether road user taxes will change travel behaviour.

16.3.1 Design of the Study The city of Copenhagen was equipped with virtual cordon rings and pricing zones from October 2001 to May 2003. Five hundred voluntary test drivers had a vehicle position system installed in their own car, making it possible for them to read the virtual pricing systems on a display. All car movements were logged in the system and a price calculated for every trip. At the end of the experiment, the test drivers got paid according to the difference in driving behaviour between the control period (which in some cases was estimated) and the test period. The selection of test drivers was based on a stratified factorial design. However, it turned out to be far more complicated to recruit the participants than anticipated: A total of nearly 26,000 people had to be contacted in order to get a proper sample of 2x200 and 1x100 households (1/3 of those who initially agreed to participate). The participants were distributed in a factorial design among income groups, commuting patterns (residence of home and work) and pricing schemes. All participants belonged to one-car families, all participants resided and/or had their workplace within the road pricing zones and all had a daily need for transport, (Nielsen and Herslund, 2002; Nielsen and Jovicic, 2003).

16.3.2 Different User Studies The participants filled out a questionnaire before the test drive and afterwards they filled out another questionnaire. In addition, they participated in a telephone inter-

322

M.B. Herslund

view which, among other things, was designed to test whether they changed attitudes towards road pricing during the test drive. Furthermore, before the test drive 300 participants had participated in a stated preference experiment consisting of questions concerning valuation of cost and length of different trips, stated preference on derivation of value of time, stated preference on time-of-day decisions and stated preference on route changes. And selected test drivers also participated in qualitative focus-group interviews. For comparison, a telephone interview with 1015 respondents living in the road pricing area was carried out before the field test in order to get a benchmark of ordinary people’s awareness and attitudes towards road pricing. All qualitative and quantitative data have now been processed and selected results are presented in the following.

16.3.3 Results from Telephone Interviews A survey based on 1015 telephone interviews was conducted in the autumn of 2001 (before starting the field test) of randomly selected residents in the capital area (Sørensen, 2003; Sørensen, 2004). The interview was structured as 20 questions related to socioeconomics and 28 questions related to non-socioeconomics (attitudes towards road pricing, tax, etc.). The respondents can be described as being geographically distributed over the capital area with one-quarter in the suburbs and the majority in the central city area. The distance to public transport was below 500 metres for most of the respondents (nine out of 10) and the public transport service level was at least three connections per hour for virtually all respondents. Half of the respondents were aged between 20 and 40 and one-quarter of the residents had a gross household income in the uppermost income category. Half of the respondents were car owners; most of these were one-car owners (45%), 8% owned two or more cars. Half of the respondents were aware of road pricing systems – dominated by female respondents (60%, 40% for male) and dominated by car ownership (60%, 40% for non-owners) and increasing by age. The rate of awareness also increased by income level, though it was not separated from education level. The understanding of road pricing systems was mainly segregated between payment for specific roads/time of day (50%), specific roads (20%) and entering a city (10%). However, it was explained to half of the population who were not aware of road user charging that road pricing is payment for use of specific roads/time of day. 22% of the interviewed persons had heard of the AKTA road pricing project in Copenhagen. The awareness of the project was slightly higher among males than females, but independent of income. Four out of five respondents had actual experience with toll payment; the fraction rising with income level. The tax system in Denmark with fixed car tax (see Table 16.1) was appreciated by 38% of the respondents and disliked by 43%, where females were

Road Pricing in Denmark – User Attitudes and User Reactions

323

slightly more in favour of the system while car owners slightly more disfavoured the system. The variable road pricing system was favoured by two-thirds (65%) of the respondents irrespective of gender, but car owners were less in favour (58%) than non-owners (72%). The favouring declined by age, but increased by frequency of the public transport service. The assessment is unaffected by mileage and income. The enthusiasm for road pricing was not fully maintained when an additional peak hour fee was in question; the fee was not further quantified. Half of the respondents supported this idea, again dominated by non-car owners (60%) over car owners (45%). Interestingly, respondents with and without flexible working hours are equally supportive. However, more respondents with flexible working hours (45%) are against than respondents without flexible working hours (38%)! As for the variable road pricing, the enthusiasm declined with income level. The belief in the effect of a peak hour fee on other car users is 40% though only 35% expect they may change their own behaviour – the latter covers geographical differences. The figures were unaffected by whether the respondent had flexible working hours. Questioned on the peak hour kilometre rate for change of travel time, one in four respondents answered the lowest option (1 DKK/km) and one in three the highest rate (5+DKK/km) without variation by gender (rate ranging from 0.13-0.67 EUR). The peak hour kilometre rate before mode change is on average higher, which covers that 20% of the respondents intended to change by 1 DKK/km, 35% do not intend to switch mode before the cost is over 5 DKK/km. Of the respondents, 53% believed that a zone-based fee is good. Car owners are more in opposition (45%) than non-owners. There are only minor deviances by gender. Of the respondents, 57% believed that a zone-based system would reduce the car use though only 46% believed they would reduce their own driving. Again, there were only minor deviances by gender. 28% of the respondents would alter their time of travel if a cordon fee of 4 DKK (lowest level) to the central city was introduced, 35% stated the cordon fee should be at least 12 DKK (highest rate). This is largely independent of gender. The pattern is similar for change of mode (fee ranging from 0.54 to 1.61 EUR). The general opinion is that the revenue from road user charging should be used to improve the public transport (extend service and reduce prices) whereas improvements of the road network have lowest priority. The majority of the respondents (58%) is not willing to accept higher taxes for road use than today; males (65%) are more in opposition than females (52%). The respondents did largely agree that the variable road pricing will diminish the traffic in city areas (62%), that road pricing is fairer than the existing system (68%) and that the environment will benefit from it (57%). A fairly high fraction of the same respondents did also agree that car users in general would have to pay more than today for car use (52%), that prices will be less transparent (58%) and an additional level of bureaucracy will emerge (63%) if a road pricing system is introduced.

324

M.B. Herslund

More than 50% of the respondents believed that variable road pricing will reduce traffic in Copenhagen and reduce the pollution related to traffic. The respondents are generally sceptical with regard to the public spending of the revenue from road pricing, as only one-third believed the revenue would be used to improve the public transport service. Hence, the general trends based on telephone interviews with 1015 respondents are that more than half of the respondents are in favour of variable road pricing as an alternative to the existing tax system. Two out of three were supportive of road pricing and slightly more than half supportive of peak hour fee and zone-based fee. Females were generally more supportive than males. However (as seen in the FORTRIN study), road pricing is thought to influence other car drivers more than oneself.

16.3.4 Pricing Schemes Three different pricing schemes were tested in the three rounds of the AKTA field study: Two of the schemes were zone-based with four different levels per km, with the most expensive in the inner city and the cheapest in the suburbs. The third scheme was a toll-based system with payment for zone crossings. The pricing varied between the peak and non-peak hours in all scenarios (see Fig. 16.2).

Fig. 16.2. The AKTA zone structure. 1 DKK = 0.13 EUR (Nielsen and Herslund, 2002)

Road Pricing in Denmark – User Attitudes and User Reactions

325

The GPS device calculated dynamically the pricing level. And for a given trip, the participant could see the pricing level (zone), be notified about zonal-crossing (at the cordon scheme) and read the accumulated cost of the trip (see Fig. 16.3).

Fig. 16.3. On Board Unit in a private car (Herslund, 2003)

Selected results from the field test are presented in the following. The high km-based pricing level (0.07 – 0.67 EUR) clearly made an impression on the participants. Even if they could not change behaviour, they examined alternative travel options before rejecting them. The participants considered changing route, mode and “occasional” trips. About 50% of the test drivers changed behaviour in some way. The main changes were new routes and for “occasional” trips new destinations, time of day (to non-peak) and to some extent fewer trips. Commuting trips were supposed to be difficult to change, e.g. shifting away from the peak hour, working at home (telecommuting) or using another mode (bicycle or public transport). The low km-based pricing level (0.07 – 0.34 EUR) was in general not sufficiently high to change behaviour, although a few participants did some minor changes when it was easy. Most participants believed that it was by chance if they travelled more or less in the low pricing period than the control period, e.g. that they had more visits to Central Copenhagen or a vacation in one of the two periods than the other. The km-based schemes were in general considered more fair than the cordonbased (from 0.13 to 1.61 EUR per crossing). The participants had nonetheless more difficulties understanding these than the cordon-based system. It is interesting to note that the more fair and economically justified the schemes, the more difficulties are experienced understanding it. Nonetheless, the km-based scheme turned

326

M.B. Herslund

out to be the most effective in terms of behavioural changes. Hence it can be concluded that x drivers testing the high km-tolls made the most changes in travel behaviour, while drivers testing the low tolls made very few changes. About 50% of the test drivers changed behaviour in some way and the main changes were: x Choice of different (cheaper) routes – mostly for non-commuting trips x Increased use of Kiss-and-ride facilities (more passengers in the test cars, these passengers choosing private and not public transport) x Changes in destination and time of day for non-commuting trips. But log data from the test period show almost no changes regarding: x Cancelling of trips x Mode choice in favour of public transport x Commuting x Shifting from peak to non-peak hours. So – commuting seems to be difficult to change and there is some inertia in changing route, mode and time for the trip.

16.3.5 Focus Groups Two focus-group interviews were carried out after the experimental rounds in AKTA. The focus-group participants were selected among drivers after the test drive. It turned out to be a little difficult to recruit participants, since the test drivers had already spent a lot of time on the experiment (Nielsen and Herslund, 2002). Ten households were represented from the second experimental round and 13 households from the third round. Both focus-group meetings were structured like the ones of FORTRIN (see table 16.2) based on a list of questions with sufficient time for free discussion and any comment from the participants. During the discussion, the moderator noticed the most characteristic subjects that were discussed and 16 of these statements were selected. Each participant was then asked to rank the five most important of these on a scale from five to one, based on how much he/she felt they had marked the discussion. The main findings of the focus-group interviews were: x The high km-based pricing scheme made a strong impression on the participants, whereas the low level did not lead to changes. x Road pricing is felt to be able to change route, mode and leisure trips, but not lead to drastic decisions (to move, sell the car, etc.). x Road pricing is regarded fairer than the present tax system in Denmark. However, it was feared that road pricing would just be an additional tax and that the revenue would not be used to improve the infrastructure.

Road Pricing in Denmark – User Attitudes and User Reactions

x x x x x

327

Participants felt that revenues should be used for infrastructure improvements in the Copenhagen region, especially to improve public transport in the areas with poor services. The GPS system provoked no great fear of surveillance. The greatest fear of the participants was that politicians would turn variable road pricing into a tax spiral. There seemed to be differences in attitudes towards road pricing dependent on where the respondents reside. The greatest effect on the driving was found when the price differentiation is based on zones and time, but road pricing was believed to influence other drivers more than the one being asked. Participants found that the tariffs of the road-pricing scenario will increase expenses, but still only have only little impact on private driving, because the price level is too low to significantly change behaviour.

It is interesting to note that the participants focussed mainly on the changes in traffic behaviour, rather than political issues. Even with the high charging level, the participants stated that they would not consider drastic decisions such as selling the car, using public transport to work, moving to another address or changing job. They were aware that they could be forced to take such decisions with a sufficiently high level of pricing, but it should be higher than in the AKTA project. As fixed car taxes are extremely high in Denmark, the marginal costs are relatively less important – even at the high pricing level. Surveillance was not really a big issue – some of the participants even stated that one can be followed by mobile telephone, bank account, video cameras on the streets, etc. and that a GPS-based system is no worse. This is quite the opposite to what role surveillance has in the press and by politicians, which, however, does not seem in line with the view of the population.

16.3.6 Key Results from Questionnaires All test drivers participated in ex-ante and ex-post questionnaires. In line with the results from focus-group interviews, these questionnaires show that road pricing in some way does affect behaviour; it is not considered as another fixed cost, but as a marginal cost that drivers respond to. The responses are even higher than if they are only considered as a marginal cost with the same willingness to pay. Again, in this part-study the participants' attitudes to road pricing were less emotional than expected (especially considering the debate in the Danish press). Half of the respondents found that the existing tax system in Denmark should be replaced by a system where you pay for the use of the car. Most participants did not consider surveillance as a problem (that the cars can be tracked by the logged co-ordinates). The possibility to control speed violations was neither considered important. The participants disagreed on whether road pricing is fair or not, including whether society becomes more class-divided between people that can pay or not and for people who cannot change behaviour (e.g. with fixed meeting

328

M.B. Herslund

hours or with children in school and day care). Preferred use of revenue seems to be improving public transport (better coverage and cheaper fares, stated by 65% of the respondents) and traffic safety (stated by 55%). Participants were asked about whether they thought they changed their attitudes towards road pricing just as changes were tracked by comparisons of before and after responses to identical questions. Almost half of the participants acknowledged they did change attitude to some extent. Approximately one-third of the participants changed attitude towards either the existing tax structure or the principle of road pricing, every fifth changed attitude towards both. Close to two-thirds changed attitude to the road pricing principle or the effect of a peak hour factor on general traffic. More than every fifth participant was indifferent towards variable road pricing though positive towards the effect of a peak hour factor. A similar pattern can be seen for the participants’ expectations on their own driving behaviour. More than every fifth participant expects it to have an impact on his own driving combined with indifferent attitude towards the existing system. Crossing the changes for the two peak hour factors, it can be seen that as many as 65% changed their attitude in the same direction for both. More than every fourth changes attitudes such that he/she was more positive towards the effect on traffic in general than on his/her own driving. Only 6% of the participants did not change attitude towards either of the tax structures. After the trial, more than half of the participants thought the present tax system was bad, irrespective of which toll schemes they tested. Furthermore, onethird of the participants thought of the present system as a bad system and of variable road pricing as a good system. Only one in 20 thought the present system was good and road pricing bad. Positive expectations of a peak hour factor on traffic in general are most outspoken for participants considering the present system as bad (about every fourth participant). Close to half of the respondents thought of the variable road pricing system as good combined with their belief of that a peak hour factor would affect traffic in general. Close to every third participant thought the road pricing system was good and expected to reduce his/her own driving provided it was implemented. One-third of the opponents expected to alter his/her driving if implemented. There seemed to be a connection between participants expecting to change behaviour themselves and expectations to changes in general if a peak hour factor was implemented. Three of five expected changed behaviour by themselves and in general. More than one in four expected no change by themselves nor in general. Furthermore, there seems to be some correlation between expectations for a peak hour factor and whether the participant expects his/her own driving to be affected. More than one in four responded to some or full extent to both, while two in five responded negatively to both.

Road Pricing in Denmark – User Attitudes and User Reactions

329

16.4 Overall Conclusions of the Two User Studies The pricing level needs to be high if road users are to change behaviour (between 1 and 5 DKK/km in the peak hours and half price in off-peak hours. 1 DKK equals 0.13 EUR). But still only few car drivers are ready to change behaviour. Road pricing is expected to influence others the most and time plus route for commuting trips are difficult to change. Congestion pricing seems to be the most unpopular part of the road-pricing concept. High pricing of private cars in city centres is ok, but an additional rush hour charge is less acceptable. And pricing should be nationwide, not only a variable tax for driving private cars in big cities. Interesting findings of both the FORTRIN and AKTA studies are that attitudes towards road pricing seem less emotional than expected. Fear of surveillance is not a big problem and the possibility to control speed violations by the GPS-based system is not considered important. Road users’ greatest fear seems to be that the government will turn road pricing into a hidden tax and that the revenue will be used for something other than to improve traffic conditions. The two Danish programmes demonstrate that it is possible to develop a road pricing system which charges a kilometre rate based on time, place and distance driven. It is also demonstrated that road user charging can, to some extent, affect the driving pattern. The GPS-based system was widely understood and accepted and a majority of road users in Denmark finds variable taxes on car driving more acceptable than fixed ones. Yet payment, security and control functions have not been described in such detail that it is assured that this is practicable, (Nielsen et al., 2003). A test of the technical design of these systems was not contained in the programmes. As these functions are vital to the feasibility of a road pricing system, there is a need for further technical development.

References Herslund, M.B. (2001). A Distance Dependant Road pricing Scheme. Concept and User Reactions. Paper presented at the 8th World Congress on Intelligent Transport Systems: ITS Based Travel Demand Management, Sydney, Australia, 30 September – 04 October. Herslund, M.B. (2003). Road pricing in Denmark. User Attitudes and User Reactions. Paper presented at the 10th. World Congress on Intelligent Transport Systems. Madrid, Spain, 16 – 20 November, paper No. 2140T. Herslund, M.B. (2004). Road User Charging in Denmark – acceptance and impact of three different charging schemes. Paper presented at the ITS Congress in Europe, Budapest, May 2004, paper no.2821. Herslund, M.B., Ildensborg-Hansen, J., Jørgensen L., Kildebogaard, J. (2001). The FORTRIN Programme. A Distance Dependent Road Pricing System – Main Report. Centre for Traffic and Transport, Technical University of Denmark, Lyngby.

330

M.B. Herslund

Kitzinger, J. (1995). Introducing Focus Groups. British Medical Journal, 311: 299-302. Kvale, S. (1997). Interview (In Danish). Hans Reitzels Forlag a/s, Copenhagen. Morgan, D.L. (1998). Focus Groups as Qualitative Research. Qualitative Research Series, vol. 16, Sage Publications. Nielsen, O.A. and Herslund, M.B. (2002). The AKTA Road Pricing Experiment in Copenhagen. European Transport Conference (PTRC). Seminar on Investment in Roads. CDROM with proceedings, PTRC. Cambridge, September. Nielsen, O.A., Jovicic, G (2003). The AKTA road pricing experiment in Copenhagen. Conference paper. Session II: Valuation/Pricing. Moving through nets: The physical and social dimensions of travel. 10th International Conference on Travel Behaviour Research. Lucerne, 10-15. August 2003. Nielsen, O.A., Kristensen, J.P., Würtz, C. (2003). Using GPS for road pricing – experiences from Copenhagen. Paper presented at the 10th. World Congress on Intelligent Transport Systems. Madrid, Spain, 16 – 20 November, paper No. 2131T. Sørensen, M.V. (2003). The Greater Copenhagen Area. Technical University of Denmark, Centre for Traffic and Transport. Sørensen, M.V. (2004). Summary of Telephone Interviews on Variable Road Pricing in Copenhagen. Technical University of Denmark, Centre for Traffic and Transport.

17 A Cost-Minimisation Principle of Adaptation of Private Car Use in Response to Road Pricing Schemes Peter Loukopoulos1, Tommy Gärling2, Cecilia Jakobsson3 and Satoshi Fujii4 1

The Swedish National Road and Transport Research Institute, Linköping Göteborg University, Sweden 3 Göteborg University, Sweden 4 The Tokyo Institute of Technology, Japan 2

Abstract In this chapter, a theoretical framework is proposed with the aim of understanding reduction or changes in private car use in response to road pricing. It is argued that economic disincentives may activate car-use reduction or change goals in individuals and households. However, for car-use reduction or change goals to occur, other travel demand management measures are needed that make alternative travel options attractive. A review and classification of these other measures is provided followed by an assessment of their potential effectiveness.

17.1 Introduction Substantial environmental and societal costs of private car use such as congestion, noise, air pollution and depletion of energy resources, are expected future conesquences of the worldwide increasing trend in car ownership and use (Goodwin, 1996; Greene and Wegener, 1997). In many urban areas, these consequences are in fact already urgent problems that need to be solved. This has resulted in proposals for a number of policy measures targeted at reducing or changing private car use. These will be referred to as travel demand management (TDM) measures (Kitamura et al., 1997; Pas, 1995), although other terms are often used such as transport system management or transportation control measures (Pendyala et al., 1997), transportation demand management (Litman, 2003) and mobility management (Kristensen and Marshall, 1999; Litman, 2003; Rye, 2002). In the next section, we briefly discuss market-based TDM measures including the focus of this chapter, road pricing. We then discuss other types of TDM measures for reducing private car use. Our argument is that an evaluation of the effectiveness of road pricing should be made relative to other TDM measures. Furthermore, evaluation of the effectiveness of TDM measures in reducing private car use need to be based on realistic behavioural assumptions. In the third section, we propose a theoretical framework drawing on such assumptions.

332

Peter Loukopoulos et al.

17.2 Market-Based Travel Demand Management (TDM) Measures Examples of market-based TDM measures include road and congestion pricing (Banister, 2003; Foo, 1997, 1998, 2000; Goh, 2002), kilometre charges (Ubbels et al., 2002), fuel taxes, parking charges (Meyer, 1999), and public transport discounts and travel vouchers (Root, 2001). Three main purposes have been identified (Lindberg, 1995): (1) finance of maintenance of old infrastructure and investment in new; (2) abatement of adverse environmental effects by reducing car use; and (3) reduction of congestion in urban areas. Market-based TDM measures are theoretically founded in classical economics: behaviour is regulated by the principle of supply-and-demand and explicit cost-benefit analysis, such that if the price of a product (i.e. transportation) increases, the demand for this product will decrease, and vice versa. Using congestion pricing as an example, by raising the price level, the congestion externality is internalised (Emmerink et al., 1995; Rothengatter, 2003). Marketbased measures have become increasingly popular in recent years, particularly amongst politicians who appear to have embraced a new competition paradigm emphasising less governmental control. In line with this, many types of regulation have been abandoned as being outdated or unnecessarily restrictive. Although there are reasons to regulate transportation to maintain quality, reliability and safety, it is believed that unnecessary regulation can be reduced, and regulation objectives can be changed to address specific problems while encouraging competition, innovation and diversity (Klein et al., 1996, 1997). Increasing popularity amongst politicians has, however, been matched by an increasing scepticism amongst researchers. For road pricing to be viewed as a firstbest instrument for tackling the problems of car use, certain requirements must be fulfilled, two of which are that (i) households and individuals maximise utility, and (ii) full information is available on all costs involved (Emmerink et al., 1995). Yet, as discussed below, the habitual nature of car use with the consequence of limiting predecisional information search renders the second assumption suspect. The assumption of (expected) utility maximisation has also been found to be violated in numerous studies (Dawes, 1998; McFadden, 1999). This particular issue will also be addressed below. Furthermore, independent lines of empirical research have revealed low price elasticities to be associated with various pricing policies (at least in the short term, although some higher elasticities have been identified in the long term) (Hensher and King, 2001; Schuitema, 2003; Sipes and Mendelsohn, 2001). Jakobsson (2004) summarises the mainly negative outcomes of a limited number of field experiments conducted with the purpose of evaluating the effectiveness of road pricing. The popularity of market-based measures such as road pricing nevertheless continues to grow amongst politicians who also see the potential for such measures to yield additional revenues, which can be used for other environmentally-friendly transport modes or for other services such as health and education (Johansson et al., 2003; Odeck and Bråthen, 2002; Rajé, 2003).

A Cost-Minimisation Principle of Adaptation of Private Car Use

333

In this chapter, we are primarily concerned with policy measures targeting the reduction of car use. Market-based TDM measures may accomplish this, though they are not the only measures, and perhaps not the most effective. There is also a need for mixing different policy measures.

17.3 Classification of Travel Demand Management (TDM) Measures There are, in fact, many other measures that may reduce the levels of car-use related congestion, noise and air pollution in urban areas. Some of them (e.g. increased capacity of road infrastructure or cleaner cars) do not involve a reduction in car use. A general assessment of the current situation is, however, that measures that manage car-use demand must be implemented (e.g. Hensher, 1998). According to this assessment, it is necessary to both reduce car use and to change it with respect to when and where car owners drive, particularly on major commuter routes during peak hours and in city centres. Litman (2003, p. 245) notes that TDM is “a general term for strategies and programmes that encourage more efficient use of transport resources (road and parking space, vehicle capacity, funding, energy, etc)”. This broad definition reflects the historical progression in TDM measures away from merely forcing people to reduce their use of the private car by changing modes of transport or by driving less. As Taylor and Ampt (2003) note, TDM measures now encompass any initiative with the objective of reducing the negative impact of the car. The definition also reflects the broader evolutionary changes in policy measures, which have progressed from the 1960s when increasing infrastructure capacity to alleviate traffic problems such as congestion was commonplace, to the 1970s where the emphasis was on improving the management of the existing infrastructure, to the 1980s and beyond when policies began to target altering travel behaviour (Bovy and Salomon, 2002; Pas, 1995). An even more recent manifestation is the attempt to change human values and mobility culture, as has occurred in, for example, some parts of Switzerland with local administrations marketing a slower lifestyle and better image for public transport (City of Zurich, 2002). There are several classifications of TDM measures. Litman (2003) distinguishes five classes: improvements in transport options; provision of incentives to switch mode; land-use management; policy and planning reforms; and support programmes. A partly overlapping set is proposed by May et al. (2003) as land use policies, infrastructure provision (for modes other than the private car), management and regulation, information provision, attitudinal and behavioural change measures and pricing. Vlek and Michon (1992) suggest that the following TDM measures are feasible ways of implementing car-use reduction policies: physical changes such as, for instance, closure for car traffic and/or providing alternative transportation; legislation; economic incentives and disincentives; information, education and persuasion; socialisation and social modelling targeted at changing social norms; and institutional and organisational changes such as, for instance,

334

Peter Loukopoulos et al.

flexible working hours, telecommuting, or “flexplaces.” Louw et al. (1998) argue that car use is influenced by policies encouraging mode switching, destination switching, changing time of travel, linking trips, substitution of trips with technology (e.g. teleworking), and substitution of trips through trip modification (e.g. a single goods delivery instead of a series of shoppers’ trips). Gatersleben (2003) distinguishes between measures aimed at changing behavioural opportunities and measures aimed at changing norms, motivations and perceptions. Partly based on the different systems of classification listed above and partly on the basis of relevance to the theoretical framework presented in the next section, we propose that TDM measures vary on a number of important dimensions including coerciveness, top-down vs. bottom-up processes, spatial scale, time scale, market-based vs. regulatory mechanisms, and impacting latent or manifest travel demand.

17.3.1 Coerciveness TDM measures vary in terms of whether the change is voluntary and within the control of car users or whether the change is forced upon them. For instance, public transport improvements or information campaigns are non-coercive TDM measures since the decision to reduce car use is left to the individual. On the other hand, TDM measures such as road closures and prohibition within city centres are highly coercive as households have no choice but to reduce car use within the designated areas. The degree of coerciveness of other TDM measures such as road pricing or parking fees depends on household income. Jones (2003), Schuitema (2003), Steg and Vlek, (1997), Stradling et al. (2000) and Thorpe et al. (2000) make a distinction between push and pull measures. Push measures discourage car use (or car ownership) by making it less attractive. Coercive measures tend to be classified as push measures. Pull measures encourage the use of alternative modes to the car by making such modes more attractive. Non-coercive measures such as cheaper public transport, new cycle lanes, or even car pooling subsidies are classified as pull measures. There is a close correspondence between push and pull measures and between TDM measures encouraging attitude change 1 and those forcing behavioural change. For example, the idea underpinning the Individualised Marketing Program (see Chapter 14) is that a lot of opposition to public transport is due to a lack of information and motivation. By bringing such information to the individual and by showing when public transport is beneficial, the expectation is that attitudes will change such that individuals become more willing to reduce the use of the car for certain trips and purposes. Forced change of behaviour, on the other hand, disregards attitudes and imposes restrictions. Forced changes may have negative side effects outweighing the expected benefits (Gärling et al., 2002b), such as costs or sacrifices that households will not 1

See Eagly and Chaiken (1993, 1998) for comprehensive reviews of research on attitude formation and change.

A Cost-Minimisation Principle of Adaptation of Private Car Use

335

accept as well as potential negative health impacts arising from increases in time pressure and stress. Additionally, the non-coercive strategies (attitude change or pull measures) may be based on untenable assumptions about how much households are willing and able to change their car use, particularly in the light of the fact that car use is frequently habitual (see Chapter 13). Such habits interfere with information search and processing of information, so that alternatives to the car remain unknown. There appears to be a consensus that a mix of measures needs to be introduced consisting of coercive measures that break habits (for example, by making car use no longer possible or prohibitively expensive) and non-coercive measures (such as increased public transport services or new routes) encouraging the use of other modes (Meyer, 1999).

17.3.2 Top-Down vs. Bottom-Up Processes Taylor and Ampt (2003) distinguish traditional top-down approaches telling people what to do from bottom-up approaches allowing people freedom in choosing to change their car use. Bottom-up approaches are referred to as voluntary travel change. Although the distinction between these and pull measures is not clear-cut, the goal of the former is to empower people to change as opposed to expecting or forcing a response to external stimuli or pressures (coercive or otherwise). A key principle is that each individual in a household defines his or her own goals in accordance with his or her own needs and existing lifestyle. This is why change must be initiated as part of a bottom-up process with households deciding whether or not they wish to participate (Rose and Ampt, 2001; Taylor and Ampt, 2003).

17.3.3 Time Scale A somewhat forgotten aspect is the variation in time scale required for both the implementation of TDM measures and for responses to various TDM measures. Goodwin (1998) notes that while many effects can be realised immediately the cumulative effects of policies may not be felt for many years. For example, prioritising public transport and new fares policies have impacts on demand within the first year of implementation, with longer-term elasticities being twice as great after 5-10 years. Road pricing schemes and changes in the costs of petrol are argued to yield small responses within the first year with a build-up of effects on car use (and car ownership) over the next 5-10 years or more. Perhaps the most obvious example of a TDM measure with long-term consequences is that of land use planning. Short-term effects of land use planning to reduce travel distances are small, with larger cumulative effects being felt only after 20 years. Related to land use planning, pedestrianisation is claimed to have an immediate effect on car traffic, but pedestrians and retailers on the other hand, are likely to require several years to adjust to changes in the situation (Goodwin, 1998). The temporal nature of a TDM measure also relates to its operational specifications. Congestion pricing, for example, typically operates during peak periods or

336

Peter Loukopoulos et al.

during the day, but not in the evening or at the weekend. Prohibition measures can also operate in a similar fashion (Cambridgeshire County Council, 2005), although most road closures tend to be permanent. Furthermore, TDM measures may affect the temporal nature of the activity per se. For example, work hour management strategies attempt to influence vehicle trip demand by reducing it or by shifting it to less-congested time periods (e.g. flexible working hours, staggered working hours, modified work schedules such as a four-day week, and telecommuting services) (Golob, 2001). Such strategies have been shown to be effective, but it is also known that any savings often generate new, longer trips for non-commuting purposes such as leisure activities or shopping/maintenance activities, many of which were previously linked via trip chains to the commuting trip.

17.3.4 Spatial Scale The scope of influence of TDM measures varies from the local to the national. Differentiated road pricing or congestion charging, for example, is a local initiative aimed at easing traffic flows in urban areas and improving local air pollution levels, as well as improving the quality of life in urban areas (Banister, 2003; Foo, 1997, 1998, 2000; Goh, 2002). Road closures and pedestrianisation measures are also local initiatives. An example of a TDM measure with a large spatial sphere of influence is a proposed kilometre charge (Ubbels et al., 2002). A further example is that of public transport discounts for certain groups such as pensioners or the unemployed. TDM measures may be initiated nationally but have local impacts. An example is legislation requiring employers to implement strategies to reduce transportation impacts of employees, suppliers, visitors and customers (Enoch and Potter, 2003; Rye, 2002). Such strategies vary from employer to employer and may include car-pooling schemes, coordination of specific transport routes with local public transport providers, and parking restrictions. An alternative conception of the spatial reach of TDM measures is whether or not they target the origin or destination of trips. For example, it is possible in a monocentric city to make car use less attractive by means of traffic calming and access restrictions in both the city centre (a common destination) and residential areas (the typical origin) (Louw and Maat, 1999).

17.3.5 Market-Based vs. Regulatory Mechanisms TDM measures also vary in terms of whether they can be classified as marketbased (e.g. road pricing based on pricing mechanisms) or regulatory (based on legislation, standards and legal principles). Examples of the latter include road closures, maximum parking ratios, enforced speed limits, and mandatory employer trip reduction programmes. Violations of such TDM measures are met with some form of punishment, and the assumption is that such regulations or laws are internalised under the threat of punishment. These policy measures are also

A Cost-Minimisation Principle of Adaptation of Private Car Use

337

referred to as command-and-control measures (Johansson-Stenman, 1999) since an authority assumes responsibility for the management of a transport system and exercises control so that, in principle, it functions effectively.

17.3.6 Impacting Latent vs. Manifest Travel Demand Variations in TDM measures in terms of the nature of the demand which they manage are seldom taken into consideration. Latent demand can be defined as the demand for services or resources that goes unsatisfied for various reasons (e.g. congestion). Road construction has historically been driven by a “predict-andprovide” approach (Vigar, 2002; Whitelegg and Low, 2003), where the argument was that latent demand should be satisfied because better and more roads were required as a matter of individual freedom (the right to use the car) and economic competitiveness (the need for efficient road links for business). However, as reviewed by Mogridge (1997), increases in road capacity do not necessarily yield faster or more efficient travel but, paradoxically, may make congestion worse. The reasons for this are argued by Downs (1992) to be due to the fact that free road space created by marginal reductions in commuting time is consumed quickly by those travelling just outside of the peak time period who shift back in; those driving on less optimal routes who take advantage of lowered congestion on the most popular main roads; and those on slower public transit modes who prefer driving if there is any more space on the road. Latent demand induced by increased road capacity also includes new vehicle trips by people who would not have otherwise made the trip or trips by drivers who select an alternate destination (i.e. shoppers who prefer a new shopping centre over the city centre). In other words, road infrastructure expansion is self-defeating, a point clearly made by Hansen and Huang (1997) who estimate that in California the five-year elasticity of vehicle travel with respect to highway lane miles is 0.6-0.7 at the county level and 0.9 at the metropolitan level. The implications are that most of the trips on new roads are trips that would not have occurred had the roads not been built. In contrast to influencing latent travel demand, many TDM measures influence manifest travel demand (i.e. actual travel). Road or congestion pricing or kilometre-based charges attempt to change the actual driving behaviour of many people by increasing the cost (Banister, 2003; Foo, 1997, 1998, 2000; Goh, 2002; Ubbels et al., 2002), or by decreasing the costs of alternative modes (Root, 2001). Road closures or prohibition make it impossible for travel demand to manifest itself in certain areas or at certain times (Cambridgeshire County Council, 2005). The initial waves of TDM measures focused on better management of existing resources (Bovy and Salomon, 2002), and thus were aimed at changes in manifest travel demand. Many recent TDM measures have also begun to influence specifically latent travel demand. One example is the attempt to alter human values and change mobility culture so that a less mobile society is not viewed as negative (City of Zurich, 2002). Another example of such a TDM measure is land use planning. Research has demonstrated that intensities and mixtures of land use significantly

338

Peter Loukopoulos et al.

influence decisions to drive alone, car pool, or use public transport (Cervero, 2002). The assumptions made by proponents of such measures are that land-use patterns influence the time cost of travel and that the variations in time cost due to land use is of sufficient size to induce changes in car use (Boarnet and Crane, 2001; Boarnet and Sarmiento, 1998). In summary, whereas the construction of road infrastructure assisted the satiation of latent travel demand by allowing it to be manifested in actual car use so that individuals could drive to their activities, land use policies promoting, for example, mixed zoning, satiate latent travel demand by bringing the activities to the individual.

17.4 Theoretical Framework The potential effectiveness of TDM measures for reducing private car use depends on how car users respond to them. In conceptualising such responses, it is important to take into account that private car use primarily results from needs, desires, or obligations to participate in out-of-home activities (e.g. Axhausen and Gärling, 1992; Jones et al., 1983; Gärling et al., 1994; Root and Recker, 1983; Vilhelmson, 1999). Therefore, as has been noted (Gärling et al., 2000; Gärling et al., 2002b; Kitamura and Fujii, 1998; Pendyala et al., 1997; Pendyala et al., 1998), car-use reduction should be viewed broadly as an adaptation by car owners to changes in travel options that potentially have consequences for their engagement in different activities and the satisfaction they experience from this. In the following, we describe a theoretical framework (Gärling et al., 2002a) that was proposed with the aim of analysing the multi-facetted nature of car users’ responses to TDM measures. We will apply the theoretical framework here as an alternative to demand theory for understanding car users’ responses to road-user charges. Based on the assumptions that (i) individuals and households are fully informed of the costs and (ii) make optimal tradeoffs between benefits and costs, demand theory predicts that increasing costs for car use will lead to a reduction of manifest travel demand or behaviour. Deviations from this prediction are, however, frequently observed (Goodwin, 1998). In order to understand inaccurate predictions of the degree of change as well as the existence of thresholds (discontinuities), delayed effects and unexpected effects on other behaviour, the process of change needs to be specified in more detail. Fig. 17.1 is a simplified illustration of the theoretical framework. Travel options are defined as bundles of attributes describing trip chains (including purposes, departure and arrival times, travel times, monetary costs, uncertainty and inconvenience). Choices of travel options are influenced by these bundles of attributes. Another determinant is the goals individuals and households have set. In selfregulation theory in social psychology (Carver and Scheier, 1998), such goals are assumed to form a hierarchy from concrete (programmes) to abstract levels (principles) that function as reference values in negative feedback loops regulating ongoing behaviour or changes in behaviour. If a discrepancy between the present

A Cost-Minimisation Principle of Adaptation of Private Car Use

339

state and the goal is detected, some action is carried out with the aim of minimising it. After implementing a road pricing scheme, a car-use reduction goal may be set if households experience increased monetary travel costs (Loukopoulos et al., 2005). On the other hand, if other changes are encountered such as decreased travel times (due to less congestion) or concomitant decreased living costs (e.g. children moving out, salary increases), no such goal may be set. In effect, a simple relationship does not exist between increasing the cost of driving and setting car-use reduction goals. A similar line of reasoning can be applied to changes in destinations, departure times or routes. INDIVIDUAL FACTORS x Family structure x Income x Work situation x Attitudes x Activity/travel pattern

ADAPTATION GOAL/ IMPLEMENTATION PLAN

TDM MEASURESS x Road pricing x Physical restrictions x Improved alternatives

TRIP CHAIN ATTRIBUTES x Purposes x Departure times x Travel times x Cost x Uncertainty x Convenience

TRAVEL CHOICE x Stay at home x Use electronic communication x Car pooling x Change attributes of trip chain concerning: o Purposes o Modes o Departure times o Destinations

SITUATIONAL FACTORS x Family logistics x Time pressure x Weather x Time of day x Day of week

Fig. 17.1. Theoretical framework (after Gärling et al., 2002a)

340

Peter Loukopoulos et al.

Needs, desires, attitudes and values that people acquire, influence the goals which people set and strive to attain (Austin and Vancouver, 1996). Such goals vary in content and intensity (Locke and Latham, 1984, 1990). Content is related to difficulty or lack of skill required to attain the goal, specificity, complexity (the number of outcome dimensions) and the degree of conflict with other goals at the same or higher levels. Intensity refers to perceived importance and degree of commitment. Research in other areas on goal setting and attainment (e.g. Lee et al., 1989) has shown that specific and more difficult goals increase the likelihood that they are attained provided that the difficulty is not beyond people’s skills. Commitment to the goal and immediate clear feedback about goal attainment are moderating factors. After having set a car-use reduction goal, individuals and households are assumed to form a plan for how to achieve this goal and to make commitments to execute the set plan. In social psychological research, this process is referred to as the formation of implementation intentions (Gärling and Fujii, 2002; Gollwitzer, 1993). The plan that is formed consists of predetermined choices contingent on specified conditions (Hayes-Roth and Hayes-Roth, 1979). In making plans for how to reduce car use, households may consider a wide range of options such as staying at home, suppressing trips and activities, using electronic means of communication instead of driving, car pooling or changing the effective choice set of travel options with respect to purposes, destinations, modes or departure times. Households may also consider longer-term strategic changes such as moving to another residence or changing work place or hours. It is hypothesised that individuals and households seek and select options that lead to the achievement of the goal they have set. We do not assume, however, that this process necessarily entails a simultaneous optimal choice among all options. Consistent with the notion of satisficing (Gigerenzer et al., 1999; Simon, 1990), it is instead assumed that options are chosen and evaluated sequentially. Experimental laboratory-based research (Payne et al., 1993) has shown that people make tradeoffs between accuracy and (mental and tangible) costs, and that these tradeoffs are frequently optimal. A vital difference in relation to microeconomic utility-maximisation theories (McFadden, 2001) is that it is not assumed that people invariably invest the required degree of effort. Whether they do so or not is, in our theoretical framework, dependent on properties of the set goal (for example, if it is vague or specific). Furthermore, if the cost of an effective adaptation is too high, even a small and specific reduction goal to which a household is committed may be abandoned or further reduced. A second important difference to utility-maximisation theories is that choices are made sequentially over time. This implies that the process of change is prolonged and fails to result instantaneously in outcomes beneficial to society. Furthermore, in the theoretical framework it is asserted that benefits (effectiveness) as well as costs of chosen alternatives are evaluated. If such evaluations indicate a discrepancy with the goal, more costly changes are chosen. Even though it has been shown that people make optimal accuracy-effort tradeoffs in laboratory experiments, we do not know whether they do so in real life when making complex travel choices. In fact, many things suggest that they do not. As has been noted, it is documented (Chapter 13) that habitual car use and other daily habits and

A Cost-Minimisation Principle of Adaptation of Private Car Use

341

routines cause inertia. Research has also demonstrated that a bias exists such that the current state is overvalued (e.g. Samuelson and Zeckhausen, 1988), thus making changes less attractive. In particular, if the car-use reduction goal is vague, evaluating whether or not a change option is effective may possibly be biased toward confirming the expectation that it is (e.g. Einhorn and Hogarth, 1978; Klayman and Ha, 1987). Furthermore, previous research has demonstrated that immediate and clear feedback is essential (Brehmer, 1995). A system for charging car use that does not provide this is likely to fail. On the basis of our theoretical framework, we posit the existence of a hierarchy of change options that vary in effectiveness and costs. In Table 17.1, we operationalise this hypothesis by specifying three different classes of potential change options together with their associated costs. We further assume that an inverse relationship exists between effectiveness and costs for these change options. As indicated in Table 17.1, the first stage involves making car use more efficient by chaining trips, car pooling or choosing closer destinations. The cost is an increased need to plan ahead. The resulting change in car use may, however, not be sufficient to achieve the car-use reduction goal. In a second stage, trips may also be suppressed in order to achieve greater reduction in car use. In addition to increased planning, trip suppression implies changes in activities. Although in extreme cases this would necessitate changes in lifestyles, the required changes are in general likely to be minor, perhaps solely involving the suppression of isolated shopping trips. Leisure activities are next most likely to be removed from the activity agenda or substituted by in-home activities. Least likely are more consequential changes in working hours or changes of job. Table 17.1. Proposed adaptations to road pricing schemes Choice options More efficient car use x Trip chaining x Car pooling x Choosing closer destinations More efficient car use Trip suppression More efficient car use Trip suppression Mode switching

Possible costs Additional planning

Additional planning Activity suppression Additional planning Activity suppression Increased time pressure Inconvenience

The car-use reduction goal may still not be attained unless other modes are chosen. For instance, since work cannot easily be suppressed, public transport may be chosen for such trips. Additional planning, increased time pressure and inconvenience are possible costs associated with switching mode. In addition, in order to alleviate a potentially harmful increased time pressure (Gärling et al.,

342

Peter Loukopoulos et al.

1999; Koslovsky, 1997; Novaco et al., 1991; Novaco et al., 1990), suppression of minor leisure activities and shopping would still be necessary. It should be noted that Table 17.1 only describes one possible operationalisation of the principle of sequential cost-minimisation of choices of change options. In fact, survey results (Gärling et al., 2000; Loukopoulos et al., 2004) suggest that the hypothesised hierarchy varies with trip purpose and household or individual characteristics. Furthermore, it is not necessarily the case that cost varies inversely with effectiveness. TDM measures may be applied or other changes (for example, residential relocation) may occur that singly or in combination facilitate less costly changes that are effective.

17.5 Implications for the Effectiveness of TDM Measures In order to understand how road pricing affects car use, our theoretical framework asserts that two processes need to be examined further. First, how do type and size of the increased costs caused by road pricing affect the size and type of car-use reduction goals set by individuals and households with different characteristics? How does road pricing compare to other TDM measures in these respects? Which of the properties listed in the previous section for describing TDM measures, if any, are related to the goal-setting process? One may speculate that coercive measures may work better than increased monetary costs in making people set caruse reduction or change goals as it is only for less wealthy individuals or households (i.e. those who experience road pricing to be coercive) that the effect would be the same. Second, even if car use is reduced or changed goals are set, it may, however, still not have the intended effect. This follows from the likely fallibility of the second process that needs to be examined: the formation and implementation of a plan to achieve the car-use reduction or change goals. It is well documented that attitudes and intentions may have a weak correspondence with actual behaviour (Eagly and Chaiken, 1993; Fujii and Gärling, 2003; Gärling et al., 1998). Several reasons for this have been identified and in this chapter we have alluded to some of them. Self-regulation theory (Carver and Scheier, 1998) highlights yet others. In particular, we assume that the principle of cost-minimisation (where cost is broadly defined) is important for understanding how households and individuals try to attain set car-use reduction goals. If the cost is too high for its given strength and size, the goal may be abandoned or changed. Evaluation of feedback about effectiveness (goal attainment) is another element. If it is delayed and vague, suboptimal adaptation alternatives may continue to be chosen. Coercive measures including road-user charges probably only affect the motivation to change. Additional TDM measures need to be implemented to reduce the cost of effective alternatives for change so that they are made more attractive. These measures include those focusing on latent demand, that is, increased accessibility without use of the car. Choice of appropriate spatial and time scales may also be important.

A Cost-Minimisation Principle of Adaptation of Private Car Use

343

17.6 Summary and Discussion In this chapter, we propose that car users respond to travel demand management (TDM) measures by car-use reduction or change goals. Setting larger goals is more likely with the more coercive TDM measures. Market-based TDM measures such as road pricing are considered to fall somewhere in between coercive commandand-control measures and non-coercive voluntary-change measures depending on individual and household wealth. We further propose that goal attainment entails that choices be made between available change options following a cost-minimising principle. In addition, we speculate about the nature of these costs. The principle implies a sequential satisficing choice process that may never achieve an optimal cost-benefit trade-off. Whether it does or not may depend on effective change options being made less costly and on immediate and clear feedback about goal attainment. Coercive TDM measures (and for some individuals and households, road pricing) do not help car users attain their car-use reduction aims or change goals. Other TDM measures targeting changed latent demand are needed for this. Any implementation of TDM measures would benefit from forecasts of their likely effects. With the proposed theoretical framework as a basis, mathematicalstatistical models of the goal-setting and adaptation process need to be developed. Such developments may draw on previous research on the Activity-Mobility Simulator or AMOS (Kitamura and Fujii, 1998; Pendyala et al., 1997; Pendyala et al., 1998). AMOS reproduces how travellers modify their activity/travel choices in response to specified TDM measures. It uses as input an observed daily activity/ travel pattern to sequentially generate adaptation options, then computes and evaluates each new activity/travel pattern.

References Austin, J.T., Vancouver, J.B. (1996). Goal constructs in psychology: Structure, process, and content. Psychological Bulletin, 80: 286-303. Axhausen, K., Gärling, T. (1992). Activity-based approaches to travel analysis: Conceptual frameworks, models, and research problems. Transport Reviews, 12: 323-341. Banister, D. (2003). Critical pragmatism and congestion charging in London. International Social Science Journal, 55: 249-264. Boarnet, M., Crane, R., (2001). The influence of land use on travel behavior: specification and estimation strategies. Transportation Research A, 35: 823-845. Boarnet, M.G., Sarmiento, S. (1998). Can land-use policy really affect travel behaviour? A study of the link between non-work travel and land-use characteristics. Urban Studies, 35: 1155-1169. Bovy, P.H.L., Salomon, I. (2002). Congestion in Europe: Measurements, patterns and policies, in: Stern, E., Salomon, I. (eds.), Travel behaviour: Spatial patterns, congestion and modelling, Edward Elgar, Cheltenham, UK: 143-179.

344

Peter Loukopoulos et al.

Brehmer, B. (1995). Feedback delays in complex dynamic decision making, in: Frensch, P.A., Funke, J. (eds.), Complex decision making: The European perspective, Hillsdale, NJ: Erlbaum: 103-130. Cambridgeshire County Council (2005). Vehicles access to Cambridge city centre. http://www.cambridgeshire.gov.uk/transport/around/city_access/ Retrieved 1 May, 2005. Carver, C.S., Scheier, M.F. (1998). On the self-regulation of behavior. Cambridge University Press, Cambridge. Cervero, R. (2002). Built environments and mode choice: toward a normative framework. Transportation Research D, 7: 265-284. City of Zurich (2002). Mobilität ist Kult(ur). Die neue Mobilitätsstrategie – Wie wir miteinander weiterkommen [Mobility is Cult(ure). The New Mobility Strategy – How We Can Move Forward Together]. Civil Engineering Office, City of Zurich, Zurich. Dawes, R.M. (1998). Behavioral decision making, judgment, and inference, in: Gilbert, D., Fiske, S., Lindzey, G. (eds.), The handbook of social psychology, McGraw-Hill, Boston: 497-548. Downs, A. (1992). Stuck in traffic. Brookings Institution Press, Washington, DC. Eagly, A.H., Chaiken, S. (1993). The psychology of attitudes. Harcourt Brace Jovanovich, Forth Worth, TX. Eagly, A.H., Chaiken, S. (1998). Attitude structure and function, in: Gilbert, D., Fiske, S., Lindzey, G. (eds.), The handbook of social psychology, McGraw-Hill, Englewood Cliffs, NJ: 269-322. Einhorn, H.J., Hogarth, R.M. (1978). Confidence in judgment: Persistence of the illusion of validity. Psychological Review, 85: 396-416. Emmerink, R.H.M., Nijkamp, P., Rietvald, P. (1995). Is congestion pricing a first-best strategy in transport policy? A critical review of arguments. Environment and Planning B, 22: 581-602. Enoch, M., Potter, S. (2003). Encouraging the commercial sector to help employees to change their travel behaviour. Transport Policy, 10: 51-58. Foo, T.S. (1997). An effective demand management instrument in urban transport: The Area Licensing Scheme in Singapore. Cities, 14: 155-164. Foo, T.S. (1998). A unique demand management instrument in urban transport: the Vehicle Quota System in Singapore. Cities, 15: 27-39. Foo, T.S. (2000). An advanced demand management instrument in urban transport. Cities, 17: 3345. Fujii, S., Gärling, T. (2003). Application of attitude theory for improved predictive accuracy of stated preference methods in travel demand analysis. Transportation Research A, 37: 389-402. Gärling, T., Kwan, M.-P., Golledge, R.G. (1994). Computational-process modeling of household activity scheduling. Transportation Research B, 25: 355-364.

A Cost-Minimisation Principle of Adaptation of Private Car Use

345

Gärling, T., Gillholm, R., Gärling, A. (1998). Reintroducing attitude theory in travel behavior research: The validity of an interactive interview procedure to predict car use. Transportation, 25: 147-167. Gärling, T., Gillholm, R., Montgomery, W. (1999). The role of anticipated time pressure in activity scheduling. Transportation, 26: 173-191. Gärling, T., Gärling, A., Johansson, A. (2000). Household choices of car use reduction measures. Transportation Research A, 34: 309-320. Gärling, T., Fujii, S. (2002) Structural equation modeling of determinants of implementation intentions. Scandinavian Journal of Psychology, 43: 1-8. Gärling, T., Eek, D., Loukopoulos, P., Fujii, S., Johansson-Stenman, O., Kitamura, R., Pendyala, R., Vilhelmson, B. (2002a). A conceptual analysis of the impact of travel demand management on private car use. Transport Policy, 9: 59-70. Gärling, T., Gärling, A., Loukopoulos, P. (2002b). Forecasting psychological consequences of car-use reduction: A challenge to an environmental psychology of transportation. Applied Psychology: An International Review, 51: 90-106. Gatersleben, B. (2003). On yer bike for a healthy commute, in: Hendrickx, L., Jager, W., Steg, L. (eds.), Human decision making and environmental perception: Understanding and assisting human decision making in real-life settings, University of Groningen, Groningen, the Netherlands: 161-182. Gigerenzer, G., Todd, P.M., the ABC Research Group (eds.) (1999). Simple heuristics that make us smart. Oxford University Press, Oxford. Goh, M. (2002). Congestion management and electronic road pricing in Singapore. Journal of Transport Geography, 10: 29-38. Gollwitzer, P.M. (1993). Goal achievement: The role of intentions. European Review of Social Psychology, 4: 141–185. Golob, T.F. (2001). Travelbehaviour.com: Activity approaches to modeling the effects of information technology on personal travel behaviour, in: Hensher, D. (ed.), Travel behavior research: The leading edge, Pergamon, Amsterdam: 145-183. Goodwin, P.B. (1996). Simple arithmetic. Transport Policy, 3: 79-80. Goodwin, P. (1998). The end of equilibrium, in: Gärling, T., Laitila, T., Westin, K. (eds.), Theoretical foundations of travel choice modeling, Elsevier, Amsterdam: 103-132. Greene, D.L., Wegener, M. (1997). Sustainable transport. Journal of Transport Geography, 5: 177-190. Hansen, M, Huang, Y. (1997). Road supply and traffic in California urban areas. Transportation Research A, 31: 205-218. Hayes-Roth, B., Hayes-Roth, F. (1979). A cognitive model of planning. Cognitive Science, 3: 275–310.

346

Peter Loukopoulos et al.

Hensher, D.A. (1998). The imbalance between car and public transport use in urban Australia: why does it exist? Transport Policy, 5: 193-204. Hensher, D.A., King, J. (2001). Parking demand and responsiveness to supply, pricing and location in the Sydney central business district. Transportation Research A, 35: 177-196. Jakobsson, C. (2004). Motivational and volitional control of private automobile use: The effectiveness of transport policies. Doctoral dissertation. Göteborg University, Department of Psychology, Göteborg, Sweden. Johansson, L.-O., Gustafsson, M., Falkemark, G., Gärling, T., Johansson-Stenman, O. (2003). Goal conflicts in political decision making: A survey of municipality politicians’ views of road pricing. Environment and Planning C, 21: 615-624. Johansson-Stenman, O. (1999). Regulating road transport externalities: pricing versus commandand-control, in: Sterner, T. (ed.), The market and the environment, Edward Elgar, Cheltenham, UK: 134-157. Jones, P. (2003). Acceptability of transport pricing strategies: meeting the challenge, in: Schade, J., Schlag, B. (eds.), Acceptability of transport pricing strategies, Elsevier, Oxford: 27-62. Jones, P., Dix, M.C., Clarke, M.I., Heggie, I.G. (1983). Understanding travel behavior. Gower, Aldershot, UK. Kitamura, R., Fujii, S. (1998). Two computational process models of activity-travel choice, in: Gärling, T., Laitila, T. Westin, K. (eds.), Theoretical foundations of travel choice modeling, Elsevier, Amsterdam: 251-279. Kitamura, R., Fujii, S., Pas, E.I. (1997). Time-use data, analysis and modeling: Toward the next generation of transportation planning methodologies. Transport Policy, 4: 225-235. Klein, D., Moore, A., and Reja, B. (1996). Free to cruise: Creating curb space for jitneys. Access, 8: 2-6. Klein, D., Moore, A., Reja, B. (1997). Curb rights: A foundation for free enterprise in urban transit. Brookings Institution Press, Washington, DC. Koslowsky, M. (1997). Commuting stress: Problems of definition and variable identification. Applied Psychology: An International Review, 46: 153-173. Klayman, J., Ha, Y.-W. (1987). Confirmation, disconfirmation, and information in hypothesis testing. Psychological Review, 94: 211-228. Kristensen, J.P., Marshall, S. (1999). Mobility management to reduce travel: the case of Aalborg. Built Environment, 25: 138-150. Lee, T.W., Locke, E.A., Latham, G.P. (1989). Goal setting theory and performance, in: Pervin, L.A. (ed.), Goal concepts in personality and social psychology, Lawrence Erlbaum, Hillsdale, NJ: 291-326. Lindberg, G. (1995). Road pricing: policy and options for the future, in: Johansson, B., Mattson, L.-G. (eds.), Road pricing: Theory, empirical assessment and policy, Kluwer, Boston, MA: 205221.

A Cost-Minimisation Principle of Adaptation of Private Car Use

347

Litman, T. (2003). The online TDM encyclopedia: Mobility management information gateway. Transport Policy, 10: 245-249. Locke, E.A., Latham, G.P. (1984). Goal setting: A motivational technique that works. PrenticeHall, Englewood Cliffs, NJ. Locke, E.A., Latham, G.P. (1990). A theory of goal-setting and task performance. Prentice-Hall. Englewood Cliffs, NJ. Loukopoulos, P., Gärling, T., Vilhelmson, B. (2005). Mapping the potential consequences of caruse reduction in urban areas. Journal of Transport Geography, 13: 135-150. Loukopoulos, P., Jakobsson, C., Gärling, T., Schneider, C.M., Fujii, S. (2004). Car user responses to travel demand management measures: Goal setting and choice of adaptation alternatives. Transport Research D, 9: 263-280. Louw, E., Maat, K. (1999). Enschede: Measures in a package. Built Environment, 25: 118-128. Louw, E., Maat, K., Mathers, S. (1998). Strategies and measures to reduce travel by car in European cities. Paper presented at the 8th World Conference on Transport Research, Antwerp, Belgium. May, A.D., Jopson, A.F., Matthews, B. (2003). Research challenges in urban transport policy. Transport Policy, 10: 157-164. McFadden, D. (1999). Rationality for economists? Journal of Risk and Uncertainty, 19: 73-105. McFadden, D. (2001). Disaggregate behavioral travel demand’s RUM side – A 30 years retrospective, in: Hensher, D.A. (ed.), Travel behavior research, Elsevier, Amsterdam: 17-63. Meyer, M.D. (1999). Demand management as an element of transportation policy: using carrots and sticks to influence travel behavior. Transportation Research A, 33: 575-599. Mogridge, M.J.H. (1997). The self-defeating nature of urban road capacity policy: A review of theories, disputes and available evidence. Transport Policy, 4: 5-23. Novaco, R.W., Kliewer, W., Broquet, A. (1991). Home environmental consequences of commute travel impedance. American Journal of Community Psychology, 19: 881-909. Novaco, R.W., Stokols, D., Milanesi, L. (1990). Objective and subjective dimensions of travel impedance as determinants of commuting stress. American Journal of Community Psychology, 18: 231-257. Odeck, J., Bråthen, S. (2002). Toll financing in Norway: The success, the failures and perspectives for the future. Transport Policy, 9: 253-260. Pas, E.I. (1995). The urban transportation planning process, in: Hanson, S. (ed.), The geography of urban transportation, Elsevier, Amsterdam: 53-77. Payne, J.W., Bettman, J.R., Johnson, E.J. (1993). The adaptive decision maker. Cambridge University Press, New York. Pendyala, R.M., Kitamura, R., Chen, C., Pas, E.I. (1997). An activity-based micro-simulation analysis of transportation control measures. Transport Policy, 4: 183-192.

348

Peter Loukopoulos et al.

Pendyala, R.M., Kitamura, R., Reddy, D.V.G.P. (1998). Application of an activity-based travel demand model incorporating a rule-based algorithm. Environment and Planning B, 25: 753-772. Rajé, F. (2003). The impact of transport on social exclusion processes with specific emphasis on road user charging. Transport Policy, 10: 321-338. Rothengatter, W. (2003). How good is first best? Marginal cost and other pricing principles for user charging in transport. Transport Policy, 10: 121-130. Root, A. (2001). Can travel vouchers encourage more sustainable travel? Transport Policy, 8: 107-114. Root, G.S., Recker, W.W. (1983). Towards a dynamic model of individual activity pattern formulation, in: Carpenter, S., Jones, P. (eds.), Recent advances in travel demand analysis, Gower, Aldershot, UK: 371-382. Rose, G., Ampt, E. (2001). Travel blending: an Australian travel awareness initiative. Transportation Research D, 6: 95-110. Rye, T. (2002). Travel plans: do they work? Transport Policy, 9: 287-298. Samuelson, W., Zeckhausen, R. (1988). Status quo bias in decision making. Journal of Risk and Uncertainty, 1: 7-59. Schuitema, G. (2003). Pricing policies in transport, in: Hendrickx, L., Jager, W., Steg, L. (eds.), Human decision making and environmental perception: Understanding and assisting human decision making in real-life settings, University of Groningen, Groningen, the Netherlands: 203222. Sipes, K.N., Mendelsohn, R. (2001). The effectiveness of gasoline taxation to manage air pollution. Ecological Economics, 36: 299-309. Simon, H.A. (1990). Invariants of human behavior. Annual Review of Psychology, 41: 1-19. Steg, L., Vlek, C. (1997). The role of problem awareness in willingness-to-change car use and in evaluating relevant policy measures, in: Rothengatter, T., Carbonell Vaya W., Traffic and transport psychology. Theory and application, Pergamon Amsterdam: 465-475. Stradling, S.G., Meadows, M.L., Beatty, S. (2000). Helping drivers out of their cars. Integrating transport policy and social psychology for sustainable change. Transport Policy, 7: 207-215. Taylor, M.A.P., Ampt, E.S. (2003). Travelling smarter down under: policies for voluntary travel behaviour change in Australia. Transport Policy, 10: 165-177. Thorpe, N., Hills, P., Jaensirisak, S. (2000). Public attitudes to TDM measures: a comparative study. Transport Policy, 7: 243-257. Ubbels, B., Rietveld, P., Peeters, P. (2002). Environmental effects of a kilometre charge in road transport: an investigation for the Netherlands. Transportation Research D, 7: 255-264. Vigar, G. (2002). The politics of mobility. Transport, the environment and public policy. Spon Press, London.

A Cost-Minimisation Principle of Adaptation of Private Car Use

349

Vilhelmson, B. (1999). Daily mobility and the use of time for different activities: The case of Sweden. GeoJournal, 48: 177-185. Vlek, C., Michon, J. (1992). Why we should and how we could decrease the use of motor vehicles in the future. IATSS Research, 15: 82-93. Whitelegg, J., Low, N. (2003). Managing transport demand in European countries, in: Low, N., Gleeson, B. (eds.), Making urban transport sustainable, Palgrave Macmillan, Basingstoke, UK: 240-251.

18 Car Users’ Trade-Offs Between Time, Trip Length, Cost and Road Pricing in Behavioural Models Otto Anker Nielsen1 and Goran Vuk2 1 2

Centre for Traffic and Transport, Technical University of Denmark Danish Transport Research Institute

Abstract Stated Preference (SP) experiments are the obvious choice in order to forecast travellers’ responses to alternatives that do not exist today. SP experiments can in addition change the variables in a controlled factorial design, whilst Revealed Preference surveys (RP) have to rely on measured observations. Preferences concerning correlated explanatory variables such as travel time and travel length can therefore often be estimated more easily by SP experiments than by RP surveys. However, respondents may not act as they claim in the interview and the design may affect the results. The chapter illustrates, based on an SP experiment compared with the AKTA data (Chapter 6), how the definition of variables can influence the results obtained. Value of travel time, VoT, (both free flow and congestion travel time), choice of time-of-day of travel and route choice are considered both without and with road pricing. It is shown that the design of the experiment seriously affects the result especially with respect to VoT. It is then shown how the income variable and distributed coefficients (taste variation) can improve the model fit and its behavioural accuracy. The model structure obtained corresponded well to the best models from the field experiment (RP) in AKTA, although the size of coefficients differed somewhat. It appears that (marginal) cost is a problematic variable compared to trip distance as the respondents had serious difficulties in estimating the cost per kilometre. Road pricing was considered a bit worse than a pure marginal driving

Mai-Britt Herslund is thanked for the work on SP design and the interviews, Christian Würtz on work collecting and processing the GPS data, Paolo Menegazzo on work regarding the route choice model estimator and data processing and Majken Vildrik Sørensen is thanked for work on the SP and model estimation, including work reported in Nielsen and Sørensen (2004) upon which section 18.4.3 is based. The AKTA project was primarily financed by the EU’s 5th Framework Programme and the municipality of Copenhagen, with co-funding of the research part by the Technical University of Denmark (DTU). Principle contractor of AKTA was the City of Copenhagen, from which Poul Sulkjær is thanked. DTU, the Danish Road Directorate and the Ministry of Transport were assistant contractors. The AKTA SP was funded by the Danish Transport Council and the remaining work on this was allowed to be finalised by self-funding from the DTU and the Danish Transport Research Institutes (DTF).

352

O.A. Nielsen and G. Vuk

cost in the road pricing situation, i.e. car users preferred paying for fuel than paying road pricing. But the total cost after road pricing was introduced was weighted less than before road pricing compared to travel time – or in other words travelling time was weighted higher. This can be explained by the fact that travellers have restrictions and inertia in their possibilities of changing behaviour as well as the fact that the car-users who still use the car are the ones with the higher willingness to pay and the travellers who change behaviour have the lower willingness to pay.

18.1 Introduction This chapter describes how the definition of cost and trip length variables in a Stated Preference (SP) experiment affects the results. The model’s ability to describe car users’ valuation of cost compared to other variables is a core issue in any transport model of road pricing as road pricing is an extra cost. This issue was examined 1) by testing different definitions of costs in a systematic way within an applied SP experiment, 2) by examining the sensibility of different estimation methods and model assumptions, among others cost-definitions, Nested Logit (NL) versus Mixed Logit (ML) and different assumptions on the random coefficients in the ML and 3) by comparison with the results from a Revealed Preference (RP) experiment, where the same persons’ cars were followed over a 4-6 month period. Some of the initial results have been presented earlier in Nielsen and Jovicic (2003). Section 18.2 presents the experimental design and section 18.3 presents some more general results. Section 18.4 presents and discusses the SP model and section 18.5 the RP model with focus on route choice, followed by a comparison of the two in section 18.6; conclusions are drawn in section 18.7.

18.2 SP Design The Technical University of Denmark and the Danish Transport Research Institute (DTF) carried out a research project concerning SP methods in parallel with the main AKTA road pricing experiment (see Chapter 6). The purpose was to evaluate SP as a method, since AKTA provides an excellent RP data survey for comparison. 200 car drivers from the second and 100 from the third round of AKTA (Chapter 6) were interviewed while they waited for a GPS unit to be installed in their cars at a garage. The SP data were accordingly collected prior to the start of the main trial period of AKTA. This reduced the cost of the SP survey and ensured that most of the 300 participants in AKTA followed the SP experiment as well. 279 interviews were successfully completed and processed. 3,388 SP records passed the data quality control. The SP questionnaire collected information about a typical trip, for example the journey to work, about which the respondent was expected to have sound knowledge.

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

353

The chosen trip was described by including origin and destination addresses, departure and arrival times as well as travel purpose. If the respondent undertook an extra activity on the way (e.g. shopping, visiting a bank) these activities were also noted. The departure time defined the trip as being peak or off-peak. The experiments considered 1) the trade off between travel time and cost; 2) choice of time-of-day travel (peak vs. off peak) and 3) road pricing scenarios. The respondents were presented to the different trade off situations in separate sections of the experiment. The travel cost was measured in monetary units as is usual in most value of travel time SP studies. For half of the respondents, the travel cost was measured in monetary units as is usual in most value of travel time SP studies, for the other half, the cost was measured in distances (kilometres). The pricing levels in the SP experiment for each respondent were the same as those which the same respondent would face in the main AKTA (RP) experiment. The SP experiments were carried out prior to the RP experiment in order to avoid the possibility that answers were influenced by the experiments. Each respondent then undertook three SP experiments before answering socio-economics questions: • The first SP experiment focused on the Value of Time (VoT). Traditional experiments examine trade offs between time and cost. However, prior experience in Denmark (Nielsen et al., 2002) suggests that some car users are not aware of costs and that most travellers consider only marginal costs (fuel). To investigate this issue further, some of the respondents were asked about their trade-off between time and cost (referred to as SP1a in the following) and some were asked about their trade-off between time and trip length (SP1b). • The second SP experiment focused on time-of-day (ToD) decisions and congestion1. The trade-off context in this respect was between travel cost and travel time, time being affected because of congestion, and time-of-day. Travellers in peak hours may choose to travel outside the peak because of lower costs and congestion. Travellers presently travelling outside the peak may choose to travel in the peak hours if the additional costs make this faster. This experiment also had either a trade-off between time and cost (SP2a), or between time and trip length (SP2b). Some respondents followed the SP2a design and some respondents followed the SP2b design. • The third SP experiment evaluated the choice between an existing trip, where road pricing had been added, and a possible alternative route, which was found together with the respondent. The trade-off was then investigated between cost, pricing, free time and congestion time, in a binary choice context where the four explanatory variables were varied.

1

It is noted that the definition of congestion time followed the definition in Nielsen et al, (2002), i.e. that congestion time is the extra expected travel time caused by congestion compared to the situation without congestion. The respondents were asked how much time their usual trip took and how much time they expected it would have taken without congestion. The difference was then interpreted as the extra time caused by congestion.

354

O.A. Nielsen and G. Vuk

18.3 Awareness of Travel Distance and Travel Time An important issue when attempting to measure the effect of road pricing schemes on user behaviour is the validity of the measured impact of the travel cost on car usage. Users may consider the cost of travel differently and thereby have a different trade off between the road pricing charge and travel cost. To discover whether any such problems exist, respondents were questioned on their awareness of actual costs for a particular trip. The questionnaire was coupled with an underlying route choice model so that the zone-to-zone distance of the trip could be calculated. Further, the total cost was calculated by a constant cost per km multiplied by the distance. The respondents were prompted to verify or correct the cost provided either in monetary units or kilometres. The rate of accepted to suggested cost is shown in Fig. 18.1, where it can be seen that the respondents frequently corrected the suggested cost and trip length. Trip lengths were corrected more often than costs (43% corrected the trip length, 18% the cost). This may be interpreted as respondents’ greater awareness of travel lengths than travel costs for their typical trip.

Fig. 18.1. Respondents’ correction of SP estimates of cost and time based on zonal data

Looking more into the corrections, it appears that the correction of length had the same average size independent of the trip length. This can be explained by how the distances and costs were generated in the underlying assignment model. Distances were calculated between zones (the error is related to the zonal connector in the from-and-to zones, respectively); hence the aggregation error was independent of the distance between the zones. The respondents were then asked what they would normally consider the cost of driving per kilometre. The typical answers were rounded values (0.5, 1 and 2 DKK were the most frequent answers), which can be seen in Fig. 18.2. The interval between 0.5 and 1 DKK equals reasonably well the marginal driving cost (about

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

355

(%)

45% of the answers) and values around 2 DKK equal average cost including capital cost, etc. (about 45% of the answers). 100,0 90,0 80,0 70,0 60,0 50,0 40,0 30,0 20,0 10,0 0,0 0,00

Percent Percent, cumulative

1,00

2,00

3,00

4,00

DKK 5,00

Fig. 18.2. Accepted cost per km shown as a cumulative and a density distribution

As respondents may consider different cost components to be included in their cost per kilometre, respondents were asked about what the cost included (Fig. 18.3). Only very few respondents claimed that fixed costs were included. There seems to be an inconsistency between the stated cost and the respondents' assumptions behind them. 6% of the participants even stated 0 (zero) DKK per km, since their cost of driving was paid for by other persons (employer, partner, friend, parents, etc.). About 4% of the answers in Fig. 18.3 were outside a reasonable cost range. About 48% stated a cost that was at a level which would include fixed costs, but only 10% stated that they had included fixed costs in their cost per kilometre estimates. Fig. 18.4 shows large disagreements between the answers provided by each person. As can be seen, as many as 40% of the respondents’ answered cost per kilometre deviated by more than 100% above the ratio of their accepted total cost and travel length. In addition to this, the 6% of the participants who did not pay their costs themselves did not answer this in the cost/km question due to the way this question was formulated. The main conclusion is that it was difficult for the respondents to measure their costs of driving in monetary units and that their answers differed, largely depending upon how the questions were presented and how they were assisted during the interview and/or survey.

356

%)

O.A. Nielsen and G. Vuk

100

Mentioned last

90

Mentioned second

80

Mentioned first

70 60 50 40 30 20 10 0 Fuel

Tyres and oil

Repairs

Insurance

Provision

Other

Acumulated (%)

Fig. 18.3. Participants stated components in their cost calculations. The answers from the respondents who stated their cost per kilometre could easily be compared to their a priori accepted trip length and cost for the specific trip (i.e. the ratio between the two)

100 90 80 70 60 50 40 30 20 10 0 -100

0

100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 1600

Deviation % Fig. 18.4. Comparison between accepted costs (e.g. 10 DKK) divided by accepted trip length (e.g. 10 km) – i.e. accepted total values – with stated cost per kilometre (e.g. 2 DKK/km which gives an inconsistence of the answer equal to 100%). Cumulative distribution in %

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

357

18.4 SP Model Estimation As a first step, a Multinomial Logit Model (MNL) was applied to the data in order to identify variables that could explain the variation in the data. The model included separate coefficients for costs and road pricing charges, travel time, split into free flow time and extra time use related to congestion, a coefficient for trip distance for the experiment where travel distances were presented instead of monetary costs and dummies for time period (in or off-peak), gender and alternative specific constants. Several different strategies for model building were followed including mixed logit formulation (normal and log-normal distributions of the coefficients), incomedependent model formulations and combinations with mixed logit and income dependencies.

18.4.1 Model Formulation (Utility Functions) Several different utility functions were used in the estimation. It was assumed that the road users chose the alternative i with the highest utility Ui . The utility was assumed to consist of a utility function Vi and an error term Hi which describe unexplained variation in the data. The error term is assumed to be Identical Independent Gumbel Distributed (iid) over alternatives (1), whereby a logit model is obtained (Ben-Akiva and Lerman, 1985). The independent variables were cost c (cost measured in monetary units or as length of the journey multiplied by a fixed cost per kilometre), road pricing r, free flow time tff, extra travel time due to congestion tcng and a vector S of socio-economic attributes s. Each variable where multiplied a coefficient ȕ, which was estimated for this variable specifically;

U i Vi  H i

ȕc c  E r r  ȕtff t ff  ȕtcng tcng  E s S s  H i

(1)

In the second step of the work, a mixed logit formulation was applied (see Ben-Akiva et al., 1993), whereby some of the coefficients were allowed to vary around their mean; this corresponds to substitution of coefficients ȕk’ in equation (2) for ȕk in equation (1), where k refers to the specific coefficient. Mixed models were estimated based on both normal and log-normal distributions assumed a priori as the mixing distributions. For distributed terms, both the mean of ȕ’ and the standard deviation were estimated (from these, the parameters for the log-normal distribution can be calculated).

Ek ' Ek  [

( 2)

The best MNL and EC models are presented in Table 18.1. Models 1 and 2 are MNL models while models 3, 4 and 5 are mixed logit formulations. In model 1, one cost coefficient corresponds to converting travel distances to cost by a fixed cost of driving

358

O.A. Nielsen and G. Vuk

(DKK/km) and assuming the behavioural impact of road pricing to be of the same scale as the cost of driving. Model 2 deals with three cost coefficients; one related to driving costs, one related to driving distances (rescaled into driving costs) and one related to road pricing. For the rescaling of distance to driving cost, a fixed cost of 0.55 DKK/km was used. Models 3 to 5 include one or more error components in their structure. The only difference between model 2 and 3 is that a random error was defined in model 3 linked to all time and cost coefficients. This is a hypothetical situation, but a useful test of whether heterogeneity exists in data. A dramatic improvement is observed in model 3, showing considerable variation in taste. Further disaggregation of the error components in models 4 and 5 gave even better results as all proved to be significantly different from zero. The best model was model 5 where error components were placed in relation to different cost coefficients, free flow travel time and congested travel time, i.e. six error components in total. Table 18.1. Estimation results from the best MNL and EC models based on the SP data File

model 1

model 2

model 3

model 4

model 5

Observations Final log (L) D.O.F. Rho²(0) 0.292

3388 -1662.6 8 0.299

3388 -1645.7 10

3388 -1571.1 11

3388 -1561.3 14

3388 -1538.1 16

0.331

0.343 drvcost -0.300 (-14.1) -0.976 (-4.3) -0.997 (-3.4) -1.50 (-2.9) cngtime -0.299 (-19.9) -0.296 (-19.1) -1.530 (-4.5) -1.48 (-3.4) -2.25 (-3.0) rdprice -0.350 (-16.0) -0.358 (-15.7) -2.08 (-4.3) -2.38 (-3.0) -3.09 (-2.9) inpeak .10 (3.4) 1.17 (3.6) 2.63 (2.4) 2.49 (2.2) 3.37 (2.0) t_malep -0.674 (-2.5) -0.705 (-2.6) -1.86 (-2.5) -1.93 (-2.2) -3.28 (-2.1) asc51 0.188 (2.4) 0.189 (2.4) 0.355 (1.5) 0.329 (1.3) 0.436 (1.5) costdst -0.157 (-4.7) -1.09 (-3.9) -1.18 (-2.9) -1.65 (-2.1) costSP3 -0.346 (-9.4) -1.96 (-4.0) -2.38 (-2.9) -2.59 (-2.8) ercmp cost1_e fftime_e cngtime_e cost4_e cost2_e cost3_e

-1.16

(-4.0) -2.00 -1.12 0.904 -1.48

(-2.7) (-3.1) (2.9) (-2.3)

-6.21 (-2.7) -1.74 (-2.7) 1.16 (2.7) -1.96 (-2.7) 7.33 (2.3) 0.924 (1.8)

The variables in the models are defined as: •

drvcost; driving cost coefficient in the SP1a (VoT experiment) and SP2a (time of day experiment, ToD).

•

fftime; free flow time coefficient.

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

359

•

cngtime; congested time coefficient (extra time due to congestion compared to fftime).

•

rdprice; coefficient for road pricing in the road pricing experiment (SP3).

•

inpeak; dummy variable from ToD experiments (SP2) saying that if the respondent originally travelled in peak hours, then when presented with an offpeak alternative he or she might (or might not) prefer to switch. A positive value means that the original time of travel (which is peak) is preferred.

•

offpeak; dummy variable from ToD experiments (SP2) saying that if the respondent originally travelled in off-peak hour, then when presented with a peak alternative he or she might (or might not) prefer to switch. A positive value means that the original time of travel (which is off-peak) is preferred.

•

t_malep; in the ToD experiments (SP2), dummy considering that men who travel originally in the peak are more willing to stay in the peak than women.

•

asc51; alternative specific constant in the SP3 placed on the original (typical) route. A positive value means that when everything else is equal, the respondents prefer the original route.

•

costdst; cost coefficient calculated using distances in SP1b and SP2b.

•

costSP3; cost coefficient in the SP3.

•

ercmp; error component coefficient applied only in model 3. The coefficient was applied for all cost and time coefficients in all SP experiments. The purpose of model 3 was to discover if the error components significantly improve the model estimations (which was the case).

•

cost1_e; cost error component coefficient applied to all cost coefficients in model 4 and to cost coefficients in the SP1a and SP2a in model 5.

•

fftime_e; free flow error component coefficient applied to all SP experiments.

•

cngtime_e; congested time error component coefficient applied to all SP experiments.

•

cost4_e; road pricing error component cost coefficient in models 4 and 5.

•

cost2_e; error component cost coefficient in model 5 applied to SP3.

•

cost3_e; error component cost coefficient (where costs are calculated on distances) in model 5. It was applied to SP1b and SP2b.

Table 18.2 shows the VoT for model 2, which is a MNL model, and Table 18.3 VoT for EC model 5. The VoT in the EC models are often calculated as the ratio between the time and cost coefficient (first rows in Table 18.3). However, the distribution of the VoT is more correctly described as the distribution of the time coefficient divided by the distribution of the cost coefficient. The mean of the joint distribution is not the same as the ratio of the means for each distribution. The ratio of two normally distributed terms is Cauchy distributed, for which no estimator for

360

O.A. Nielsen and G. Vuk

the mean exists. The interpretation of this is that when the denominator (the tail of the distribution of the cost coefficient) approaches zero, then the VoT limit is infinite. If the distribution contains negative values, then the VoT is negative, which is of course illogical. Nonetheless, due to software limitations the present study assumed normally distributed VoT. The second row of VoT in Table 18.3 was first simulated with the denominator (cost distribution) truncated at zero. However, as this skewed the cost distribution too much, the distribution was then truncated symmetrically around the mean with the left truncation at zero and the right at twice the mean. Nevertheless, when the numbers drawn at random approach zero, the VoT still approaches infinitely. Accordingly, a second truncation was done on free flow VoT over 250 DKK and congestion VoT over 400 DKK. The same truncation was used in the AKTA RP study to deal with participants who showed lexicographic behaviour concerning time, i.e. that they minimised time with no consideration of cost whatsoever. The cut-off values were determined at the point which was twice the maximum of the VoT for the nonlexicographic participant with the highest VoT in the RP experiment. The numerator (the time coefficient distribution) is assumed to have a central mean, which is why truncation or simulation should not be necessary. This was validated by simulation, after which the mean value was used. The simulated values from this approach were in general larger compared to the traditional estimation (Table 18.3). It should be expected that the larger the variation of the random coefficient, the larger the difference on the VoT estimates. However, this also increases the likelihood of truncation and the cut-off of maximum VoT values. Hence, it is not certain that the VoT would increase (the congestion time in the road pricing SP decreases). Table 18.2. Value of Time (VoT) in DKK/hour in model 2 (MNL model) Time component Free flow time Congested time

Cost/time SP

27.3 43.9

Length/time SP 0.55 DKK/km 0.7 DKK/km SP RP 70.3 89.5 113 144

Road pricing SP 31.9 51.3

Table 18.3. Value of Time (VoT) in DKK/hour in model 5 (EC model) Calculation method Ratio of coefficients Simulation

Time component Free flow time Congested time Free flow time Congested time

Cost/time SP 20.2 30.3 39.4 45.5

Length/time SP 0.55 DKK/ 0.7 DKK/ km SP km RP 54.5 69.4 81.8 104 108 131

Road pricing SP 34.7 52.1 44.1 49.6

In SP3, driving costs were presented together with road pricing. This gave a higher VoT (numerically smaller cost coefficients) than in SP1 and SP2, i.e. the respondents' willingness to pay for time savings increases. However, road pricing had a higher coefficient than marginal cost (both coefficients are still more positive than in experiments without pricing), which indicates that road pricing nonetheless is considered to be worse than marginal cost.

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

361

The congested travel time is weighed more negatively than free flow travel time in all models, as could be expected (which is consistent with earlier Danish studies, e.g. Nielsen et al., 2002, as well as the references provided in Nielsen, 2004). The trip length based experiments showed a much higher VoT than the trip cost experiments. The results are illustrated with the low cost/km (0.55 DKK/km – basically a very economical car) and the value used in the RP (0.7 DKK/km – an average car and fuel price as at the time when the experiment took place). The interpretation may be that respondents in the cost experiment state their believed willingness to pay, whilst they in real life primarily want to minimise time and the trip length versus time experiment describes this trade off better. The model also contained a number of dummies. The off-peak dummy revealed a tendency to not change time-of-day travel (inertia). This may explain a relatively small change in the time-of-day of trips in the main AKTA experiment (Nielsen and Jovicic, 2003). It is not surprising that peak hour drivers want to stay in the peak; they have already accepted extra time due to congestion. However, it is surprising that nonpeak drivers want to stay out of the peak, all other things being equal. This indicates that they have chosen the non-peak as the best time-of-day for their specific trip, rather than to avoid congestion in the peak. The Asc51 variable shows inertia to change route compared to the usual route, which may be logical since this indeed is the preferred choice by the respondent. T_malep (ToD – experiment) shows that males are less willing to change ToD than females (contrary to the result of Bonsall et al., 1998). This is surprising, since women could be expected to have a more constrained time schedule than men (as women more often collect children from kindergartens than men, buy food, etc.).

18.4.2 Income Effect Models Since willingness to pay may depend on income, different parameterised models tested this. The more traditional way of modelling income effect is by segmenting the sample by income classes. Several splits into classes within the answer intervals between 100, 200, 300, 400, 500, 750 and 1,000 thousand DKK yearly gross household income for 2001 was tested. However, data only allowed estimation of two cost coefficients (Table 18.4); for income groups up to DKK 400,000 and above DKK 400,000. 70% of the sample (195 respondents) belonged to the first income group while the rest (84 respondents) belonged to the high income group. The income effect could not be estimated reasonably in the SP1b and SP2b experiments where travel distances were presented to the respondents, as it appeared that lower income group respondents have a higher VoT than those with high income which is nonsense.

362

O.A. Nielsen and G. Vuk

Table 18.4. VoT in income dependent EC models (below/over DKK 400,000) calculated as ratio of coefficients. The values marked in brackets [ ] are the VoT in the similar models without income dependencies SP1a and SP2a (cost/time trade offs)

SP1b and SP2b (length/time trade offs)

SP3 (road pricing experiment)

Free flow travel time

18.9/22.8 [20.2]

Not significant [54.5]

32.1/44.8 [34.7]

Congested travel time

28.4/34.3 [30.3]

Not significant [81.8]

48.2/67.2 [52.1]

The last model types estimated in the study were two models with increasing VoT as a function of income, either with an additional Ec/I term or Ec/I*c (the latter removing the choice of unit problem in the coefficient). However, both formulations had low tvalues as well as illogical signs of the coefficients.

18.4.3 Alternative Model Formulations In addition to the model results presented above, several other models were tested (Nielsen and Sørensen, 2004). The data set is remarkable in the sense that extensive data is available for each respondent since two pure RP and the combined RP/SP questionnaires were available and combinable for all respondents (a unique ID followed all participants through the test). Thus, several segmentation criteria were possible. The tested models included adding a dummy for peak travel (fixed and distributed coefficients), combinations of gender and peak (fixed and distributed coefficients) and log-normal distributed cost coefficients (either same distribution or different parameters). Furthermore, a sequence of models where one or more cost coefficients were deflated by income was tested. First, linearly increasing coefficients were tested, then dummies for levels in a search for polynomial relations. At this point, revealed income relations had coefficients of illogical magnitudes or signs or they were statistically insignificant; the best model from previous with ‘just’ two income levels of cost coefficients prevailed. Finally, a model combining the search for non-linear relations in combination with different patterns for different cost components resulted in a model where the coefficient for road pricing increased with higher income groups. However, the likelihood value for this model (-1642.65) was only slightly better compared to the MNL model (-1645.71).

18.5 The RP Route Choice Model The behaviour of the car drivers in the main AKTA experiment was monitored by a GPS unit installed in the car. The actual choice of the drivers could hereby indirectly

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

363

be measured by 1) interpreting the sequence of GPS points (coordinates), 2) matching these to a digital road map and 3) summarising the variables along the observed route. However, the alternatives that the car driver may have considered could not be observed, but had to be generated synthetically before the model estimation (Nielsen and Jovicic, 2003). The first step in the model estimation based on GPS data was to estimate a route choice model based on this interpreted dataset. This model is described in the following.

18.5.1 Utility Function The route choice model was based on a linear utility function (3), as in the SP model. However, the error term is the sum of error terms at the arcs a along the specific route R. Since routes can be correlated (sharing links), H cannot be assumed to be independently distributed among alternatives – as in the multinomial logit model (Ben-Akiva et al., 1985) and neither can it be assumed identically distributed since splitting an arc into two in the digital map would imply a twice as high error term. Gamma distributed error terms overcome this if the variance is set to be proportional to the mean, since the Gamma distribution is non-negative, additive and overcome the overlapping route problem (Nielsen et al., 2002). The coefficients E could also follow distributions (error components as in equation 2). However, no distribution was assumed a priori (the empirical distributions were revealed). The variables are defined as in section 18.4.1. The kl is a constant factor for driving cost per kilometre, since no information on car-specific driving costs was available,

UR

VR  H R

E c ˜ rdprice  E l ˜ kl ˜ length 

E tff ˜ fftime  E tcng ˜ cngtime 

¦H

(3) a

aR

18.5.2 Estimation The estimation of the route choice model was carried out by running an all-or-nothing route choice model several times for each trip per person. Each run used different combinations of the coefficients in the utility function. The sums of the coefficients were restricted to one, since it is the ratios between the two that determine the choice. This lowered the possible number of combinations, which were pre-defined in a factorial design (Nielsen and Jovicic, 2003) to avoid building the calculation graph dynamically for each run. This speeded up the calculation time in the software (ArcGIS, see www.esri.com). In the initial analyses, 48 different combinations of coefficients were run for each trip (Menegazzo, 2003). This number of combinations was chosen as the best compromise between accuracy and calculation time. The best fit(s) to the observed route was recorded in each case. The fit was measured as the ratio of the length of the trip which had been fitted to the observed route. Since the network contains 350,000 links, this task was quite demanding. In most cases, several combinations gave the

364

O.A. Nielsen and G. Vuk

same fit (in the extreme case where a path is both the shortest and fastest). Each fit was then weighted proportionally to the number of best fits for the route it tried to match. A pilot study examined whether an added error term could improve the results (as in formula 3). However, this was seldom the case. The analyses were accordingly conducted using deterministic utility functions for each trip, as this improved calculation times dramatically. The interpretation must be that most of the heterogeneity can be explained by differences in the coefficients in the utility function and that the remaining unexplained variation cannot be explained by a distribution around the deterministic utility function. The RP model was based on a marginal cost calculation of fuel (8.5 DKK/l, 12 km/l) which resulted in a cost per kilometre of 0.7 DKK, which was higher than the SP (which is why the results in Table 18.3 and 18.4 were scaled to the RP values as well for the purpose of comparison). Using a lower or higher cost per km scales the VoT linearly which is why the same assumptions must be made to compare the results.

18.5.3 Within Person Variation The utility functions for all trips for each car could be compared in the RP experiment. About 2/3 of the participants had fairly consistent preferences, where the most utility maximising (rational) participant (car) in the experiment got the best route fit for exactly the same combination of coefficients (of the 48 different possible) for all trips. It was in general possible to fit the respondents’ routes quite well for cars with consistent preferences (in the 70-100% match interval). However, some participants had a very wide range of preferences. At the same time, it was difficult to match the routes (40-60% best match interval). It seems that these participants did not know the network very well. Manual analyses of samples of routes showed that they often followed illogical routes (i.e. not explainable after any reasonable criterion). It appears that more work needs to be done to explain the route choice of this group if this is at all possible. The best matches for a random participant in the experiment are shown in Fig. 18.5, where the preferences vary considerably. Using the most likely combination of coefficients (No. 47) may not be a good idea in this case, since it is the weighting of time components only (lexicographic behaviour on the edge of the distribution). It is better to use the mean of the values (used in the following) or the median of the distribution.

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

365

1,8 1,6 1,4 1,2 1 0,8 0,6 0,4 0,2 0 1

3

5

7

9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47

Fig. 18.5. Participant with varying preferences. The numbers at the x-axis indicate the combination of coefficients in the factorial design from 1 which is lexicographic in terms of length to 48 which is lexicographic in terms of time. The Y-axis is the likelihood of occurrence measured in %

It was examined whether there was any relationship between the best match to observed routes and socio-economic variables, i.e. if special segments of the population act more rationally than others. Such relationships could not be found. There was a weak relationship between the fit to observed routes and VoT though, but not significant (t stat in regression -1.16). The higher the value of time the less fit (decreasing from 70% in average to 60% for those with highest value of time). The interpretation can be that the persons with lowest VoT are most cost-aware and thereby examine the options and network more carefully.

18.5.4 Between Person Variation The distribution of the preferences between persons can be revealed by comparing the mean of each person's preferences. Fig. 18.6 and 18.7 show the distribution of value of travel times in the route choice models. The assumption in Multinomial Logit Models (MNL) that the coefficients are fixed clearly does not hold in this case. This was confirmed in the SP-based models, where the error component models had much better log-likelihood values than in MNL. The empirical distributions in the RP based models are by definition non-negative, as the coefficients have to be positive to make the route choice model work. The distributions are skewed to the right (the mean is larger than the median), and they look log-normal – although the number of observations is too small to determine this with certainty. Some of the users have very high value of time, which is due to lexicographic or near lexicographic behaviour (time minimisation only). The observations appear bundled at some few levels of outliers, which is because the factorial design of the search algorithm had fewer gridpoints in this area of the function.

O.A. Nielsen and G. Vuk

16 14 12 10 8 6 4 2 0

100% 80% 60% 40% 20% 245

230

215

200

185

170

155

140

125

110

95

80

65

50

35

5

0% 20

Frequency

366

VoT

16 14 12 10 8 6 4 2 0

100% 80% 60% 40% 20% 365

335

305

275

245

215

185

155

125

95

65

35

0% 5

Frequency

Fig. 18.6. Frequency and cumulative distribution of free flow value of time in DKK (7.3 DKK equals approximately 1 EUR). Mean VoT = 73 DKK median = 58 DKK

VoT

Fig. 18.7. Frequency and cumulative distributions of value of congested time. Mean VOT = 119 DKK median = 92. Note the different scale on the x-axis

The VoT from the RP based model is about one-third lower than the SP trip length experiment based on the same cost per kilometre. On the other hand, the RP VoT is about twice as high as the SP cost experiment. If the empirical distribution is fitted with a normal distribution, the best fit will most likely be around the mass of the distribution and with a right tail fitted to the high values. This again will imply a tail on the left side as well, since the normal distribution is symmetrical. The VoT calculation by simulation in Table 18.3 truncates the denominator, which to some extent takes care of this and should make the VoT estimates comparable, i.e. the last two rows in Table 18.3 should be compared with the mean VoT in Fig. 18.6 and 18.7.

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

367

120%

18 16 14 12 10 8 6 4 2 0

100% 80% 60% 40% 20% 2,40

2,20

2,00

1,80

1,60

1,40

1,20

0% 1,00

Frequency

Another issue is whether the two distributions are independent. The ratio between the two would then contain values below one. As can be seen from Fig. 18.8, this is clearly not the case. The smallest ratio is 1.2, i.e. the VoT for congestion time is always higher than for free flow time. The mean and median ratios are 1.6, which indicates a nearly symmetrical relationship between the two time components which was confirmed in a scatter diagram, not shown in this chapter. Basically, this illustrates that if a person has a high free flow value of time, then the value of congestion time is even higher. If a person has a small free flow value of time, then the congestion value of time is bigger than this. The ratio is in both cases symmetrically distributed around 1.6.

R a tio

Fig. 18.8. Frequency and cumulative distribution of the ratio between free flow value of time and congested value of time (shown in the X-axis – where ‘,’ should be understood as ‘.’). Mean = 1.6 median = 1.6

A regression between the two time-components gave 43.9 in t statistics for the gradient of the line (see Fig. 18.9). Hence, it can be seen that the assumption used in many ML models that the error components are uncorrelated with each other is clearly not valid in this case since there indeed is a clear relationship as shown in the figure.

18.5.5 Estimation of the Impact of Road Pricing on Traffic The estimation of route choice under the influence of road pricing was done in the same way as the model estimation in the control period. Only the high kilometrebased scheme has been analysed. It turned out to be difficult to estimate a utility function with one further variable than time due to current software limitations.

368

O.A. Nielsen and G. Vuk

VoT Congestion (DKK

Another approach was applied instead, where the model was re-estimated assuming the same coefficient for road pricing as the marginal cost (length multiplied by 0.7 DKK per km). This resulted in a slightly lower VoT than in the control period, i.e. a higher response on road pricing than if it was assumed equal to marginal cost. This is different from the SP, where the respondents tend to state less change in behaviour than if cost was equal (both the rdprice and costSP3 coefficients in the SP were lower, which equals a higher VoT than from the coefficient without road pricing, drvcost). In addition, the SP model contained a number of dummies for inertia, which would suppress changes further, while this in the RP is built into the coefficients. It should be noted though that the SP road pricing experiment was based on time-of-day decisions (but in a joint model estimation over all 3 experiments), whilst the RP data analysed all time periods. However, it is surprising that the road pricing in fact leads to such big changes in behaviour in the main RP experiment.

600

Observations Predicted Y

500 400 300 200 100 0 0

50

100

150 200 250 300 VoT Free Flow (DKK/h)

Fig. 18.9. Correlation between free flow VoT and congestion VoT (DKK per hour). The slope was tested with a t-stat of 43.9

The opposite was the case in a comparable experiment (Bonsall et al., 1998). However, the pricing schemes were defined differently, the urban settings were different and the pricing level lower. The high kilometre-based scheme in AKTA varied from 1 DKK per kilometre to 5 DKK per kilometre in the peak hours. The marginal driving cost was assumed to be 0.7 DKK (0.55 DKK in the SP). The increase in cost is accordingly quite large. The

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

369

cost level for public transport was about 1 DRK per kilometre at the time of the experiment (the average of the zonal-based price system).

18.5.6 Income Effect

VoT (DKK / hour)

Fig. 18.10 illustrates the VoT as function of household income, where the values for each income group have been averaged. The figure shows an increasing relationship. However, somewhat surprisingly with the highest VoT level at the third highest income level. This can be due to other socio economic attributes (e.g. higher time budget constraints in families with children, while the average age of the highest income categories must be assumed higher), or higher fixed budget constraints (e.g. housing); or the explanation may simply be the small sample size.

90 80 70 VoT 60 50 40 30 20 10 0 0

200

400

600

800 1000 Income (1,000 DKK)

Fig. 18.10. Values of travel time as a function of yearly cross-household income (1,000 DKK)

The figure explains the problems with the SP model estimation in section 18.4.2., i.e. where the only possible model had one VoT level under 400,000 DKK and another over, which seems to be the best specification. However, it does not explain why an SP-based model with a linear increasing VoT could not be estimated, except that the SP data may not include enough observations to estimate this interrelationship. This is similar to the SP estimation where the alternative specific constant is somewhat similar to the intercept. The relationship between income and VoT has substantial variation within each income category (Fig. 18.11). The regression is significant when the line is forced

370

O.A. Nielsen and G. Vuk

through the origin, but not otherwise. The slope had here a t stat of 12.8. However, if an intercept is introduced the slope is no longer significant. It is reasonable to assume that zero income implies zero VoT, i.e. no ability to pay.

Value of free flow time (DKK/h) )

300 250 200 150 100 50 0 0

100000 200000 300000 400000 500000 600000 700000 800000 House hold income (gross, DKK)

Fig. 18.11. Regression between Value of Time and income

18.5.7 Other Explanatory Variables In contrast to the SP experiment, it was possible to estimate some weak relationships between the value of time and other socio-economic variables based on the RP data. There was an increasing VoT for respondents who had fewer restrictions on time of the trip, i.e. from 60 DKK with 1) strong restrictions to 80 DKK with 4) no restrictions, over 2) to some extent and 3) to limited extent. The t-stat for the slope of this relationship was 1.74. The relationship is perhaps surprising as one could expect higher VoT with more time restrictions. An explanation can be that low income groups may have more time restrictions (e.g. fixed hours of work) or families with small children a tighter budget. The same tendency was the case for both free flow time and congestion time. Another finding was that when household size increased, VoT decreased (from 100 DKK with one person to 50 DKK with six persons, t stat of the slope -2.09). The reason here must be tighter budgets for children families and maybe also less income, since income increases with age. The value of travel time increased with age from 50 DKK for the 20-30-yearold group to 95 DKK for the 60-70-year-old group (t stat of the slope 1.96).

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

371

18.6 Comparison Between SP and RP Models It was obvious, that the participants had little idea of the real marginal or average costs of driving. Depending on the formulation of the question, different answers were given which were inconsistent with each other. The value of travel time from the SP experiments that traded off cost with time differed by a factor of 2-3 from the RP data. The SP experiment which traded off trip length and times was in the same magnitude as RP, although not equal. Using error components improved the SP based models significantly. A substantial heterogeneity of the coefficients was found (high variance compared to the mean). Similar large heterogeneities were found in the RP data. The RP data indicates that the VoT follows log-normal distributions and that the VoT of different timecomponents are highly correlated. About one-third of the participants in the RP experiment had non-explainable behaviour, whilst the others clearly were utility maximisers. The SP-based logit models assume utility maximisation with an error term. Since the choices here are binary, the unexplained variation will be taken care of by the alternative specific constants and the error term. The RP data showed an increasing VoT with income. This could be assumed to be linearly increasing or clustered around one level below 400,000 DKK in household yearly cross income and another level over. The SP model could only be estimated as a two-level VoT segmentation as there were fewer observations than in the RP data set. A split at 400,000 DKK gave the best fit, which was confirmed by the RP-based model. It can therefore be concluded that both methods led to the same main causal relationship, but that the RP model could refine the functional form of the utility function compared to the model based on the SP experiment. It is not easy to compare absolute numbers of VoT from the RP and SP models due to different assumptions of costs per kilometre as well as problems with calculation of VoT from the model with normally distributed coefficients in the utility function. However, it appears that the SP in the same formulation as the RP (trip length-time trade offs) overestimates VoT slightly. The ratio between different timecomponents (free flow time and congestion) remains fairly equal and the magnitude of heterogeneity is similar as well. The SP indicates that the value of road pricing is higher than the value of marginal driving cost (fuel mainly). However, both cost coefficients became less negative compared to the situation without road pricing (VoT increased). The RP on the other hand showed that the participants changed behaviour slightly more (i.e. a lower VoT) than expected, assuming that the value of money was equal. It must be concluded that respondents tend to underestimate their behavioural changes due to road pricing in the SP experiment compared to the field experiment, where the money is real. Alternatively, they may answer strategically in the SP. This is surprising since they could be expected to change their behaviour less in real life due to timeconstraints (constraints in their time-budget), habit or lack of reasonable alternatives (Bonsall et al., 1998).

372

O.A. Nielsen and G. Vuk

It was possible to investigate the issues of heterogeneity, income-dependent VoTs and other socio-economic variables in more depth using the RP data instead of the SP data. Although RP data is more difficult for model estimations due to a correlation between variables that cannot be controlled as in the SP, the RP data contained far more observations per participant. The SP consists of 3 sub-experiments with 6 tradeoffs per participant. Whilst the RP contained between 200 and 1,000 trips per car, each trip with an individually estimated utility function, which resulted in about 100,000 observations in the present model. It should be noted that the budget for the full RP AKTA experiment was over 10 million DKK, whilst the AKTA SP budget was only about 0.5 million DKK. The RP contains much more information, but the SP provides – if formulated as length versus time experiments and EC models – good indicators of the behavioural responses.

18.7 Summary and Conclusions The chapter has presented some results and analyses concerning inconsistencies in respondents' answers depending on how questions are formulated, variables defined and whether it is a Revealed or Stated Preference experiment (RP or SP). All questions were asked to the same group of participants, i.e. the same person faced different questions within the same SP, followed by the RP over a minimum of a 16week period (two experiments of 8 weeks each). As the RP data was collected by the GPS technology, the behaviour could be measured very accurately. The respondents’ evaluations of driving cost were obtained by various approaches. They were asked to confirm or modify cost and trip length estimates. Although costs a priori were calculated as a function of the trip length, only 18% of the participants modified this whilst 43% modified the length. They were then asked about their average driving cost per kilometre. This turned out to be highly inconsistent with the cost per length they had accepted before during the same interview/experiment (often by more than a factor 2). Finally, they were asked which elements they had included in the calculation of cost per kilometre. This again was for many respondents highly inconsistent with their prior answers. It appears that car users have very vague and erroneous knowledge of the cost of driving, whilst they have a fairly good knowledge of the travel length and time. This is critical when car users’ responses to road pricing are estimated and modelled, i.e. the predicted impact of a pricing scheme may depend on the way the questions are formulated. The SP experiment then focused on Value of travel Times (VoT), where half of the respondents chose between time and length and the others between times and cost. Cost was defined proportionally to the trip length and estimated based on the respondents usual trip. The resulting VoTs differed considerably by the two methods with up to a factor 3. However, the ratio between free flow VoT and congestion VoT was comparable. The SP model based on length versus time trade-off and cost-estimate based on true marginal driving costs gave almost identical VoTs as the AKTA RP data. If both costs and travel times are changing, this ratio is especially critical. A simultaneous change of both variables is likely to be the consequence of a road pricing

Car Users’ Trade-Off Between Time, Length, Cost and Road Pricing

373

scheme, where cost increases whilst traffic and hereby congestion and travel times decreases. The SP also investigated road pricing for existing trips, which was then charged. This led to a higher VoT than the RP route choice experiment (about 50% higher), whilst the coefficient of road pricing was about 20% higher than the pure cost coefficient. The interpretation must be that car users dislike road pricing more than other out-of-pocket costs but that their behaviour nonetheless changed to a less degree than expected. This must be interpreted as some inertia for changing behaviour (which was also described by a dummy coefficient in the SP model), this, however, is somewhat counterintuitive compared to the higher value of money (coefficient) for road pricing. The AKTA RP provides a clearer conclusion; the participants changed their behaviour to a lesser degree (about 20%) than expected if the value (coefficient) on road pricing where the same as marginal driving cost. The estimated utility functions confirmed this both in their coefficients and value of time. The interpretation must be that respondents in SP may overestimate their changes in behaviour – or answer strategically – but when faced with real life restrictions in time and space they retain their existing pattern (perhaps due to habit, convenience or lack of knowledge of alternative options). All models were initially estimated as traditional logit models, followed by a reestimation in a Mixed Logit framework (ML). The ML models based on SP were estimated by maximum simulated likelihood. Different distributions were applied – first normally in different combinations on the coefficients, then log-normal and finally simultaneous distributions (to allow for correlation between the distributions of the coefficients). The different assumptions did alter the results, but not to a large extent. Models with random coefficients (ML) improved the models significantly and there are therefore no reasons to assume that a model with fixed coefficients (MNL) is valid. The RP data contained enough information to estimate a model per participant. The empirical distributions of the coefficients within and between respondents could accordingly be derived. Statistical distributions were fitted to this, where the lognormal gave the best result. In this context, it is interesting to note that some of the car users had lexicographic behaviour – i.e. that the cost coefficients were zero within their available choice set and the VoT was therefore in principle infinite. The distribution was accordingly skewed more to the right than the log-normal. Another conclusion from the RP data was that about 2/3 of the car users had a clearly utility maximising behaviour, i.e. that their choices could be explained with a good fit by a utility function where the coefficients between trips had a reasonably small variance. However, the remaining 1/3 of the cars had a very high variance of the coefficients – they sometimes minimised costs and at other times time – at the same time as a less good fit was obtained in relation to the observed routes. One explanation may here be that each car could be used by different members of a given household. Both the RP and SP models were tested for income effect. Both as dummies, piecewise coefficients and a full specification of income effect in the indirect utility function following a micro economic framework (cost-dependent cost-coefficients). By experimenting with the interval, the SP model could be estimated with VoT in two

374

O.A. Nielsen and G. Vuk

intervals (all other models gave illogical signs and small t-values). The RP data could reveal VoTs for each of the 9 income intervals for which people have provided information. The most reasonable functional form to explain this reproduced the findings from the SP. The overall conclusion is that car users are much better at estimating trip length than cost. There are many heterogeneities in behaviour and preferences, which can be fitted better by log-normal distributed coefficients than by normally distributed ones. The distributions of different time components are most likely very correlated. SP can lead to results of the same magnitude as RP if it is designed the right way. However, in the worst cases there could be a substantial difference between the results and even the signs of the impacts could differ.

References Ben-Akiva, M., Lerman, S.R. (1985). Discrete Choice Analysis; Theory and Application to travel Demand. MIT Press, Cambridge, MA, USA. Ben-Akiva, M., Bolduc, D., Bradley, M. (1993). Estimation of Travel Choice Models with Randomly Distributed Values of Time. Transportation Research Record, 1413: 88-97. Bonsall, P., Cho, H-J., Palmer, I., Thorpe, N. (1998). Experiments to determine differences in drivers' response to a variety of road user charging regimes. The PTRC conference. Cambridge UK, September. Menegazzo, P. (2003). Estimation of route choice models on GPS data. Joint master thesis between the University of Padova, Italy and the Technical University of Denmark. Nielsen, O.A. (2004). Behavioural responses to pricing schemes: Description of the Danish AKTA experiment. Journal of Intelligent Transportation Systems, 8(4): 233-251. Taylor & Francis Ltd. Nielsen, O.A., Jovicic, G. (2003). The AKTA road pricing experiment in Copenhagen. 10th International Conference on Travel Behaviour Research. Proceedings, session 3.2 Valuation/Pricing. Lucerne, Switzerland, August. 2003. Nielsen, O.A., Sørensen, M.V. (2004). Sensitivity of variable definitions in SP analyses – An empirical study of car users’ evaluation of length, cost and time. World Conference on Transport Research Society (WCTRS) Proceedings, D01 Paper 1407, July 7th, Istanbul, July 4-8. Nielsen, O.A., Frederiksen, R.D., Daly, A. (2002). A stochastic multi-class road assignment model with distributed time and cost coefficients. Networks and spatial economics, 2: 327-346. Kluwer.

19 The Impacts of e-Work and e-Commerce on Transport, the Environment and the Economy Andy Lake HOP Associates, Cambridge, England

Abstract This chapter provides an overview of recent research into “Virtual Mobility”, set against the context of European policy development. Based on current work with the UK Department for Transport, it explores the various impacts of e-work (telework) and e-commerce on transport, the environment and land use. In an area where there is much speculation, the approach taken is to distinguish between empirical data and conjecture, showing where significant evidence-based results have been obtained and where conclusions can be drawn to influence transport policy and analysis. This chapter also highlights where further research needs to be undertaken, and suggests an alternative approach to the “substitution versus generation” debate.

19.1 Being Active Without Moving Historically, we have needed transport for carrying out many of the key activities in life – working, shopping, learning, having fun and undertaking many kinds of social interaction. Either we need to travel to do them, or we have needed to have physical objects sent and received. The new information and communications technologies (ICT) allow us to change the way we undertake these activities. “Telework” or “e-work” allows us to work from anywhere – the commute trip to the office is not a necessity. E-commerce allows us to buy and sell online, without necessarily having to undertake a physical journey. It is important for transport analysts, planners and policy-makers to understand the transport effects of these new technologies. The key questions for the transport analyst are: x How is travel replaced, generated or modified by applications such as e-work, e-commerce and other electronically mediated activities? x How is the movement of goods replaced, generated or modified by applications such as e-commerce and electronic service delivery? Over the past decade or so, a kind of sceptical orthodoxy has arisen in these matters, dismissing telework as “the future that never arrives” or proclaiming that ICT applications generate as much or more travel than they eliminate (Black, 2001) which, amongst other targets, debunks claims for ICT-based transport

376

Andy Lake

substitution. Less provocatively, influential US pioneers of research into telework such as Salomon and Mokhtarian have modified some of their more upbeat early findings to conclude in favour of complementarity between online applications and physical travel. This view has arisen in reaction to earlier over-excitable claims from futurists and business interests in the 1970s and 1980s. There is, then, a key research task: to steer a neutral course and establish whether online applications – in the current jargon “e-work”, “e-commerce” and “e-services” – can or cannot be shown to have a quantifiable traffic reduction effect. Quantifying the transport impacts also relates directly to the issue of whether these e-applications contribute to environmental sustainability. There is increasing emphasis in European and national policies on the contribution that ICT can make to reducing the need to travel and “cleaning up” economic growth. The goal is a “sustainable information society”, where economic growth is decoupled from transport growth (e.g. European Commission, 2000a; European Foresight for Transport, 2002).

19.2 The UK Study In order to begin to find a way through the claims and counter-claims, the UK Department for Transport (DfT) has undertaken a project (Lake and Charrett, 2002, and Lake, 2003) to build a “knowledge base” of research literature examining the effects. The work has been carried out by HOP Associates and the Transportation Research Group at the University of Southampton. Results can be found online at www.virtual-mobility.com. The project tackled these questions through a critical review of existing studies, focusing on the evidence for transport impacts. While the value of more speculative studies is accepted, the focus of the research was to determine what is actually known, and what remains conjectural. So for the purposes of this study, a premium has been put on studies that include robust empirical data. As far as possible, data sources and consistency of methodology were evaluated, and an assessment made of the accuracy of the results and the transferability of the research into the UK situation. It was important at the outset to distinguish the scope of the field. For the majority of transport engineers and researchers, ICT is about managing transport through technology, rather than replacing it. This study focused on “virtual mobility”, which overlaps into a number of other existing fields of study. The figure below sketches out the field of the study.

The Impact of e-Work and e-Commerce on Transport

377

Fig. 19.1. The field of study

As this is an interdisciplinary field, criteria were developed to assess strengths and weaknesses of the factors considered – understanding of transport, of technology, and of wider socioeconomic factors. The primary focus of the study was on e-work and e-commerce, with reference also to the potentially important area of e-services, although in this latter area there is an absence of empirical studies related to transport effects.

19.3 E-Work E-work encompasses a number of different fields, which may be summarised as: x Telework o home office/home as base o telecentre-based o satellite office/local office/facilities exchange o mobile teleworking o working form client sites x

Online collaboration (replacing meetings)

x

Videoconferencing

x

Remote diagnostics and monitoring

x

Electronic service delivery/customer service (i.e. changes in location for workers when e-services and e-commerce are introduced)

378

Andy Lake

Changing working practices and business processes to introduce any of the above necessarily has an impact on travel behaviour. However, it is studies of homebased and telecentre-based telework that provide by far the largest amount of empirical data about the relationship between ICT-based activity and travel behaviour. Probably one reason for the concentration on these forms of e-work is that journeys are relatively easy to count - at least in terms of commute trips replaced – and can very easily be incorporated into surveys and travel diaries. The very concept of “telecommuting” – the more common term in the US – emphasizes an impact on the home-to-work commute trip. The other forms of e-work listed above have received little research coverage so far. The evidence from implemented schemes is overwhelmingly of significant travel reductions per teleworking occasion (which to some extent is self-evident), and per teleworker. The following table gives examples of the travel reduction found in studies using travel diaries: Table 19.1. Examples of travel reductions reported Study Hamer et al. (1991)

Travel reduction per teleworker 17% total trip reduction 26% reduction in peak-hour car travel

Koenig et al. (1996)

27% total trip reduction Non-commute trips increase by 0.5 trips per person per day

Surrey County Council (1999)

Average19% reduction in trip length to work for staff using telecentre

Glogger et al. (2003)

19% total trip reduction 43% reduction of commute trips

Strict comparison is difficult, as studies tend to take different approaches to recording and interpreting data, and in particular in factoring in non-work trips and travel by other household members. But there is now a healthy number of robust studies that provide clear evidence of an immediate transport reduction effect. This effect endures when direct rebound effects are taken into account (which we discuss below in the “rebounds” section). In the studies as a whole, it is possible to identify a typical range of 20-50 miles saved (PMT) per teleworking occasion, and a typical range of 1300-3500 miles saved per teleworker per year (with calculations in studies usually based on teleworking for between 1 and 2.5 days per week according to the particular implementation). There are a number of studies that have shown higher than average reductions:

The Impact of e-Work and e-Commerce on Transport

379

Table 19.2. Travel reductions per teleworker Study Mitchell and Trodd (1994)

Commute travel reduction per teleworker 42 miles per teleworking occasion 113 miles per week

Lyons et al. (1998)

58 miles per teleworking day

Mokhtarian et al. (1997)

51 miles per teleworking occasion (centre-based) 34 miles per teleworking occasion (home-based)

Hills et al. (2003)

90 miles per week for car users 124 miles per week for rail users

Hopkinson et al. (2001)

90 miles per week (full-time home-based) (= 18 miles per day)

Mileage reductions are found whether teleworkers are home-based or centre-based. The unexpectedly high mileage saving from the centre-based survey is attributed to the teleworkers who used the centre having even longer commute trips than their home-based colleagues. Table 19.2 includes two teleworking case studies where higher mileage savings are achieved – at the telecoms giant BT (Hills et al., 2003) and for call centre agents at the AA, providing a roadside assistance service (Hopkinson et al., 2001). In both cases, the teleworkers work from home for more days per week. At BT, extensive provision of remote working technology and a culture that encourages remote work help to achieve these higher levels of commute reduction. The AA case is of their "virtual call centre" where staff work entirely from home – a comparatively rare practice as yet. It is also worth noting that the AA call centre workers are non-typical as teleworkers in other ways, being female, on lower income and with previous average commute journeys pretty much at the national average. This is a reminder that the majority of studies reflects patterns of early adoption, and that future patterns may well be different. An assumption found more in the transport literature than in the technological or business literature is in the tendency to treat telework as a single phenomenon regardless of the technologies and business processes involved. The quality of the telework experience would appear to have an impact on the frequency of teleworking. For example, the AA virtual call centres are set up with seamless access to all the office systems, to allow 100% home-based teleworking. At the European Commission (European Commission, 2000b, three categories of telework (“hard”, “medium” and “soft”) were introduced on the basis of selected staff’s need to telework. Different levels of technology and access to office systems were provided accordingly. In each case, frequency of telework occasion is directly linked to technology provision. In turn, this impacts on the travel substitution potential. This is an area that needs to be studied more closely. Various attempts have been made to aggregate potential travel reductions through telework to the city, state or national level. While they often raise interesting conceptual issues, they are almost entirely unconvincing due to data problems and methodological weaknesses.

380

Andy Lake

A kind of methodological orthodoxy has emerged for speculative studies, based on estimating “telework penetration” in the workforce and levels of teleworking (the numbers who would be teleworking at any particular time). As large-scale studies of teleworkers and their travel behaviour have not been undertaken, a variety of proxy indicators have been used – numbers of white-collar workers in census data (e.g. Gillespie et al, 1995), home PC ownership, Internet connections, numbers of self-employed, growth of services sector, and so forth. At some point in the mid-1990s, influential US researchers proposed a notional ceiling of 40% of the workforce as the upper limit for teleworkers. This is frequently cited, but it is a far from robust figure. The figure of 1.5 days per week as an average for the number of teleworking occasions is somewhat more robust, based on measured studies of implementations. But studies also show that it is difficult to generalise about how that will translate into substituted trips. Similarly, calculations about new trips generated that should be deducted from the savings are far from convincing at this aggregate level, and “back of the envelope” calculations are frequently cited as gospel. A good example concerns the extra trips generated due to urban sprawl, caused by teleworkers moving further from their workplace (US Department of Energy, 1994). Though the authors of this notional clawback make it clear this is guesswork, it reappears as authoritative in many other studies. In terms of total reduction potential, conservative commentators put estimates in the region of 1-2% (e.g. Mokhtarian, 1998; Dublin Transportation Office, 1998), while the most upbeat of the recent studies forecasts that in 2010 homeworking and telecentre working will reduce car commuting by 15% from its forecast level, business travel by 5%, and car shopping trips by 10% (Dodgson et al., 2000). This range of forecasts highlights the need for further research, new forms of data collection and new approaches to modelling. Probably one of the greatest weaknesses in our understanding of the impact of telework on transport so far has been an almost exclusive focus on the commute trip. A few studies have looked at the impacts on business travel. Some measurement has been done of the transport effects of introducing teleworking for field staff (e.g. Jupp, 1998; Hills et al., 2003). This is an area urgently in need of further research.

19.4 E-Business and e-Commerce The study evaluated literature examining e-business impacts on transport, but here we will comment primarily on findings relating to e-commerce. E-business describes the range of online business processes within companies or encompassing their external relationships with partners, customers and suppliers. In one sense, e-commerce is a particular subset of e-business activities focusing on marketing, sales, purchasing and order fulfilment.

The Impact of e-Work and e-Commerce on Transport

381

E-business processes may have a transport effect, primarily by improving efficiency in inventory management and logistics, and by moving the paper trail of business organisation online. In our field of study, business-to-consumer (B2C) e-commerce has the added interest of impacting on both passenger travel and freight movements. B2C operations will in most cases be supported by business-to-business (B2B) ecommerce operations as well. There are, however, very few studies of e-commerce impacts that include robust empirical data. In the absence of numbers, there is a growing literature focusing on the principles underlying the interaction between ecommerce and transport, and making projections based on various scenarios. Millard-Bell, in an article in Local Transport Today, cites studies suggesting increases in the number of light van movements of up to 17% (Millard-Bell, 2000). But the article concluded that it was difficult to quantify the exact impact of ecommerce on freight movements as it was still in its infancy. In some ways, this sums up the current state-of-the-art. The following paragraphs summarise some of the speculative and semi-speculative findings that illustrate the wide variance of opinion. Cairns et al. (2004) report that replacement of the private car by delivery vehicles for food shopping could reduce food shopping travel by 70%. For nonfood shopping, 64% of online shoppers report that it had saved them a trip. Other work (e.g. Easley and Easley, 2000) suggests that there could be substantial increases in delivery vehicle movements as a result of increased e-activity. Travel survey data from both Sydney and South-East Queensland in Australia, however, show that the majority of shopping trips are short trips. According to Nariida Smith, ‘It is less likely that 5-minute dashes for a loaf of bread or carton of milk will be replaced by online orders, except to customers willing to pay a premium’ (Smith et al., 2001). Intuitively, this sounds correct. However, this remains conjecture, and in a fast changing world it can be dangerous to base predictions on current practices and expectations. Who 30 years ago would have predicted that millions of people would order home deliveries of products like pizzas? Round-the-clock convenience does appear to be something people are willing to pay for. Some literature discusses the ways in which companies that are starting to develop B2C e-commerce might develop in the future (e.g. Marker and Goulias, 2000; Browne, 2001; Hultkranz and Lumsden, 2001). There is a general assumption that growth in ICT will continue to open up new markets and increase revenues. Browne predicts that B2B e-commerce will lead to new group distribution strategies, becoming more demand responsive with shared information leading to a reduction in overall aggregate inventories. The potential impact of ICT on warehousing has been raised, with a reduction predicted as companies ship directly to customers. In Finland, an early Internet adopter, demand for warehousing space is declining, with a concomitant rise in the mean distance travelled by vehicles between warehouses and retailer (Kilpala et al, 2000). The number of retail stores is also diminishing as planners opt for over-sized trade centres. This decreases the distribution of freight by trucks, but increases the mileage of private cars.

382

Andy Lake

Furthermore, it has been suggested that an increase in Internet shopping may render conventional supermarkets unprofitable leading to three possible outcomes: x Abandoning retail stores and delivering from warehouses. x Abandoning large retail stores and adopting a boutique strategy. x Maintaining large retail stores and competing on price. The first two options in particular have significant land use implications (Marker and Goulias, 2000). Several commentators also suggest that the impacts of e-shopping may not be as great as suggested. Shopping is a social and recreational experience that cannot be replaced by e-shopping, although some kinds of shopping are more attractive than others (Gould and Golob, 1997). Shopping involves information gathering as well as the actual purchase and e-shopping may not be able to supplant other means of information gathering, although information requirements for grocery shopping may be limited. Salomon was one of the first of many to express the view in this context that humans need to travel out of boredom or a sense of exploration (Salomon, 1986). Any reduction in travel through technology may be offset by increased latent travel demand. Telecommunications may generate trips that would not have occurred without them. But, as with telework, even 18 years after Salomon’s observation, there is little beyond speculation about trip generation effects. The conjecture may be plausible, but it remains conjecture. One study that does attempt to provide comprehensive insights into the traffic effects of e-commerce (as part of a comprehensive approach to the effects of all ICT applications) is the Motors and Modems Revisited study (Dodgson et al., 2000). Reduction (from predicted levels) of car travel for shopping is forecast to be 5% by 2005 and 10% by 2010. Delivery trips, primarily by light goods van, are predicted to increase by 0.25% and 0.5%, and freight movements by HGV are expected to decline by 17% and 18.5% by these dates. How these figures are reached is not entirely clear. In the case of the decline in shopping trips, an extra weighting appears to be given to the rise and potential of new technologies. In this respect, most other commentators are more cautious, foreseeing other obstacles to their use as well as greater trip-generation potential. The HGV figures are based on a comparatively greater emphasis on the dematerialisation of products, ICT-based efficiencies in the supply chain and logistics, and displacement of some road freight to rail and water-based transport as ICT helps these alternative modes to become more competitive again – usually in the form of combined transport options with ICT smoothing out the cross-mode logistical difficulties. The comparatively few extra trips for light goods vehicles seems to hinge on the principle that most extra goods ordered online will be carried by existing services, adding relatively few trips and implying greater efficiency. Many commentators predict that not all businesses using e-commerce will wish to develop their own logistics capabilities, leading to a growth in 3PL (third party logistic) providers. Marker and Goulias (2000) have stated that 3PL companies could consolidate deliveries for B2C transactions, although the number of customers who shop from multiple stores, or the number of customers each store has, is unknown. Hopkinson and James (2000) suggest a quantitative threshold for

The Impact of e-Work and e-Commerce on Transport

383

the take-off of B2C e-commerce, which will trigger major changes in distribution patterns. It may also be the case that there is a qualitative threshold, where online activities become a more natural thing to do. A weakness of many studies seems to be an assumption that if you have a computer and telecommunications all the options are open. This is clearly not the case: issues such as pricing plus level, quality and reliability of service, etc. are very important. Studies need also to analyse patterns of market penetration – something which so many of the early ecommerce entrepreneurs and investors failed to do successfully. In some ways, in the heady pioneering days of e-commerce, it may be rash to build scenarios around “first wave” practice. For example, the WebVan experience is cited by some commentators as one possible model for future practice. WebVan was a third party provider, and unfortunately one of the more spectacular victims of the bursting of the “dotcom bubble”. The downturn in its fortunes undermines the credibility of the business model and the projected transport effects built on it. However, a similar implementation by other entrepreneurs in more propitious times may prove to be more successful. The lack of sophistication mentioned above in studying telework is generally evident in the study of e-commerce. While some studies have noted the different models of distribution, few have noted the variety of different business implementations of e-commerce, e.g.: x Amazon – completely online sales operation, sourcing products from other producers for repackaging and distribution x Dell – online sales operation, producing own product to order x Tesco – existing retailer supplementing bricks and mortar operation, fulfilling online orders from designated local stores and warehouses x Next/Debenhams – existing retailers supplementing bricks and mortar operation, fulfilling online orders from centralised warehouse also used for mail order operations x Dematerialised product providers – buying of music, e-books, software, etc. online with fulfilment and delivery also online So there is no simple model for determining transport impacts – transport from manufacturer to wholesaler, to retailer to customer – all of which will vary according to both the kind of product and the business model. What is required is a period of monitoring, clearly differentiating between the different models and their diverse effects, so that analysis and policy-making have robust foundations.

19.5 Rebounds Crucial to understanding the overall impact of ICT on transport is the need to understand the nature and extent of any rebound effects. Two factors have hindered this so far: x A tendency to prefer plausible speculation to measurement

384

x

Andy Lake

A lack of clarity about how far to cast the net when fishing for rebound effects

In study after study that we reviewed, measured implementations of telework contained a cautionary but conjectural “epilogue”. This normally says something to the effect that while certain travel savings had been recorded, factors x, y and z could lead to a reduction of these benefits. These rebound issues can be summarised as: x To what extent will new trips be made by the home/telecentre worker during the course of the day that would otherwise not have been, or by other family members using the car? x How proportionately will transport substitution affect different traffic modes (e.g. will regular public transport users become occasional car users)? x To what extent will latent demand be realised by other road users taking advantage of "liberated" road space? x Will ICT in due course affect location decisions so that people will tend to live further from their places of work, and therefore make fewer, but longer trips, and contribute to urban sprawl? It is important to note that in most studies these are posed as questions, and the effects are almost never measured. The minority of telework studies that attempt to tackle these questions by counting alternative trips for the teleworker and their households nevertheless still show a net reduction of trips even allowing for compensatory trips. For example, a study of US telecentres showed that on average teleworkers increased their number of home-to-work trips because they were more likely to go home for lunch. Nonetheless, their total mileage was substantially reduced, as the telecentre was much nearer home than the “normal” workplace. More interesting than the simple plus-and-minus “substitution versus generation” game is analysis of the redistributive effect on household activities and associated travel. A spatial analysis of the activity space of telecommuters and their household members by Saxena and Mokhtarian (1997) found that on telecommute days around 30% of activities (including shopping and leisure) are performed closer to home than on non-telework days, and that destinations on telecommuting days are more evenly distributed in all directions around the home, whereas a majority of destinations on commuting days are oriented toward the work location. Other studies have found similar, and sometimes counter-intuitive results, with trips for the whole household reducing after teleworking begins (e.g. Hamer et al., 1991). A recent study in the Munich area by Glogger et al., see chapter 21, found significant redistribution effects in household activities – redistribution of tasks by household members, redistribution of timing of some activities, and some changes of mode for some trips. Similar phenomena are found in the BT case study (Hills et al., 2003). It is also sometimes forgotten in studies into ICT/transport impacts that all demand management measures have rebound effects, but these are generally not the subject of close scrutiny. Questions are being asked of the impacts of e-work and e-commerce that are not being asked of other mode-shift measures.

The Impact of e-Work and e-Commerce on Transport

385

Someone who cycles or catches a bus to work will need replacement trips for shopping, as much as an e-worker. Each alternative mode involves leaving a car at home, potentially for use by other household members. Amongst these options, public transport is unique in that it requires by its nature additional trips by another mode to be generated in order to use it (unless the worker lives and/or works at a public transport depot!). We could define public transport as “modes of travel that take you from where you are not, and deliver you to where you don’t exactly want to be” and so is likely to generate complementary car travel on many occasions. There is substantive evidence that the promotion of park-and-ride facilities generates additional car travel (e.g. Parkhurst, 2000): nonetheless it is considered beneficial to reduce car trips from congested urban areas even if it risks increasing them in less congested rural areas. Marshall and Banister’s (2000) comparative study of travel reduction strategies is instructive in this regard. This examines teleworking alongside other intended travel reduction measures such as cycling measures, park and ride, improved parking information, city centre restrictions and changes in urban form – all of which have both benefits and negative rebound effects When considering rebound effects, the wrong approach is to tally the positive impacts and the negative rebounds and simply put them in the balance to see whether it all adds up to the initial e-activity being worthwhile. The point of policy is to identify and maximise the positive impacts, while identifying and combating any negative impacts.

19.6 Environment Many studies indicate a reduction in energy consumption as a result of telework. Usually this is a direct extrapolation from reduced car use, calculated either from directly observed monitoring in case studies or from application of hypothetical reduction in car use applied at a local, regional or national level using transport statistics and/or models. Many individual studies combine an element of the latter approach on the basis of “if this pattern were followed at a national level…” In Japan, teleworking has been incorporated as a measure to be promoted in order to help the country meet its emissions reductions targets under the Kyoto protocols. In the US, studies by the US Department of Transportation (1993) and the US Department of Energy (1994) both attempted to quantify the total amounts of pollutant emissions that might be avoided by the US as a whole given predicted levels of telecommuting uptake. The latter study also attempts to factor in allowances for urban sprawl, though it is not clear what this calculation is based on. It is interesting that some studies working on this basis attempt quite detailed calculations. For example, the Department of Energy study attempts to factor the impacts of more cold starts and shorter journeys in reducing the assumed beneficial effects of teleworking in reducing pollutant emissions. While the principle may be

386

Andy Lake

important, the margin of error in the base figures is so wide that such tinkering at the statistical edges is far from convincing. A study at the Department of Employment in Sheffield (Wright, 1997) adopted a distinct approach, measuring total energy consumption in the office, in teleworkers’ homes and in their vehicle use. This found that home working produces around 80% reductions in energy use in an “ideal scenario” – one where not only the commute trip is eliminated but where the office space is also decommissioned. The study is useful in three respects. First, it measures every aspect of energy consumption at home, on the road and in the office. Secondly, it serves as a reminder that the assumed economies of scale that offices theoretically achieve tend not to occur. This is mainly due to systems and equipment being always on – lighting, heating, ventilation, IT, photocopiers, drinks machines, etc. – whereas in the home these are only on when needed. The extra lighting and heating etc. often mentioned for home-based teleworking was found to be less than expected. Thirdly, around two-thirds of an office worker’s energy consumption is attributable to travel. The Sheffield study appears to be one of a kind. Though the 80% reduction figure may raise some eyebrows, and would require extensive demolition of offices under some kind of teleworking command economy to achieve, the study has clearly taken a rigorous and all-inclusive approach to the measurement of energy consumption. Further studies of this nature, in a range of contexts, would be welcome. A field of study very much in its infancy is illustrated in the works of Forseback (2000, 2001) for the European Commission. These look at the wider environmental impacts of the “Information Society” as a whole. The studies include numerous case studies of ICT-based processes, products and services reducing resource and energy consumption, and various approaches to measuring the impacts. The resource use, energy consumption and pollutant by-products of transport are one element of the wider picture. The case study approach makes for interesting reading, though it raises questions about the underlying methodologies of the case studies cited. These works also explore and attempt to quantify aspects of ICT use not covered in the other literature on e-commerce and ICT. In particular, the “dematerialisation” of products and services raises important issues that require further study. That is, products or services that previously were of a physical nature requiring physical movement to access or distribute are being converted to or supplanted by new ICT-based alternatives. Examples such as the replacement of answering machines by voicemail are given. Here, a physical product brought into the home is replaced by an electronic service located at the exchange, and requiring no transport to implement it. One can think of other examples that are in principle measurable in terms of transport effects: the closure of branch offices and demise of travelling representatives in the financial services sector, being replaced with call centre based services. Just beginning are major developments in the music industry that most commentators think will lead to major changes in the distribution of musical products for sale. The bulk production of CDs and distribution from factory to retail outlet is

The Impact of e-Work and e-Commerce on Transport

387

expected to be largely replaced by online products and electronic distribution over the next few years. It would be advantageous to undertake research into such phenomena to monitor their effects as they evolve. Some recent comparative case studies of actual and hypothetical e-commerce implementations have adopted a life-cycle analysis approach (e.g. Williams, 2002; Reijnders and Hoogeveen, 2001). In these, transport is one component in total energy usage. Studies so far tend to indicate that online retail operations as a whole use as much energy as traditional operations. They indicate that customer travel does decrease – but additional freight movements plus other items such as additional packaging mean that there is no overall reduction. However, while this approach has considerable benefits, the increased freight and reduced customer travel seem as much to stem from assumptions as from observation. In general, commentators conclude that transport substitution through e-work and other aspects of ICT use will have only a very limited effect in reducing energy use and pollutant emissions. Better home insulation and more fuel-efficient vehicles are amongst the comparisons made as better options for improving environmental performance. And, of course, a key issue is that these ICT innovations also take place in the context of the pursuit of economic growth.

19.7 Economic Growth, Transport Growth and European Policy Promoting ICT has become a key plank in European policy and in the policies of most national states. This is partly in terms of its expected benefits for job creation, learning, creating more efficient administrations, etc. Information Society policy is now enshrined in an Action Plan, eEurope 2005: An Information Society for All (European Commission, 2002). In addition to these benefits, ICT has now become the glue that holds together arguably incompatible policies and aspirations. European economic policy is committed to continuing prosperity and economic growth. Transport policy is committed to increased mobility, not least with regard to European enlargement and the expectations of the accession states for increases in trade with other member states. The links between increased mobility and economic growth have long been recognised, and key instruments of economic development have been (and still are) the building of new infrastructure for transport. At the same time, European policy, not least in the transport arena, is committed to sustainability, and aspires to decoupling economic growth and transport growth. In this respect, ICT is coming to be invoked almost as a magic wand that has the potential to reconcile these policies. “Virtual mobility” at a conceptual level continues the linkage between increased mobility (access to more places, and increased speed of access, almost to the point of instantaneity) and increased prosperity. Similarly in the UK, the DfT and a large number of local authorities are promoting telework as a component of company Travel Plans. It is probably fair to say that the policies are running ahead of the evidence. Although the benefits from increases in e-work may be comparable to potential

388

Andy Lake

benefits from mode-shift transport measures, increases in physical transport arising from increased prosperity are set to continue for the foreseeable future. However, some commentators note a correlation between increased ICT-based economic growth and a relative slowdown in energy consumption (ABARE, 2001; Romm et al., 1999). The Internet economy does, it is claimed, deliver increases in GDP at lower rates of increase in energy consumption. Economic analysts ABARE, on the basis of current data, have modelled a scenario where an increase in the uptake of e-commerce of 50% above a projected reference case results in an increase in world GDP of 3.9%, but an increase of energy consumption of only 1.5%. This is an area where one needs to proceed with caution, as the evidence is at best circumstantial. Other commentators have attempted to show a correlation between increased telecommunications use and increased use of physical transport (Plaut, 1999). Both kinds of approach are full of assumptions, and causality is inappropriately inferred from contemporaneous trends, with insufficient demonstration of necessary connection. The measured studies of e-working would, however, lend some support to the view that similar or higher levels of productivity can be achieved with lower levels of transport energy. The appropriate conclusion from this, however, is not that decoupling is being achieved, but that there are new dimensions to the relationship between mobility and economic growth.

19.8 Future Research Directions One of the reasons why policy tends to revolve around hopeful generalisations, and forecasters are often at sea in predicting trends and impacts of e-applications is that there has been no systematic approach to building up, piece by piece, our understanding of the impacts of ICT on transport. Big-brush terms like ICT have a certain convenience, but are often unwieldy in terms of analysis. Research into ework is a little more mature than research into e-commerce, reflecting its longer history as a practical proposition. But research into the transport effects of other eapplications, such as e-learning, telemedicine and online administrations, is pretty much non-existent. ICT encompasses a wide range of applications, and researchers need to unbundle the concept in order to research the wider picture. What is needed is an identification of discrete areas where empirical data can be gathered – areas such as home-based e-work, mobile e-work, e-shopping, telemedicine, etc. In each of these areas – where in turn there will be numerous different models of implementation – evidence needs to be gathered to answer the questions about the extent of the phenomenon, trends in uptake, limits to growth, characteristics of user profile, technology variations, direct transport impacts and rebounds in terms of new trips and activity redistribution. The importance of this is that while gaining a big picture view of the relationship between telecommunications and travel may be academically interesting, at a policy level it is no help at all. For example, if it can be shown that home-

The Impact of e-Work and e-Commerce on Transport

389

based e-work has a net impact of reducing travel, appropriate policies need to be developed to encourage it. It would be a missed opportunity if it were to be neglected or even constrained, on the grounds that other ICT applications are believed to have a different effect. Appropriate policies need to be developed for each application. This unbundling approach is also important when analysing practices within a particular field of ICT application. For example, most of the studies we have reviewed on the transport effects of e-commerce have to some extent hedged their bets, saying, “if e-commerce is organised like this, it will have these effects, but if it is organised like that, it will have those effects”. At present, these kinds of statement are not very helpful. But after research has established what “these” and “those” effects are, the different models of practice become very significant for developing targeted policies. Research, for example, may identify the value or otherwise of having community distribution points to enable B2C e-commerce (Hopkinson et al., 2000). There are also particular implications for Information Age transport modelling. Work is being done to build up a picture of the types of factors that need to be measured in order to model the effects of ICT (e-work, e-services, ecommerce) on residential and business location. Particularly noteworthy are the studies by Golob (2000), Ben-Akiva et al. (1996), Handy and Mokhtarian (1995) and Mokhtarian (1998). The principles and issues outlined in these studies provide helpful guidance for developing new modelling processes. Existing models are based on historic data – typically fairly old data – and none of it relating to how ICT use replaces, generates or redistributes travel. Golob and Regan (2001) have suggested that “in-home travel” will be so important that: ‘practitioners will come to realise that most household travel survey data collected to date will be of little use in forecasting impacts of telecommunications on travel.’

19.9 Conclusion Our conclusions from the project can be briefly summarised as follows: x There is a growing body of empirically robust studies that demonstrate a travel reduction effect from certain forms of e-work: home-based and centre-based telework (other forms being almost completely unresearched). x Where rebound effects have been measured concerning new trips by teleworkers and their households, the effects are complex, but do not eliminate, and sometimes add to, the travel savings. x Research into e-commerce is in its infancy. The common sense consensus that LGV movements will increase, and passenger shopping trips decrease, needs to be supported with observed data. x The views expressed by some commentators that wider rebound effects will compromise the benefits, or that e-applications generate significant amounts of travel, are almost entirely based on plausible conjecture. The empirical work to demonstrate these effects has yet to be done.

390

x x

x

Andy Lake

All transport impacts, whether reduction, generation or redistribution are important for transport analysts and policy-makers. There is insufficient evidence as yet to show that ICT can lead to a decoupling of economic growth and transport growth. But there are indications that economic growth can be achieved at lower levels of associated transport growth. The way forward for research is to focus on gathering new data, and projects should focus on discrete applications – such as mobile telework, particular forms of e-commerce, etc. – rather than on broad-brush speculation, of which there is already more than enough.

References ABARE (2001). The New Economy and the Energy Sector: assessing the economic impacts. ABARE, Canberra. Ben-Akiva, M., Bowman, J.L., Gopinath, D. (1996). Travel Demand Model System for the Information Era. Transportation, 23(3): 241-266. Black, W (2001). An Unpopular Essay on Transportation. Journal of Transport Geography 9(1): 1-11. Browne M. (2001). E-commerce, Freight Distribution and the Truck Industry Discussion Paper presented to 4th ACEA (Association des Constructeurs Européens d' Automobiles) Scientific Advisory Group Meeting. ACEA, Brussels. Cairns, S., Sloman, L., Anable, J., Kirkbride, A., Newson, C., Goodwin, P. (2004). Smarter Choices – Changing the Way We Travel. UK Department for Transport, London. Dodgson, J., Pacey, J., Begg M. (2000). Motors and Modems Revisited: The Role of Technology in Reducing Travel Demands and Traffic Congestion. Report for RAC Foundation, London. The earlier report by the same team was Motors and Modems (1997). Dublin Transportation Office (1998). Telecommuting: The Shortest Route to Work. Dublin Transportation Office, Dublin. Easley R.B., Easley S.C. (2000). Issues and Recommendations for E-Commerce Impacts on Urban CVO (Commercial Vehicle Operations) Deliveries and Parking. Paper presented at 7th ITS World Congress, Turin, Italy, November. European Commission (2000a). Towards a Sustainable Information Society. Commission, Brussels.

European

European Commission (2000b). Evaluation of the Telework Pilot in the Information Society Directorate-General and the Employment Directorate-General. European Commission, Brussels. European Commission (2002). eEurope 2005: An Information Society for All – An Action Plan. European Commission, Brussels.

The Impact of e-Work and e-Commerce on Transport

391

European Foresight for Transport (2002). What Influences Mobility and the Transport Sector? A view from non-transport fields. Report of the FORESIGHT for TRANSPORT project summarising the output of the Semmering thematic consultations, June 3-7, 2002. Forseback, L. (2000). Case Studies of the Information Society and Sustainable Development. European Commission, Brussels. Forseback, L. (2001). The Knowledge Economy and Climate Change. European Commission, Brussels. Gillespie A., Richardson R., Cornford J. (1995). Review of Telework in Britain: Implications for Public Policy. Report prepared for UK Parliamentary Office of Science and Technology. Glogger, A.F., Zwangler, T.W., Kard, G. (2003). The Impact of Telecommuting on Households’ Travel Behaviour, Expenditures and Emissions (2003). Paper presented at TRIP conference February 2003, Copenhagen. Golob, T. (2000). Travelbehavior.com: Activity Approaches to Modeling the Effects of Information Technology on Personal Travel Behavior. A Resource Paper for IATBR 2000, 9th Conference of the International Association for Travel Behavior Research Gold Coast, Queensland, Australia 2-7 July, 2000. Golob T., Regan A. (2001). Impacts of Information Technology on Personal Travel and Commercial Vehicle Operations: Research Challenges and Opportunities. Transport Research 9C: 87-121. Gould J., Golob T. (1997). Shopping Without Travel or Travel Without Shopping? An Investigation of Electronic Home Shopping. Transport Reviews, 17(4): 355-376. Hamer, R., Kroes, E., Van Ooststroom, H. (1991). Teleworking in the Netherlands: an evaluation of changes in travel behaviour. Transportation, 18: 365-382 Handy, S, and Mokhtarian, P.L. (1995). Forecasting Telecommuting – An exploration of methodologies and research needs. Transportation, 23(2): 163-190. Hills, S., Hopkinson, P., James, P. (2003). SUSTEL Case Study UK 2 – BT Workabout. Deliverable of EU SUSTEL project, IST-2001-33228. www.sustel.org Hopkinson, P.,James, P. (2000) Virtual Traffic - the Impact of E-Commerce on Logistics and the Implications for Sustainable Development, in: Wilsden, J (ed.): Digital Futures. - The impact of e-commerce on society and the environment Earthscan, London. Hopkinson, P., James, P., Mayurama, T., Selwyn, J. (2001). The Impacts of Teleworking – A Study of AA Employees. Unpublished research report. UKCEED, Peterborough. Hultkranz O. and Lumsden K. (2001). E-commerce and Consequences for the Logistics Industry. Article published in proceedings for Seminar on “The Impact of E-Commerce on Transport”, Paris, 5-6/6 2001. Jupp, S. (1998) Miles Better. Case study of Hertfordshire County Council Trading Standards at www.new-ways-of-working.co.uk

392

Andy Lake

Kilpala H., Seneviratne P., Pekkarinen S. (2000). Electronic Grocery Shopping and its Impact on Transportation and Logistics with Special Reference to Finland. Paper presented at 79th Annual Meeting of the Transportation Research Board, January 9-13, Washington, DC. Koenig, B.E., Henderson, D.K., Mokhtarian, P.L. (1996). The Travel and Emissions Impacts of Telecommuting for the State of California Telecommuting Pilot Project. Transportation Research 4C(1): 13-32. Lake, A.S. (2003). Travel Reduction and Teleworking: what we know and what we don't. www.flexibility.co.uk/issues/transport/travsubsummary.htm Lake, A.S., Cherrett, T.J (2002). The Impact of Information and Communications Technologies on Travel and Freight Distribution Patterns: Review and Assessment of Literature (2002). Report and database available online at www.virtual-mobility.com. Lyons, G.D., Hickford, A.J., Smith, J.C. (1998). The Nature and Scale of Teleworking's Travel Demand Impacts: Insights From a U.K. Trial. Proceedings of Third International Workshop on Telework: Teleworking Environments, Turku, Finland, 1-4 September, 312-330. Marker J., Goulias K. (2000). A Framework for the Analysis of Grocery Teleshopping. Paper presented to 79th Annual Meeting of the Transportation Research Board, January 9-13, Washington, DC. Marshall, S., Banister, D. (2000). Travel Reduction Strategies: Intentions and Outcomes. Transportation Research, 34A(5): 321-338. Millard-Bell, A. (2000). White Van Gridlock or a Boon for Traffic Reduction – how will ecommerce affect transport? Local Transport Today, 297, 31.8.2000. London. Mitchell, H., Trodd, E. (1994). An Introductory Study of Telework Based TransportTelecommunications Substitution. UK Department of Transport, London Mokhtarian, P.L. (1998). A Synthetic Approach to Estimating the Impacts of Telecommuting on Travel, Urban Studies, 35(2): 215-241. Mokhtarian, P. L., Ho, C., Hung, S. W., Lam, T. B., Raney, E. A., Redmond, L. S., Stanek, D. M., Varma, K. V. (1997). Residential Area-Based Offices Project: Final Report on the Evaluation of Impacts. Research Report UCD-ITS-RR-97-17, Institute of Transportation Studies, University of California, Davis. Parkhurst, G. (2000). Influence of Bus-Based Park and Ride Facilities on Users' Car Traffic. Transport Policy, 7(2): 159-172. Plaut, P.O. (1999). Do Telecommunications Reduce Industrial Uses of Transportation? World Transport Policy and Practice, 5(4): 42-49. Reijnders, L., Hoogeveen, M.J. (2001). Energy Effects Associated with E-Commerce. Journal of Environmental Management, 62(3): 271-282. Romm, J, Rosenfeld, A., Hermann, S. (1999). The Internet Economy and Global Warming: A scenario of the Impact of E-Commerce on Energy and the Environment. Center for Energy and Climate Solutions, Washington.

The Impact of e-Work and e-Commerce on Transport

393

Salomon, I. (1986). Telecommunications and travel relationships: A review. Transportation Research, 20A(3): 223-38. Saxena, S., Mokhtarian, P.L. (1997). The Impact of Telecommuting on the Activity Spaces of Participants and their Households. Geographical Analysis, 29(2): 124-144. Smith, N., Ferreira, L., Mead, E. (2001). E-business Impacts on the Transport System, Transport Impacts of E-business. Series of working papers for Australian National Transport Secretariat, Brisbane. Surrey County Council (1999). Evaluation of the Epsom Telecentre. Cited at ww.flexibility.co.uk/issues/TS_travsubsummary.htm U.S. Department of Energy (1994). Energy, Emissions, and Social Consequences of Telecommuting. Washington. US Department of Transportation (1993). Transportation Implications of Telecommuting. Washington. Williams, E.D. (2002). Environmental Evaluation of b2c E-Commerce in Japan. Proceedings of the 2002 IEEE International Symposium on Electronics and the Environment. San Francisco. Wright, A. (1997). Saving Energy through Teleworking. Flexible Working, 2(2): 14-16.

20 A Web-Based Study of the Propensity to Telework Based on Socio-Economic, Work Organisation and Spatial Factors Lasse Møller-Jensen1, Chris Jensen-Butler2r, Bjarne Madsen3, Jeremy Millard4 and Lars Schmidt4 1

Department of Geography and Geology, University of Copenhagen, Denmark Department of Economics, University of St. Andrews, Scotland 3 Centre for Regional and Tourism Research, CRT, Bornholm, Denmark 4 Danish Technological Institute, Århus, Denmark 2

Abstract This paper discusses estimates of the extent of teleworking in Denmark and reports on a study aimed at identifying the determinants of the decision to practice teleworking. The study is based on a panel dataset obtained through a major commercial opinion poll and market research agency using a web-based survey. There are 2,680 respondents from the panel. The data set is used i) to identify the probable determinants of the choice to become a teleworker and the determinants of intensity of teleworking, ii) to make a first estimate of the consequences of teleworking for travel activity and iii) to make a first estimate of the long-term effects of teleworking on choice of place and type of residence and place of work. It appears that factors related to the place and organisation of work are the most important in explaining both propensity to telework and its intensity.

20.1 Introduction The results presented in this paper are part of a major research effort aimed at identifying the potential contribution of teleworking to the reduction of emissions to the environment in the transport sector. This research project has a number of specific aims. 1. Estimates will be made of the extent of teleworking in Denmark 2. Estimates will be made of the effects of teleworking on the demand for transport, in particular, use of the private car. 3. Changes in transport activity will be related to changes in emissions to the environment.

r

After this chapter was written Chris Jensen-Butler has died.

396

4.

5. 6.

Lasse Møller-Jensen et al.

Estimates will be made of the determinants of the decision to practice teleworking and the determinants of the intensity of teleworking activity amongst potential teleworkers. This will lead to forecasts of future levels of teleworking activity. The effects of teleworking on local, regional and national economies, through reduction of commuting costs, congestion and emissions will be estimated using and local economic model.

The present paper deals primarily with question 4 and, to a lesser extent, with question 1.

20.2 Teleworking: Theoretical and Methodological Issues The principal methodological issues relate to the definition of teleworking, identification of the extent of teleworking, identifying substitution and complementary effects with respect to transport, modelling telework take-up, the treatment of technology and modelling the local and regional economic effects of telework.

20.2.1 Defining Teleworking There is no easy operational definition of teleworking, which can include a range of work-related activities, such as home-based work using ICT (information and communications technology), work at home using ICT, with a permanent place of work elsewhere, work in telecentres and work using mobile telephony and Internet access. Not all work at home is telework. Furthermore, some people no longer have a fixed workplace outside the home which is their own. Vilhelmson & Thulin (2001) provide data from the Swedish national ICT survey estimating the following: Home-based work Commuting-based work Mobile work Other

4.0% 86.3% 6.9% 2.8%

They estimate that 4.8% of all employed persons are teleworkers (at some unspecified interval), mainly in the group Commuting-based work. They define telework as regular work done at a location other than the ordinary fixed place of work. A similar definition is adopted here, where the prime interest is the commuting-based workers, who sometimes work at home using ICT. This definition excludes mobile teleworking and work in telecentres. This more restrictive definition is necessary for operational reasons, as problems of definition and data collection otherwise would get severely out of hand. However, there still remains

Web-based Study of Propensity to Telework

397

the problem of measuring the frequency of teleworking and the relationship of this frequency to the definition of telework.

20.2.2 The Extent of Teleworking Estimates of the extent of teleworking vary considerably as a few examples show. NERA (2000) provides an estimate that 1998 levels of telecommuting are between 0.5 and 5.0% of the UK working population and that on any one weekday 0.1 to 1.0% of the UK working population actually is engaged in teleworking. Vilhelmson and Thulin (2001) estimate that about 5% of all gainfully employed persons were in 1999 regularly engaged in telework and about 2% were actually engaged in this activity on any weekday. The Danish Ministry of Science and Technology (Ministeriet for Videnskab, Teknologi og Udvikling, 2002) estimates that in 2001 28% of Danes with access to the Internet use ICT to work at home. 59% of all Danish families are estimated to have Internet access. This implies that 17% of all families telework at some point in time. Whilst this figure is in accord with other estimates (European Directory of Teleworking 2000 for example), it does seem to be a very optimistic estimate. Where small numbers in a large population are involved and where there is no obvious stratification criterion, the problems of collecting data on teleworking are substantial, which, together with problems of definition, is a fundamental reason why substantially different results are obtained in different studies.

20.2.3 Teleworking and Savings in Transport Effort A number of studies have estimated savings in transport through teleworking. Balepur et al. (1998) suggest that a telecentre project in California reduced total weekly car travel by the order of 18%, whilst another study (Varma and Mokhtarian, 1998) found a reduction in car travel of the order of 11-12% in another study of Californian telecentres. The interest in telecentres reflects the serious problems of obtaining data for teleworking at home, where there is no obvious method to identify teleworkers. NERA (2000) have provided estimates of reductions in car commuting of 11% by 2007 and 22% in 2017. Car business travel was estimated to be reduced by 6.5% in 2010. These estimates include allowances for secular growth in traffic. The key issue is to what extent teleworking is a substitute and to what extent it is a complementary activity in relation to transport. There is considerable literature on this subject (Batten 1989, Salomon 1996, 1998, 2000, Mokhtarian and Salomon 1997). A key problem in investigating the extent to which substitution occurs is the conceptual question of complementary effects. Teleworking can have several effects: 1. 2.

Reduction in use of private and public transport by the teleworker. Reduction in use of private and public transport by other household members.

398

3.

4. 5.

Lasse Møller-Jensen et al.

Increase in use of private and public transport because of work requirements when partly undertaking work at home using ICT (for example, greater flexibility at work which generates more trips). Increases in numbers of non-work related trips by the teleworker because of greater flexibility of working patterns. Increases in numbers of trips by other household members, either because of greater flexibility or because of availability of a vehicle.

These questions can only be answered satisfactorily by difficult and costly travel diary studies. On balance, it seems that the literature indicates that there is some evidence for a limited substitution effect from teleworking, but that the substitution effect has not been so powerful that it has given rise to substantial numbers of and growth in teleworkers. Vilhelmson and Thulin (2001) suggest that this failure to live up to expectations has been because researchers have tended to view teleworking in isolation. If, instead, research took as its point of departure what motivates travel and communication using ICT, in an activity-based perspective, where activities are viewed in time and space, better understanding of the nature and growth potential of teleworking could be gained. This suggestion is reinforced by a small number of studies suggesting that the importance of the substitution and complementary effects is related to type of region and geographical context. In more peripheral, less densely populated and rural areas, complementary effects are more important, whilst substitution effects are stronger in urban areas.

20.2.4 Modelling the Take-Up of Teleworking There are a few proposals in the literature for methods to model the take-up of teleworking. Mokhtarian and Salomon (1994) have set out the basic theoretical issues in a constrained utility maximisation framework. Mokhtarian (1998) proposes a simple multiplicative probability model identifying key factors at each stage of the take-up model. The expected number of people who telecommute at any given time (T) can be estimated by: T=ExAxWxC Where E is the average number of people employed in a given time period, A is the number of workers who are able to telecommute, W is the proportion of those who can telecommute, who want to telecommute, C is the proportion of those who choose to. It follows from this that the average number of people teleworking on any given day is: O=TxF where F is the frequency of teleworking (fraction of a five-day working week). This simple multiplicative model is then translated into reductions in transport

Web-based Study of Propensity to Telework

399

effort by a multiplicative extension of the model relating teleworking to transport effort. However, the model depends entirely upon the definition of probabilities, and some investigation of the role of exogenous variables in determining these probabilities is necessary. This is an issue addressed by the present paper, using the multivariate logit model. The interpretation of the coefficients of the logit model is difficult, as the dependent variable is the logarithm of the odds of the occurrence of an event (teleworking).

20.2.5 Technology Issues A variety of technological issues affect the take up and growth of teleworking. These can be grouped under three main headings. Access technologies The literature considers at some length the role of access technologies in driving the growth of teleworking. Many models assume that Internet access is a precondition for teleworking. However, this may not be the case, as the decision to obtain Internet access may be closely related to telework. In particular, the transition from ISDN to Broadband (ADSL) technology is regarded as being a key driver (NERA, 2000). This view is highly technology-determined. Organisation of work Changes in the organisation of work, including degree of work flexibility, participation in strategic decision-making and leadership models are also forms of technology, often ignored by more hardware-driven views (Vejrup-Hansen, 2000). Activity Patterns In a broader sense, treating activity patterns as technology places teleworking inside a broader theoretical framework, as argued by Vilhelmson and Thulin (2001). In this approach, the impact of a wide variety of changes in activity patterns can be related to teleworking take-up and intensity.

20.2.6 Long-Term Effects A series of important issues arises when considering the long-term effects of teleworking. Changes in commuting costs, including congestion costs will, in the long run, affect behaviour with respect to choice of place of residence and place of work. However, this is only one factor affecting these future choices. The nature of the pattern of settlement and urban development which will emerge from the Information Society is one of considerable importance, both for the future pattern of economic development and not the least future energy consumption and emissions to the environment (Graham and Marvin, 1996).

400

Lasse Møller-Jensen et al.

20.3 The Determinants of Teleworking and Transport Substitution Effects The study presented below addresses these issues. A set of background variables is examined in relation to whether or not teleworking is practiced and these variables are related to the intensity of teleworking, amongst teleworkers identified. The data are a panel data set obtained through a major commercial opinion poll and market research agency. All members of the panel are Danish users of the Internet. There are 2,680 respondents from the panel, whilst the response rate is unknown. This number is reduced to 2,210 when respondents who are not working at present are removed. Responses from the panel are obtained via the Internet. Amongst these Internet users, a subset undertakes telework. Caution must be exercised with respect to estimating absolute figures in the Danish population, as there are grounds for believing that the selection of the panel involves bias, as more motivated Internet users will tend to be panel members. However, comparisons within the panel would seem to be more defensible. The questionnaire permits identification of the 1,009 members who state that they telework. Comparisons can be made with those who do not. This questionnaire does not permit identification of travel patterns of other household members and differences in these when teleworking occurs, compared with travel patterns when there is no teleworking. Thus, the main use of the data set is i) to identify the probable determinants of the choice to become a teleworker, the determinants of intensity of teleworking, ii) to make a first estimate of the consequences of teleworking for travel activity and iii) to make a first estimate of the long-term effects of teleworking on choice of place and type of residence and place of work. The next stage of the study will involve a much larger data set on travel behaviour for a representative sample of the population, collected by Statistics Denmark. This data set involves travel diary type of information for all household members.

2.4 The Consequences for Travel of Working at Home There are 1,009 persons who state that they sometimes telework. This number is reduced to 946 when those who only work at home are removed. Table 20.1 shows, for all teleworkers, their personal use of different means of transport, measured as an average number of minutes for a) the last working day when the respondents worked at home and b) the last working day when they did not work at home.

Web-based Study of Propensity to Telework

401

Table 20.1. Average times spent on different types of transport by teleworkers on days when they do and do not telework

Walking Bicycle Own car Other car Bus Train

Non-teleworking day Mean minutes N 221 20 179 33 468 68 121 62 59 49 72 63

Teleworking day N Mean minutes 209 21 166 28 370 45 76 59 32 34 18 64

Table 20.2. Total number of minutes spent by all teleworkers in the sample on different days

Walking Bicycle Own car Other car Bus Train

Non-teleworking day Total Minutes 4372 5956 32030 7480 2919 4571

Teleworking day Total Minutes 4475 4705 16466 4467 1087 1151

% change 2.4 -21.0 -48.6 -40.3 -62.8 -74.8

There are clear differences between travel activity measured in minutes, particularly with respect to use of own car and use of bus and to a lesser extent use of another person’s car. This suggests a certain potential for reduction of travel to the extent that teleworking encourages work at home. Table 20.2 shows for the entire panel the total number of minutes used, by means of travel, on days when respondents worked at home and when they did not. This table shows that substantial reductions in travel time by car, bus and train (of the magnitude of 50%) occurred when working at home. The biggest absolute reductions were for the car. It should be stressed that changes in use of means of transport relate only to the respondent and not to his or her household, thus complementary effects of teleworking are not properly considered. Furthermore, these savings only apply on days when teleworking occurs, thus their absolute effect is substantially reduced when weighting for frequency of teleworking is applied. However, there does seem to be a substitution effect between telework and transport effort, for the respondents at least, though this may be modified when household travel patterns are taken into account.

20.5 Determinants of Teleworking Respondents were asked whether or not they teleworked at all. Teleworking was defined in a limited manner, as work undertaken at home with partial or continuous use of a computer with Internet links. This excludes other sorts of telework, including work undertaken in telecentres and mobile teleworking. A range of explanatory variables can be related to this yes/no question.

402

Lasse Møller-Jensen et al.

A number of bivariate relationships were examined. Concerning teleworking by sex the results show that while 45.7% of all respondents have undertaken some form of telework, men are more likely to be teleworkers than women as 50.4% of men are teleworkers whilst 39.6% of women telework. This difference is real as Ȥ2 is significant at the 1% level. Concerning teleworking by age the results show that the probability of teleworking increases with age up to 50 years. 25.8% of the 1630 year old respondents telework, whilst 45.5% of the 31-40 year olds and 54.6% of the 41-50 year olds telework. This falls to 49.4% of those over 50 years of age. Again, these differences are statistically significant. When looking at teleworking by family type the results show that teleworking occurs with the highest frequency (49.1%) amongst families with children at home, where there are two adults. Families with only one adult and no children at home have the lowest teleworking frequency (38.4%). Perhaps most surprising, is the relatively low frequency (41.9%) of single parents who telework and the relatively high frequency of couples without children who telework (45.3%). Overall, differences are not substantial, but they are statistically significant. Looking at teleworking by type of employment the results show that selfemployed persons have a greater propensity to telework (66.5% of the selfemployed) compared with employees (44%). As will be seen later, this is almost certainly associated with the organisation of work and related flexibility issues. Again, this difference is significant at the 1% level. The frequency of teleworking varies markedly by sector. Financial and business services are clearly overrepresented (50.3%), followed by public and private service (46.5%). Other sectors are under-represented, the lowest being industry (34.5%), construction (38.6%), energy and utilities (40%), transport, post and telecommunication (40.9%) commerce, hotel and restaurants (41.8%), the primary sector, including agriculture, (44.2%). However, it is perhaps surprising how relatively important teleworking is in traditional sectors. Concerning teleworking by education, the results show that there is a clear relationship between educational level and frequency of teleworking. 65.4% of respondents with university degrees telework, whilst only 30-32% of respondents with a basic school education or a trade qualification telework. In general, the propensity to telework increases with level of education. This is perhaps a surprising result, given the assumption that routine office tasks are usually considered to be suitable candidates for work at home. Concerning teleworking and organisation of work, the results show that the degree of influence on the way in which the respondent can plan his or her own work is an important determinant of the propensity to telework. 54.1% of those who reply that they have considerable influence on planning of their own work also telework, whilst only 5.3% of those who reply that they have almost no influence undertake teleworking. When looking at teleworking and planning and organisation of others’ work, the results show that 61.8% of respondents who plan and organise work tasks for others also telework, whilst 31.1% of those who have no leadership responsibility telework. Concerning teleworking and strategic decision-making roles, the results show that 67% of the respondents who participate in strategic decision-making in their own organisation telework, whilst only 31% of those who do not participate undertake teleworking.

Web-based Study of Propensity to Telework

403

The results of the bivariate analysis report also on the relationship between size of firm (numbers working at the same place of work) and propensity to telework. In general, it can be seen that there is little variation in propensity to telework by size of firm, when the smallest firms (0-1 employees) are removed. The relationship between the length of the respondents’ working week in hours and propensity to telework is such that 27.9% of those who work less than 37 hours a week telework, whilst 59.2% of those who work between 38 and 50 hours telework and 77.8% of those who work over 50 hours a week, telework. Clearly, teleworking seems to extend the number of working hours. 100 90

% teleworking

80 70 60 50 40 30 20 10 0 10-20 20-25 25-30 30-35 35-40 40-45 45-50 50-55 55-60 60-65 65-70 70-75

>75

monthly income (1000)

Fig. 20.1. Teleworking by gross personal monthly income

Fig. 20.1 shows the relationship between respondents’ gross personal income and propensity to telework. There is a clear and positive relationship between income and telework, except in the highest income groups, where teleworking becomes less important. Concerning car ownership, the results show that whilst only 32.2% of respondents who live in a household without a car telework, this increases to 53.8% for respondents living in households with at least two cars. Single car households occupy an intermediate position. Teleworking appears, therefore, to have a certain potential to change travel patterns where the car would be otherwise used. The bivariate analysis also included issues related to the residence of the respondent: Concerning size of residence (in square metres), the results show that 33.3% of respondents living in the smallest residences (under 80 m2 ) telework, whilst 62.4% of respondents living in the largest residences (over 200m2) telework. As the size of residence increases, so does the propensity to telework. Concerning existence of a dedicated room for telework, the results show that 52.7% of respondents who have a dedicated room at home telework, whilst 41.8% of those

404

Lasse Møller-Jensen et al.

who do not have a dedicated room telework. Concerning type of residential area, the propensity to telework by degree of rurality has been analysed. The probability of teleworking is similar in rural and non-rural municipalities. Likewise, there seems to be little difference between propensities to telework in municipalities of different sizes. The observable differences for these two variables are not statistically significant.

20.5.1 Multivariate Relations As a number of the relationships identified above are almost certainly spurious, a multivariate logit model was used to identify causal relationships in relation to adoption of teleworking. The qualitative dependent variable is dichotomous, being the answer to the question whether or not the respondent sometimes is involved in teleworking. Table 20.3. A multivariate logit model with teleworking as the dependent variable Variablesa SEX AGE EMP FAMP FAMK IND EDU PLAN LEAD ORG INC CAR ROOM SIZE WEEK FIRM RURAL Constant a) Variables:

B -0.122 .008 .157 -.020 .178 -.047 .316 .790 .068 .345 .021 .055 .283 .003 .037 .000 -.283 -6.892

S.E. .104 .005 .218 .134 .111 .103 .047 .114 .073 .078 .005 .086 .106 .001 .007 .000 .155 .551

Wald 1.369 2.391 .521 .022 2.568 .205 45.968 48.391 .860 19.709 18.033 .418 7.110 5.980 27.634 .093 3.344 156.473

Df 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1

Sig. .242 .122 .471 .883 .109 .651 .000 .000 .354 .000 .000 .518 .008 .014 .000 .761 .067 .000

Exp(B) .885 1.008 1.170 .980 1.195 .954 1.371 2.203 1.070 1.412 1.021 1.057 1.327 1.003 1.038 1.000 .754 .001

Sex (binary), Age (interval), Employment (binary: employee=1, self-employed = 0), Family type P (binary:1=single, 0=couple), Family type B (binary:1=children at home, 0=no children at home), Ind (sector, binary: 1=financial and business service, private and public service 0=all others), Edu (interval: length of education, 1-4), Plan (interval: degree of influence on planning of work 1-3), Lead (interval: degree of influence on planning of others’ work 1-3), Org (interval: degree of involvement in strategic, decision-making 1-3), Inc (interval: gross personal income), Car (interval: car ownership 0-3), Room (binary: dedicated room for teleworking: 1=yes, 2= no), Size (interval: of residence, m2), Week (interval: average weekly work time), Firm (interval: number of employees at place of work), Rural (binary: 1= rural municipality, 2=urban municipality).

Web-based Study of Propensity to Telework

405

The result is shown in Table 20.3. R2 is low at 0.274, which is not uncommon in models based upon individual-level data and the variables which are significant, using the Wald test, at the 5% level are EDU (educational level, + sign), PLAN (degree of influence on planning of one’ own work, + sign) ORG (degree of involvement in strategic decision-making, + sign) INC (income, + sign) ROOM (existence of a dedicated room, + sign) SIZE (size of residence, + sign), WEEK (number of working hours in a the week, with a positive sign). The model has been rerun with these variables alone, with the same pattern of significance levels and no change in R2. It seems that the most important variables related to the probability of being a teleworker are those related to the way in which work is organised and the associated industry technology. Independence in planning one’s own work and involvement in strategic decision-making in the firm, together with a longer working week are highly significant. Planning and organising the work of others is not significant, suggesting that teleworking is not directly associated with leadership functions. Educational level, a closely related variable, is also significant, as is income. Finally, two variables related to place of residence appear significant, size of residence (m2) and presence of a dedicated room. In order to validate these results, an OLS regression model using intensity of teleworking (measured by average hours per week working at home on line) as the (interval-level) dependent variable was tested. The same variables as used in the logit model were chosen as independent variables, with the addition of estimated distance to work (DIST: interval). Table 20.4 shows the results of this analysis. R2 is 0.167 and the significant variables at the 5% level are AGE (+ sign) EMP (+ sign, indicating that self-employed are more likely to telework) EDU (educational level, + sign), PLAN (influence on planning of one’s own work, + sign), WEEK (weekly hours of work, + sign) and DIST (distance to work, + sign). In relation to intensity of teleworking, the variables PLAN and WEEK, relating to the nature and technology of work reappear, as does EDUC, Educational level. What is perhaps interesting is the positive relationship between intensity of teleworking and distance to place of work. On the other hand, the coefficient of the variable DIST suggests that an increase of one kilometre’s commuting distance increases the average amount of teleworking by only 1/50 of an hour. Age has now entered as a significant variable. These results seem to provide some validation of the logit model, though they do imply that the set of variables which determine the probability of becoming a teleworker are fewer and more strongly related to the nature and organisation of work and the technology involved than the variables which determine intensity of teleworking. The implication of these findings is that forecasting of future telework frequencies should involve consideration of changes in the nature of organisation of work and related industrial technology, rather than variables related to the individual characteristics of teleworkers and the localities in which they live.

406

Lasse Møller-Jensen et al.

Table 20.4. An OLS regression model with intensity of teleworking as the dependent variable Model (Constant) SEX AGE EMP FAMP FAMK IND EDU PLAN LEAD ORG INC CAR ROOM SIZE WEEK FIRM RURAL DIST

Non-standardised coefficients B Std. error -10.444 3.369 .360 .607 .006 .033 -5.900 1.139 .944 .776 1.224 .634 -.529 .592 .587 .266 1.680 .800 -.588 .420 .522 .429 .004 .027 -.396 .497 -.275 .591 .001 .005 -.318 .037 .000 .001 -.121 .894 .002 .006

Standardised coefficients Beta .019 .069 -.170 .043 .066 -.029 .071 .065 -.048 .044 -.050 -.029 -.015 .051 .296 -.018 -.004 .126

t -3.100 .592 2.100 -5.179 1.218 1.930 -.894 2.206 2.100 -1.402 1.216 -1.357 -.798 -.466 1.377 8.657 -.602 -.136 4.098

Sig. .002 .554 .036 .000 .224 .054 .372 .028 .036 .161 .224 .175 .425 .641 .169 .000 .547 .892 .000

20.6 Long-Term Effects of Teleworking The opportunity to take up teleworking may potentially influence future decisions concerning choice of workplace as well as residence location and possibly residence size. This has been examined using a separate set of questions within the questionnaire. Table 20.5. Teleworking and choice of future location of residence

Will influence choice of location of residence

Yes very important Yes, but less important No

Total

Count Row% Col% Count Row% Col% Count Row% Col% Count Row% Col%

Intensity of teleworking (hours last 5 work days) 4 19 47 28.8% 71.2% 4.2% 9.6% 79 119 39.9% 60.1% 17.3% 24.3% 359 323 52.6% 47.4% 78.6% 66.1% 457 489 48.3% 51.7% 100.0% 100.0%

Total 66 100.0% 7.0% 198 100.0% 20.9% 682 100.0% 20.9% 946 100.0% 100.0%

Web-based Study of Propensity to Telework

407

Table 20.5 shows the respondent teleworkers’ view of to what extent the opportunity to work at home has affected their choice of location of their residence. 4% of respondents state that there has been an effect on their choice. It can also be seen that of those who responded positively to this question of locational choice tend to be those who work more intensively with teleworking. These 4% divide equally into two groups where one group states that they have moved closer to place of work, whilst the other group states that they have moved further away. Concerning choice of size of residence, 3.4% answer positively that the opportunity to telework has affected choice of size of residence, again mainly those who work more intensively with teleworking. Concerning future choice of location of residence, given the opportunity of teleworking, the results show that 7% answer that it will have considerable influence and 21% answer that it will have some influence. Again, those who work most intensively with teleworking tend to state that it will have the greatest effects and again, the number stating that they will move further away is approximately the same as those who state that they will move closer. However, the average stated distance for those who will move closer is 18 km and for those who will move further away it is 39 km. Table 20.6. Teleworking and choice of place of work

Changed workplace to increase teleworking

Total

Yes

No

Count Row% Col% Count Row% Col% Count Row% Col%

Intensity of teleworking (hours last 5 work days) 4 5 24 17.2% 82.8% 1.1% 4.9% 452 465 49.3% 50.7% 98.9% 95.1% 457 489 48.3% 51.7% 100.0% 100.0%

Total 29 100.0% 3.1% 917 100.0% 96.9% 946 100.0% 100.0%

Concerning choice of place of work, Table 20.6 shows the respondents’ answer to the question of whether they have changed their place of work because of the opportunity to telework. 3.1% reply positively, again predominantly those who work more intensively with teleworking. The numbers who have moved closer is approximately equal to those who have moved further away. Table 20.7 shows the percentage of respondent teleworkers who are considering changing their place of work in the future in order to have a greater possibility of working at home. 6% answer positively to this question, again distributing themselves evenly between those who say that the distance between home and work will increase and those who say it will decrease. In conclusion, it seems that the opportunity to telework will potentially affect long-term decisions concerning place of residence and place of work, but that it is impossible to say whether the result will be a spatially more or less dispersed pattern, as these sorts of decisions are complex.

408

Lasse Møller-Jensen et al.

Table 20.7. Teleworking and choice of future place of work

Will change workplace to increase teleworking

Yes

No

Total

Count Row% Col% Count Row% Col% Count Row% Col%

Intensity of teleworking (hours last 5 work days) 4 23 34 40.4% 59.6% 5.0% 7.0% 434 455 48.8% 51.2% 95.0% 93.0% 457 489 48.3% 51.7% 100.0% 100.0%

Total 57 100.0% 6.0% 889 100.0% 94.0% 946 100.0% 100.0%

20.7 Conclusion This empirical study presents results obtained from data derived from a web-based questionnaire sent out to a panel of Internet users in Denmark. The results indicate that there seems to be considerable potential for reduction of the use of the private car, looking alone at the behaviour of the teleworker. Variables which influence the decision to telework at all have been identified and a multivariate analysis indicates that those variables related to the organisation and flexibility of work and its integration with Information and Communication technology are the important determinants. This has substantial implications for the design of forecasting of teleworking frequencies. Finally, the long term effects of teleworking on location of place of residence and place of work are difficult to identify. The general finding is that at present it seems that teleworking will have limited effects on these choices.

References Balepur, P.N., Varma, K.V., Mokhtarian, P. (1998). Transportation impacts of centre-based telecommuting: interim findings from the neighbourhood telecentres project. Transportation, 25: 287-306. Batten, D. (1989). The future of transport and interface communication. In: Batten D., Thord R. (eds.) Transportation for the future. Springer Verlag, Berlin. Graham, S., Marvin, S. (1996). Telecommunications and the city. Routledge, London. Ministeriet for Videnskab, Teknologi og Udvikling (2002). Regionernes IT status, 2002, Ministry of Science and Technology, Copenhagen. Mokhtarian, P. (1998). A synthetic approach to estimating the impacts of telecommuting on travel. Urban Studies, 35(2): 215-241.

Web-based Study of Propensity to Telework

409

Mokhtarian, P., Salomon, I. (1994). Modeling the choice of telecommuting: setting the context. Environment and Planning A, 26(5): 749-766. Mokhtarian, P., Salomon, I. (1997). Modeling the desire to telecommute: the importance of attitudinal factors in behavioural models. Transportation Research A, 31(1): 35-50. NERA (2000). Motors and modems revisited. National Economic Research Associates, London. Salomon, I. (1986). Telecommunications and travel relationships: a review. Transportation Research A, 24: 247-255. Salomon, I. (1998). Technological change and social forecasting: the case of telecommuting as a travel substitute. Transportation Research C, 17-45. Salomon, I. (2000). Can telecommuting help solve transportation problems? In: Hensher D.A., Button K.J. (eds.), Handbook of Transport Modelling. Amsterdam. Varma, K., Mokhtarian, P. (1998). The Trade-Off Between Trips and Distance Traveled in Analyzing the Emissions Impacts of Center-Based Telecommuting. Institute of Transportation Studies, Working Paper Series UCD-ITS-REP-98-16, Institute of Transportation Studies, UC Davis. Vejrup-Hansen, P. (2000). Det fleksible arbejdsmarked. DJØF-forlaget, Copenhagen. Vilhelmson, B., Thulin, E. (2001). Is regular work at fixed places fading away? The development of ICT-based and travel-based modes of work, Sweden. Environment & Planning A, (33): 10151029.

21 The Impact of Telecommuting on Households’ Travel Behaviour, Expenditures and Emissions Andrea F. Glogger, Thomas W. Zängler and Georg Karg Technische Universität München, TUM Business School, Germany

Abstract The demand of individuals for travel increases steadily. As a consequence, traffic increases as well, especially by car. Due to the negative side-effects of traffic on the environment, e.g. air quality, a change in travel behaviour is required. Innovative concepts are called for. Recently emerging information technologies offer chances to replace to a certain extent physical by virtual travel and therefore reduce traffic induced emissions. The research project presented here examines an option that was developed for work-related travel: telecommuting. The objective of this project is to analyse the effect telecommuting has on travel behaviour of telecommuters and their household members, the households’ expenditures in money and time and trafficinduced emissions. To determine the effects of telecommuting, a pre- and post-treatment survey was carried out using detailed travel diaries and questionnaires. Analysis of the survey leads to the following results: telecommuting changes the distribution of trips over the day. Peak hour traffic is avoided, and the total number of trips and distance travelled decrease. Expenditures in time and money change accordingly. The amount of traffic-induced emissions of greenhouse gases and air pollutants decreases.

21.1 Introduction In a market economy, households have various options to earn income and purchase goods. The activities to generate and spend income can in general not be undertaken at households’ places of residence. Usually the household members have to travel from their home to their employers’ locations or the locations of

This study was made possible by the support of the German Federal Ministry of Education and Research and the City of Munich. The bpu Unternehmensberatung consultancy introduced telecommuting into the surveyed enterprises and was involved in the development of the questionnaires. bpu and NFO infratest carried out the fieldwork.

412

A.F. Glogger, T.W. Zängler and G. Karg

purchase. Even leisure activities like meeting friends or relatives, or visiting natural and cultural sites require travel. In Western societies, the mobility of people arising from earning and spending income is a main driver for the standard of living of private households. At the same time, this mobility is also responsible for problems like overcrowded commuter trains, congested streets and polluted air resulting from travel. Except for CO2, these negative impacts of traffic manifest themselves, especially in conurbations (MOBINET, 2002). To date, it is generally accepted that this development is not sustainable. Therefore, a change in travel behaviour is required. To change travel behaviour, one has to take into account that frequent travel on regular trips (for example, travelling to and from work every day) becomes script-based or habitual (Gärling et al., 2001). Thus, the performance is carried out without much deliberation (Lotan, 1997; Ouellette & Wood, 1998; Gärling & Golledge, 2000; Gärling et al., 2001; Gärling et al., 2002; Gärling & Axhausen, 2003). For this reason, traditional instruments of traffic management have not been able to change travel behaviour in such a way that its negative effects on the environment can be reduced. Innovative concepts of mobility management are therefore required. This is all the more important because the demand of individuals for travel is increasing steadily. In this context, the emerging information technologies offer opportunities to reduce traffic while maintaining mobility. Information technologies have this potential as they enable people to substitute some physical travel and therefore reduce traffic and its associated negative effects (Glogger, 2002). Work-related travel is a major source of traffic. With information technology, travel-related work can be reduced by telework. Various kinds of telework have to be distinguished. In this study, we consider part-time (‘alternating’) telework (Empirica, 2000), referred to as telecommuting. The employee works part of the week in his or her office at the employer’s location and part of the week at his or her home. Since telecommuting is a promising concept for mobility management (Rye, 2001), its introduction gave rise to numerous travel studies with the objective of evaluating the impacts of telecommuting on travel behaviour (Wermuth et al., 1984; Quaid & Lagerberg, 1992; Kitamura et al., 1991; Axhausen, 1995; Axhausen et al., 2002). One of the first studies (Nilles et al., 1976) analysed a project on telecommuting in a “remote work centre” in Los Angeles in 1973. Their study states that telecommuting in a tele-centre saves petrol by reducing commuting distances. The “State of California Telecommuting Pilot Project” resulted in various publications on the interdependence of telecommuting and travel (e.g. Kitamura et al., 1991; Pendyala et al., 1991). Their main findings were the reduction of peakperiod trips, a significant decrease in the total amount of trips and distance travelled, and reduced action spaces for telecommuters. Additionally, a rescheduling and reallocation of activities was noted. According to this project, household members do not increase their car use if additional family cars become available. In 1991/92, the “Puget Sound Telecommuting Demonstration” took place (Quaid

The Impact of Telecommuting on Households’ Travel Behaviour

413

& Lagerberg, 1992). The effects of the tele-centre implemented in this project were analysed by Henderson & Mokhtarian (1996). Travelled distances were significantly reduced. Furthermore, Mokhtarian (1997) used two telecommuting programmes in San Diego, California, to document and illustrate a variety of transportation-related impacts of telecommuting. She also reported a reduction in the aggregated distance travelled without generation of any new trips. Yet she also argued that telecommuting did not motivate individuals to give up a car, and it might also support long-distance residential relocation. Europe can benefit from the detailed telecommuting research undertaken in the USA. However, local and contextual factors, like differences between countries with respect to infrastructure and general travel culture, have to be taken into account (NERA, 2000). Hamer et al. (1991, 1992) focused on the Netherlands and measured direct and indirect effects of telecommuting on telecommuters and their household members. Their findings confirm the already reported significant reduction of travel distances. However, Hamer et al. stress that especially trips using public transport are eliminated. After the telecommuters started telework, the mobility of other household members was reduced as well. In Germany, Vogt et al. (2001) analysed the travel behaviour of telecommuters in urban and rural areas. According to their study, travel distances were reduced due to telecommuting and peak-period trips were increasingly avoided. Summing up the findings of these studies, there is evidence for some substitution of physical by virtual travel. However, the socially desired effect of substitution is not the only possible interdependence between physical and virtual travel (Mokhtarian, 2000; Salomon, 2000). Different studies have given varied results and there has been much debate on the interrelations between physical and virtual traffic, as well as on the trade-offs within households (Hamer et al., 1991 and 1992; Niles, 1994; Nilles, 2000; Empirica, 2000; Mokhtarian, 2000). The research programme, MOBINET (1998-2003), sponsored by the German Federal Ministry of Education and Research, contains a project which examines among other things the effect of telecommuting on travel in the Greater Munich Area (City of Munich and eight neighbouring districts). The City of Munich itself is with 310 square kilometres and 1.3 million inhabitants the most densely populated town in Germany. The Greater Munich Area has a population of 2.4 million people with 1.3 million employees in 2000. In recent years, there have been two developments in the Greater Munich Area which are favourable for the introduction of telecommuting. On the one hand, the service sector has expanded and now dominates the economy. In this sector, work progress can often be transported via information and communication networks. On the other hand, the average home to work distance increased in recent years due to continuing suburbanisation. Thus, people become more open to suggestions for reducing the strain of commuting every day. The objective of the project is to analyse in the monocentric conurbation of Munich the effects which telecommuting has on: – Travel behaviour of telecommuters and their households, – Households’ expenditures in money and time and

414

–

A.F. Glogger, T.W. Zängler and G. Karg

Traffic-induced emissions of greenhouse gases and air pollutants.

This chapter is structured as follows. First the models and the method used in the empirical analysis are described, then the results are presented and discussed, and finally conclusions are drawn.

21.2 Model Two models were used to describe and analyse travel behaviour of telecommuters and their household members. The first model (Fig. 21.1) is the adapted Socioeconomic Model of Travel Behaviour (Zängler, 2000). It describes the travel behaviour of individuals: – At the micro level with respect to the mode choice, extent, activity and the socio-economic characteristics of the telecommuter, his or her household members and their household, – At the meso level with respect to accessibility to shopping facilities, social, cultural, economic institutions and businesses, – At the macroscopic level with respect to macro-economic and institutional characteristics.

Household Household member m Household member 2 Household member 1 (Telecommuter) TRIP Mode of Transport Time & Distance Mode of Communication

Activity Type

Location

accessibility to shopping facilities, social, cultural, economic institutions and businesses

micro level meso level macro level

socio-demographics, state, law and norms, society, culture

Fig. 21.1. Socio-economic Model of Travel Behaviour (adapted from Zängler, 2000)

The Impact of Telecommuting on Households’ Travel Behaviour

415

The second model (Fig. 21.2) focuses on the interrelations between physical and virtual mobility.

In a free society, households have access to physical and virtual mobility. To become physically mobile, they use the available network of transport (Salomon, 2000). To become virtually mobile, household members use the available communication network, e.g. telephones, Internet or faxes. In the context of telecommuting, the relation between physical and virtual mobility can be described as follows. Without telecommuting, employees travel to work every day and use the available network of transportation to reach their office. When telecommuting, they commute less frequently between home and office via the network of transportation; instead they become virtually mobile and ‘travel’ via the network of communication. Both networks are also used by the household and employer to carry out other activities to reach other locations.

other activities/locations

NC Virtual Mobility

Household h

Physical Mobility

Virtual Mobility

t1

Employer

t0

Physical Mobility

NT other activities/locations NT

network of transportation

NC network of communication

t 1 with telecommuting t 0 without telecommuting

Fig. 21.2. Interdependence between Physical and Virtual Mobility

Theoretically, the interrelationships between physical and virtual mobility can be modifying or neutral. In the case where they are neutral, no effect can be measured. In the case where physical and virtual mobility modify one another, there are in general three different effects to be considered (Moktharian, 2000; Salomon, 2000): 1.

Substituting/replacing physical by virtual mobility or vice versa

416

2. 3.

A.F. Glogger, T.W. Zängler and G. Karg

Generating virtual or physical mobility or both “Changing responsibilities in household duties and/or shifting mode or time of transport”.

Usually, combinations of all three effects are likely (Moktharian, 2000). These effects can express themselves qualitatively (e.g. modal shifts or modified travel patterns) or quantitatively (e.g. a change in the total distance travelled), or both.

21.3 Method In the years 2000 (t0) and 2001 (t1), a survey of the travel behaviour of telecommuters and their household members was carried out in eight big enterprises in the Greater Munich area. All employees who were to start telework were asked to take part in the study. It was set up as a pre- and post-treatment survey (before and after the introduction of telecommuting). The telecommuters and their household members of driving age were asked to fill out a detailed 2-day travel diary, a personal and a household questionnaire (adapted from Zängler, 2000; Axhausen, 1995 and 2002; Jones et al., 1985). Each person had to report his or her travel in the travel diary. The reported ‘mobility day’ started at 4 a.m. and ended at 4 a.m. the next day. Trips were recorded by the day of the week, the starting and finish time of the trip, the distance covered and the means of transportation used. Questions were asked about accompanying persons, subjective reasons for the choice of the mode of transport and subjective impressions of the trip. Activities were recorded in detail with respect to purposes, destinations, importance, time limits, and stops (intermediate or final). The personal questionnaires consisted of detailed questions on the expected influence of telecommuting on use of time, distances and the household budget, the telecommuter’s professional as well as private life, questions on the preferred mode of transport and the person’s driving licence, access to private means of transportation, and habitual travel behaviour. The household questionnaire was completed by the person familiar with the general characteristics of the household. These general characteristics are location and type of the place of residence, access to different social, cultural and economic institutions of daily life, the organisation of household duties (e.g. shopping, child care), availability of electronic equipment, car ownership, access to public transport, and the total income of the household. Household size and socio-demographic variables describing household members are also included.

The Impact of Telecommuting on Households’ Travel Behaviour

417

21.4 Results In the following, the characteristics of the sample are described, followed by resulting changes in travel behaviour, expenditures measured in both money and time as well as emissions.

21.4.1 Sample Characteristics In the first round, the main unit of the survey at t0 consisted of 227 future telecommuters and the members of their household with a total response rate of 46%, in the second round 51% of these respondents took part. In both rounds, 16% (36 telecommuters and 29 household members) participated. Gender was almost equally distributed over the sample with 54% female and 46% male telecommuters, 52% female and 48% male adult household members. The average age of t1 was 39 years for telecommuters and 40 years for household members. The average household size was three persons per household, which is above the average in Munich. Children lived in 69% of the households.

21.4.2 Travel Behaviour In the following, the impact of telecommuting on the temporal distribution of physical mobility and on cumulated distances is described for telecommuters and household members alike. The significance tests used were the McNemar test for the temporal distribution of physical mobility and the Wilcoxon test for trip distances. The non-parametric test was used due to the fact that the data are not normally distributed1. The most likely indicator to change after the introduction of telecommuting is the temporal distribution of trips over the day (Pendyala et al., 1991). Fig. 21.3 describes the temporal distribution of trips for telecommuters respectively their household members before and after the introduction of telecommuting. The x-axis represents 24 hours and is subdivided into 96 time intervals of 15 minutes. The yaxis shows the relative frequency (in per cent) of physically mobile persons (telecommuters or household members) of the corresponding target group. Fig. 21.3 a respectively b refer to telecommuters respectively household members and show the temporal distribution of physical mobility over the day before (t0) and after (t1) the introduction of telecommuting. Fig. 21.3 c respectively d refer to telecommuters respectively household members in t1 and show their temporal distribution of physical mobility on a telecommuting day versus an ordinary commuting day.

1 The McNemar test is a sign test used when binomial variables are analysed in a dependent sample. It relies on the change of the proportion of frequencies between the first and the second phase (Sachs, 1992).

418

A.F. Glogger, T.W. Zängler and G. Karg

The introduction of telecommuting leads to the following effects. Among telecommuters (cf. Fig. 21.3 a), the share of mobile persons decreased in the morning and evening rush hours in t1. This observed change on the temporal distribution of physical mobility of telecommuters is significant2. Among household members (cf. Fig. 21.3 b), the share of physically mobile persons is less in t1 than in t0. However, the observed change is not significant and the curve progression of their temporal distribution in the course of the day does not change (qualitatively). A large part of the decrease in physical travel is due to the telecommuters’ change in travel participation. However, household members cut their travel time and travelled distance by one-third as well. When interpreting these findings, underreporting – as observed in the State of California telecommuting pilot project (Kitamura et al., 1991) – has to be considered. However, Hamer et al. (1992) also reported that household members of telecommuters are more at home after the introduction of telecommuting. After the introduction of telecommuting (t1) a comparison of telecommuting and commuting days demonstrates the following effects. Among telecommuters (cf. Fig. 21.3 c)3,the share of mobile persons during a telecommuting day is considerably less than during a commuting day. Moreover, there is a clear difference between the temporal distribution of physical mobility on a telecommuting and commuting day resulting in a smaller share of mobile persons during the morning and evening rush hours. Among household members (cf. Fig. 21.3 d), the share of mobile persons and their temporal distribution on a telecommuting and commuting day is practically equal. For telecommuters, there is a decrease in total number of trips by 21% and in work trips by 46%. Households cut their number of all purpose trips by 18% and work trips by 33%. The telecommuters’ share of commuting distance by car increases from 38% to 45%. In addition, the number of car trips for leisure purposes and the total distance travelled for leisure purposes increase. At the same time, the household members’ share of car trips for leisure purposes decreases from 64% to 51% (see Table A21.1 and Table A21.2).

2

The differences between the shares of physically mobile tc in t0 and t1 are highly significant from 06:15h to 07:30h, from 16:00h to 16:45h, from 17:00h to 18:00h (McNemar, p