In Perpetual Motion

In Perpetual Motion" Travel Behavior Research Opportunities and Application Challenges Related Elsevier books DAGANZ...

Author: Hani S Mahmassani

39 downloads 1527 Views 21MB Size Report

This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form

DOWNLOAD PDF

In Perpetual Motion" Travel Behavior Research Opportunities and Application Challenges

Related Elsevier books

DAGANZO (ed.)

Fundamentals of Transportation and Traffic Operations

ETTEMA &TIMMERMANS (eds.)

Activity-Based Approaches to Travel Analysis

GARLING (ed.)

Theoretical Foundations of Travel Choice Modelling

ORTOZAR (ed.)

Travel Behaviour Research: Updating the State of Play

STOPHER & LEE GOSSELIN (eds.)

Understanding Travel behaviour in An Era of Change

HENSHER & BUTTON (eds.)

Handbook of Transport Modelling

BREWER et al. (eds.)

Handbook of Logistics and Supply-Chain Management

BUTTON & HENSHER (eds.)

Handbook of Transport Systems and Traffic Control

HENSHER (ed.)

Travel Behaviour Research: The Leading Edge

Related Elsevier journals

International Journal of Transport Management Editor: Mike McDonald Transport Policy Editor: Moshe Ben-Akiva, Yoshitsugu Hayashi and John Preston Transportation Research Part A: Policy and Practice Editor: Frank Haight Transportation Research Part B: Methodological Editor: Frank Haight

For full details of all transportation titles published under the Pergamon imprint please go to: www.transportconnect.net

In Perpetual Motion" Travel Behavior Research Opportunities and

Application Challenges

EDITED BY Hani S. Mahmassani

University of Texas, Austin, USA

2002 PERGAMON An Imprint of Elsevier Science

A m s t e r d a m - London - New York - Oxford - Paris - Tokyo Boston - San Diego - San F r a n c i s c o - S i n g a p o r e - S y d n e y

ELSEVIER SCIENCE Ltd The Boulevard, Langford Lane Kidlington, Oxford OX5 1GB, UK

© 2002 Elsevier Science Ltd. All rights reserved.

This work is protected under copyright by Elsevier Science, and the following terms and conditions apply to its use: Photocopying Single photocopies of single chapters may be made for personal use as allowed by national copyright laws. Permission of the Publisher and payment of a fee is required for all other photocopying, including multiple or systematic copying, copying for advertising or promotional purposes, resale, and all forms of document delivery. Special rates are available for educational institutions that wish to make photocopies for non-profit educational classroom use. Permissions may be sought directly from Elsevier Science Global Rights Department, PO Box 800, Oxford OX5 1DX, UK; phone: (+44) 1865 843830, fax: (+44) 1865 853333, e-mail: [email protected]. You may also contact Global Rights directly through Elsevier's home page (http://www.elsevier.com), by selecting 'Obtaining Permissions'. In the USA, users may clear permissions and make payments through the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; phone: (+ 1) (978) 7508400, fax: (+ 1) (978) 7504744, and in the UK through the Copyright Licensing Agency Rapid Clearance Service (CLARCS), 90 Tottenham Court Road, London W1P 0LP, UK; phone: (+44) 207 631 5555; fax: (+44) 207 631 5500. Other countries may have a local reprographic rights agency for payments. Derivative Works Tables of contents may be reproduced for internal circulation, but permission of Elsevier Science is required for external resale or distribution of such material. Permission of the Publisher is required for all other derivative works, including compilations and translations. Electronic Storage or Usage Permission of the Publisher is required to store or use electronically any material contained in this work, including any chapter or part of a chapter. Except as outlined above, no part of this work may be reproduced, stored in a retrieval system or transmitted in any form or by any means, electronic, mechanical, photocopying, recording or otherwise, without prior written permission of the Publisher. Address permissions requests to: Elsevier Science Global Rights Department, at the mail, fax and e-mail addresses noted above. Notice No responsibility is assumed by the Publisher for any injury and/or damage to persons or property as a matter of products liability, negligence or otherwise, or from any use or operation of any methods, products, instructions or ideas contained in the material herein. Because of rapid advances in the medical sciences, in particular, independent verification of diagnoses and drug dosages should be made.

First edition 2002 Library of Congress Cataloging in Publication Data A catalog record from the Library of Congress has been applied for. British Library Cataloguing in Publication Data A catalogue record from the British Library has been applied for.

ISBN:

0-08-044044-4

The paper used in this publication meets the requirements of ANSI/NISO Z39.48-1992 (Permanence of Paper). Printed in The Netherlands.

CONTENTS Acknowledgements

ix

Foreword Hani S. Mahmassani

xi

Section 1. Response to New Transport Alternatives and Policies

1

Chapter 1. Setting the Research Agenda: Response to New Transport Alternatives and Policies Peter Jones

3

Chapter 2. Living Models for Continuous Planning Andrew Daly

23

Chapter 3. Household Adaptations to New Personal Transport Options: Constraints and Opportunities in Household Activity Spaces Kenneth S. Kurani and Thomas S. Turrentine

43

Chapter 4. Responses to New Transportation Alternatives and Policies: Workshop Report Martin E. H Lee-Gosselin

71

Section 2. Dynamics and ITS Response

79

Chapter 5. Dynamics and ITS: Behavioral Responses to Information Available from ATIS Reginald G. Golledge

81

Chapter 6. Research into ATIS Behavioral Response: Areas of Interest and Future Perspectives Ennio Cascetta and Isam A. Kaysi

127

Section 3. Telecommunications-Travel Interactions

141

Chapter 7. Emerging Travel Patterns: Do Telecommunications Make a Difference? Patricia L. Mokhturian and Ilan Salomon

143

vi

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges

Chapter 8. Transport and Telecommunication: First Comprehensive Surveys and Simulation Approaches Dirk Zumkeller

183

Chapter 9. Telecommunications-Travel Interaction: Workshop Report David A. Hensher and Jackie Golob

209

Section 4. Travel Behaviour-Land Use Interactions

221

Chapter 10. Travel Behavior-Land Use Interactions: An Overview and Assessment of the Research Susan L. Handy

223

Chapter 11. Comparative Neighborhood Travel Analysis: An Approach to Understanding the Relationship Between Planning and Travel Behavior Roger Gorham

237

Chapter 12. Towards a Microeconomic Framework for Travel Behaviour and Land Use Interactions Francisco J. Martinez 26 1 Chapter 13. Land Use-Transportation Interactions: Workshop Report Ed Weiner and Roger Gorham

277

Section 5. Time Use

287

Chapter 14. Emerging Developments in Time Use and Mobility Nelly Kays and Andrew S. Harvey

289

Chapter 15. Time Use and Travel Demand Modeling: Recent Developments and Current Challenges Eric I. Pas 307 Chapter 16. Time Use: Workshop Report Ryuichi Kitamura

333

Contents Section 6. Travel Behaviour Measurement

vii 339

Chapter 1%

Current Issues in Travel and Activity Surveys Tony Richardson

341

Chapter 18, Motivating the Respondent: How Far Should You Go? Peter Bonsall

359

Section 7. Methodological Developments

379

Chapter 19. Recent Methodological Advances Relevant to Activity and Travel Behavior Analysis Chandra R. Bhat

381

Chapter 20. The Goods/Activities Framework for Discrete Travel Choices: Indirect Utility and Value of Time Sergio R. Jara-Diaz

415

Chapter 21.

Integration of Choice and Latent Variable Models Moshe Ben-Akiva, Joan Walker, Adriana T Bermardino, Dinesh A . Gopinath, Taka Morikawa, and Amalia Polydoropoulou

43 1

Chapter 22. Methodological Developments: Workshop Report Juan de Dios Orttizar and Rodrigo Garrido

47 1

Section 8. Forecasting

479

Chapter 23.

Forecasting the Inputs to Dynamic Model Systems Konstadinos G. Goulias

48 1

Chapter 24. Uncertainties in Forecasting: The Role of Strategic Modeling to Control Them Charles R a m

505

Chapter 25. Forecasting: Workshop Report Kostadinos Goulias

527

...

viii


Section 9. Microsimulation of Travel Activities in Networks

531

Chapter 26. Activity-Based Travel Behavior Modeling in a Microsimulation Framework Eric J. Miller and Paul A. Salvini

533

Chapter 2 7. Complexity and Activity-Based Travel Analysis and Modeling Pia M. Koskenoja and Eric I. Pas

559

Chapter 28. Microsimulation: Workshop Report Kay Axhausen and Ram Pendyala

583

ACKNOWLEDGEMENTS This volume represents the culmination of a collective undertaking that has drawn on the talents and energy of many individuals. From the initial planning stages of the Austin lATBR conference, to the detailed logistical aspects of organizing it successfully, to the production of the various post-conference publications, which have included several special issues of leading Transportation journals, I am grateful for the contribution of many dedicated colleagues, students and associates. Colleagues serving on the steering committee of the conference provided much wisdom, counsel and support through various stages of this process. This distinguished group included Kay Axhausen, David Hensher, Ryuichi Kitamura, Martin Lee-Gosselin, Juan de Dios Ortuzar, Eric Pas, John Polak, Peter Stopher and Ed Weiner. They were instrumental in helping define and refine the focus themes for this volume, as well as in suggesting the best authors for the commissioned resource chapters. I was truly fortunate to have such talent to rely on. I am also grateful to the many referees who helped at various stages of the selection process, for the conference, the various post-conference publications, as well as the present volume. This entire undertaking, including this volume, would not have been possible without the financial support of the US Department of Transportation, the Southwest University Transportation Center of Excellence (SWUTC), and the resources of my institution. The University of Texas at Austin. Those who attended the conference will undoubtedly recall the key role that my administrative assistant, Anne Suddarth, played in this endeavor as Conference Coordinator. I have the utmost gratitude to Anne for her dedication and tremendous effort in all stages of the conference as well as in the planning and preparation of this volume. Anne moved on to bigger and better challenges prior to seeing this particular project to completion. I am fortunate that Rebecca Weaver-Gill ably took over completion of this project, including final preparation of the camera-ready manuscript.

This Page Intentionally Left Blank

FOREWORD

Hani S. Mahmassani

Travel behavior interacts in a deep way with how we work, play and go about pursuing the various activities that make us a society and an economy. The diversity of forces that affect our economic and social activities, and the lifestyles we choose to pursue, also affect directly and indirectly how we travel. Social, intellectual, economic and technological forces are continually interacting with and affecting the spatial and temporal patterns of activities in which people and businesses engage. Technological developments in the production, dissemination, and consumption of information continue to amaze social observers and commentators. These developments have important implications for how we use our time, and how we go about engaging in the various work, sustenance and leisure activities that constitute our daily existence. That we travel to enable and accomplish our activity patterns is a longestablished tenet of the travel behavior "paradigm". But travel for its own sake, for the purpose of leisure and discovery, is also an activity that goes back to time immemorial. This book is intended to provide an authoritative assessment of the state-of-the-art developments in travel behavior research and applications, and identify the principal emerging trends, challenges and opportunities in this important area of transportation research. It is an outgrowth of the Eighth Meeting of the International Association for Travel Behavior Research, held in Austin, Texas in September 1997. The Austin meeting was intended as a milestone event in terms of defining cutting-edge problems and developments in this area, and providing both a snapshot, and entry point, as well as a foundation for future developments likely to take place over the next decade. This volume serves as the principal vehicle for accomplishing the above agenda. It is not a Proceedings collection in the traditional sense; rather, it is a focused collection of both commissioned as well as contributed chapters that achieve the desired objective of producing an authoritative document for the travel behavior research field. With continuing developments in the technological, methodological and policy realms, the time is opportune for a major assessment of accomplishments, current trends, and future directions. To accomplish this objective, this volume is organized around nine major themes

xii


selected to reflect areas of future opportunity and growing professional interest, rather than to mirror the traditional mature areas of past endeavors. For each theme, one or two resource chapters have been prepared by the best-known scholars in the respective field of research or application. When several perspectives were relevant to a particular theme, two resource chapters were commissioned. For example, in the emerging field of time use, a chapter was invited by the leading sociologists working in the area of time use, to share their fundamental insights and approaches with the travel behavior community. At the same time, the leading researcher in travel behavior who was trying to incorporate time use ideas, the late Eric Pas, provided the other resource chapter. Taken together, these provide an invaluable and unique entry point to this growing area of investigation. Naturally, each chapter was carefully refereed, and extensive comments provided by other leading researchers were incorporated in the final manuscript. In addition to the resource chapters, which form the backbone of this volume, one or more examples of best or most innovative research were selected for each theme. Extensive refereeing and revision resulted in the chapters included in this volume. In addition, for most themes, the Austin meeting produced a report based on workshop deliberation during and after the meeting, aimed at identifying key challenges facing research and practice in the next decade. The workshops conducted their deliberations over several days, and in several instances produced milestone documents for their respective themes. This volume contains the last contributions of a dear friend and colleague, and truly influential thinker in the travel behavior field, the late Eric Pas. As noted, he is the author of one of the two resource chapters highlighting developments and challenges in the area of time use research. At the time of the conference, Eric produced a document that he labeled as a work in progress. Unfortunately, his untimely death kept it in that form. In many ways, this area of investigation remains very much a work in progress. Rather than edit that chapter, or seek to have it completed by one of his close colleagues, I have chosen to leave it in "raw form", as a testimonial to Eric's contribution, and as rare glimpse into the mind of a genuine contributor and thinker to this field. As such, it reveals the germination of several important ideas and concepts, and research directions in their early stages. It is a fitting testimonial and recognition that this field, to which Eric has dedicated his professional career and in which he was invested personally and socially, remains very much a work in progress. Eric is also the co-author of a second chapter, with Pia Koskenoja, on "Complexity and Activity-Based Travel Analysis and Modeling" (Chapter 27). In this case, the co-author was responsible for the final revision of the chapter. That work is also indicative of important new

Foreword

xiii

directions that Eric's work was beginning to chart, along particularly challenging dimensions of activity and travel behavior modeling. Below is a description of the key themes, which form the different parts of the book.

ORGANIZATION OF THE BOOK The volume kicks off with an excellent tour of the major policy and societal drivers underpinning much of the current interest in travel behavior research. These are the challenges that necessitate better and deeper understanding of travel behavior, and provide much of the motivation for the substantive and methodological developments that are highlighted in this volume. Peter Jones shares his unique clarity of insight into these complex issues, and crisply identifies the key questions that must be elucidated by travel behavior researchers to help guide policy. He identifies seven key areas in which new transport alternatives and policies are motivating new research into travel behavior. These areas include: changes in road capacity, traffic restraint measures, new modal alternatives, information provision, tele-services, mobility management, and land use policies. This chapter may well prove to be the research agenda for travel behavior research into the new century. To further explore the methodological aspects of how to model the response of potential tripmakers to proposed new transport alternatives and technologies, a second resource chapter was prepared by Andrew Daly, based on his extensive professional experience in advising agencies on such strategic questions. The chapter provides a valuable resource and practical insights for researchers and professional developers of model systems for forecasting the demand for new modes and transport alternatives that did not exist at the time the modeling system is developed. Chapter 3 provides an example of research on user responses to new transport options, specifically adoption of small "green" vehicles intended to reconcile mobility needs with environmental considerations. The chapter illustrates the complex interactions between activity patterns and travel behavior in a spatial context. Chapter 4 is the closing chapter in this theme. It presents the collective deliberation of researchers during and after the Austin meeting, on the challenges and implications for behavioral theory, survey methods and modeling, of major contemporary shifts in the demographic, social, technological and political contexts for personal travel. Considerable wisdom on how to approach the future is conveyed in this chapter.

xiv


The second part of the book addresses the rapidly evolving theme of Dynamics and ITS (Intelligent Transportation Systems) Response. This theme considers behavioral responses to real-time information through advanced traveler information systems (ATIS), including various forms and sources of traveler information. Research along this theme is also concerned with within day and day-to-day choice processes, and other aspects of travel behavior dynamics. Chapter 5, prepared by Reginald Golledge, offers an extensive state-of-the-art assessment of developments, from a variety of disciplines, that are relevant to this theme. It is complemented by a systematic discussion of the behavioral dimensions of ITS response, and the substantive and methodological research opportunities offered by this more recent area of travel behavior. While initiated as a workshop report, this chapter, prepared by Ennio Cascetta and Isam Kaysi, has grown into a full-fledged resource chapter that nicely complements and augments Golledge's contribution. The third theme represents another area where telecommunications technologies present strong interactions with travel behavior. Titled Telecommunications-Travel Interactions, the intent of the theme is to extend the scope of interest beyond telecommuting, which has been the primary focus of most previous work in travel behavior in this general area. With the provocatively titled "Emerging Travel Patterns: Do Telecommunications Make a Difference?", Patricia Mokhtarian and Ilan Solomon, who have individually and jointly pioneered and defined this area of investigation, collaborated on a seminal resource contribution (Chapter 7) for this volume. This chapter is as close as our field is likely to get to a definitive, authoritative assessment of where we are, and where we are likely to go in understanding the interaction between telecommunications and travel. With continuing rapid development and deployment of telecommunication technologies, coming unto third generation (3G) broadband wireless access to the Internet, there will likely be no shortage of challenges and opportunities to understand travel and telecommunication behavior jointly in the context of activity participation and scheduling over time and space. An example of such joint investigation is provided by Dirk Zumkeller in Chapter 8, illustrating a novel survey approach used in Germany and Korea. The workshop dealing with this theme proved highly successful, producing an excellent document (Chapter 9) carefully put together by David Hensher and Jackie Golob. The chapter articulates important conceptual notions that form the basis of an advanced methodological framework to jointly examine telecommunication and travel activities. In particular, the concept of the content mix of an activity, and the resulting mixed content framework, hold intriguing promise for further development of this theme. The fourth part of the book addresses a relatively new area of interest in the travel behavior community, namely Travel Behavior-Land Use Interactions. This theme addresses the effects of land-use characteristics and the physical environment on travel behavior, including

Foreword

xv

pedestrianization and bicycle use, traveler attitudes and perceptions associated with neighborhood characteristics, and the potentials to use land-use policies to bring about changes in travel behavior. To reflect the wide range of theoretical, methodological and disciplinary perspectives that bear upon this theme, two resource chapters are included. The first (Chapter 10), by Susan Handy, presents the issues primarily from the perspective of the planning community, reflecting concerns for the built environment, land use policy and neighborhood accessibility. These themes are amplified in the research presented by Roger Gorham in Chapter 11, which examines the effect of neighborhood characteristics on travel behavior. Chapter 12 contains the second resource paper, prepared by Francisco Martinez. It presents more of the microeconomic perspective underlying much of the mathematical models of land use and location processes, considered jointly with the demand for travel. The interaction of land use and travel behavior and the demand for transport presents a fertile field of scientific and methodological investigation for the travel behavior research community. The diversity of disciplines and perspectives, as well as the range of policy questions that motivate such research were encouraged to interact and exchange views during the Austin meeting, resulting in the workshop report prepared by Ed Weiner and Roger Gorham, and included in Chapter 13. As activity-based approaches to travel behavior and demand analysis have continued to gain acceptance beyond the research community and into the practicing community of planners and demand modelers, the natural connection between activity participation and time use is the natural frontier to conquer. Research into Time Use forms the fifth theme of the volume. It addresses emerging developments and multi-disciplinary perspectives on the analysis and prediction of travel behavior in the context of tripmaker time allocation to various activities. Time use research has been of interest to sociologists for many years, and a learned society exists to deal specifically with time use issues. The Austin meeting marked the first formal attempt at a rapprochement of intellectual perspectives between the time use and travel behavior research communities, as the focus of interest converges on understanding time use and allocation, albeit with different ultimate objectives. This volume features a resource chapter co-authored by two of the leading scholars on time use research, primarily from a sociological and anthropological perspective. Nelly Kalfs and Andrew Harvey collaborated on this effort, presented in Chapter 14. A second resource chapter (Chapter 15) was prepared by the late Eric Pas, as noted earlier; this chapter constitutes his last known work to be published, and reflects the best thinking in the travel behavior community on the issues that consideration of time use introduces to understanding mobility, and on the challenges and prospects of integrating this aspect of human behavior in models of travel behavior. These same issues were addressed and debated in a workshop setting, resulting in Chapter 16, prepared by Ryuichi Kitamura.

xvi


Part six of the book addresses Travel Behavior Measurement issues and techniques, and related analysis methods, with a focus on novel, non-traditional and technology-assisted measurement techniques. Measurement of actual behavior, associated perceptions and underlying attitudes is a critical requirement for understanding and modeling travel behavior. Chapter 17, by Tony Richardson, provides the resource contribution to this topic in this volume, highlighting current issues in travel and activity surveys. One of the most vexing issues faced by survey designers and analysts is to obtain the cooperation of survey respondents. Various schemes have been devised over the years to "incentivize" potential respondents. But how reliable are responses supplied primarily for the purpose of earning a prize? This issue is addressed by Peter Bonsall in Chapter 18, in a study that has already been widely discussed in the travel behavior research community. It is another important contribution in this volume, and is likely to become a classic in the study of survey methods. Measurements of travel behavior often form the basis for the development of mathematical models to represent users' choice behavior along the various dimensions of travel and activity participation. As noted in virtually all the other themes in this volume, growing complexity of the travel behavior processes of interest to researchers and policy-makers, and increasing sophistication in the ability to measure dynamic aspects of travel and activity at the micro scale give rise to interesting methodological challenges for analysis and model development. A comprehensive tour of pertinent developments and applications of econometric and psychometric methods and statistical modeling techniques to travel and activity behavior is provided by Chandra Bhat in Chapter 19, which forms the resource contribution to the Methodological Developments theme of part seven of the volume. Much attention in that chapter is devoted to the structure of individual discrete choice models, under different behavioral assumptions and data generating processes. The chapter provides an excellent entry point for econometricians interested in assessing the prevailing level of sophistication in methodological approaches used in travel behavior research. Focusing on the theoretical microeconomic underpinnings for model specification and interpretation of estimated model coefficients, Sergio Jara-Diaz discusses an important topic for the evaluation of transport alternatives, namely value of time estimation, in the context of an activity-based framework (Chapter 20). Chapter 21, by Moshe Ben-Akiva and several collaborators, presents a model framework and specification that integrates discrete choice models with latent variable formulations. Proponents of different methodologies tend to hold strong opinions about the relative merit of their preferred approach; the methodological developments workshop engaged in spirited debate, but eventually produced a report (Chapter 22, prepared by Juan de Dios Ortuzar and Rodrigo Garrido) that reflects the intellectual

Foreword

xvii

dynamism of this field, and conveys the promise of many more challenging debates in the years to come. One of the practical objectives that motivates much methodological development in travel behavior analysis is the application of the behavioral models to forecast the future demand for travel under different scenarios. Part eight of this volume has as its theme the Forecasting process, including the rationale and procedures for forecasting future demand using dynamic models of travel behavior in a network context, and the relation of such forecasts to the policy decision process. Such models provide new capabilities to examine the system's evolution under alternative travel policies instead of focusing on a single time point in the future, and recognize more explicitly the interaction between policy decisions and alternative future paths. The scope of this theme also includes the input forecasts required to drive most travel forecasting models, and methods by which to develop such forecasts. This theme is addressed comprehensively in the resource chapter (Chapter 23) prepared by Kostas Goulias. Forecasting would not be a challenge without uncertainties to contend with. Chapter 24, by Charles Raux, discusses a strategic decision-making approach for controlling forecasting uncertainties. A workshop report, prepared by Kostas Goulias, is included as Chapter 25. The last section of the book deals with an emerging methodological theme that is growing in significance with the need to apply micro-models of travel behavior for policy assessment and impact forecasting. The Microsimulation of Travel Activities in Networks theme addresses theoretical and methodological issues in the development of microsimulation procedures for the application and implementation of activity-based approaches and other emerging and existing travel behavior analysis methods. Eric Miller and Paul Salvini provide a valuable resource contribution in Chapter 26 that should serve as an excellent entry point into this topic area for travel behavior analysts who may have had only limited exposure to this methodological domain. While not strictly a microsimulation application. Chapter 27, a collaboration between Pia Koskenoja and the late Eric Pas, provides important new directions on how to approach complexity and activity-based travel analysis and modeling. Workshop participants tried to grapple with many issues associated with the application of microsimulation approaches, and tried to transfer some of the experience developed with these methods in the area of traffic simulation to that of activity and travel behavior modeling. The report from these deliberations, prepared by Kay Axhausen and Ram Pendyala, is included as Chapter 28.

xviii


IN CLOSING Travel behavior research has emerged into a vibrant field of intellectual inquiry, pursued by a truly cross-disciplinary community of scholars, who identify with the International Association for Travel Behavior Research as a primary home for the dissemination and cross-fertilization of these ideas across traditional disciplinary boundary lines. Through the interaction of disciplines like economics, psychology, sociology, statistics, artificial intelligence, management science, urban planning, geography and transportation systems engineering emerge new ideas and new approaches to grapple with the complexity of travel and activity behavior, and to address important challenges in the transportation policy arena. As we continue to travel together, this field, like its subject matter, will likely remain in perpetual motion. This volume is intended as no less than a milestone documenting the direction and gradient of the dynamic evolutionary path of the travel behavior research community.

SECTION 1 RESPONSE TO NEW TRANSPORT ALTERNATIVES AND POLICIES


In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.

SETTING THE RESEARCH AGENDA: RESPONSE T O NEW TRANSPORT ALTERNATIVES AND POLICIES

Peter Jones

INTRODUCTION The last fifty years has seen a period of relatively consistent growth in car traffic in North America and Western Europe. This has been associated with the gradual diffiision of car ownership to the large majority of households, the construction of high speed and high capacity inter-urban (and in some cases urban) road networks, and the decentralisation of much economic activity. During this period the transport options have remained broadly similar (though with improvements in quality, speed and variety), and most of the time there has been a general acceptance that road traffic will continue to increase, and that this is both inevitable and - in most cases, on balance - desirable. Now, however, we are entering a period of much greater uncertainty, where past trends are not a reliable guide to the fiiture. Many sections of road are becoming heavily congested for much of the day, thereby inhibitingfiirthergrowth; concerns about the environmental impacts of traffic are increasing (particularly in relation to air quality, traffic noise and CO2 emissions); many cities are actively trying to constrain car use; and developments in telecommunications are beginning to have major impacts, both on the operation of the transport systems and, more generally, on the ways in which people organise their lives. The second half of the twentieth century has been associated with a step change in accessibility by road: the construction offi-eewaysand motorways has typically halved journey times by car, and

4


in congested cities like London reduced them to one-third of previous levels. But, after decades of continuous improvements to inter-urban road networks, demand is now outstripping capacity (with little prospect of a major new round of highway construction), to the extent that trends are being reversed: journey times in some areas are becoming longer and travel times much less predictable. The UK parliament has recently passed two traffic reduction acts. The Road Traffic Reduction Act 1997 requires local authorities to assess the 'carrying capacity' of local roads and to take measures to limit traffic levels in areas where there are major environmental or economic negative externalities from road traffic; where action is not recommended, this has to be justified. The Road Traffic Reduction (National Targets) Act 1998 requires Central government to set national traffic reduction targets, or introduce other measures to reduce the adverse impacts of road traffic growth. Cities in several European countries are now looking seriously at traffic restraint measures, either via selective vehicle bans or using some form of electronic road pricing. Under legislation currently before the UK parliament, local authorities will be empowered either to charge vehicles individually for the use of existing roads, or to levy an annual charge on each private car parking space reserved for employees of that company. In both cases, the aim is to reduce traffic levels and to raise additional revenue to fiind improved transport alternatives to the private car. In parallel with this, the range of new altematives to the conventional privately owned, petrol/diesel ilielled car is growing rapidly; ranging from electric vehicles, to light rapid transit systems and automated people movers, and the formation of neighbourhood 'car clubs'. Information about current traffic conditions and transportation options is becoming available on a scale hitherto unknown - with uncertainties as to whether this will lead to greater stability or instability in network conditions. And telecommunications is offering 'virtual' mobility as an alternative to physical travel, through tele-working, tele-banking, tele-shopping, etc. Both in policy and behavioural terms, we are rapidly moving into uncharted waters. As a consequence, the modelling and evaluation frameworks that were suited to trend extrapolation and assessing 'more of the same' are inadequate to address many of these emerging transportation planning issues. More than ever before, what is needed is an understanding of travel behaviour: what motivates people to travel? how will travellers respond to the various new products and measures intended to influence their behaviour? will the public accept proposed restrictions on car use, and efforts to get them to substitute telecommunications for travel? All this needs to be set in the broader context of the kind offiituresthat societies wish to achieve.

Setting the Research Agenda: Response to New Transport Alternatives and Policies In the UK, for example, there has been much talk on the part of senior transport professionals, as well as the prime minister, about the need to fundamentally change lifestyles. Part of this debate is in the wider context of Local Agenda 21, but if taken seriously it would have major implications for transport and travel behaviour. This chapter first identifies seven areas where research analysts are being confronted by new transportation policies and alternatives, namely: changes in road capacity, traffic restraint measures, new modal alternatives, information provision, tele-services, mobility management, and land use policies. In each case, it discusses the questions they raise in relation to travel behaviour, assesses some of the evidence to date on responses and identifies some key unresolved questions for future research. The chapter then considers some of the implications of this changing research agenda for travel behaviour methodology, in particular for: conceptual/analyticalframeworks,modelling requirements, data needs and evaluation.

NEW T R A N S P O R T A T I O N A L T E R N A T I V E S A N D P O L I C I E S

Changes in Road Capacity While historically the assumption has been that, in increasing road capacity, the traffic engineer is simply catering for the growth in traffic demand, not influencing it, there is some evidence to suggest that supply does influence demand - as economists would expect it to do. For example: (i) The 'gridlock' that has been forecast to result in some European cities if road capacity were not increased in line with the growth in car ownership has generally not occurred. (ii) In urban areas where capacity has been taken out of the road network (either for emergency repairs or as apart of a planned re-allocation of roadspace), it appears that a part of the displaced traffic has 'disappeared' from the network. Cairns et al (1998) identify over 60 case studies world-wide where such traffic reduction has taken place. On average, 40% of traffic disappeared from the affected sections of the network, with a net reduction of 25% when displacement effects are taken into account. (iii) Where road capacity has been increased substantially in higher density areas, with significant levels of suppressed demand, there is evidence of traffic growth above trend ( S A C T R A , 1994); in one corridor into central London, for example, the construction of a high capacity motorway appears to have led to a 70% increase in traffic levels compared to comparable corridors into London.

5

6


The SACRA study identified a number of behavioural mechanisms that might account for the observed increases in traffic levels, including: route switching (with a possible increase in route distance), time switching, a change in mode or destination, or the generation of new, out-of-home activities. The study concluded that more research is needed to identify the relative importance of these traveller responses in different circumstances. At a more macro level, there is evidence of two phenomenons that can assist in accounting for this elasticity between demand and supply. First, empirical evidence of an apparent constancy in travel time budgets, at an aggregate level (though this seems to disappear once data is disaggregated by person type). On average, travellers seem to spend about an hour per day travelling. Figures from German national travel surveys (Brog and Erl, 1996) show a general consistency in average daily time budgets between different urban areas, with a slight upward drift in values over time. In addition, despite the very different travel conditions in the West and the former East Germany, time budgets in aggregate are very similar. The UK data shows a similar pattem (Noble and Potter, 1998), with just under one hour of travel per person per day, on average. In addition, both studies report very little change in average trip rate over time - resulting in a broad stability in average travel time per trip, but a significant increase in average trip length. Accepting the validity of the evidence (though we do not fully understand the underlying behavioural mechanism), then better roads will enable an increase in average vehicle speeds and hence result in a greater distance to be travelled per trip for the same time expenditure. In the aggregate this leads to an increase in car traffic on the improved network. Conversely, we would expect that more congested roads would result in shorter trips and less traffic in aggregate. This also suggests that the analysis of travel behaviour should be carried out at the level of daily travel, not at the individual trip level. Second, there is the 'Downs-Thomson Paradox' (Mogridge, 1990), which states that in areas where there is suppressed demand for car travel and a good rail-based public transport service, then there will be an equilibrium between the average door-to-door speed by private car and by public transport. This was demonstrated by Mogridge for radial trips into the centre of London and Paris, and has recently been independently confirmed in journey time studies into central London, carried out for the government. The implication of this finding is that, in cities with good rail networks, an important 'source' or

Setting the Research Agenda: Response to New Transport Alternatives and Policies 'sink' for generated/suppressed car traffic is the parallel journeys made by rail. This is the nature of the paradox, since it implies that the best way to increase road network speeds is to invest in better public transport; this then raises the equilibrium speed on both networks. Conversely, increasing road capacity will attract users from rail, leading to service cut-backs on that mode, and ultimately a lower equilibrium door-to-door travel time on both modes. Some Key Research Questions: • How are travel patterns influenced by road network capacity? • What are the long term impacts on behaviour of changes in capacity? • Are there behavioural mechanisms that can account for the travel time budget phenomenon?

Traffic Restraint Measures This aspect of transport policy perhaps confronts more directly than any other many of the implicit assumptions built into past analyses and modelling efforts: instead of catering for past trends, how can we change them? It forces consideration of the full range of possible behavioural responses, from route shift through modal shift and destination switch, to trip re-timing, trip consolidation/re-chaining and on to trip suppression. Many of these responses are poorly handled by - or may be entirely absent from - existing travel demand models. It also confronts the analyst with issues of public acceptability, thereby broadening the notion of behavioural 'response' to encompass attitudinal factors as well as observed changes in behaviour. And raises basic questions concerning the ways in which we evaluate transport measures. While there is little empirical evidence of the effects on travel behaviour of introducing a road pricing scheme that has the objective of reducing traffic levels (unlike the Norwegian schemes, where revenue-raising is the primary objective), the various stated preference studies that have been conducted tend to give a consistent picture (Jones, 1992). The preferred driver response is re-routing (where scheme design permits this), followed by trip re-timing. If destination switching is a realistic option (depending both on the trip purpose and the location of alternative attractions), this may occur in preference to modal switch, unless the existing destination is of high quality and access using alternative modes is attractive. In situations where alternative modes are poor, then significant trip suppression may occur. As has been noted, an important aspect of traffic restraint that has to be taken into account is

7

8


public acceptability. Generally, there is greater support for physical or regulatory measures over pricing, mainly on equity grounds. But the details of the design can have a major influence on public attitudes, as well as the uses to which the net revenue is put. Typically in the UK, support for urban road pricing doubles (from 30% to 55-60%) if the money is ring-fenced for transportrelated projects that the public supports (Jones, 1995). The recent adoption of 'value pricing' in the US (Berg et al, 1999) is well aimed at maximising public support: drivers are provided with a choice (paying versus a free lane), they derive some personal benefit from the payment (i.e. saving time) and the money collected is used tofimdthe improved facility. Some Key Research Questions: • How is travel behaviour differentially affected by the kinds of restraint measure that are introduced (physical, regulatory, pricing)? • How do different forms of traffic restraint affect car ownership and longer term residential and employment location decisions? • How are public attitudes affected by the type of restraint measure adopted and the rationale behind implementation?

New Modal Alternatives The development of new modal alternatives to the private car exposes weaknesses in our understanding of travel choices, since there is little or no past experience on which to base the analytical work. In the case of drivers switching to an improved public transport service, for example, there is circumstantial evidence to suggest that prospective travellers prefer a rail-based over a bus-based system, even though the performance characteristics of the two may be indistinguishable in conventional generalised cost terms (Luk et al, 1998). Here the challenge is to uncover the other variables that influence choice (e.g. ride quality, greater perceived reliability) and incorporate these into the analysis. Some of the factors that seem to affect attitudes and behaviour have little to do with the usual measures of travel utility; for example, some UK work into preference for light rail suggests that it is the perceived status and permanence that are important factors attracting certain groups of car drivers (see Transport 2000, 1998). The work on electric cars has shown the importance of image and has also brought out other issues not dealt with in conventional travel analyses. In particular, the need to take into account the distribution of distances driven by cars in a day, and the 'safety margin' that prospective purchasers require in order to feel confident when relying on an electric vehicle with a relatively

Setting the Research Agenda: Response to New Transport Alternatives and Policies slow recharge/refliel time - even though its range may only be exceeded for a few days in a year (Golob and Gould, 1998). Several countries have also seen policy efforts to encourage greater walking and cycling, not only in the context of encouraging more sustainable travel patterns, but also as part of health sector policies to increase physical fitness and psychological well being. Results of such initiatives have been mixed, and walking and cycling behaviour are probably the least well understood facets of travel behaviour. Some Key Research Questions: • What are the 'missing' variables in mode choice models and how can they be measured? • What are the factors affecting the mode-specific constant in mode choice models? • Which factors influence attitudes towards and use of walking and cycling? • How do we address the need to carry out tour- or daily travel-based analysis of the scope for using new travel modes with different performance characteristics?

Information Provision Advances in the applications of transport telematics over the last few years have highlighted a major deficiency, both in basic economic theory and in modelling applications. Each has assumed that, when making choices, travellers have 'perfect' information about the transport options and their performance (or, in the random utility model, that perceptions are distributed around the true values). Thus, before substantial work could be undertaken to show the behavioural impacts and benefits of improved information, it has been necessary to develop models that are based on the assumption of travellers having imperfect information. The main sources of travel information are to be found at trip origins or destinations (in-home/at workplace), at transport interchange points, in-vehicle and at the roadside. A number of studies, particularly from the United States, have shown the considerable network benefits that can result from providing a limited number of drivers with accurate information about network conditions (e.g. Al-Deek et al, 1998; Mahmassani, 1997). Work on network information provision raises difficult ethical issues that have not hitherto affected transport professionals in such explicit fashion. In particular, issues relating to withholding information, either from certain groups of drivers (in order to maximise network benefits), or regarding certain parts of the network (e.g. residential streets) in order to protect neighbourhoods from through traffic.

9

10


Less work has been done on the provision of pre-trip information, though in principle the potential benefits are much greater than for the provision of in-trip information. As well as rerouting, there is scope to re-time trips, change destination or mode, or cancel the trip entirely (Jones and Cassidy, 1995). In relation to real-time information about public transport services, at bus stop or in station, there is little evidence of a major effect on travel behaviour, although the stress associated with waiting seems to be reduced considerably, thereby lowering the additional weight applied to waiting time in generalised cost formulations. However, a recent demonstration project providing real-time bus information on the Internet, does appear to have substantially increased bus patronage. Some Key Research Questions: • What are the key traveller information needs, and the most appropriate mechanisms for delivery? • What are the effects of improved information provision on attitudes and behaviour? • Does inappropriate forms of information provision lead to network instability, and how can this be avoided? • Which new evaluation procedures are needed in order to capture the full range of benefits of ITS applications?

Tele-services There is a rapidly growing literature on the impacts of tele-commuting, tele-shopping and telebanking on aspects of travel behaviour, but as yet there is no general consensus on the consequences of these developments for the overall level of trip making. Some studies have claimed that tele-commuting could eliminate peak period road congestion (e.g. NERA, 1997). However, others have been more cautious in their claims. While substituting telecommunications for travel undoubtedly eliminates or reduces the travel associated with that particular work activity at that time, there may be secondary effects that largely offset this (Mokhtarian, 1997). In particular: (i) In the longer term, part-time tele-commuters may choose to locate further from their workplace, substituting fewer, longer trips for daily shorter ones, so that total VMT on a weekly or monthly basis may not be reduced, (ii) Non-work travel may increase and take up some - or all - of the saved commuting time: if the 'constant travel time budget' hypothesis were strictly applied, then this would be the

Setting the Research Agenda: Response to New Transport Alternatives and Policies

11

logical outcome, (iii) A car no longer required for the daily commute may instead be used by another household member, who previously travelled on foot or by public transport. In addition, there may be wider environmental effects that negate the VMT savings made by the primary traveller; for example, spending extra time at home (instead of at the workplace) may push up energy consumption in the form of extra heating, lighting and air conditioning. Wider issues may also arise in relation to some other tele-activities. For example, home deliveries made in response to tele-shopping may consume more fuel and create more pollution (especially particulates) than the shopper would have done in making the trip him/herself While some of these effects have been successfully analysed at the trip/no trip level, a fuller understanding of impacts is likely to require a daily or weekly analysis framework, and some consideration of the wider impact on household travel and activity patterns. Some Key Research Questions: • How does tele-working affect longer term decisions about residential location? • What are the impacts of the various kinds of tele-services on overall household activitytravel patterns? • Are there impacts of tele-services on the numbers or types of cars owned?

Mobility Management Mobility management is a European term used to denote measures designed to reduce the traffic intensity of land use activities at major traffic generating sites, such as workplaces, shopping centres, schools or football grounds; they may be aimed primarily at employees, customers and/or visitors. In parts of Western Europe, considerable use is now being made of this measure as a contribution to limiting traffic growth (Bradshaw and Jones, 1998). The stimulus for implementation either comes from local government (primarily concerned with reducing local traffic congestion, air pollution, etc.); or from the companies themselves. The latter may wish to reduce their impact on the local neighbourhood, or make more productive use of their site and reduce the provision of car parking spaces, or to expand their operation without making increased parking provision. The kinds of measures involved include: improved cycle provision, better access on foot, limitations on parking provision, flexible work hours, organised car sharing schemes, works

12


buses, special ticketing deals and better information about public transport services. They may be introduced in conjunction with a local 'travel awareness' campaign, often organised by the local authority (Ciaburro ^r«/, 1994). These measures are generally initiated by the site managers or owners and normally involve only limited investment in new infrastructure, although they may be implemented in partnership with other organisations (such as local councils or public transport operators) who are able to invest in substantial transport schemes. In the United States such initiatives are usually refereed to as Travel Demand Management and are generally limited to joumey-to-work initiatives. In Europe, Green Travel Plans may cover all trip purposes, and also seek to reduce the numbers of goods and service vehicle trips. The success of these measures seems to depend in large part on the commitment of management to implement such policies and to take the lead in changing their own behaviour (Rye, 1997). Implementation of supporting measures by local authorities and transport operators can also assist. Some Key Research Questions: • How do travellers respond to changes in working practices; for example, does the introduction of flexi-time help to reduce or increase car dependence? • What are the effects of small changes in transport supply (e.g. provision of cycle lockers) on modal choice, and how can cost-effective packages of measures be developed? • What is the influence of publicity and changing images/fashions on travel attitudes and behaviour? • Do work-based initiatives have any influence on other aspects of travel behaviour (either at the individual or household level)?

Land Use Policies Recently there has been renewed interest in the UK, in parts of the US and in some other countries concerning the role that land use policies can play in reducing car dependence and the volume of car traffic. Such policy measures include higher density development, mixed-use development (instead of land use zoning) and well integrated public transport and walking/cycling networks (IHT, 1999). Some European cities are also experimenting with 'car free' housing developments. The transport benefits of such land use polices are felt to lie in invoking two behavioural


13

responses. Either in the shortening of the trip lengths of journeys that continue to be made by car (since drivers have a larger range of destinations closer to home); or by encouraging car drivers to switch to alternative, more sustainable modes that can be provided more competitively in such situations - because activities are close together (so making walking or cycling an attractive alternative), and/or because they are concentrated in corridors where public transport can offer an attractive service. One UK study estimated that the consistent application of such land use policies over a number of years could achieve a 15% reduction in traffic levels, compared to the trend projection (DoE/DoT, 1993). A small UK survey found lower levels of car ownership and use among matched households in areas where a high proportion of facilities could be reached within convenient walking distance (Walker, 1997). Taking a longer term perspective, there are unresolved questions about the responsiveness of land use patterns to transport system provision. If travel by car becomes significantly slower or more expensive, how will the property market respond? In general, both the micro-level effects on household activity/travel patterns, and the aggregate impacts on traffic volumes in an area are poorly understood. Some Key Research Questions: • Can higher density and mixed-use developments reduce car dependency and use? • Do car drivers use the greater density of opportunities to select from a wider choice set without an average reduction in trip length, or do they make shorter car trips, or substitute other modes in some cases? • Do higher density developments lead overall to a reduction or increase in road congestion and air pollution?

IMPLICATIONS FOR METHODOLOGY Here we consider, in turn, the state of knowledge and practice in four areas of methodology (concepts, models, data, and evaluation), and the priorities for further research.

Conceptual/Analytical Frameworks Like any others, transport problems are perceived and debated by politicians, the public and professionals within the context of a particular paradigm. This has a strong influence on the way

14


the issue is addressed: the kinds of problems that are identified, their diagnosis and on the remedial measures that are recommended. For example, the widespread adoption of 'free market economics' principles in many countries has led to a particular interpretation of the deficiencies in transport provision, and to remedies that involve the deregulation and privatisation of many transport services. In terms of our understanding of traffic and travel demand, there has been within the transport profession, a close relationship between developments in modelling and in conceptual thinking (Jones, 1983). The early post-war interest in expanding the road network to accommodate the growth in car use resulted in simple vehicle-based forecasting models. Initially these just applied growth factors to major road links, then later shifting attention to the network level and developing a three-stage vehicle generation, distribution and assignment modelling package. Subsequently, in urban areas, there was pressure to expand the scope of the models. First, to take into account the travel needs of non-car owners when planning fiiture transport investment, by incorporating a modal split segmentation into the trip generation sub-model. Later, to try and actively encourage a shift from car to other transport modes for certain trips, by building a separate mode split sub-model to facilitate the analysis of this behavioural response. In order to make these modifications, a shift in the basic unit of analysis was required: from vehicles to person trips. Since it was now recognised that the underlying rationale for travel, and for investment in transport infrastructure, is to move people (and goods) not vehicles per se - the latter in the main simply provide a means to an end (accepting that some people enjoy dxw'mgper se). Several travel behaviour researchers have argued for many years that there is another, more fimdamental unit of analysis, behind the person trip: this is based around activity-based analysis. Here the starting point is to assume that, in the main, people do not travel just for the sake of doing so, but in order to move from one location to another so that they can perform a set of activities. Such a paradigm is consistent with the growing emphasis on accessibility rather than mobility, and provides a logical way of handling trip chaining or trip consolidation, and the role of telecommunications or to-home services as a substitute for travel. As policy thinking broadens, the pressure for activity-based analysis will continue to grow. Unfortunately, until recently work on developing practical activity-based models has been slow, due to a combination of inertia, complexity and limited fiinding. The latter problem has eased in recent years and some of the papers at the Austin lATBR conference show the considerable recent progress that has been made. A number of operational models have taken a partial step beyond using the trip as the dependent


15

variable, by developing tour-based models. At face value, it is surprising that mode choice models were ever developed within a trip-based paradigm, since where travelling by car is an option, the choice of mode is invariably based the complete home-based tour. It is only when the historical evolution of transport planning is considered (as outlined above), that this anomaly makes any sense at all. However, in the UK at least, it is not the issue of mode choicQ per se that has resulted in pressure to shift to tour-based modelling, but rather the need to model the effects of time varying urban road pricing charges. Here it has been recognised that drivers will take into account both inbound and outbound charges in making their travel choices, and that any shift in trip timing to minimise charges has to take account of the consequences for the whole trip tour. First applied in the UK to model the effects of road pricing charges in London, it resulted in the calibration of a tour-based time-shifting sub-modelfi*oma stated preference survey (Polak et al, 1993). Attempting to model modal choice within an inappropriate trip-based paradigm is not the only example of the practitioners' perception of reality being constrained and distorted by prevailing paradigms and model structures. For many years the 'conventional wisdom' in the UK was that new roads do not generate traffic, because the traffic assignment models used in forecasting were based on fixed origin-destination matrices (even though the original model developers recognised that this was a simplification). It took a major government review (SACTRA, 1994) to demonstrate - what ordinary members of the public had been arguing for many years - that in high density areas with suppressed demand, new roads can generate significant amounts of new traffic. It is thus crucial that travel behaviour analysts find ways of codifying our accumulated knowledge about behavioural decision-making, particularly where these insights cannot as yet be fully incorporated into operational models.

Modelling Requirements As the understanding of travel behaviour has increased over time, the demands put on models have expanded considerably, in four main respects: (i) The use of more complex representations of travel behaviour as the dependent variable, reflecting the aspects of behaviour that policy is trying to influence, (ii) The requirement to forecast a wider range of consequences of changes in travel behaviour, from network congestion to noise and air pollution, and from CO2 emissions to impacts on land use patterns and local economic activity, (iii) The incorporation of a larger set of relevant independent variables, that guide and hence

16


account for the observed changes in behaviour, (iv) The inclusion in the models of a much wider range of policy variables and other stimuli of interest that might trigger changes in travel behaviour, hi addition, there has been growing interest in dynamic modelling approaches across the spectrum, from assignment modelling to long term scenario modelling. These various challenges have been met in a variety of ways, including: 'bolting on' new submodels and routines to existing model systems; adding a time series element; developing simplified strategic models, or task-specific models (e.g. parking models); or by using a hierarchy of models (e.g. the London road pricing research used a three-tier modelling approach). For some recent developments, see Wachs (1996) and some of the papers presented at this conference. However, in the wider policy and transport-planning environment, transport models have come in for considerable criticism in several countries. On the one hand, for being too simplistic in their assumptions and replication of decision making; and, on the other hand, for being too complex for non-professionals to understand and have confidence in. This double criticism has led many community groups, at least in the UK, to become very suspicious of transport models and to reject their findings. This is a very serious issue, particularly in a situation where governments are attempting to introduce restrictive policies that will only be implemented and work successfiilly with the active support of local communities. There are several strategies that could be adopted to meet this difficult challenge, in particular by: (i) Making greater use of micro simulafion modelling approaches. Here the decision rules and constraints can be set out in a more transparent way and the approach offers techniques that lend themselves to handling greater detail and complexity (such as including precise timing constraints) (ii) Involving local communities in the process of model construction, through their participation in data inputs from local surveys and by inviting them to suggest variables to be included in the models and some of the policy options to be tested. (iii) Ensuring that more of the user interfaces are graphically based, and that ftiture scenarios are presented in the form of simulated environments wherever possible. One of the main attracfions of network simulation models such as TRANSMS and PARAMICS, both to politicians and the public, has been their ability to present their outputs in a visually attractive manner. Through such means, models might become better accepted by the public and politicians. They


17

could form an integral part of local consultation processes, and so help to inform public and policy thinking, in a constructive manner.

Data Needs Data needs for analytical and modelling purposes broadly mirror those noted above in relation to modelling requirements. The demands on travel-related data have become more numerous and more complex over time, in response to our recognition of the greater complexity of the issues involved, and an increasing sophistication in the policy questions asked of modellers. For example, the dependent travel variable began as a simple vehicle flow measure, and now routinely requires data on many facets of each trip made by the respondent (purpose, mode, time of day, duration, origin/destination, parking characteristics, cost, etc). There is also growing pressure to provide multi-person, multi-day and multi-period data sets. Activity-based analysis adds to this pressure by requiring data on all the activities carried out at one location (not just the 'main' purpose of a trip), including activities within the home. Some models also require information about rejected behavioural options that were within the respondent/s choice set which can add many-fold to the data collection task. The set of relevant independent variables has also grown over time, hi addition to a wider range of individual and household-level variables, there is growing recognition of the need for information about the factors that constrain the choices which travellers can make when confronted with a policy measure or other external change. For example, this includes information on the precise characteristics of the trip (whether the traveller was carrying heavy luggage or materials, number of passengers, constraints on arrival time, etc), the prevailing background conditions (e.g. weather conditions) and on a number of objective or subjective factors that constrain choice. There is also increasing recognition of the importance of attitudinal factors in influencing travel choices, which adds a new dimension to data collection. The range of potential external factors that might influence travel behaviour has also broadened. Taking just the transportation system characteristics, there is a wide range of variables that are now thought to be affecting travel choices (e.g. timetabling, comfort, image, security) and a growing number of different modes of transport from which to choose (including hybrid buses, electric cars, new light rapid transit systems). Pricing and regulation add new variables to the analysis, as does the interest in using non-transport measures to influence travel behaviour (e.g. activity re-timing, substitution of tele-services for travel, mixed land use developments). Research has highlighted the need to take account of perceptions of attributes as well as their objective

18


measurement, if we are to understand travel choices. This has further increased respondent burden (Ampt, 1997). The pressures for increased quantity, variety and quality of data collection have been recognised and debated in conferences such as that recently held at Eibsee. At the same time, cost and practicality issues are becoming significant constraints on data collection efforts. The resolution of this problem is likely to involve greater selectivity, coupled with the use of a wider range of data collection techniques, and the increasing re-use of existing data sets. Some recent advances in travel survey methods are described in Lee-Gosselin et al (1998). For example, selectivity in data collection can be achieved by first identifying key variables through carrying out a phase of exploratory, qualitative research. With time, if accumulated knowledge were better codified, this could provide guidance appropriate to different circumstances. Developments in technology and survey methodology may also assist in the quest to obtain additional data while minimising respondent burden. For example, GPS data loggers can automatically record vehicle location and movement characteristics, only requiring respondents to add trip purpose-related information. There is also scope for further developing stated preference techniques, perhaps in closer conjunction with gaming simulation methods. At the same time, some national and local agencies are making basic travel data sets more readily available for secondary analysis.

Implications for Evaluation Relatively little effort has been devoted by travel behaviour researchers to making improvements to transport evaluation procedures, perhaps because this is seen as the economist's role, rather than something that engineers and planners should become involved in. However, as a consequence of this neglect, the sophistication of the evaluation procedures seems to lag behind our understanding of travel behaviour. In the same way as our modelling and conceptual paradigms influence how the transport analyst views and models the world, so the evaluation procedures have a strong influence not just on which schemes score highly and are implemented, but also on the types of options that are developed for assessment. For example, within the cost-benefit approach, recent increases in the value attached to human


19

life in some countries has had a strong influence on the kinds of road schemes that are currently being designed and recommended for funding. Similarly, were countries such as the UK to put a monetary value on reliability that comes anywhere close to the high weighting given to this factor in attitudinal research, then this would have a major influence on the kinds of scheme proposed average time savings would no longer be the overriding design criterion. As was noted previously, current evaluation procedures are not designed to deal with the assessment of traffic reduction policies. Although savings in travel time count as benefits, we are not certain how to handle reductions in the number of trips, or to take into account the improved quality of the travel experience in different environments. An attempt was made to get around this problem in a cost-benefit evaluation of a traffic reduction scheme in Oxford (Jarman, 1994). Here cost benefits included reduction in the need for stone cleaning of ancient buildings (due to reduced pollution levels), improved health (reduced hospital costs), increased tourism spend in the city centre (due to longer time spent in the car-fi*ee environment), and an 'amenity value' based on an imputed willingness-to-pay from a proportion of drivers using the city centre who said that they would accept increases in travel time in order to improve the city centre environment. It is in part the changing requirements of the evaluation process that are putting pressure on modellers, particularly to forecast a wider range of secondary impacts of changes in travel behaviour. However, in the UK at least, the results of conventional cost-benefit evaluation processes often lead to recommendations that are at odds with what the public and politicians regard as the preferred solutions. The adoption of a multi-criteria analysis framework might alleviate the problem to some extent, by allowing different interested parties to put their own weights on the different variables; but the problem goes deeper than this. Increasingly, because of the debate about sustainability, the primary aim of transport policy is no longer seen as being to provide for unlimited increases in vehicle traffic and movement. This has two implications. First, classic indicators such as increasing traffic volumes, higher trip rates and longer travel distances do not provide an appropriate measure of whether transport policy is succeeding: it is no longer unambiguously clear whether policies which result in more or in less of these traditional measures of travel are to be preferred. Second, transport is being viewed as offering one set of tools for achieving more basic social and economic goals relating to sustainable economic development and the enhancement of urban quality of life; relying on travel impact measures to assess schemes are only of limited relevance in this context. In the UK this latter issue is currently being addressed in the 'Civilising Cities' initiative, funded

20


by a motoring-sponsored charity and the Department of Environment, Transport and the Regions (see University of Westminster, 1998). The aim is to demonstrate how transport measures can be designed to contribute to improving many aspects of urban life beyond the normal remit of the transport professional, including health improvements and reductions in street crime. The switch to an activity-based paradigm would assist in addressing many of these issues, by offering a new set of measures against which to judge the success of transport - and other - policy initiatives. For example, travel time budgets are already used in some countries as social measures of quality of life, and the balance of time spent in-home and out-of-home could provide another measure (e.g. minimising out-of-home time on obligatory activities and maximising time spent on discretionary activities). In the future, one of the central objectives of transport policy might be turned on its head and become one of minimising the amount of travel required to achieve a given set of activities. It is also likely that attitudinal measures will play a greater role in the future evaluation of transport-related policies, based on levels of traveller/customer satisfaction both with the policies themselves and with their impacts (e.g. resulting levels of traffic noise). This would be consistent with a switch to a more market-oriented approach to transport provision, which is being encouraged by governments in several countries.

CONCLUSIONS This chapter has identified a number of key challenges for travel behaviour researchers that address major transport policy concerns that are to be found internationally, and which have implications for all aspects of methodology. One common theme to emerge is the need to look in a more holistic way at household travel patterns, whether to study the potential impacts of electric cars, tele-services or road pricing schemes. The chapter strengthens the case for activity-based analysis and modelling, and highlights the need for parallel improvements in data collection and evaluation procedures.

REFERENCES Al-Deek, H. M., A. J. Khattak and P. Thananjeyan (1998). A combined traveller behaviour and system performance model with advanced traveller information systems. Transportation Research 32A (7), pp. 479-493. Ampt, E. (1997). Respondent Burden: Understanding the People We Survey! Resource Paper, Raising the Standard, an International Conference on Transport Survey, Quality and Innovation, May, Grainau, Germany.


21

Berg, J. T., K. Kawada, M. Burris, C. Swensen, L. Smith and E. Sullivan (1999). Value Pricing Pilot. TR News, September/October, Washington. Bradshaw, R. and P. Jones (1998). TDM trends in Europe. International Association of Traffic and Safety Sciences Research, 22 (1). Brog, W and E. Erl (1996). Changing Daily Urban Mobility. ECMT Round Table 102, Paris. Cairns, S., C. Hass-Klau and P. Goodwin (1998). Traffic Impact of Highway Capacity Reductions: Assessment of the Evidence. Landor Publishing, London. Ciaburro, T., P. Jones and D. Haigh (1994). Raising public awareness as a means of influencing travel choices. Transportation Planning Systems 2 (2), pp. 5-21. DoE/DoT (1993). Reducing Transport Emissions Through Planning. Departments of the Environment and Transport, HMSO, London. Go lob, T. and J. Gould (1998). Projecting use of electric vehicles from household vehicle trials. Transportation Research 32B (7), pp. 441-454. IHT (1999). Guidelines for Planning for Public Transport Developments. The Institution of Highways and Transportation, London. Jarman, M. (1994). Valuing wider environmental benefits from an urban traffic restraint package. Transportation Planning Systems 2 (2), pp. 23-37. Jones, P. (1983). A New Approach to Understanding Travel Behaviour and Its Implications for Transportation Planning. PhD Thesis, Imperial College, London. Jones, P. (1992). Review of available evidence on public reactions to road pricing. Report to the London Transportation Unit, Department of Transport, July 1992. Jones, P. (1995). Road Pricing: the Public Viewpoint. Road Pricing: Theory, Empirical Assessment and Policy eds. B. Johansson and L.G. Mattsson. Kluwer, London. Lee-Gosselin, M., P. Bonnel and C. Raux (1998). Eds, Special Issue: Extending the Scope of Travel Surveys, Transportation 25 (2). Luk, J., N. Rosalion, R. Brindle and R. Chapman (1998). Reducing Road Demand by land Use Changes, Public Transport Improvements and TDM Measures - A Review. ARR 313, Australian Road Research Board, Melbourne. Mahmassani, H. S. (1997). Dynamics of commuter behaviour: recent research and continuing challenges. Understanding Travel Behaviour in an Era of Change, eds P. Stopher and M. Lee-Gosselin, Pergamon, Oxford. Mogridge, M. J. H. (1990). Travel in Towns: Jam Yesterday, Jam Today and Jam Tomorrow? Macmillan Press, London. Mokhtarian, P. L. (1997). The transportation impacts of telecommuting: recent empirical findings. Understanding Travel Behaviour in an Era of Change, eds P. Stopher and M. Lee-Gosselin, Pergamon, Oxford. NERA (1997). Motors or Modems? National Economic Research Associates, for the RAC, London. Noble, B. and S. Potter (1998). Travel patterns and journey purposes. Transport Trends, Government Statistical Service, London. Polak, J.,P. Jones, P. Vythoulkas, R. Sheldon and D. Wofinden (1993). Travellers Choice of Time of Travel Under Road Pricing. Report to the UK Department of Transport. Rye, T. (1997). Implementing Workplace Transport Demand Management in Large Organisations. PhD Thesis, Nottingham Trent University, England. SACTRA (1994). Trunk Roads and the Generation of Traffic. Standing Advisory Committee on Trunk Road Assessment, HMSO, London. Transport 2000 (1998). The Case for Quality Public Transport, London.

22


University of Westminster (1998). Civilising Cities: the Contribution of Transport and Land Use. Phase One Report: Study Definition and Work Programme. RAC Foundation for Motoring and the Environment, London. Walker, H. (1997). Mixed Use Development as an Agent of Sustainability. Reclaiming the City Mixed Use Development, ed. A. Coupland. E & FN Spon, London. Wachs, M. (1998). Ed. Special Issue: A New Generation of Travel Demand Models. Transportation 23 (3).


LIVING MODELS FOR CONTINUOUS PLANNING

Andrew Daly

INTRODUCTION Travel demand modelling systems often represent a substantial investment in money and time by a planning authority, construction agency or transport operator. While modem techniques of data collection, model development and implementation can help in reducing the costs, it remains inevitable that the modelling of a large and complicated transport system in any detail requires a large and correspondingly complicated model. Once a travel demand model system is in existence, therefore, the client organisation (government, constructor or operator) will wish to retain the model in operation to investigate the planning issues that may arise over the subsequent months and years. Often, the construction of a model system will be motivated by the needs of a particular planning project; for example, in The Netherlands, the National Model (LMS) was set up to meet the needs of the Second Transport Structure Plan. Even in these cases, however, the model system can have a substantial value after the completion of the specific project for which it was developed and in appropriate circumstances much more value can be extracted from the initial investment than was obtained by the specific project in question; the LMS has now been in continuous development and use for other studies for 11 years, both before and after the completion of the Structure Plan for which it was developed, and both application and development are expected to continue for some years yet. In this context, it is natural to consider how to maximise the possibilities for use of a model system after completion of the project for which it was developed. What are the characteristics

24

In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges

of a model that make it re-usable in this way? What can be done to extend the capabilities of a model for future use? Can a model be adapted to future circumstances that are different from those for which it was developed? In the limit, when is it possible to use an existing model and when is it necessary to develop a new one? In summary, to what extent can existing models be used to evaluate policy different from that incorporated in the model design? The objective of this chapter is to describe demand model structures and extension procedures that tend to extend the period and scope of applicability of models. The models in question are those applicable to urban, regional and national planning issues but also to corridor studies such as those concerned with major infrastructure. Examples are primarily drawn from the author's experience and focus particularly on the modelling of passenger transport, although similar techniques can be used for freight modelling. The issue of the interaction of analysis and planning over an extended period is considered by Manheim (1979). He describes a cyclical process in which modelling interacts with ongoing planning and evaluation of policy; the point of view is that of designing the entire transport planning system. In this chapter, the narrower standpoint of the capabilities of the model is considered. The novel features of the work described in this chapter are those for retaining modelling assets from one cycle of the planning process to the next and the ways in which these possibilities can be enhanced. The chapter does not focus on technical development but on practical possibilities for improving the applicability of models. The approximate maximum likelihood method presented is believed to be novel, but the remaining procedures have been used in previous studies, although perhaps not in the context in which they are presented here. Throughout, the emphasis is on practical methods - almost all of the methods presented have been used in practice - and effective approximations are often suggested rather than theoretically exact procedures where the latter are not known, are excessively complicated or require difficult programming or processing.

THE PROBLEM It is assumed for the purposes of exposition that a travel demand model is to be developed for a client by a development team to meet an explicit series of policy objectives and that a series of

Living Models for Continuous Planning

25

applications of the model is then to be made to support decision-making with respect to those objectives. Subsequent to the initial applications, it will not be a great problem to apply the model to investigate policies that vary to a moderate extent from those considered in the model development. However, the applicability of the model for more widely varying objectives may be called into question for several reasons. •

•

•

•

Policy changes that the client wishes to investigate but that it had not specified prior to model development can arise for a wide range of reasons: because of proposals put forward by pressure groups, changes in the balance of financial and environmental objectives or simply that a particular construction project has advanced in its life cycle and a shift is required from the yes/no decision to focus on detailed questions of project design and pricing. There may even have been a lack of foresight on the part of the client or the development team or it may not have been possible for time or resource reasons to complete the original model development. These are merely examples from a long list of circumstances in which new policy issues can arise beyond those envisaged when the model was developed. In addition to policy changes by the client or its critics or competitors, there may of course be important changes in the exogenous variables that are relevant to the market in which the client is operating. For example, ftiel prices may change or taxation regulations may make certain types of trip (e.g. international shopping expeditions) more or less attractive. Market changes may occur, in particular the appearance of new competitors in a market; these may well be brought about because of technological innovations, such as the appearance of fast ferries in a sea corridor, or of altemative-ftiel vehicles in the car market. Alternatively, level-of-service changes may occur or be considered, e.g. price changes because of a price war or government regulation, that suggest that application of the model would take it outside its range of applicability. Time passes, as ever, bringing about changes in the size and nature of the transport market and initiating the suggestion that the model is out of date. Social norms may change, bringing about changes in attitudes with respect to noisy vehicles (e.g. fast ferries) or those using cleaner fiiels. Technology advances within the modelling environment, so that it may be advantageous to adapt the model to take account of improvements in computer hardware or software or to take advantage of advances in modelling knowledge that were not available when the original development took place.

In any of these cases the modeller may well be faced with the question of the applicability of the (expensive) model system in changed circumstances. The objective of this chaper is to

26


describe a range of techniques tliat the modeller can deploy to maximise the possibility that a model system can be applied with reasonable reliability in changed circumstances. A key component of the modeller's armoury in such circumstances is the collection of new data. The next section of this chapter considers the possibilities for new data collection: the use of Stated Response data, the application of aggregate data (e.g. counts) and the use of Revealed Preference data, particularly from cheaper on-board surveys. Essentially, these techniques are used to enrich the original data base from which the model was developed. Examples are given of the application of each of these techniques. The fourth section is concerned with the ways in which policy can be formulated to facilitate model application. The objective is to present policy in terms to which the model can respond, while ensuring its proper interpretation, whether that policy is that of the original client or of another agency, such as a pressure group. A number of techniques for policy formulation are described and comments given on their success in meeting various criteria. Finally, the main conclusions of the chapter are brought together.

DATA ENRICHMENT The issue considered in this section is that of a model system that has been developed from a given data base, which may be extended to allow the model to be operated in changed circumstances. Three types of data commonly considered for this purpose are discussed: Stated Response data, aggregate data such as counts and en route or intercept surveys. Home interview surveys, e.g. with trip diaries, are less commonly used for this purpose because they are expensive and because they are difficult to focus onto specific types of behaviour. This process can be considered to be that of data enrichment: the original data used to develop the model is enriched with subsequent information designed to focus on issues that have become interesting because of subsequent developments.

Stated Responses In this context, the term Stated Response (SR) is used for a range of interview techniques including stated choice, preference rankings and contingent valuations (e.g. 'transfer price').


27

Many of these techniques are often classified as 'Stated Preference' (SP), a term we shall reserve for stated choice data. These techniques have in common that they place an interview respondent in a hypothetical situation closely related to the actual choice situation and ask for his or her preferences between real and hypothetical alternative journeys - possibly including the actual journey - designed to maximise the information obtained. A close relationship between the hypothetical situations put to the respondent and journeys that have been made is essential to maintaining credibility for the respondents and thus obtaining reliable results from an SR exercise. In contexts where developments are anticipated but have not yet taken place, or are proposed but not yet accepted in the political process, SR is the natural way to investigate likely changes in demand that will or would follow. The new features of the market can be explained as clearly as possible to respondents and their stated responses obtained. However, because SR gives the possibility of obtaining multiple responses from each individual, making it a relatively cheap procedure, it is also often used when developments have already taken place. Applications of SR. Applications of SR procedures for model updating and extension can be made to deal with at least five different types of problem. • The analysis of new policy that had not been considered in the initial development of the model and which most likely is not in operation anywhere in the study area. Here the objective of the SR experiments is to attempt to discover the responses of travellers to changes in variables that had previously remained constant or which had not been incorporated in the model at all. For example, road pricing is not incorporated as an explicit policy option in many of the model systems in use for major conurbations. The introduction of payment on specific links of the highway network will cause changes in route choice, mode choice etc. but the extent of these changes cannot be determined from actual behaviour when no such system is in operation. In this context, an SR experiment can be used to investigate the extent of behavioural response to the cost variable that was previously present only in the form of the cost of ftiel etc. for driving. • Similarly, an SR experiment can be used to throw light on the influence on behaviour of changes in variables that take them outside the range within which the model was estimated, to values that may be considered beyond the range of reliability of the model. This situation will often arise when very low prices are being considered, whether because of government policy (as in the case of the new student tickets in The Netherlands, allowing students to travel at zero cost at some times) or because of a

28

In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges price war (as in the case of day trip travel across the English Channel after the opening of the Channel Tunnel).

•

Another context in which an SR experiment can be useful is when an aspect of behaviour has to be added to the model that was not included in its initial specification. For example, an analysis of time-varying road pricing may make it necessary to introduce into the model a description of car drivers switching times to avoid having to pay the road pricing charge. In the Netherlands National Model, not only was it necessary to introduce this aspect of behaviour, but also to extend the model ftirther to allow consideration of the purchase of monthly passes, allowing free use of the highways for which there was otherwise a charge. Two separate SP exercises were used to set up models of these two aspects of behaviour.

•

The introduction of new alternatives into a transport market will naturally require a travel demand model to be adapted to accommodate them. When the new alternatives have not yet appeared in the market, SR may be the only approach to modelling demand for them. Even when the new alternatives are in operation, SR may present a useftal means of modelling the demand for them, although in this case it would be preferable also to utilise Revealed Preference (RP) information on the total use (i.e. counts) of the new alternatives, as described later in this section. In several important studies of major infrastructure, for example in studies of the Channel Tunnel or of Scandinavian fixed links, as well as in urban studies of light rail systems, the forecast of demand for the new alternatives is the main focus of the study. In other cases, such as forecasting for fixed links when new ferry services enter the market, such as the fast catamarans that are now being used on a number of European sea crossings, the new services are incidental to the main interest. In either case, the same SR procedures can be used to estimate the demand. Finally, SR data can be used to update a model when it is believed that a number of its parameters, e.g. those determining the trade-off between time and cost, have become out of date. While this process does not change the specification of the model in terms of the alternatives and variables represented, the updating of the parameters can take account of increasing incomes and changes in preferences arising for other reasons. Practical studies of this type are less common.

•

Thus in a wide range of contexts, the applicability of a model can be substantially enhanced by the collection of SR data, provided this data can be integrated into the existing model structure. Integration of SR-Based Modelling. When an SR survey has been conducted for one of the reasons described above, its integration into the existing model system will require a number of adjustments of various kinds to be made to that model and possibly also to the SR data itself


29

One approach is to re-estimate the model in its entirety, adding the new SR data to the existing data bases used for estimating the current model. When new alternatives are being modelled, this approach will typically require a three-stage estimation procedure as set out by Daly and Rohr (1996) in which a model is estimated from merged data sets (often using the method of Bradley and Daly, 1991, for combining SP with RP data), then corrected in two ftirther estimation stages which correct the alternative-specific constants in the model: a) to accommodate the best available information on the current shares of the existing alternatives; and b) to correct the SP estimates of the shares for the new alternatives, given the corrections made in stage (a). The two supplementary estimation stages (a) and (b) are referred to as the 'A' and 'B' runs respectively. This approach has been used in several practical studies. It may be necessary to follow the ftill 'A' and 'B' run procedure even in cases when new alternatives are not being introduced into the model. For example, if new variables are being entered into the model, or if the formulation of a variable is being changed, for example to model an extended range of that variable, it will be necessary to adjust the alternative-specific constants in the model and both 'A' and 'B' runs may be necessary. An important issue when introducing new alternatives into the model is how these should be structured relative to the existing alternatives. For example, when the model is of the common hierarchical logit form, the question arises as the issue of where in the 'tree' structure the new alternatives should be introduced. When this problem can be foreseen, the SR survey can be extended to investigate choices involving existing alternatives as well as new alternatives to give insight into the relative cross-elasticities of the choices between existing and new alternatives to the choices between existing alternatives. To give a concrete example, suppose interest is in the switching of time of travel in response to time-specific road pricing. An SP survey can give insight into the importance of the road pricing charge relative to congestion differences and other preferences. However, if the survey is restricted to car driving alternatives only, it will not be possible to give any information about the relative magnitude of switching between times of travel and, say, mode switching. However, by introducing a mode choice possibility into the SP this problem can be solved. This approach has been used to estimate complicated model structures with several hierarchical levels.

30


An important consideration in analyses combining SR data with an RP data base is to obtain a correct overall model. It is inevitable that SR data will contain a number of biases relative to RP data. It is important to remember that the RP data is the best - perhaps the only - view of the current market situation, so that the basic description of the market must be related to that given by the RP data. This means that the overall market shares of each alternative in the model must be related to the description given by the most recent RP data. Further, the impact of changes in explanatory variables must be put on a scale that is consistent with the overall scale against which behaviour is measured. Most often, this means that the RP data must be used to define the overall scale of the model as well as the base market shares. Although exceptions could be found, most often the RP data should be considered as the best overall description of reality and SR data should be seen as very approximate in describing the overall scale but usefiil for giving insight into specific aspects of behaviour. A ftirther issue that arises in mixing data sets is that of consistency across time. Most simply, when costs are represented in different data sets attention will be needed to the issue of inflation. However, particularly in the case of new travel alternatives attitudes may change over time as awareness increases. The model may well have to be adjusted to take account of shifts of this nature only, as has already been mentioned, but even when there is some other prime motivation, changes in attitudes may well remain important. Examples of this type of change have been observed with respect to fast ferries in sea crossing markets. Finally, it is necessary to draw attention to the repeated measures problem: that most (if not all) SR surveys collect multiple responses from each individual respondent. This means that the data derived from such surveys cannot be analysed using naive methods. Some progress has been made in developing reasonably simple methods for handling this problem (e.g. Cirillo et al., 1996, Ouwersloot and Rietveld, 1996; see also McFadden, 1996). However, these methods have been applied only to single data sets containing multiple responses per individual; the necessary extensions to deal with combining such data sets with other data have not yet been made. Implementation Issues. A final set of issues arising in the integration of SR data into a forecasting model concern the context in which the extension of the model is to be set. These issues primarily concern the necessary simplification of the choice context for SR interviewing. For example, a typical form of SR interview for obtaining information about a new alternative is to ask respondents to take part in a binary SP experiment in which they compare their current alternative with a new alternative. To use this data in a model in which there are more than two alternatives - some old, some new - requires adaptation of the choice model. Simply


31

stated, the probabilities calculated in a binary experiment are not appropriate for predicting choice among many alternatives. Similarly, it may be necessary in an SR experiment to omit some of the detail of even a binary choice. For example, the interview context may limit the interview time and hence the explanation of the choice that can be given. For long and complicated journeys involving several changes of mode it may not be possible to specify full detail of the access modes and the respondent must be assumed to have a reasonable idea how he or she would access a known airport, for example. However, when choice is finally being forecast, the details of the access modes will normally require to be included, perhaps to ensure consistency with RP modelling. This change will require a further adjustment to the model. Thus while SR surveys can form a useful basis for extending a model to take account of new aspects of the choice situation, a series of issues arise in integrating the data both in modelling behaviour and in setting up a forecasting model. These issues can be overcome, but care is required in ensuring that the modelling is consistent at all stages.

Aggregate Data An important an cheap source of data for updating and adjusting models is aggregate data. In particular, count data relating to the changing market shares of the main alternatives is an effective source for keeping limited aspects of the model up to date. Some data of this type can even be collected automatically, e.g. from traffic counters or from ticket sales records. In the latter case, there may well be other information attached to the record, e.g. the ticket type, which can be of further importance in giving information about the segmentation of the market. The main information content of count data of this type is to indicate the market shares for each alternative and the total market size. Adjusting a model so that it gives overall market shares equal to those observed is usually a matter of adjusting the alternative-specific constants. Any convenient process may be used for the adjustment, often trial and error is as effective as any other. However the adjustment is done, alternative-specific constants that give correct aggregate shares satisfy conditions closely related to the first-order conditions for maximum likelihood^ so that the degree of sophistication of the calculation process should not be a concern.

32


Aggregate observation of the total traffic in the market will obviously be a suitable data source for adjusting a model of growth or generation. Indeed, identifying separately the change in demand due to exogenous growth (e.g. from an increase in average incomes) and that due to endogenous generation (e.g. due to improved levels of service) can present difficulties when several variables change simultaneously. Observation of a series of demand levels over an period of time can help to identify these effects separately and therefore allow the analyst to adjust each model appropriately. Another application of a series of observations over time is to study the development of the market for a new alternative. By postulating simple theories about changes in market shares and testing these against the observations, understanding of the dynamic development of the market can be obtained and forecasts developed for the likely levels of demand to be achieved in the future. Particularly in the case of commercial investment decisions, in which the initial rate of return is of importance, understanding the rate of development of the market is of considerable importance. Examples of this type of application are the use that has been made of tabular data from the national travel survey for the updating and regional calibration of national and regional models in The Netherlands. Similarly, use is made of count data to adjust models of ferry choice in sea crossing corridors; in this context, counts of both passengers and of cars can be used to ensure that car occupancy is correct in each corridor.

En Route Surveys A third type of data that is very useful in understanding changes in behaviour over the period since the initial development of the model is RP data collected by interviews with travellers. Particularly when the analyst is working for the operators of a new transport alternative, it can be cheap and convenient to collect data from the users of that alternative. When data from all the alternatives is available, it is a relatively simple matter to update the model. When data is available from one alternative only, it would seem that only a general picture of the market for that alternative can be obtained. But even if it is sampled from one alternative only, such data can be used to make a reestimation of the entire model, as illustrated by the following example of modelling for a sea crossing. The example exploits the binary structure that can be introduced by considering the


33

alternative for which en route data is available against the set of other alternatives. More complicated analysis could be used to deal with other cases and an exact maximum likelihood estimation is also possible, although this is much more complicated. Suppose the existing model works through a sample enumeration procedure based on OD records, with expansion factors, drawn from a previous survey, which perhaps was used for the original model estimation. This model is considered as a binary logit model giving the probability p^ of choosing the new alternative, say a fixed link in competition with ferry systems, chosen with probability pf by log (Pt / Pf)

=

Pi . AV + p2'

where AV = ( Vt - Vf)

as a function of the measured utilities V^ and Vf of the fixed link and ferry systems and parameters P which are to be estimated. These Vs could be defined by an existing model, of which this model is then a generalisation; for the original model p = 1 .^ Let us also assume that we have used the information about total current market shares to amend the model (i.e. the Vs) so that P2 = 0 is consistent with that information when Pi = 1. If we could observe choices for both fixed link and ferry system, we could estimate the binary model, deriving estimates of the ps that might be different from 1 and 0. The new estimate of Pi would amend the elasticities, giving an improvement to the model. (A new estimate of P2 would also be determined to keep the total market shares right.) One way of estimating these values would be by maximum likelihood, in which case the first-order conditions for optimality are Sj. Wj.. ptj.. AVj.

=

Ej. ^r • ^tr • ^ ^ r

SrWf.ptr

=

SrWi-.Str

(^^r Pi)

and (for P2)

where Wj. is the expansion factor for each record r of the data base giving the total market and h\x is the hypothetical observed choice, taking the value 1 if r chooses the fixed link and 0 otherwise. The data needed to calculate the left sides of these equations can be derived from the existing model, with an appropriate growth assumption applied to adjust the w's up to the year of the en

34


route survey. Estimates of the right sides can be derived from the observations of fixed link use and their utilityfiinctionsin the existing model, which would give us Sg Wg . AVg and

£§ w^

for these observed fixed link records s and since 5^^ is uniformly 1 for those who choose the fixed link they are estimates of the numbers required. The procedure above can readily be generalised to estimate multiple P's, e.g. to update time and cost elasticities independently. It requires only a comparatively minor amendment to standard estimation software, but can only be applied in this simple form to linear logit models, i.e. not to structured logit or other more complicated forms. This is because it relies on the 'sufficient statistics' of linear logit, which do not exist for more complicated models. Much more complicated analysis procedures are then needed for models of those types.

Other Approaches Other approaches to model updating can be considered that involve the collection of no data, or very little data, in the specific market of interest. These approaches chiefly involve model transfers, which can be made on the basis of using a small amount of local data to 'calibrate' a model developed elsewhere, or simply applying, by analogy, model results or observed results from other locations. A limited literature exists describing techniques for model transfer, which attracted research interest around 1980. Many of these are similar to those for model updating ('temporal transfer') described above, see Daly et al. (1983) which references other publications. In this work the idea is to use limited local information to exploit a detailed model developed elsewhere; a hierarchy of methods is presented, from which a choice can be made depending on the extent of the local data. Analogy procedures can be very rough-and-ready. An evaluation such as 'a good information system would be worth two minutes of travel time' gives an opportunity to evaluate the impact of information improvements, but this approach depends solely on the quality of the assumption on which it is based.


35

Analogy and transfer approaches can be useful in special circumstances, particularly when time is short, but usually a more acceptable method, based on more substantial local data collection, can be justified.

POLICY FORMULATION The issue addressed in this section is how policy can be formulated for analysis by a model, particularly when the model has not been developed to deal with policy of that type. It is convenient to discuss this issue by considering as an example a specific model system through which many types of policy have been assessed. The Netherlands National Model, the LMS, was initially used to support decision-making concerning the overall framework of national policy, which was then set out in the Second Transport Structure Plan. Subsequently, more detailed plans were needed for a number of specific aspects of policy, such as for policy implementation by regional authorities, for the railways and for the management of existing transport infrastructure. In particular, the detailed implementation of road user charging systems, such as 'Road Pricing', was of interest. The example of the LMS is interesting because in its relatively long life (12 years of application) a number of different issues have arisen, leading to different approaches to extending the model. It is useful to note that the original objective of the model was to help in determining alignments for the strategic road and railway networks of the country; a number of such studies have been made but these are of limited interest for this chapter because they require no specific extensions to the capability of the model. It is important to be reasonable in attempting to extend the scope of application of a model. There is a point at which fiirther extension is not realistic and it is better for some groups of applications to develop an independent model system. This situation has arisen with respect to the LMS in two applications areas. First, the application for regional policy, in particular the evaluation of specific projects, has been tackled by setting up a series of regional models for different parts of The Netherlands. The models set up - collectively described as New Regional Models (NRM) - have a common base in their description of traveller behaviour; this base is taken from the national LMS (see Gommers and Pommer, 1995). However, the local inputs of transport networks, zonal data, etc. can be much more detailed than would be possible in the LMS. Moreover, the technical

36


separation of the models also makes possible an institutional separation that fits with the organisation of responsibilities between centre and regions in The Netherlands. Second, while the LMS has been used for several applications for Dutch Railways (NS) and was extended to meet NS' requirements, ultimately a considerable extension in the level of detail was required to deal with train-specific issues. In particular, the combinations of choice of access and 'egress' modes, the choice of stations and the choice of train service (when several operate in parallel) required a degree of extension that was not reasonable within the LMS. Accordingly, the 'ProMiSe' model was developed (Cohn, et al., 1996) to incorporate these features. However, like the regional models, ProMiSe makes extensive use of LMS features, in this case for describing travel modes other than the train; unlike NRM, it uses the same zonal system as LMS, facilitating data exchange. In future, ProMiSe and LMS may become even more consistent. Apart from these two important examples, it has generally proven possible to extend the LMS to deal with the policy issues that were required. An important example was the assessment of time-dependent road user charging, some variants of which can be called 'Road Pricing'. It was intended that the policy would suppress car driver trips and in particular peak hour car driver trips, and would therefore have a positive impact on emissions and congestion, as well as on the transport budget. The transfer of trips from the peak to off-peak helps in reducing congestion, but, because peak-hour travel becomes more attractive for certain classes of traveller (broadly, those with a high 'value of time'), the impact on emissions is less than could be expected. A forecast was therefore required of transfer between time periods. Time switching behaviour was obtained from an SP survey, conducted in two parts to look at the separate impacts of road pricing and congestion on drivers' choice of time of travel. Analysis of all the SP survey material produced a single model of choice of time of day, as a function of cost and congestion differences. This model was then integrated into the LMS at an appropriate point in the existing model structure. The integration required a number of adjustments to the existing model, e.g. to take account of differing congestion at different times of day and to deal efficiently with the interaction of demand and network models that leads to equilibrium congestion (see Daly et al., 1990). In application the extended model gave usefiil results, showing that the equilibrated impact of road pricing in reducing peak-period traffic was approximately half what might have been predicted with a na'iVe approach. Subsequently, a further SP survey studied the likely take-up of 'passes' which might be offered at a monthly price to allow unlimited use of the tolled highway system. This model was also


37

integrated into the LMS in a similar way (although at a quite different point in the structure) to the time-of-day model. A quite different approach was adopted for the assessment of a policy of physical restraint on parking. In order to assess the impact of this policy an iterative procedure was adopted in which the first-round numbers of car driver trips arriving in each zone were compared with the available capacity, which in some zones was quite drastically reduced by the policy. For overcapacity zones the car driver and car passenger arrivals were then diverted to neighbouring zones (where these were not also affected by capacity limits) or to alternative modes. This procedure was found to give a good assessment of the impact of this important policy instrument, which could change the mode split substantially in city centres although it was of little use in outlying areas. Another approach was used to assess the impact of 'telematics' (teleworking, teleshopping, etc.). The overall impact of these developments had been assessed in an earlier study and the requirement for LMS was to attribute these overall impacts to changes in mode split, trip lengths etc.. This was achieved through calibrating additional terms in the utility functions to achieve the required overall effect, using the model only to distribute the impacts across modes, regions and segments of the population. Anticipated changes' in working practices, such as increased part-time working, whether by reducing hours per day or reducing days per week, were also modelled. Here the approach was to compare people in the base year who had different working practices and apply the different travel patterns of part-time workers etc. more widely in the future. Many of these changes affected the travel frequency for work and other purposes. A much more detailed study was made to assess the impact of traffic management measures aimed at maximising the use of existing highway capacity (Bakker et al, 1995). Here, a series of measures had been put forward, such as ramp metering, flow homogenisation, reserved lanes for specific vehicle groups and various kinds of provision of information to drivers. The aim of the analysis was to assess the impact of these measures on congestion at a national level and to predict changes in traffic flows between classes of roads. There was also an interest in any mode switching effect that the measures might cause. Each of the measures was discussed in detail with traffic engineers familiar with the results of the pilot projects that had been conducted for most of the proposed measures. In each case, the impact was translated into terms that the model could accept, such as changes in capacity or speed on network links. To obtain a full assessment of the impacts, it was necessary to improve the assignment algorithm

38


to give a better interpretation of the time lost in standing traffic, a common feature of the network in peak conditions and the impact this could have on crossing streams. The results, given in more detail in the paper referenced, indicate the relative efficiencies of each of the measures and the overall impact of a 'package' comprising a carefully chosen mix of several of them. The mode split and wider consequences of these policies were minimal. Impacts that were easier to incorporate in the model were those concerning fuel price and efficiency. A policy was set out providing for an increase in fuel prices in real terms. The cost impact of this policy on traveller behaviour could be translated directly into the model, which contained a cost variable. However, there is a trend towards the use of more fuel-efficient vehicles and it is expected that this trend will continue, partly under the influence of the policy that has been adopted on fliel pricing. Thus the cost impact of fuel price changes, with its impact on mode split, trip lengths etc., has to be reduced to account for increasing efficiency, although this in turn has a welcome effect on emissions, which are not handled directly in the LMS itself Many other policy measures were also modelled. Often, the interpretation was a case of finding some 'common currency' in terms of which a policy could be presented in the model. For example, in developing the complete policy package to be recommended in the Second Transport Structure Plan, it was necessary to take account of a number of minor but relevant elements such as the following: • the impact of improved information systems for public transport, chiefly a free telephone line with complete national information on all public transport services; the system itself has been evaluated (Kroes et al., 1994) but a simple measure was required of its effect on overall travel patterns and in particular on mode split; • the impact of improved priorities for cycling and walking, in particular by changed setting at traffic lights; • the impact of changed accessibility of public transport systems through increased residential densities and specific zoning systems for residential and commercial development (part of the well-known ABC policy). The impact of each of these measures was expected to be small and it was not required that the LMS make a specific evaluation of them. However, it was necessary that they should be included in an overall assessment of the impact of transport and planning policy on traffic levels. In each of these cases, an estimate was made of the impact of the proposed policy in terms of the components of travel time of the respective mode. The impact of each of these policies was generally judged to be appropriately estimated. While this procedure would not


39

be adequate for the evaluation of the specific policy, it was accepted as appropriate for the assessment of these policies as 'background' to the more important policy that was the main subject of the assessment. Many other policy measures have been evaluated. However, those presented here give an idea of the range of policy and the methods by which it has been interpreted for implementation in the model. For other model systems, the specific policy to be evaluated will be different in detail. However, the range of possibilities evaluated with the LMS indicates that, with imagination, a model can be adapted to cope with a wide range of possibilities. The applications of the LMS in these ways have generally retained the confidence of the professionals working with the model, both in the development team and in the client organisation, and the adaptations have thus enhanced the value of the model enormously. Other model systems can generally also be adapted in ways that are similar in principle, although different in detail, providing the design of the model does not impose too many restrictions.

CONCLUSIONS Many possibilities exist to preserve the investment that has been made in setting up a modelling system. By exploiting these possibilities the analyst can extend the effective life of a model as well as increasing its value substantially. The need for extension of the model can arise from a need to consider new policy, changes in the market or changes in the composition of the travelling population. Of course there are limitations in the extensions that are usefully made: when much more detail is required in some parts of the model, while other parts can be simplified, an independent model may well be more appropriate. An important means by which models can be extended is by the collection of new data, which is essentially an enrichment of the original data from which the model was created. Particularly useful types of data in this context are: • stated responses, which can give insight into many aspects of behaviour, including those not yet available in the market place, but require some care in both analysis and implementation to ensure that proper attention is paid to the respective roles of SR and RP data;

40


•

aggregate data, usually counts, which can be used effectively to ensure that the model predicts the correct total market volume and that the shares of each of the alternatives is correct; • en route RP surveys, most conveniently collected in collaboration with transport operators, which give some detail about the traveller and the development of the market. An approximate maximum likelihood method was presented for the analysis of this type of data that allows the complete re-estimation of a model on the basis of new data collected for one alternative only. The formulation of policy for implementation in the model was considered with reference to examples of implementations of the Netherlands National Model. A range of different methods has been used to interpret the different policies. These methods illustrate the range of methods that can be used to apply models to test the application of policy, when that policy cannot be expressed exactly in terms of the variables explicit in the model.

ACKNOWLEDGEMENT I am grateful for the comments on an early draft by an anonymous reviewer.

REFERENCES Bakker, D. M., P. H. Mijjer, A. J. Daly and F. Hofman (1995 ). "Prediction and Evaluation of the Effects of Traffic Management Measures on Congestion and Vehicle Queues", presented to Seventh World Conference on Transport Research, Sydney. Bradley, M. A. and A. J. Daly (1991 ). "Estimation of Logit Choice Models using Mixed Stated Preference and Revealed Preference Information", presented to 6th. International Conference on Travel Behaviour, Quebec. Cirillo, C, A. J. Daly and K. R. Lindveld (1996). "Eliminating Bias due to the Repeated Measurements Problem in SP Data", presented to PTRC European Transport Forum. Cohn, N. D., A. J. Daly, C. L. Rohr, W. Oosterwijk, T. van de Star and A. Dam (1996). "ProMiSe: Policy-Sensitive Rail Passenger Forecasting for the Netherlands Railways", presented to PTRC European Transport Forum. Daly, A. J., H. F. Gunn, G. J. Hungerink, E. P. Kroes and P. H. Mijjer (1990). "Peak-Period Proportions in Large-Scale Modelling", presented to PTRC Summer Annual Meeting. Daly, A. J. and C. L. Rohr (1996 ). "Forecasting Demand for New Travel Alternatives", presented to Conference; Theoretical Foundation for Travel Choice Modelling, Stockholm.


41

Daly, A. J., H. F. Gunn, P. Barkey and H. D. P. Pol (1983)"Model transfer using data from several sources", PTRC Summer Annual Meeting. Gommers, M. A. and J. F. Pommer (1995 ). "The Dutch Regional Model System: Applications and Development", presented to Seventh World Conference on Transport Research, Sydney. Kroes, E. P., H. van der Loop and H. F. Hofker (1994 ). "A new service for travel information about public transport in The Netherlands: initial effect and analysis of the market", presented to PTRC European Transport Forum, M. L. Manheim (1979). Fundamentals of Transportation Systems Analysis, Vol. 1: Basic Concepts, MIT Press. McFadden, D. L. (1996). "[..]", presented to Conference; Theoretical Foundation for Travel Choice Modelling, Stockholm. Ouwersloot, H. and P. Rietveld (1996). "Stated Choice Experiments with Repeated Observations", J. Transport Economics and Policy, pp. 203-212. Exactly the same as maximum likelihood conditions only when the model is a simple logit model. ^ The V for the ferry system could represent a 'logsum' variable from a model of choice of ferry route or could even include other modes, such as air travel.



HOUSEHOLD ADAPTATIONS TO NEW PERSONAL TRANSPORT OPTIONS: CONSTRAINTS AND OPPORTUNITIES IN HOUSEHOLD ACTIVITY SPACES

Kenneth S Kurani and Thomas S. Turrentine

ABSTRACT We investigate the potential use of small, electric vehicles to substitute for a portion of household travel demands, with the goal of reducing car size and emissions for many trips, thus reducing parking needs, road sizes, and air and noise pollution. We investigate with households whether new types of small, cleaner vehicles are convenient, practical, and attractive given their current travel patterns and future lifestyle goals, and explore what household lifestyles and public policies make it sensible to invest in such a specialized vehicles. In particular, we work with households to discover a likely sphere of their activities around homes and transit stations which fits the capabilities of small electric vehicles.

INTRODUCTION Automobiles have come to dominate personal travel in many countries and threaten to dominate in many more. For years in the U.S., transportation, energy, and land use policies encouraged households to use automobiles to expand both the psychological and geographical space in which they travel on a day-to-day basis. At the same time, walking and cycling declined, and transit failed to compete with cars to provide access to this wider and more complex personal space. Unfortunately, this car dominance has not come without a price, and autos have many costs including injuries and deaths from accidents, diminished health and

44


deaths from emissions, noise, inefficient land use, and absolute reductions in the mobility of persons without access to cars. Efforts are underway to reverse this trend toward unimodal household travel, to restore walking scale neighborhoods, encourage bike riding, and improve transit. However, the limited spatial and temporal reach, as well as cargo capabilities, of walking, biking and transit mean they can only make limited impacts on auto use. We investigate here another approach to reducing the negative impacts of automobility—^the use of small, electric vehicles to substitute for a portion of household travel demands, including improving access to transit, with the goals of reducing car size and emissions for many trips, thus reducing parking needs, road sizes, and air and noise pollution. Conventional motor vehicles of today are capable of carrying four or more people, accelerating quickly to 60 mph, and cruising comfortably at 75 mph. This combination of attributes is desirable for some trips, but necessary for few. As long as all vehicles are expected to serve all trips, large powerful vehicles will be preferred. But this all-around capability comes at a cost, not only in terms of the direct costs of vehicles, fuels, and road space, but also external environmental costs and the indirect costs of maintaining an autocentric transportation system. We perceive though that multiple vehicle ownership by households allows an increasing number of households the flexibility to specialize their vehicles. Almost 40% of U.S. households own 2 vehicles and an additional 20% own 3 or more (U.S. Federal Highway Administration, 1990). Moreover, for most trips and households, large, full-powered vehicles are not necessary. Approximately half of all trips are less than 5 miles in length. Further, they are made by a person traveling alone at a relatively low average speed (EPA, 1992). We investigate whether households view new types of small, cleaner vehicles as convenient, practical, and attractive given their propensity to own multiple vehicles, current travel patterns and future lifestyle goals, and explore what household lifestyles and public policies facilitate sensible investment in such a specialized vehicles. In particular, we work with households to discover a likely sphere of their activities which fit the capabilities of small electric vehicles (EVs). These goals indicated the use of intensive data and interactive gaming techniques. We've worked with a variety of interview techniques based upon what Lee-Gosselin (1996) calls Interactive Stated Response (ISR) approaches in several sets of experiments, including PIREG (Turrentine, et al 1992; Kurani, et al 1994), neighborhood electric vehicle (NEV) users in a demonstration project (Kurani, et al 1995), and most recently with station car users in a demonstration project. In PIREG, the gaming interviews were conducted in hypothetical situations (i.e., the households did not actually drive an EV). In the NEV and station car

Household Adaptations to New Personal Transport Options

45

studies, participants used a small electric vehicle in demonstration test periods varying from one week to three months.

AN ACTIVITY ANALYSIS FRAMEWORK The work over the past thirty years on household activity analysis provides us with concepts and tools to evaluate consumers' possible responses to these, and other, transportation innovations. We adapted techniques pioneered at the Oxford Transport Studies Unit to explore households' activity spaces (e.g., Jones, et al, 1983). These techniques included travel diaries, timelines, and maps of their activity locations that we used to reflect back to households their travel through of their travel in interactive stated response interviews. These data, as well as technical information, video images, and test drives of electric vehicles were used to create gaming scenarios and hypothetical choice experiments. The activity analysis framework demonstrates that each single trip is dependent on choices made about previous trips and on trips still to come. This point is central to the choice of the activity analysis framework for our EV, NEV, and station car market research. Activity analysis provides a structure in which to explore the meaning of travel constraints within a household's entire set of activities and travel tools. The variety of vehicle types we explored and the differences between implementing very different vehicles in existing urban development or in urban structures specifically designed for them presented a potentially confusing array of research possibilities. We use the concept of a household activity space to provide a unifying thread. This concept and its application to each of the market studies and demonstration elements will be described in detail later, but briefly we describe a household's activity space as: the household members' activities; • the time schedule of those activities; • the geographic location of those activities; • linkages between activities; and • the modes and routes used to access those activities. Hagerstrand's typology of constraints (described below) differentiates transportation and communication tools according to their capabilities to mediate distance and time. The driving range limit on a single charge and the length of time it takes to recharge impose a new capability constraint on the EV, as compared to conventional vehicles. Gasoline (and diesel) vehicles and their ubiquitous fuel stations provide practically limitless range. Providing the information context for households to competently imagine how they would incorporate a

46


vehicle of limited range into their stocks of vehicles is the core of the designs for all of our studies. Our initial research questions were the following: Will households create EV activity sub-spaces? (Or more generally, will households create sub-spaces for any new transportation option?) • Is the existence ofanEV activity sub-space a sufficient condition for a household to include EVs in their choice sets for their vehicle purchase decisions? The remainder of this chapter is devoted to answering these questions for EVs, NEVs and station cars. In the following section we develop the concepts from the activity analysis paradigm that we use throughout, in particular, we present our rationale for focusing on the new daily distance constraint posed by EVs and for discussing electric vehicles in terms of the access they provide to an EV activity sub-space. Similarly, we explore the range, speed, and size constraints of NEVs and the alternative vehicle ownership arrangements offered by station cars.

Important Concepts from Activity Analysis Activity analysis is distinguished from other transportation research paradigms by its emphasis on travel as a derived demand that exhibits daily and multi-day patterns, related to and derived from differences in life style and activity participation across the population (Jones, et al, 1990). The intellectual roots of activity analysis included studies of human geography that delineated systems of constraints on activity participation in time-space (Hagerstrand, 1970) and identified patterns of behaviour across time and space (Chapin, 1974). We use Hagerstrand's system of constraints to define the new capability constraints that EVs might represent. Once we have defined the new constraints, their effects on household travel behaviour are explored in the market research and vehicle demonstrations described in the following sections. Hagerstrand (1970) introduced time-space prisms—bounded areas of time and space in which it is possible for a person to exist. Within these prisms, individuals foWov^ paths of actual timespace locations. Central to defining the shapes and sizes of these prisms and the paths through them, Hagerstrand proposed a typology of constraints: capability constraints, coupling constraints, and authority constraints. Capability constraints arise from biological requirements and the tools available to an individual. Some capability constraints, notably biological constraints such as sleep and sustenance, follow the individual throughout their time-space path, but are typically satisfied at a single, home location and require a certain minimum amount of time.


47

How Capability Constraints Subdivide the Time-Space Prism. We stated above the premise that households have, or will construct, portions of their time-space prism that they access by different travel modes. The origin of this premise lies in the fact that different travel modes impose different capability constraints on our movement across space and through time. We travel by a combination of certain physical functions and tools—walking, bicycles, buses, autos, etc. We communicate either directly through our senses or by communications technology. Thus the time-space prism through which an individual moves can be divided into regions of varying accessibility, depending on her physical capabilities and the availability to her of different travel and communication tools. Central to our research is an examination of whether such sub-divisions are simply an analytical tool for understanding travel, or an actual organizing principle used by households. We start with the assumption that the EV is a new tool to mediate distance—but it is a limited tool compared to the capabilities of a full-size ICEV. Because an EV can only be driven a relatively short distance before requiring a lengthy recharge time, there is the potential that it allows access to only a limited part of a household's time-space prism. Whether a household will consider buying an EV will depend in part on whether that household can access desired paths through its time-space prism using an EV in conjunction with other travel tools available to the household. While capability constraints define the extent of our time-space prisms, paths inside that prism are determined in large part by coupling and authority constraints. Coupling constraints "define where, when, and for how long, the individual has to join other individuals, tools, and materials in order to produce, consume and transact." (Hagerstrand, 1970) For example, employment may require that we interact with other people and tools on a particular schedule at one or more locations. Authority constraints define domains within the time-space prism to which an individual either controls the access of other individuals or to which his access is controlled by others. Empirical research has shown that household travel can be explained by this framework of constraints. For example, Kitamura, Nishi and Goulias (1990) show that choices of timing and location for non-work activities by commuters are consistent with a set of hypotheses based on the constraints Hagerstrand proposes. Those authors found that coupling constraints (shop opening times) and authority constraints (work start times) severely limit the number of nonwork trips made before work. Because of authority constraints and capability constraints, nonwork activities made during work-time are tightly clustered in space around the work location and tend to be either work-related trips or trips to eat.

48


Time-Space Prisms and Household Activity Space. Our use of the phrase activity space to describe the sets of activities that households access is based on definitions used by Horton and Reynolds (1971) in their initial development of an analytical framework to examine the effects of urban spatial structure on individual behaviour. If Hagerstrand defined the limits of the time-space prism, then Horton and Reynolds provide additional insight into how households choose paths within the prism. They define objective spatial structures as the location of a household relative to the objective locations of potential activities and an associated objective attractiveness of each location. The household's action space is defined as that group of all locations or nodes within the objective spatial structure for which the household has both information and a subjective utility. Finally, the household activity space is defined as the subset of all locations in the action space with which the household has direct contact as the result of day-to-day activities. Thus a household's activity space is a set of realized paths through Hagerstrand's time-space prism.

New Travel Constraints, New Travel Opportunities We explored with households how they respond to new constraints and opportunities on their activity space represented by EVs, NEVs, and station cars. EVs may impose a distance and time constraint, yet allow home recharging. To these, NEVs add top speed limitations, as well as passenger and cargo space restrictions. Station cars add new vehicle ownership arrangements. Household adaptations to these new travel tools may take the form of rejection of a vehicle that embodies new constraints, changes in their activity space, or the ability to recognize that despite different capabilities than conventional gasoline vehicles, the new vehicles types impose no binding constraints on household time-space paths. Electric Vehicles. We briefly review here how we apply the concepts discussed above to the specific problem of studying household response to EVs, NEVs, and station cars. We have already given the example of how the range and recharging characteristics of EVs impose new capability constraints on such vehicles. We also point out that the new capability to recharge an EV at home eliminates one trip-activity pair that is required for a gasoline-powered vehicle—the trip to the gas station to refuel. We note this seemingly minor change to a household's overall activity pattern is perceived by some households to be a significant benefit. There has been an overwhelming preoccupation with the effect that limited range will have on the market for electric vehicles. We have explored this preoccupation elsewhere and greatly discounted the impact of a daily range limit on one vehicle in households that own more than


49

one vehicle. Driving range limits on one vehicle appear to cause insurmountable problems in only a few of the increasing number of multi-vehicle households (Kurani, et al 1994; Turrentine and Kurani, 1995). In the discussion that follows, we will show that for many, if not most, multi-vehicle households a range limit of between 40 and 120 miles on one vehicle imposes no significant new constraint on the households ability to access its activity space. No search of their action space is required to find new activity locations, no activities are rescheduled or canceled. The primary adaptive strategy is occasional negotiation between household members for their mode of travel. Neighborhood Electric Vehicles. The general defining characteristic of NEVs is their specialization for local travel.^ As such, they can have more limited range and lower top speeds than EVs, and thus lower energy storage and power requirements. Consistent with low energy and power requirements, NEVs will be small. They will likely accommodate two or three persons plus storage space, though some may be larger to accommodate families with children. We envision that NEVs will range from top-end vehicles that are intended to travel on arterial streets at speeds of up to 45 mph, to bottom-end NEVs, with top speeds of about 25 mph.^ NEVs might have separate right-of-ways, only mixing with other motor vehicles in specialized circumstances, such as streets with vehicle speed and size limits. We hypothesize NEV purchase decisions will be predicated on households' assessment of how tightly a NEV restricts activity choice. Whether a household is willing to include a NEV in its choice set of vehicles it will potentially buy will depend in large part on whether the NEV is seen to provide access to some meaningful set of household activities, a set of activities we call the NEV activity sub-space or more simply, the NEV space. NEVs, because of their very short daily range, low top speeds and small payload and passenger capacities, represent a new travel tool with many potential capability constraints. These capability constraints may act to restrict the choice of activities that could be accessed in a NEV. Driving range limits may be rele-vant to analyses of the market for NEVs because these vehicles will have even shorter ranges than larger, freeway-capable EVs. NEVs also represent a possible new time constraint because of their limited top speed. The choice of activities accessed in a NEV may be circumscribed by how long it takes to get places as well as how far those places are from home or each other. These two constraints act to reduce the size of the time-space prism it is possible for an individual to access. The limit on the number of passengers may impact activity access through the coupling constraints created by the need for one person to provide transportation services to other household members, co-workers, or clients. If the payload and passenger capability constraints

50


do not allow a driver to provide needed or expected transportation services to another person, this creates conflicts through the coupling constraints to those dependent travelers. The range, speed and size capability constraints may conflict with authority constraints. As an example, perhaps the only way an adult in the household could drive a NEV to work would be to leave earlier to arrive on time (workplace authority constraint and speed capability constraint causes earlier departure from home). Now suppose this person is also responsible for delivering children to daycare (a coupling constraint). The daycare center may not open in time (an authority constraint) for the adult household member to both leave early to arrive at work on time and yet leave late enough to deliver the children to daycare as it opens. In this hypothetical case, the new capability constraint on vehicle speed, the existence of a coupling constraint to another household member, and the conflict between authority constraints imposed by work place and day care schedules renders a time-space path via a NEV unfeasible. Despite such possible limits, an NEV may represent a highly valued travel tool. If within the context of multiple travel tools, the household can construct a NEV space, then such a vehicle may be seen as a way to maintain the high accessibility and mobility of multi-vehicle ownership at a reduced cost over owning and operating yet another conventional gasolinepowered vehicle. Station Cars. Station cars are perceived by their proponents as relaxing some of the capability constraints of traditional transit service. As most homes and activity locations are not within easy walking distance of a transit station, access to and egress from transit are important considerations. Studies done for the San Francisco Bay Area Rapid Transit (BART) District indicate potentially large numbers of new riders if parking problems at stations can be solved, if people who do not live near BART stations can be provided convenient access, and if reverse (urban to suburban) commutes can be encouraged. BART is currently running a smallscale station car demonstration to assess whether the station car concept can address these three issues. To the extent that station cars are either freeway-capable EVs or are NEVs, they will impose many of the same capability constraints as discussed for those vehicle types above. Through their relationship to the household as a short-term, but frequent, rental vehicle rather than a privately owned vehicle, station cars may either relax existing authority and coupling constraints, or impose new ones. They may provide a lower cost option than ownership of another household vehicle to access specific activities. Their value may be assessed by households according to both how many activities are made accessible via transit and how many are accessible directly via the station car itself


51

STUDIES REVIEWED The results presented below are drawn from several studies conducted by the authors over the past six years. All but the station car study were part of two multi-year research efforts to examine consumer markets for EVs and NEVs. The series of EV market studies began in 1991 and included these studies: a ride-and-drive clinic of alternative fuel and electric cars held at the Rose Bowl in Pasadena, California (Turrentine, et al, 1992a); • phone interviews of EV owners in California (Kurani and Turrentine, 1993); • interactive stated response interviews with California residents in Sacramento, Santa Clara, and Riverside counties (Turrentine, et al, 1992b; Kurani, et al 1994); a mail survey of California residents in the Sacramento, San Francisco Bay Area, Los Angeles, San Diego and Fresno urban areas (Turrentine and Kurani, 1995; Kurani and Turrentine, 1996). The series NEV studies, all reported in Kurani, et al, (1995), included the following: case studies based interviews, focus groups and field observations of the "golf cart communities" Sun City, Arizona and Palm Desert, California; ride-and-drive clinics in Sacramento and Davis, California in which people drove and reviewed a wide variety of electric vehicles, including NEVs; vehicle trials in Sacramento and Davis, California in which households who were given use of a NEV for a one-week period kept travel diaries and participated in interactive stated response interviews; and • the same statewide mail survey of household electric vehicle purchase intentions as shown above for EVs. The results reported here for station cars are based on a small set of interactive stated response interviews and focus groups conducted with participants in the ongoing BART station car demonstration in 1996/97.

52


SPECIALIZATION OF TRAVEL MODES BASED ON THE SUB-DIVISION OF HOUSEHOLD ACTIVITY SPACE

Evidence for Pre-Existing Travel Mode Specific Activity Sub-Spaces Recalling our two central questions, note that the first does not ask whether households will search their action space for new activity locations to accommodate EVs or any other new transportation tool; nor does it ask what activities they are willing to give up in order to accommodate an EV. It asks the more general question whether they can create, by any means, a sub-set of their activities to which the new travel tool allows them access. Implied in this is the fact that other available travel tools provide them access to whatever activities are not in \\iQEV space. We find evidence that mode specific sub-spaces have been created in households that already use distinct travel options (e.g., a conventional car and an EV, car and bicycle, car and transit). In households that currently access all their away-from-home activities by automobiles, imagining and creating sub-spaces of their activities that they can access by combinations of familiar and new travel options allows them to assess the new travel options in a manner that allows the evaluation of both practical travel implications and larger lifestyle issues. Electric Vehicle Owners. In the sample of EV owners we interviewed, we observed efforts to change activity locations and the explicit formation of EV spaces (Kurani and Turrentine, 1993). Still, the proportion of EV owners who reported changing activity locations to accommodate their EV was smaller than those who made no such changes. Fifty-nine percent of the EV owners said they made no adjustments; 36 percent indicated they had made some changes. EV space formation was observed in vehicle use planning behaviour. The most often mentioned accommodation to the EV was the need to plan which vehicle (EV or gasolinepowered) to use for certain days or certain trips. For almost a third of these people, this change involved an active decision process; for another third, it involved a simple trip-by-trip rule of thumb—if the trip is within the EV range, the EV is used. Only 11 percent said they had changed the location of common activities. They shopped closer to home or planned shopping trips to stores or malls with electrical outlets at which they could recharge. The mundane nature of these changes is one explanation why EV range capability did not affect gasoline vehicle use in EV owners households. For most EV owners, their activity space appears to be segmented—they access by EV all of their activity space that they can; activities outside the EV range are accessed by the gasoline vehicle.


53

We found that their aggregate use of the household's gasoline vehicle was largely independent of aggregate EV use. Total miles traveled per year in the gasoline vehicle was not related to percentage of household travel apportioned to the EV, the total miles driven in the EV, or the EVs speed capability. These results indicate that while characteristics of the EV do determine EV use, they do not affect the household's total use of their gasoline car. Households who put relatively few miles on their EV do not necessarily put more or less miles on their gasoline car than households which put many miles on their EV. This result is consistent with the hypothesis that EVs fill a specific proportion of the households activity space according to the characteristics of the EV. Interactive Stated Response Interviews (PIREG). We observe that in the ISR interviews, households acted conservatively with respect to changes in activity locations, timing, sequence, and duration. Most actions taken to accommodate an EV involved little adjustment to existing time-space paths, but were more likely to involve an infre-quent change of travel mode, e.g., swapping for the gasoline-powered car with another of the household's drivers or returning home to switch to an unused gasoline vehicle (Turrentine, et al, 1992; Kurani, et al 1994). The EV space most of these households created exactly overlapped the existing subspace assigned to an existing conventional vehicle except for those few trips or series of trips whose total distance is beyond the EVs range. Assignment of the EV to particular trips was usually made through the choice of a primary driver of the EV. This choice was in large part determined by the household's choice of which vehicle in their current fleet would be replaced by an EV for the games we played with them in the interview. The household often chose a comfortable range for the EV that allowed one driver to accomplish most of their travel days. The important policy-related finding was that these comfortable ranges were far shorter than the minimum acceptable ranges implied by econometric models estimated on hypothetical choice data (e.g., Morton, et al (1978); Beggs and Cardell (1981); Calfee (1985); Bunch, et al (1993)). From these PIREG interviews, we extracted two variables related to the households' overall activity spaces that determined the comfortable range capability for the hypothetical EV (Kurani, et al, 1994; Turrentine, et al, 1991). These two variables were the routine activity space defined by that set of activities that the household accessed on a daily and weekly basis (including all the other associated dimensions—location, mode and route to access, etc.); and a critical destination that a household member felt they must be able to reach even if the "unlimited range" gasoline vehicle was not available. Household Neighborhood Electric Vehicle Trials. We observed segmentation of household activity spaces by travel mode in our household NEV trials. During our NEV research, we

54


conducted a small-scale demonstration project in which households in both Davis and Sacramento, CA drove an NEV for one week. During this trial week, drivers in the household recorded all their travel in diaries. For each trip, drivers recorded all dimensions of the activity space. Respondents also drew the route of each trip on a map. The diaries served as the basis for creating summary maps of their week's travel to be used in interactive gaming interviews that explored the households* travel, their assessment of the NEV, and initial NEV purchase intentions. In Davis, we found evidence of household activity sub-spaces accessed by different travel modes, but not in Sacramento. Some of our sample households in Davis used bicycles extensively. These households consistently accessed a distinct set of activities by bicycle from those they accessed by car. Their bicycle and automobile activity spaces had regular and welldefined (if sometimes over-lapping) boundaries. The creation of mode-defined activity subspaces did appear to be an organizing principle for travel in these households. The households in Sacramento exhibit the unimodal travel characteristics discussed in the introduction to this chapter. None of these households used transit (either bus or light-rail). Few walked, and those that did usually only infrequently accessed only one or two activities In those households in Davis that used bicycles as travel tools, the bicycles provided access to work, grocery and other shopping activities, especially if they were located on the university campus or in the contiguous downtown. Trips to run errands during the day before returning home were made on foot or bike, if bike was the mode taken to work. While cars and bikes were both used for local trips, only cars were used for out-of-town travel. In addition, many Davis house-holds noted they had altered their lifestyle to reduce the number of local automobile trips and had consciously chosen to cycle as much as possible. The specific existence of a bicycle activity space does indicate that households will create activity sub-spaces distinguished by travel modes, but does not itself appear to be positively associated with desire to buy a NEV. The NEV space must be sufficiently distinguished from both the bicycle space and car space to trigger a positive purchase intention. In several households this means the NEV must be sufficiently larger and faster than a bicycle, while remaining sufficiently less expensive than an automobile. For many of our Davis households, a vehicle must have the requisite speed and range to be able to reach nearby towns before it is sufficiently distinguished from a bicycle to be seriously considered for purchase. This performance level is well beyond the NEV examples that these households drove in our demonstration. Golf Carts as Transportation Tools. In the towns of Palm Desert, California and Sun City, Arizona, we also found evidence that house-holds created distinct sets of activities that they


55

accessed by distinct modes. Households in both case study towns had constructed golf cart activity spaces. In some of these households, this activity space was quite simple, consisting of two types of activity locations—home and golf courses. This does not mean that playing golf was the only activity accessed by golf cart. Golf courses often served as the center of other social and recreational activities for these households. In many households, including some who did not play golf at all, the golf cart provided access to a wide variety of activities, e.g., social, shopping, personal and professional business. The existence of a set of activities regularly accessed by golf cart and the reported purchase of a golf cart to replace a full-size car both argue for the existence of mode-specific activity sub-spaces in these households. Palm Desert has built a network of golf cart specific roadway infrastructure to facilitate cart travel, but we saw carts used as general purpose transportation tools in Sun City too, where there is no specialized infrastructure to accommodate low-speed vehicles. At least as important as infrastructure were the characteristics of the cart drivers. Most households who drive golf carts in both towns were subject to few authority or coupling constraints—as retired adults without children at home, they were subject to fewer of the imposed schedules that jobs and children impose. The limits of those golf cart spaces are determined by attributes of the vehicles, the transportation infrastructure, and the activity choices of the household. The capability constraints on speed, dis-tance and payload restrict the distance people are willing to travel and their sequences of activities. The more important constraint on how far people will travel in their carts was top speed and not driving range. Distance (driving range) was less relevant to travel choices in these communities where all the daily activity locations were within a few miles of home.

How Much of the Households Activity Space is Contained in the NEV Sub-space? Having established that do create distinct, mode-defined sets of their activities to incorporate EVs and NEVs into their set of travel tools, we next explore for NEVs how "large" such activity sub-spaces are and the impact of the NEVs' capability constraints on the households' ability to access desired activities. We examine the total amount of travel by our NEV sample and how this travel was apportioned to different modes. In this way, we begin to assess households' abilities to form NEV spaces. Further, we establish the relevance of those subspaces to the households' total activity space. Trips and miles traveled in the NEVs. As expected based on their limited speed and range, NEVs replaced a greater proportion of trips than of miles in almost every household in both

56


Davis and Sacramento. The NEV accounted for an average of 19% of the total distance households traveled during their diary week, yet they were driven for 41% of trips. The percentage of household trips for which a NEV substituted ranged from a low of 10% to a high of 72%. The percentage of miles for which a NEV substituted ranged from a low of 6% to a high of 43%. The number of miles traveled by a household in a NEV varied from a low of 12 miles to a high of 106. As we also expected, there was a moderately strong statistical correlation between the number of trips and the number of miles. However, there was no such association between the proportion of trips and the proportion of miles. That is, the households moved through such different activity spaces that the substitution of a NEV for a given proportion of trips did not lead to a predictable proportion of miles the NEV travels. A comparison of the number of trips and miles traveled by travel mode highlights the extent to which personal, motorized transport dominated travel in these households. The total number of trips made by all 15 households in the NEV trials was 824. Across the sample, 86% of all these trips were made by some form of personal, motorized travel. Across the sample, the NEV substituted mostly for motor vehicle travel. Of the 364 NEV trips, 240 were trips that would have been made in a gasoline car if the NEV was not available. Many of the rest were NEV "test drives" for family, friends, and neighbors. Forming Neighborhood Electric Vehicle Spaces. The greater the performance differences between the EV and gasoline vehicles (i.e., the more the EV is an NEV), the more the household's successful adaptation depended on their defining subspaces within their overall activity space that were distinct from any existing set of activities allocated to an existing vehicle. In the formation of these sub-spaces, we observed different types of constraints that form their boundaries—capability constraints (range, speed, as well as passenger and cargo capacity), as well as coupling and authority constraints, determined the boundaries of the NEV space and the overlap of this space with other mode sub-spaces. We identify four types of activities important to mode-defined activity sub-spaces: 1) Activities on the boundaries of each mode defined sub-space; 2) Activities within the sub-space; 3) Activities outside the sub-space; and 4) Activities that belong to more than one mode-defined sub-space. The activities at the boundaries between sub-spaces highlight important constraints. The other three types of activities determine the value a household might place on the travel mode that defines any given sub-space. Almost every household in Davis indicated nearly the entire town was accessible to them in the NEV, but any activity location outside town was not accessible because of the capability


57

constraints imposed by the NEVs driving range and top speed constraints. Parts of town to which they did not travel in the NEV were seldom visited by any mode. The question remains, to what extent do the boundaries of Davis, which formed the boundary of the objective spatial structure that can be accessed in a NEV, coincide with the boundaries of the household's routine activity space? That is, we expect the boundary of the NEV space of Davis residents to be no larger than the city limits. What we must determine is: • the importance of activities beyond the town's boundary to the lifestyle choices of the households and how they access those activities; • whether they construct a set of travel tools for accessing activities within the boundaries of Davis that includes a NEV; and how they differentiate the use of NEVs, cars, and bikes within the town limits. Our sample of households in Davis contains some households that leave Davis less than weekly and others that travel beyond the city limits daily. However, this simple distinction alone does not determine which households construct a useful NEV space from those that do not. The frequency of such trips, the usual travel mode, the activity for which the trip was being made and the household's vehicle holdings all contributed to whether a vehicle that was limited to in-town travel alone would be seriously considered. Within the spatial boundaries of their routine activity space, almost every household discovered activities for which some capability constraint other than driving range eliminated some activities from their NEV space. The limiting constraint was almost always the passenger or pay load capacity. Several passenger trips to chauffeur children or other family members, trips that linked chauffeuring and shopping, and trips to haul bulky items could not be made in the NEVs. These trips could not be made as single-purpose trips or as extended tours. In multicar households, another vehicle was available to make these trips and the possibility exists a NEV could displace one of the existing vehicles in the households' current holdings. In households who own only one car, the NEV would have had to have been an additional vehicle. Of the activities in Davis to which the NEV did provide access, it competed with cars and bikes. These trips included those to the university campus, downtown, some shopping, the post office, bank and other local activity locations. Trips that could not be made in a NEV included trips out of town, trips within town that required travel on one of the few streets with a speed limit in excess of 35 mph, and trips with more than one passenger. The NEV most often replaced bike trips when the main reason for riding a bike was environmental, not convenience. Participants perceived that the NEVs provided "guilt-free" driving, an opportunity to take a "car" without worrying about polluting the environment. If the bike was used for a given trip because it was more convenient than a car, households speculated that

58


bikes would be chosen over NEVs too—even if the NEV was a permanent part of their repertoire of travel tools. For example, the university does not allow motor vehicle traffic in the campus core. For university employees who live in Davis and work in the campus core, travel to work by bicycle is generally faster than by car and parking is easier and free. Under these conditions, a NEV that is subject to the same access restrictions and parking fees as a standard automobile was unlikely to substitute for their bike. But the NEV was also perceived as less safe than automobiles by almost all the drivers. Safety concerns were expressed both regarding the vehicle's size and acceleration capability. Compared to a bike, the NEV was more comfortable and could carry larger loads, but only inclement weather rendered the NEV superior to a bicycle for many households. Bicycles were often more convenient because NEVs were subject to the same restrictions as cars, especially on parking. Bicycles were certainly perceived as less expensive. There was no consensus perception of the relative safety of the NEVs and bikes. Some people felt much safer in the NEV, others felt safer on their bikes. Changing routes to activities. The inability of NEVs to travel on freeways, urban expressways and other high speed roads required several participants to search for alternative routes to activities that were otherwise within the driving range of the NEV. Whether an acceptable alternative could be found was crucial to households' ability to create a NEV-space. In an attempt to deal with the NEVs speed limitations, many Sacramento participants altered their regular routes, leaving the freeway to drive on surface streets. In addition to searching for entirely new routes, households switched to alternate routes that were already used occasionally. For example, one driver's usual pattern in her gasoline car was to commute to work on surface streets, but to return home on a route that included a freeway because it passed by her usual grocery store. In the NEV, she commuted both to and from work on her usual surface street route. Thus she changed her route home from work and to the grocery store to a route she regularly used for another trip. Land use and infrastructure limits on route choice. A household's ability to find an alternative route was heavily influenced by the types of roads available and the speed and acceleration capabilities of the NEV. In Sacramento in particular, we see the effects of existing urban infrastructure on households' ability to create a useful NEV space. Some households were located in residential enclaves surrounded by high-speed roads. The high-speed streets served as barriers to access to all but a limited number of activities. Unfortunately, these land use and transportation infrastructure patterns are typical of the majority of the suburban communities surrounding downtown and midtown Sacramento.


59

This land use pattern is typical of suburbs throughout California and the nation. Virtually all retail and commercial activity near these neighborhoods can only be accessed by entrances from the arterial roads. Within this land use pattern, households were almost completely unable to construct a meaningful NEV space. In one household that lives in such a suburb, the NEV was relegated to visiting garage sales and other trips within the neighborhood. Its one substantive use was for the work commute of the female head of household, but as this trip was usually made by walking or carpooling, substitution of a NEV would have few positive household travel or policy impacts.

Household Response to a Distance Constraint on One Household Vehicle Within the one-week trials, we did not expect to observe "stable" adaptations to distance budgets. Households did not always explore the full extent of the NEV's range. We expect that in the long-run, the actual NEV space that these households create will be different from those in the one-week trials. Therefore, we do not analyze the extent of those spaces per se, so much as we examine the adaptive strategies used to begin to create those activity sub-spaces. A short driving range on a vehicle imposes a distance budget on household travel choices. Just as households have different financial budgets, households will have different responses to this new budget. An important tradeoff between whether to drive the NEV or a car centered around driving range and the amount of pre-trip planning required to adapt to this new capability constraint. Adaptations to the distance budget included pre-planning the day to decide which driver (if any) would take the NEV, switching vehicles between drivers during the day, planned changes in trip linkages, unplanned interruptions of trip linkages and daytime recharging. Because of the short range of the vehicles and the long charging time, travel had to be planned in advance. With no place to recharge besides home (and occasionally work), last minute changes in plans required greater attention to whether the NEV was sufficiently charged than would be paid to whether the car had enough gasoline. Along with planning individual trips, daily travel plans had to be made conservatively; attempting too many trips or trips of too great a length might mean being stranded. The decision as to which person would drive the NEV was based on the expected activities for each person that day, moderated by any underlying propensity for unexpected daily variation, e.g., last minute trips. Typically, multi-driver, multi-vehicle households decided at the beginning of the day who would drive the NEV that day. Unexpected daily variation in activities can cause the NEV to be left home altogether. Some participants took their car to work rather than the NEV on one or more occasions because the possibility existed that they might have to stay late or run errands during or after work. Also, the desire to link several

60


activities can cause the NEV to be left home or change the intended sequence of activities. Alternatively, if the NEV is taken on a series of linked trips, the list of intended activities to be linked in any given excursion from home can be changed to accommodate the NEV. These changes can be planned or spontaneous. Daily activity planning, strategies to deal with day-to-day variability in travel, and day-time recharging are likely long-term adaptations to the distance budget imposed by the driving range limits of NEVs. On-going interruptions of activity sequences are less likely to be tolerated. Vehicle specialization makes the choice of which vehicle to drive—^the NEV or a gasoline car—easier, as does the availability of other viable travel alternatives to a household vehicle.

The Impact of Authority and Coupling Constraints on Travel As we have discussed, range and speed capabilities determined which activities could be contained within the NEV space, while the overlap between the NEV-space and the car and bike sub-spaces was often determined by the passenger or payload capacity of the vehicle. In combination with limited passenger capacity, coupling constraints to other household members or persons outside the household (e.g., chauffeuring children, doing errands, carpooling to work, going to parties with friends) often determined whether the NEV was used for a particular series of trips. Although travel surveys indicate many of us travel alone in our cars for most trips, the inability to travel with another person was often as important a constraint on use of a NEV as was range or speed. The number and age of children in the household affected the ability of the NEV to fulfill certain "chauffeur children" trips. The following example highlights the NEVs passenger and cargo limits. One driver often stops at daycare to pick up her young son and then stops at the grocery store before going home after work. In the NEV, this sequence of activities was not possible because there was not room for her son, all his gear, and the groceries. Since she had to return directly home from daycare to switch to a car before making the grocery shopping trip, the NEV was much less convenient than her own car—she prefered to take her car for the entire day rather than make the extra trip to accommodate the NEV. Another household also indicated the NEV was too small for them and their young children. However, in this household, vehicle size per se was rarely the problem. Most trips for which the NEV was too small were also to locations too far away. In still another household, the passenger limit came into play when ftjlfilling carpool obligations to other households, specifically, when taking their daughter and her friend to school. Parents who are unwilling to leave young children at home, couples who do errands together, even a desire to take a pet along for the ride—all


61

required more seats or cargo space than the NEV provided. In these cases, activities that could otherwise easily be accessed in the NEV based on its range and speed were instead accessed in a gasoline car. To provide a sense of the extent to which authority and coupling constraints affect household travel mode choices, we asked households to record in their travel diaries whether each trip could have been made at a different time and how long they had known they would be making this trip. Trips were also coded to indicate whether they were made solely to provide transportation service to another person, i.e., the driver would not have made the trip at all if not for the need to deliver another person to an activity. By examining the answers to these questions, we can summarize how the authority and coupling constraints shaped households travel choices. First, we examine how much of the aggregate travel by households in the NEV trials sample was subject to authority and coupling constraints. We summarize responses to questions as to whether trips could have been taken at another time. Two-thirds of all trips were either themselves constrained to a particular time or linked to another trip that was constrained to the time at which it was made. While this fact does not summarize all the reasons why these trips could not be made another time, it does illustrate that within our sample, most travel took place under authority and coupling constraints that require that most trips be made within specific windows of time, and that the constraints on these trips can affect other trips made in sequence with the constrained trip. When we examined the time for which people had known they would be making trips, we see that a great deal of travel was planned far in advance—44 percent of all trips had been scheduled to take place for many days. Also, there was a relationship between how long trips have been expected and the flexibility of their timing. Fifty-five percent of all trips that could not be made at another time had been anticipated for several days. In contrast, forty-five percent of all trips that could have been made at another time were made at the last minute. Authority constraints lead to routines in household behaviour that are manifested by travel that is known and scheduled far in advance. The relationship between whether a trip could have been made at a different time and the time for which the trip had been expected is summarized in Table 1. A chi-square test performed on the data in Table 1 rejects the null hypothesis that trip scheduling and flexibility were independent. Among trips that had been expected for many days, the number of trips that could not be made at another time far exceeded what we would expect under independence—^trips that are inflexible in their timing tend to be expected in advance.

62


Table 1 How Long Trip Had Been Expected By Whether Trip Could be Made at Another Time Can trip be made at some other time? Trip anticipated for how long? No Yes Total 292 days before 38 330 last night 58 41 99 today, earlier 50 63 113 today, last minute 103 106 209 Total 503 248 751

The role of authority and coupling constraints on trips to engage in different activities is highlighted in Table 2. The number of trips to different destination types is cross-tabulated by how long the trips to those destinations had been expected. Almost twice as many trips to work or school had been anticipated for several days than expected under the null hypothesis. Further, three-fourths of all work and school trips had been anticipated for many days. In contrast, only 2 of 41 trips made to dine had been expected for many days, far fewer than the expected number of trips under the null hypothesis of independence.

Trip anticipated for how long? days before last night earlier today last minute Total

Table 2 Activity Type by How Long Trip Had Been Expected Activity Type at Destination Personal Social or Work or Dining School Business Shopping Recreatio n 2 8 39 51 91 4 16 14 15 12 17 13 17 7 13 28 25 35 12 28 41 112 60 106 128

Total 191 61 67 128 447

Serve Passenger Trips—a Coupling Constraint. "Serve passenger" trips are trips made by one person solely to provide transportation services for another person. These trips most often arise


63

out of household responsibilities that are a form of coupling constraint. For example, adult household members who, on their way to work, deliver children to school or daycare, are making a serve passenger trip—^they would not be driving to the school or daycare center unless they were providing a ride for the children. Most serve passenger trips in our NEV sample were subject to rigid time constraints. In Table 3, data on whether a particular trip was a serve passenger trip is cross-classified by whether that trip could have been made at another time. A total of 141 "serve passenger" trips were made. Of these, 128 trips could not have been made at another time; only 94 are expected under the null hypothesis of independence. Thirteen of the "serve passenger" trips could have been made at another time; 47 would be expected. The chi-square test on this data is significant. When coupling constraints result in one household member providing transportation services to others, the transportation provider becomes subject to the authority and coupling constraints of the travel-dependent person. Table 3 Serve Passenger Trips by Whether Trip Could Be Made Another Time Could trip be made at another time? Serve passenger trip? No Yes Total: 244 631 No 387 13 Yes 128 141 257 772 Total 515

J

Data on whether a trip was a serve passenger trip and how long the trip had been expected to be made are cross-tabulated in Table 4. The chi-square test on these data also rejects the null hypothesis of independence. Serve passenger trips tended to be expected well in advance. The coupling constraints that resulted in serve passenger trips tended to produce routines in which those trips were anticipated for several days. Table 4 Serve Passenger Trips by How Long Trip has been Expected How long as trip been expected? Many Last Early Last Serve passenger trip? Days Night Today Minute 100 No 256 81 186 15 Yes 78 18 27 Total 334 99 115 213

Total 623 138 761

64


Station Cars and Other Short Term Rental Vehicles In addition to our work on EVs and NEVs, we have been examining consumer response to "station cars", a type of short term rental vehicle made available at transit stations to improve access possibilities, or capabilities, of transit trips, and thus expand the client base of the transit system. In addition to households' existing activity space, station cars might provide access to a new set of activity locations around transit stations that lie beyond the normal walking area around a station. Several station car demonstration projects are in process. These demonstrations are designed to test new vehicle, reservation and communication technology, as well as the concept of transit linked and shared used vehicles. We have been working on an electric station car program operated by the Bay Area Rapid Transit (BART) District in the San Francisco Bay Area. Stations cars are similar to NEVs with respect to vehicle design issues, (e.g., small size, lower speeds and battery driven electric designs for emissions benefits) but are linked to a transit station. In concept, station cars are made available through various ownership plans, from leasing to very short term rentals. In the BART demonstration so far, station cars are made available through monthly leases, in some case shared by two carpooling drivers. Participants lease use of a station car either at their home or work end transit station. At the home end, the station car is designed to provide clean transport between home and the BART station, and can be used as well as for other trips at home, including weekends. Other participants lease a station car to use at the work end of their transit commute, to get from the station to work and to run errands during the day at work. In the case of the home end, the station car may supplant the need for a second car by households. The electric station cars in the BART project were used by participants for periods of between three months and one year. The vehicles had special parking and charging locations at particular transit stations used by the participants. We have so far been unable to assess consumer response to short term rental systems within this demonstration project. We will be testing some aspects of such a electronic, self-rental system in a new demonstration during the coming year. While our analysis of this project is far from complete, we observe similar spatial issues to the NEV demonstration, particularly at the work end, although the household may be expanding their activities at the work end rather than redefining a sub-space. The two seat cars were quite practical at the work end, allowing for a rapid commute link between the station and work, errands around the work location and transit stations as well as lunch dates away from work, a limitation for many transit users who do not having shopping and lunch restaurants within


65

walking distance of their work location. In fact, the very short wheel-base of the vehicles in the BART program allowed for parking in spaces no other vehicle except motorcycles could use. At the home end, most of the households with children found the limited passenger capacity a constraint. Not only were there trips which could not be made because of this constraint, but often, linked chains of travel required cargo flexibility to complete the chain. Range was not a problem—new versions of the vehicles had close to seventy miles of range. Households enjoyed the use of the vehicle for its novelty value and perceived environmental benefits, but it was less practical because of cargo limits. Additionally, such a vehicle has a practical and cheap alternative—a used conventional vehicle, which can be parked for free at the BART station. When dealing with the idea of short term rentals situation, at the home end, households were resistant to the idea, primarily because such an arrangement imposes a severe capability constraint on the household—^the vehicle is not available for any other household uses that involved "serve passenger" and other family trips.

CONCLUSIONS

Markets for Electric Vehicles, NEVs and Station Cars Our results show that consumers do explicitly create specialized sub-sets of their overall activity spaces defined by their mode of access. Some households will do this for small vehicles for use within existing roadway infrastructure. Specialized infrastructure—such as a small dedicated roads, specialized parking, and away-from-home recharging services—can increase the number of activities contained within these activity sub-spaces. Given the high percentage of household trips we observe in these specialized zones, and the demand for more sustainable cities through cleaner air, quieter vehicles, and reduced congestion, as well as other policy instruments such as specialized drivers licensing, we see great potential for household specialization of vehicles. Our findings suggest that in many communities, households may find a specialized, small vehicle can practically be incorporated into their vehicle fleet either as an additional vehicle or by displacing an existing vehicle. In particular, policies which foster local planning and infrastructure that facilitate use of small EVs can increase their practicality. As these vehicles are clean and quiet, there will be significant benefits to communities for reducing emissions, especially cold start emissions, reducing pedestrian hazards by reducing vehicle mass, reducing vehicle noise, and reducing park-ing space, and perhaps create safe community access zones for new drivers or seniors with reduced driving capabilities.

66


We have argued in previous works that one of the primary policy-related findings of our approach to household markets for EVs is that driving range is much less an issue than reported in a variety of other studies. If an EV is expected to displace an existing vehicle from multi-vehicle house-holds, it does need to share the passenger, payload, and amenity characteristics of conventional vehicles. However, a driving range limit of between 40 and 120 miles on one vehicle does not itself create a significant new constraint to multi-vehicle households' ability to access to their existing activity spaces. Home-rechargeable EVs relax a significant constraint in some households by eliminating the need for refueling trips. Finally, the mix of conventional and gasoline vehicles within a households vehicle holdings allows convenient access to their activity space and the expression of a wider variety of lifestyle goals. The things which distinguish NEVs and station cars are their small size, reduced speed and perhaps electric drive train and batteries. The most basic description of such vehicles is that they are designed to access a localized objective space. Thus an important determinant of their use will be the roadway infrastructure of that area. If use within a proscribed area does not require freeway speeds, the next most important capability constraint will be passenger load, the importance of which will vary between whether the base for such a vehicle is a home, or a transit station (or other non-home base) In the home-based applications, there is a greater need for cargo and passenger loads within the local home area, given the way such cars are used. This of course varies from household to household, as some retired households, single person households and other small households will find two passenger vehicle adequate. While travel behaviour studies seem to indicate a preponderance of single occupant cars on the road, and so indicate a single-passenger car is of interest, our work shows that access to the local area around homes demands greater cargo and passenger flexibility because of the wide variety of activities and coupling constraints imposed by children. It is precisely in this area around home that the passenger and cargo needs are greatest. Thus NEVs (and possibly station cars) will need to be in the 2+2 passenger seating configuration. We note that much of the roadway and recharging infrastructure that can foster NEV use is also in accordance with public policy emissions goals, households' lifestyle goals, and community-based land use plans, programs for seniors and other travel dis-advantaged groups, and improved transit use. Unfortunately, modem suburban neighborhood designs which make walking and bicycling difficult, also make the specialized access routes for NEVs or other specialized small cars difficult to implement. If access from a neighborhood to the shopping mall must be made by high speed arterials, then bicycle, walking and small, low speed vehicles are limited to the neighborhood only. In the past few years, the U.S. federal Inter-Modal


67

Surface Transportation Efficiency Act (ISTEA) has provided an opening to infrastructure to assist pedestrian and bicycle mobility. It will be difficult to enact similar legislation aimed directly at creating NEV friendly neighborhoods until there is a proven market. However, there are opportunities in particular communities to leapfrog existing standard land-use and infrastructure designs and to provide specialized infrastructure with federal funding to assist these local experiments. In particular, Congestion Management and Air Quality (CMAQ) funds have provided the type of infrastructure developments like what would be needed for small electric vehicles. The one trend which indicates the growth of markets for such vehicles is increasing diversification of household vehicle holdings. Households increasing purchase diverse vehicle types, such as sports utility vehicles, minivans, sports cars and pick-ups to increase the versatility of their private fleet. The automobile now stands as an extremely versatile, one size fits all tool. It may be desirable and even necessary to discourage this sort of one size fits all versatility to encourage investment in specialized, small vehicles which can reduce vehicle impacts on local areas.

Methodology The study of possible futures around new types of vehicles presents many methodological challenges. We have reported on several studies in which we utilized detailed household activity data and intensive interview techniques to create contexts for households to evaluate new travel tools. The concept of an "activity sub-space" fits well with the theoretical framework provided by Hagerstrand. We can identify the features of new travel tools into whether they impose new capability, authority and coupling constraintson household activity participation, or create new opportunities in activity space formation. This in turn allows us to conduct our analysis in a more systematic fashion, differentiating between the hypothetical aspects of gaming techniques in our interviews. We are able to construct hypothetical situations for participants that correspond more specifically to important aspects of technology. In particular, we have found that much attention must be given to potential reorganization of trips and household conceptualization of space, as they assess the utility of these new types of vehicles. These are expensive choices for households, affecting critical lifestyle goals. Simple interviews and survey techniques are likely to be invalid. We foresee using computer aided interviews in the future that make such spatial interview games more sophisticated and for which we can save the evolving concepts of each household sub-space for even further analysis of the decision processes we have described only briefly here.

68


REFERENCES Beggs, S. D. and N. S. Cardell (1980). "Choice of smallest car by multi-vehicle households and the demand for electric vehicles". Transportation Research A, 14A, pp. 380-404. Bunch, D. S., et al (1993). "Demand for clean fueled vehicles in California: A discrete-choice, stated preference survey". Transportation Research A, 27A, pp. 237-53. Calfee, J.E. (1985). "Estimating the demand for electric automobiles using fully disaggregated probabilistic choice analysis". Transportation Research B, 19B, pp. 287301 Chapin, F. S. (1974). Human Activity Patterns in the City: Things People do in Time and Space. London: John Wiley and Sons. EPA Report #420-R-93-007 (1993). cited in E.W. Johnson. "Taming the Car and Its User: Should We Do Both?" Bulletin, The American Academy of Arts and Sciences. Vol 46, No. 2, November, 1992, pp. 13-29. Hagerstrand, T. (1970). "What about People in Regional Science?" Papers of the Regional Science Association, v. 24 pp. 7-21. Jones, P.M., M. C. Dix, M. I. Clarke and I. G. Heggie (1983). Understanding travel behaviour. Aldershot, U.K.: Gower. Jones, P.M., et al (1990). "Activity Analysis: State-of-the-Art and Future Directions." in P. Jones (ed.) Developments in Dynamic and Activity-Based Approaches to Travel Analysis. Aldershot, U.K.: Gower. Horton, F. E. and D. R. Reynolds (1971). "Effects of Urban Spatial Structure on Individual Behavior." Economic Geography. 47:1 pp. 36-48. Kitamura, R., et al (1990). "Trip Chaining Behavior by Central City Commuters: A Causal Analysis of Time-Space Constraints." in P. Jones (ed.) Developments in Dynamic and Activity-Based Approaches to Travel Analysis. Aldereshot, U.K.: Gower. Kurani, K. S. and T. Turrentine (1993). "Electric Vehicle Owners: Tests of Assumptions and Lessons on Future Behavior from 100 Electric Vehicle Owners in California." University of California, Davis: Institute of Transportation Studies. Kurani, K. S., et al. (1994). "Demand for Electric Vehicles in Hybrid Households: An Exploratory Analysis." Transport Policy. 1:4, 244-56. Kurani, K. S., et al. (1996). "Testing Electric Vehicle Demand In 'Hybrid Households' Using A Reflexive Survey". Transportation Research D. v. 1 n.2. Lee-Gosselin, M. E. (1995). "The scope and potential of interactive stated response data collection methods." Presented at the Conference on Household Travel Surveys: New Concepts and Research Needs. Irvine, CA. March 12-15. Also in press in the Transportation Research Record. Morton, A., et al. (1978). "Incentives and acceptance of electric, hybrid and other alternative vehicles", Cambridge, MA: Arthur B. Little. Turrentine, T., et al. (1992a). Market Potential of Electric and Natural Gas Vehicles. Davis, California: Institute of Transportation Studies, University of California. UCD-ITS-RR92-8. Turrentine, T., et al. (1992b). A Study of Adaptive and Optimizing Behavior for Electric Vehicles based on Interactive Simulations Games and Revealed Behavior of Electric Vehicle Owners. Available from University of California, Davis: ITS-Davis RP-24-92. Turrentine, T. and K. S Kurani (1997). "Using reflexive methods to identify and estimate markets for novel transportation products: Household markets for electric vehicles." Submitted to Transportation.


69

U.S. Federal Highway Administration (1990). Summary of Trends. Washington, D.C., p. 14.

' Several authors have struggled with imposing a technological definition on NEVs. In fact we ourselves have previously defined NEVs as vehicles whose top speed capability restricts them from high speed routes such as freeways, highways, and expressways. Ultimately however, it is the household that defines whether a vehicle is an NEV through its use of the vehicle. ^ The U.S. National Highway Traffic Safety Administration has issued for comment a new low-speed vehicle definition. This definition states the maximum speed capability of vehicles in the new class cannot exceed 25 miles per hour.



RESPONSES TO NEW TRANSPORTATION ALTERNATIVES AND POLICIES: WORKSHOP REPORT

Martin E.H. Lee-Gosselin

OBJECTIVES AND ORGANISATION OF THE WORKSHOP This workshop was designed as a forum to discuss the implications for behavioural theory, survey methods and modelling, of major contemporary shifts in the demographic, social, technological and political contexts for personal travel. Of course, this was not the only workshop in the Conference to consider such implications: for example, another workshop focussed on recent telecommunications developments. Our focus was broad and the workshop benefited from an excellent mix of contributions from 17 participants representing eight countries, and covering response to these major shifts at different scales and from different perspectives. Indeed, the breadth of the theme motivated the commissioning of two resource papers (included in this volume). Peter Jones (Chapter 1) provided a substantive overview of contemporary new transportation alternatives and policies, and their implications for conceptual and analytical frameworks, models, data and evaluation methods. Andrew Daly (Chapter 2) reviewed the state of the art of the simulation of alternatives and policies, using "living" models for continuous planning. Eight workshop papers and two short communications were presented during the workshop. Several papers were challenges to rethink some underlying assumptions about how individuals go about setting their level and type of use of different modes. It is also interesting to note that

72


half of the papers touched on the possibility of reducing travel demand, which was also a major theme in the Jones resource paper. They fell into four groups, which were discussed in the following order: 1. Public and private perceptions of public and private modes 2. On reducing the need for (motorised) travel 3. Personal strategies for car-use 4. Balancing emerging needs of user segments After a discussion of scope and the expectations of participants, the workshop chose to examine recent developments and the priority research agenda around each of these themes in turn. The thrust of the workshop was slightly more substantive than methodological, but this distinction does not wholly capture the nature of the debate. A substantial part of the discussion concerned the theory of behavioural change, and the extent to which the wide range of shifts presented in the papers were addressable with our theoretical frameworks. The common threads that were sought for the two plenary reports formed the basis of this summary.

CONCEPTUAL CONSIDERATIONS Early in the workshop, the participants sought to set some limits and to agree on some working definitions and distinctions. These are best summarised as our responses to two questions. First, what are the legitimate roles of the travel behaviour researcher in the face of new alternatives and policies? The consensus favoured an essentially "humanistic" response and a reinforcement of the activity-based paradigm: to understand how changes in the transport system fit into people's lives, not how people's lives fit into changes in the transport system. For some, understanding response was enough; for others there was a clear intention to influence behaviour, notably around reducing motorised travel in pursuit of environmental and health benefits. There was general agreement that our role was at least to help decision-makers understand: • which transport changes are "in synch" with the population • possible adoption paths (so how enable, overcome barriers) • better ways to evaluate the impacts (anticipated and unanticipated) of new alternatives once adopted • how best to involve the customer at each stage of planned interventions

Workshop Report-Responses to New Transportation Alternatives

73

Secondly, what do we mean by "new"? We found it useful to distinguish between changes occurring to households and individuals and those occurring outside the home. Household and personal changes refer, in particular, to people undergoing important transitions in lifestyle, such as the formation of a new household, a birth or a shift in career. Outside changes include options, (e.g., a new cycleway, an expanded river crossing or a new airline route) or combinations of circumstances (e.g., a critical deterioration of regional air quality). Either may be accompanied by regulatory interventions such as pricing or parking restrictions, as well as knock-on effects such as land development or the marketing of new holiday packages. Another distinction concerned how the various elements of policy response are delivered: pull (creating/enabling alternatives) versus push (mostly regulation), although it was noted that, increasingly, these are combined. Finally, we observed that common to all new situations are decisions and responses that are made under more than usual uncertainty.

MOST IMPORTANT DEVELOPMENTS IN THE PAST 3-5 YEARS Within the broad scope described above, the participants identified the following trends which they considered to be among those that have most influenced travel behaviour in recent years (no order of effect implied): • An increasing number of countries are seriously concerned about the negative effects of motorized traffic growth, and in some countries (notably in Europe) there is the political will to consider trying to reduce it A rapid increase in motorisation or road transport in developing countries An increase in car ownership in countries with large populations and thus a large potential increase in the world fleet of motor vehicles An increased diversity in the motor vehicle fleet (some larger vehicles, some smaller) An increase in number of households, and reduction in average household size, in developed countries New market structures for dispersed activities, contributing to an increase in the average length of some types of journey In the USA, for the first time ftmds can be transferred from highway to transit budgets An increase in telecommunications and complementary hi-tech systems, such as Intelligent Transportation SystemsA'ransport Telematics The advent of successftil alternative car ownership schemes

74


RESEARCH AGENDA The participants chose to organise their insights from the papers and discussions under two rubrics, in both cases in the form of recommendations for the travel behaviour research agenda.

Facing Up to the Complexity of Behaviour Under Novel Situations The workshop cited a number of examples where the theoretical frameworks were insufficiently nuanced or flexible for the complexity inherent in conditions of more than usual uncertainty. Priority was given to five issues: The need to find out if preferences actually exist. The centrality of the notion of preference sometimes leads researchers to assume that a given preference exists, well understood by the holder and is "waiting to be observed". An example was cited from research on acceptance of novel gaseous- and electric-fuelled vehicles: in a market survey, respondents were expected to choose between different driving ranges (in terms of miles or kilometres) without any reference to methods and availability of refuelling, or any consideration of the respondent's travel patterns, hi this situation, revealing responses such as "I don't know - I guess I want what I have got now, whatever that is..." should not be dismissed as unusable, or simply coded as if the respondent had expressed the average range of a gasoline-powered vehicle. This example also reminds us of a more general caution about inertia: that reactions to one's "own" existing behavioural alternatives tend to be dominant when choice sets are evolving. Seeking a better understanding of lifestyles and their link to global attitudes (e.g. environmental concern). In the first of three issues around attitudes, it was felt that observed global attitudes may be taken too much at face value. In part, this is a problem of instrumentation: people with different types of education and knowledge may use terms differently, especially if they wish to appear socially responsible. A more meaningful approach would be to examine how people interpret what they actually do and how they live. An example of one attitude study was given in which a segment was identified that rated high on environmental concern, but in reality felt that they had "done their bit" by a single gesture buying a more fuel-efficient car; their observed travel behaviour was quite similar to groups less sensitive to the environment, and the potential for future behavioural shifts was much lower than their "green" global attitude implied.


75

Investigating how global attitudes interact with specific attitudes to transport options. Attitudes and opinions towards particular transport options are often treated in isolation from the motivational and cognitive factors that underlie the choices of adaptation strategies by households and individuals. In some cases, these factors simply go unobserved, in other cases phenomena such as social dilemmas may not be ftilly appreciated. Participants felt that travel behaviour researchers had not yet developed a adequate body of theory to link global and specific attitudes. Understanding the evolution of attitudes and values. There is a frequent assumption in transport that attitudes do not change over time, yet other fields such as health behaviour recognise stages of attitudinal change. Identify and benefit from key moments, seeking out 'ripenessft)rchange. In principle, much could be learnt under conditions of actual or incipient change. Of course, finding ways to identify segments that are "ripe for change" is recognised in some travel survey sampling strategies, both for efficient stratification and as an input to dynamic models. Once again, there was a sense that our field needs a theory of the mechanisms of response when people are faced with new options and circumstances.

Some Priorities for Improved Methodology The following eight issues received much of the attention of the workshop in the time available, although participants felt that a longer list of methodological priorities could be justified: Taking lessons from attitude research fi)r the application of Stated Prefr re nee (SP) methods. The literature on attitudes suggests that they explain, at best, 25% of the variance in most behavioural responses and possibly less in novel situations - should we expect any better from SP? It was necessary here to distinguish the use of SP to infer the utilities of attributes among relatively constrained choice sets, from the use of SP or any other technique to understand why people perceive attributes in different ways. For example, some people may show fear of tunnels, but to make sense of this it is necessary to find out what, according to them, constitutes a "tunnel". This led to seeking ways to observe a broader range of behavioural responses. This implies considering a wider range of Stated Response (SR) methods, with different levels of

76


involvement of respondents in the definition of possible behavioural outcomes and constraints. For example, choice experiments can draw on increasingly sophisticated SP methods, but can also include "reflexive" surveys in which respondents observe and record their own exploration of alternatives. Reflexive methods are feasible on representative samples, as Tom Turrentine demonstrated. At the same time, intensive interactive methods on small samples or in laboratory settings are sometimes the only feasible approach to explaining the complexity of behaviour in novel situations. Providing practitioners direct output from in-depth interactive methods. The tradition of our field is to use in-depth and qualitative methods to improve the specification of quantitative data collection and models. This is of great value, but there is also much value in direct outputs to policymakers from these methods. These include "narrative" reality checks on policy, and a better understanding of how and why people respond in novel situations, drawing on global and specific attitudes, biographies, and "parables" - the symbols and stories people use to explain and justify what they do or might do. It was suggested that SP-based model outputs could be combined with data from other SR methods and not just with revealed choice data. On a related point, the participants called for methodological research to compare outputs from different Stated Response methods applied to the same question. The sense here was not only to understand how the way you ask the question influences the response (see, for example, Staffan Widlert's memorable experiment with variants of SP), but also how to develop complementary information. Recommended use of natural and "real world" experiments to observe behaviour. Some dramatic events, such as earthquakes, have led to important insights about adaptation and spatial behaviour, but there is a need to find ways to get researchers and client agencies to see tracking the consequences of "ordinary" unanticipated shifts as part of the legitimate toolbox. With appropriate caution about novelty effects, a number of methods such as continuous "background" data collection (especially panel surveys) can help, including in situations of planned change where evaluation is weak or absent. The workshop also expressed the need to define segments in new ways. In part, this was to widen the scope of population segments which merit attention a priori, for example to understand the travel needs of children, women or the elderly. But it was also a recognition that practitioners could benefit from a more flexible "bottom up" approach, including segmentation by propensity to change.


77

Designing models around more realistic, more dynamic behavioural mechanisms. In part, this implies a shift to models, which incorporate understanding of decision processes. It also implies applying new sources of data on change to the best of existing models whose performance over a substantial period is known; part of the research agenda is knowing when extending a heavily vested model is no longer justified.

CONCLUSION The priorities identified above testified to a growing desire to for our field to understand adoption paths and barriers to new alternatives and policies, and not simply perform the forecasting of states. In some of our countries, the evaluation of potential alternatives has tended to be limited to narrow set of indicators, such as the value of travel time savings. This workshop argued in favour of drawing on a broader set of behavioural science methods to build an evidence base in which the whole promises to be much more than the sum of the parts.

ACKNOWLEDGEMENTS Lisa Buchanan and Liz Ampt (Steer Davies Gleeve) acted as workshop recorders. This report is based upon their notes and the written feedback from the workshop members, notably brief written commentaries on the first report of the workshop to the conference plenary, which was part of the workshop process. Participants included: J. Acha-Daza, L. Ampt, O. Adnan, L. Buchanan, A. Daly, B. Faivre D'Acier, A. Fujiwara, T. Garling, J. Garvill, P. Jones, K. Kurani, M. Lee-Gosselin, H. Lubis, Y. Shiftan, T. Turrentine, and T. Victorine.


SECTION 2 DYNAMICS AND ITS RESPONSE



DYNAMICS AND ITS: BEHAVIOURAL RESPONSES

TO INFORMATION AVAILABLE FROM AXIS

Reginald G. Golledge

INTRODUCTION The 21^^ Century promises some exciting changes in the way people do business and how they undertake travel. "Real-time" communications and interactions are a distinct characteristic of the technological society, which is quickly emerging at global, national, regional and local scales. One component of this future society that every day approaches reality, is the use of real-time telecommunications based traffic and transport system controls. The successful development of an Intelligent Transportation System (ITS) depends on the capability of incorporating a vast amount of information about the location of facilities which generate travel, realistic representation of elements of the transportation network on which travel occurs, and a knowledge of driver behaviours under different information or stimulus conditions. ITS's are being designed to utilize advanced communication and transportation technologies to achieve traffic efficiency and safety. Their goal is to encourage freely moving traffic, lower congestion, reduce stress and hazard for drivers and passengers, and allow the traveler to maximize access to information about the system and its operation so that intelligent driver decisions can be made about which parts of the system to use for which particular trip purpose at various times of the day. ITS components include Advanced Traveler Information Systems (ATIS), Automated Highway Systems (AHS), Advanced Traffic Management Systems (ATMS), Advanced Vehicle Control Systems (AVCS), and Advanced Public Transportation Systems (APTS).

82


AXIS, ATMS and AVCS are generally considered to be components of In-Vehicle Navigation Systems (IVNS). IVNS uses advanced information and communication technology to manage traffic, advise drivers, and control vehicle flow. AXIS is targeted to assist drivers in trip planning destination selection, congestion avoidance, selection of departure times, route choice, and to assist navigation. Chen and Mahmassani (1993) point out that four classes of AXIS are often recognized, ranging from class 0 (open, loop static system) to Class 4 (closedloop dynamic system, enabling 2-way communication between driver and traffic control center). Obviously an IXS must be muhifaceted and be able to integrate all these components listed above. Xhe optimal result is informed intelligent driver decision making within an information framework that is as close to reality a possible, and in as real a time as possible. Xhus IXS is designed as a dynamic system in which reaction to temporary or permanent system shocks can be facilitated in real time. Xransportation science has the expressed goal of increasing accessibility for all groups of people as they travel through their daily environments. Xo achieve this goal, there must be significant contributions to research on transportation system architecture, technology development, policy formation, and operational tests of various systems which represent the components of IXS. In this chapter I focus mostly on one aspect of the dynamics of the transportation systems—Advanced Xraveler Information Systems (AXIS). Xo deal with this topic comprehensively we have to examine both the demand side and the supply side of transportation modeling. In recent decades, there has been a paradigmatic shift in transportation needs, with driver decision making and behaviour receiving much attention and complementing the network and system oriented research that focused on route modification or construction. In this newer approach, demand side modeling has dominated, particularly over the last two decades, and a hoard of travel behaviour models based on various logit and stochastic formats have been offered and tested, many of them with considerable success (for recent reviews see Rosenbloom, 1978; Stopher et al, 1996; Institute Xransport Engineers, 1989; Mahmassani et al, 1996, Fan et al., 1992). Xhe goal of demand-side models is to better distribute traffic both spatially and temporally, on the existing road network. Currently, some of the most promising results for understanding how people behave within a transportation system have derived from work on the activity based approach (Kitamura, 1987; Stopher and Lee-Gosselin, 1997). But, despite the fact that this demand side of the modeling equation has expanded by creating new and better data sources (e.g. from survey and panel data (Axhausen and Garling, 1992) or from micro simulations (e.g., Adler, Recker and McNally, 1992 A, B, C;

Dynamics and ITS: Behavioural Responses to Information

83

1993), and as a greater understanding of the travel decision process has been achieved, there has been a tendency to somewhat neglect the task of updating the supply side both in terms of estimating how driver responses to AXIS affect the provision of available route space in networks, and in terms of pursuing the task of selecting route-based data models that conform more to user-cognitions of routes and networks. For example, many behavioural travel models are still tied to simple planar network representations of road systems. Their deficiencies are well known (Goodchild, 1998), and will be explored later. On the supply side there is an emerging need to capture the representation of a transportation system in a way that reflects how people perceive it and consequently use it (i.e. to extend the demand side behavioural approach to include a network representation mode consistent with human perceptions and cognition of network structure and its connectivities, and to expand the supply side by creating representations of road systems using database formats that are consistent with the way people view them). Specifically, humans tend to view environments as collections of objects, so object-oriented representations of networks may prove desirable and useful. To this extent different types of data models are being used, the two most common being the node-link model (e.g. in ARCINFO/NETWORK) and the object-oriented model (e.g. using object-oriented software such as VERSANT for a suitable database such as ETAK's MAPINFO). In this chapter I present comments on both demand and supply areas of behavioural travel modeling.

BACKGROUND The past decades have seen a paradigm shift in transportation planning from the construction of new infrastructure (supply) to the more effective management of travel demand. Part of the reason for this shift was the recognition that building new highways was only a temporary measure to relieve movement problems such as congestion. The shift to travel demand management as a significant traffic control strategy has consolidated in the last decade. Both the spatial and the temporal dimensions of travel behaviour are being examined and increasing emphasis has been placed on flex-time working hours, telecommuting, and an increasing concern with the in-car dynamic reception and use of Advanced Traveler Information Systems (ATIS). These measures are designed to facilitate movement through existing systems by: a.) reducing travel demand through the suppression and selective elimination of trips; b.) targeting single occupant vehicles at peak period commuting times, and reducing traffic volume on key links in the period of peak vehicle flow; c.) reducing driver frustration and stress along with affecting traffic flow by providing timely in-car, en-route or pre-travel information about

84


hazards such as congestion, construction, or accidents; and d.) examining in more detail the activity patterns of individuals and households to more completely understand features such as the allocation of resources (e.g. vehicles) among household members, the timing of household activities, and the significance of multipurpose and multi-stop trips (trip chaining) in the episodic activity behaviour of household members. In particular, the latter trend has attempted to treat travel behaviour in more realistic terms: this has required a search both for new data that is being produced either by survey and panel research or by travel simulators, and new types of travel demand models (see Mahmassani et al. 1996; Stopher, 1996; Jones 1990; van Aerdeetal., 1996).

FORMS AND S O U R C E S

OF TRAVELER INFORMATION

More than three decades of research in behavioural travel modeling has explored different ways to collect information about traveler behaviour. Questionnaire surveys distributed to drivers at selected places along routes provided the first comprehensive databases for the Chicago Area Transportation Study (CATS), Pittsburgh Area Transportation Study (PATS) and many others. Such surveys collected data on personal and family characteristics of drivers and passengers, demographics, trip types, mode of travel, trip purpose, and trip frequency. The data was usually generated for traffic zones rather than individual locations, and graphic "desire-line" maps of origin and destination flows at different temporal intervals, provided a strong visual description of traffic flow (although somewhat simplified as crow-fly distances rather than flows related to specific routes were used). As survey research itself began to be established more as a procedural science, and as sampling procedures became better established, surveys became more robust and comprehensive. While mechanized traffic counts continued to provide a measure of gross vehicular movement at specific locations within a transportation network, surveying became the dominant means for providing information about the traveler, travel preferences, route selection procedures, reasons for destination choice, and so on, which provide the bases of much of the data collected about travel behaviour and travel activities today (Brog et al., 1985). These in-car, on-the-spot, or mail surveys have been supplemented by a variety of alternate but complementary ways of collecting information about travel and travelers. Extremely detailed travel diaries kept by a specially commissioned sample group and focused on specific aspects of urban travel (e.g., commuting, shopping, recreational activities, etc.) began to dominate the data collection process. A comprehensive overview of these approaches were collected in Ampt, Richardson and Brog (1985) in which a range of "new" survey methods in transportation were reviewed, along with a discussion of sampling procedures and analytical techniques. The result provided an important reference source for


85

collection of detailed transportation data. Examples of data sets based on travel diaries that have yielded significant insights into travel and traveler behaviour include the Swedish Travel Survey (Hanson and Hanson, 1981) and more recently, the Portland Travel Survey. A cursory view of the journals Transportation Research or Transportation Science over the last decade shows an increasing concern with data reliability and validity and consequently detailed discussions of the data collection procedures used for specific purposes, (e.g.. Pas and Koppelman (1986); Kitamura and Van der Horn (1987); Taylor, Young, Wigan, and Ogden, (1992); Duncan, (1987); Mahmassani, Joseph and Jou (1992)). For example, Mahmassani et al. (1992) attempted to capture the day-to-day dynamics of user behaviour in a commuting context using a two-stage survey of commuting habits in the North Dallas transport corridor. This involved first distributing a brief two-sided questionnaire to 13,000 households in the selected area, followed by a more detailed activity diary for a selected portion of this sample which was for recording commuting trips from home and returning to home. Data were also collected on trip chaining, departure time and path choice. Such two-stage procedures have proven to be cost-effective ways of collecting large quantities of data. Given this emphasis on travel demand as a derived demand our understanding of the complex travel behaviours associated with movement in cities has expanded. It has been suggested that an activity-based approach to travel behaviour requires an understanding of human decision making processes associated with travel (Brog and Erl, 1981). This is a substantially different direction from that previously adopted for the development of the "four stage procedure" of transport modeling where statistical associations, rather than behavioural relationships, were the primary concern in model development. As well as concentrating on human decision making and choice processes, the activity approach recognizes the essential role of capacity, coupling and authority constraints (Hagerstrand, 1970; Lenntorp, 1976) on the trip-making behaviour of households. Advances that have resulted from the activity-based approach include data on perceptions, attitudes, and stated preferences, along with time use and episodic interval information that both supplement and complement existing revealed preference and statistical estimation methods (Stopher and Lee-Gosselin, 1996). Of course a significant component of the disaggregate and activity-based domain has been due to the considerable advances in computational capabilities and the innovative methodologies such as Computational Process Models (e.g. Garling et al. 1994; Liesser and Zilbershatz, 1988; Kwan, 1995; Recker, Root and McNally, 1984) and Dynamic Microsimulation models that have been developed specifically for Intelligent Transportation Systems (e.g., Adler et al., 1992 a, b, 1993; Koutsopoulos et al., 1194; 1995; Chen and Mahmassani, 1993).

86


Apart from working within the idea of constrained activity spaces, this research began emphasizing behavioural changes or behavioural dynamics represented by human decisionmaking and choice behaviour when confronted with changes in the travel environment. These changes could vary from the process of switching between driving alone and carpooling to work, to the more real-time adjustment of changing destinations, changing routes, substituting destinations, changing the time scale at which activities are undertaken, deleting and delaying activities as a response to information about changing travel environments or delaying departure times. Other travel behaviour features that have come under investigation include trip chaining, scheduling of activities over a time span rather than a single time, substituting out-ofhome for in home activities (such as might be the case with two-person-working households who begin dining out, instead of eating at home in the evenings), and an emphasis on household members' life cycle stages (Mahmassani and Herman, 1990; Stopher, 1997; Garling and Hirtle, 1990; Janelle et al. 1998). One of the characteristics of the activity approach is that it extends interest in what is going on beyond the physical nature of the trip itself. Individual cognitions (e.g., cognitive maps), person-to-person and among-household relationships, and other coupling phenomena (e.g., working out with a friend, sharing rides to a transit terminal) as well as phenomena such as variability in path selection criteria for different trip purposes, all come into play in the attempt to understand the reasons behind movement. This detailed personal and small group knowledge is often extremely difficult to obtain and to code and process (Janelle et al., 1998). Nevertheless, there is substantial evidence that the benefits associated with adopting the methods, the ideas, and the concerns of this general behavioural approach appear to more than compensate for those difficulties, particularly by giving increased knowledge of the nature and structure of decision and choice processes of travelers. Transportation researchers who use activity approaches usually differentiate between those that are routinized and those that result from deliberate choice. For example, routinized trips (spatial habits) imply that people tend to use the same mode for each trip, to leave their home or work at approximately the same time, to arrive at a work or other destination at approximately the same time, to follow the same route on consecutive trips and to minimize active decision making while in the process of trip making. Such trips often include commuting (work trips), trips for religious purposes, and trips for various professional reasons including those that are medical or health related. Even though the times at which these activities take place may differ substantially, as will their episodic frequencies, the tendency is still to expect that a routinized procedure will be followed. Other trips (such as food shopping at a major supermarket) may be routinized to a lesser extent in the sense that such a visit may be routinely


87

included as part of a trip chain from work to home depending on the number of working persons in that household. In the latter case, specific path segments may still be followed on a regular basis. But trip making on any specific day can also be conceptualized as a process of deliberate choice. Travel plans may be developed prior to the initial trip from home in which a preliminary scheduling of activities is set up (Garling and Golledge, 1989, 1993; Axhausen and Garling, 1992). However, as the dynamics of daily living have to be accounted for, including both events that are beyond the traveler's control (e.g. congestion and hazards), and those that are within the traveler's control (e.g. departure time and destination choice), routinized patterns may break down and dynamic decision making may have to be invoked. The result is that somewhat different paths may be selected on the "to" and "fro" segments of a trip; trips may be rescheduled and reordered in terms of a single day or multiple day events; and low priority scheduled events may be eliminated as time taken to complete higher priority events exceeds expectations (e.g., because of congestion, parking problems, and so on). Some trip purposes (e.g. social or recreational trips to meet with friends, or for dining away from home), and many business trips, may be rescheduled with different episodic intervals or frequencies, different lengths or durations, different destinations, different priorities, different behaviour sequences, different path selection criteria, and different probabilities of being conducted as either single purpose or multi-purpose (i.e., chained) trips. Modeling routinized choices has achieved considerable success using discrete choice models, dynamic Markov models, and variations of spatial interaction models (Stopher and LeeGosselin, 1997). Less success has traditionally accrued when trying to model behaviours resulting from deliberated choice. The conventional model bases of routinized aggregate behaviours (e.g., spatial interaction or entropy models, optimized network travel behaviour models, and so on) have been supplemented by compositional and decompositional choice models, compositional and decompositional preference models, models of variety seeking behaviour, computational process models, and dynamic microsimulation models (for a review see Golledge and Timmermans, 1988; Timmermans and Golledge, 1990). Much of the recent work in travel behaviour modeling has been pursuing the difficult task (originally identified by Root and Recker, 1983) of combining destination choice, departure time, waiting time, frequency and duration of activity participation, activity mix, activity priorities, and scheduling, all within a single conceptual framework. All these decisions were

88


seen to be interrelated and partly dependent on each other. Root and Recker suggested that by focusing on the scheduling process and developing a model that was reactive to changes in choices as environmental conditions changed, a more realistic and more effective examination of travel behaviour would take place. These suggestions spawned the development of computational process models, which consisted of sets of interacting computer programs capable of relating elements of real and cognized environments with factors influencing choice of destination, household preferences for scheduling activity sequences, and a selection of coupling, capability, and authority constraints (Recker, Root, and McNally, 1984 A, B; Garling et al. 1989; Miller, 1990; Axhausen & Garling, 1992). Much of this work, however, continued to rely on utility maximizing assumptions. Recent developments of behavioural travel models have tended to use a boundedly rational or satisficing context (Mahmassani and Chang, 1988; Supemak, 1992). Computational process models (CPM) allowed the researchers to focus on and include interdependent choices. This was facilitated by developing cognitively based models that allowed differential acquisition, storage, and retrieval of information and recognition that tradeoffs would have to take place between the accuracy of recalled information about environmental features (e.g., locations, hours of business, and remembered paths) particularly in terms of effort expended and expected satisfaction. Operational CPMs (e.g., SCHEDULER, Garling, Kwan and Golledge, 1994; GISICAS, Kwan, 1995) are productions systems implemented as computer programs. They offer a testbed for assessing the consequences of different policy measures, or as mechanisms for facilitating the development of different hypotheses about travel behaviour, particularly with regard to intermittent traffic blockages. They also allow one to use a variety of behavioural assumptions in order to determine most probable outcomes.

DRIVER

SIMULATORS AND COMPUTATIONAL PROCESS MODELS

As the potential for ATIS to influence driver behaviour and traffic conditions in a network has become more obvious, a number of research efforts have focused on examining the impacts of real-time traffic condition information on dynamic driver behaviour. Because few ATIS systems have been implemented in the real world, much research has concentrated on using computer-based interactive simulation rather than working with real-time field studies. Driver simulation procedures differ from the more classic revealed preference studies in that, while the former require individuals to answer hypothetical questions about technologies they have yet to experience, in a driver simulator subjects are given the opportunity to experience


89

different traffic or information scenarios such that their decision and choice processes are revealed by the consequent actions they take. Those examining driver behaviour under AXIS conditions in recent years include Bonsall and Parry (1991), Ayland and Bright (1991), Koutsopoulos et al, (1994), Chen and Mahmassani (1993), Vaughn et al. (1993), and Adler, Recker and McNally (1992 A, B: 1993; Hu et al., 1993). Adler et al. (1992 A, B: 1993) also have developed an interactive computer-based simulator named FASTCARS. Its purpose is to gather data for estimating and calibrating predictive models of driver behaviour under conditions of real-time information. It was written in Turbo Pascal, and designed to run on a 386-series PC running at least 33 megahertz and equipped with VGA graphics and a voice adapter. The authors claim that it is not a pure driving simulator, but rather simulates real-time travel decision-making conditions. It presents a series of possible environments tied to the experience that a subject has with the basic environment and studies temporal and spatial factors such as perceptions of speed and volume, time lapse, network familiarity, information acquisition, and travel goal specification and evaluation. The simulator encompasses the entire driving process from pre-trip planning to arrival at the destination. During the trip, players are required to make a range of choices, including specification of goals, rerouting where necessary, changing lanes, and making decisions as to whether or not to use specific information technologies, such as in-car guidance or advisory signage. Pre-trip planning involves selection of departure time and initial route choice. In addition, travel objectives for each trip must be specified. During post-trip debriefing, subjects evaluate their success in meeting their pre-trip goals. FASTCARS uses a node-link network model. Travel takes place on a link-by-link basis, ignoring system-wide traffic and focusing on traffic around the player. Traffic conditions are displayed visually on a screen and include a network viewer, a control panel, roadside information viewer, and the in-vehicle navigator. In the network viewer, players are provided with a bird's-eye view of a one-mile stretch of a road section with cars displayed as small rectangles moving in lanes. The driver's vehicle is shown as a solid bracketed rectangle. Each lane has a specific speed with speed increasing as one moves to the left. Players control lane changing and road changing via keystrokes. At crossstreets turns are indicated by arrows at either end of the street and the name of street and turning direction is indicated. All turns from freeways are made from the right-hand lane, but on surface streets, the lane closest to the turn direction is selected. Surface streets are distinguishable by traffic signals and lower speeds. Specific cycles in the traffic signals control the delay period at intersections. FASTCARS offers three types of ATIS: variable message signs (VMS), highway advisory radio (HAR) and in-vehicle shortest time navigation systems (IVNS). VMS are displayed at

90


specific freeway locations and provide information on local traffic conditions on the current link. Information is then given to the driver about the condition of the next link. HAR allows players to activate prerecorded radio messages containing relevant information on highway conditions and on the availability and accessibility of alternate routes if a diversion appears to be necessary. IVNS offer the driver a source for finding the shortest path to their destination. No information on traffic conditions are given, simply directions as to when to turn and what streets to follow. The critical decision made by the driver is whether to accept the FVNS guidance or not. Information provided includes the next intersection or freeway exit, expected shortest travel time to the destination, and distance from the current location to the destination via shortest time path. Researchers at MIT have also developed an interactive simulator to facilitate data collection and calibration of a route choice model. Their simulator uses fuzzy set theory, fuzzy control, and approximate reasoning (Koutsopoulos, et al., 1993). The model is loosely based on a previous study of the dynamics of driver behaviour under conditions of provision of real-time information (IGOR: Interactive Guidance On Routes, Bonsall and Parry, 1991). IGOR simulates en-route travel through a network and emulates an in-vehicle navigation system to provide players with real-time route guidance so that drivers' compliance with guidance advice can be evaluated. The quality of advice is manipulated, and the relationship between advice quality and advice acceptance was determined. The Koutsopoulos et al. (1994) model enhances the use of the interface, allows for modeling different operating conditions, improves the information provision capabilities available to the driver, and accounts for the driving task. A graphic display shows a car moving through a network and information is presented through a roadside display/broadcasting system as well as a graphical in-vehicle information window. Chen and Mahmassani (1993) also developed a simulator that integrates a traffic simulator program and offers the capability for multiple driver participants (DYNASMART). This models pre-trip planning, en-route travel, and post-trip evaluation. Pre-trip planning involves selection of a departure time and a path. At the selected departure time, players see a display of the network with expected travel time for specific routes. An option is provided to allow departure on time or to delay the trip. The initial route is selected at the time of departure. The explicit purpose behind building this simulator was to examine the behavioural processes underlying commuter decisions on route diversions including en-route and day-to-day departure time and route choices as influenced by the provision of real-time traffic information. Three components are visually displayed by this simulator in en-route conditions: a network illustrator, a legend window for explaining color codings, and a real-time message display. Real-time updates of vehicle location are provided and where turn decisions have to be made.


91

this information set is analyzed to determine whether the current route will be continued or an alternate path selected. The emphasis in the Texas simulator is on investigating day-to-day adjustments. It allows manipulation of departure times, offers a capability for real-time interaction with and among multiple driver participants within a traffic network, and considers both system performance as influenced by driver response to real-time traffic information and driver behaviour as influenced by real-time traffic information. The Texas simulator actually simulates traffic conditions, for its engine is a traffic flow simulator and ATIS information generator that displays information consistent with the processes actually taking place in the simulated traffic system. This dynamic approach allows the researcher to investigate day-to-day evolution of individual decisions under different information strategies. Thus, driver learning behaviour is allowed; this provides a longer-term dimension to simulation of driver behaviour than is possible with other models. Thus, in the Texas situation, subjects and simulator may combine in a loop whereby trip makers may revise their decisions from one iteration day to the next (i.e., spatial learning is allowed). The authors suggest that in this way it would be possible to examine convergence to an equilibrium, stability and evaluation of benefits following shifts in user trip-timing decisions. In particular Chen and Mahmassani (1993) conducted experiments in three categories: a) pre-trip and en-route path selection only; b) pre-trip departure time and path choice and en-route path selection; and c) pre-trip departure time and path choice, real-time departure time adjustments and en-route path selection. Subjects were required to "drive" a vehicle from a peripheral location to the central business district through a network corridor. Subjects were provided with real-time traffic information before each trip and on the basis of this information selected a departure time and a specific path. These were fed into the simulator and path assignment model. Vehicles are then moved along the selected path according to prevailing traffic conditions on the link the vehicle occupies at some particular time. Real-time traffic information is provided at nodes where a switch to an alternative route is possible and the decision is made whether to stay on the current path or make such a switch. At the end of the trip feedback is offered on the consequences of decisions, and provide the basic information for new decisions concerning the next day's trip. The simulation was written in FORTRAN and run on an IBM RISC system/6000 acting as the host computer. The critical features of the Texas system are that: a) the simulator has multi-user capabilities; b) it is dynamic in that all user responses directly influence prevailing traffic conditions; and c) it can be run in real-time such that every simulation time step conforms to the speed of the host computer's clock. Two alternative rules are available in the user decision component for both en-route path switching and initial route selection: a deterministic choice rule and a boundedly rational rule. Under a deterministic rule the simulated commuter will always select the best path in terms of least cost

92


or least travel time from the current position to the destination. An alternative boundedly rational rule was operationalized by Mahmassani and Jayakrishnan (1991) and assumes that a driver will switch from a current path to the best alternative only if the improvement in the remaining trip time exceeds some threshold expressed either in absolute terms or relative to the remaining trip time. Background traffic is considered to be simulated traffic that interacts with the participants' vehicles in the same corridor network. There were 10,800 simulated vehicles, some of which were not designed to switch routes because they were not equipped with traffic advisory units or their drivers were assumed not to rely on real-time information. Using different market penetration scenarios, different quantities of vehicles were equipped with the ability to access information en route and consequently to switch routes according to the proscribed behaviour rules. Using driver simulators then provides data that acts as a basis for the development of user response models that in turn influence simulation assignment tools and their evaluation of network performance under conditions of real-time information provision (Mahmassani, 1996). Although currently state-of-the-art in terms of the types of information that can be obtained from driver responses to changing real-time traffic conditions, the various simulation experiments are not intended to totally replace actual field demonstrations and tests, but to provide information on what conceivably may happen as different traffic condition and information flow parameters are manipulated. Solving such issues is critical to the further development of IVHS technologies. Adler et al. (1993) and Adler (1997) suggest that real-world implementation of ATIS will involve multiple media formats which are both auditory and visual as well as varied message contents, route guidance and traffic condition information, and information display formats. Some information may be most suited to roadside posting and be passively available to all drivers even without appropriately modified in-car vehicle guidance systems. VMS, for example, would be available to most drivers but would require active acquisition via advisory radio signals. This in turn requires a deliberate effort by drivers to tune to the advisory radio station. Yet other information will be available only to a subset of drivers who will pay extra for this service but also have to make an active decision as to when to acquire it. The cost of an in-vehicle guidance system for general consumption is still being explored and simulation experiments are still being undertaken to determine drivers' willingness to both acquire and use ATIS of different complexity. The Adler, Recker and McNally implementation of FASTCARS (Freeway and Arterial Street Traffic Conflict Arousal and Resolution Simulator) was specifically designed to address this question.


93

For diversion from a pre-selected route to occur, Adler et al. (1993) found several significant variables, including: (1) perceived travel speed—a sharp decrease in travel speed increases frustration and anxiety levels and is often a precursor to diversion; (2) average link speed—as these averages decrease diversion behaviour becomes more probable; (3) VMS signs on freeway links broadcasting messages about severe congestion downstream also trigger diversion behaviour. As well as these link-specific variables of speed and VMS, other factors here noted including experience with the current path vs. experience with other paths, and road type (with most diversions being made from freeway segments to other freeways or arterials). The number of diversions already undertaken also appeared to influence the probability of making a new diversion, as did the magnitude of the remaining distance to destination and the road type that constituted the current link of the path being followed. Adler et al. (1993) show that the benefits of using an IVNS were most realized by low-familiarity drivers who generally made poor initial route selections and used inefficient en-route diversion strategies. One fiirther critical result from this study was the conclusion that drivers who are most familiar with the network and the types of conditions that might be experienced on any given layout segment (e.g., commuters) were better able to anticipate and react to normative congestion and were less likely to rely on real-time route guidance information. They indicated higher preferences towards HAR over IVNS. Route guidance information was found to be more significant for drivers with lower familiarity profiles. Some significant uses of travel behaviour simulators include analysis of commuter behaviour for investigating diversion tendencies and information acquisition (Jou and Mahmassani 1996). Longitudinal studies of driver behaviour under recurrent congestion conditions (Koutsopoulos and Lotan, 1990) emphasis on pre-trip information processing using ATIS to help determine departure time and initial route selection (Mahmassani, 1996, Mahmassani, Hatcher and Caplice 1996: Mahmassani and Chen, 1991) and continuing efforts to make HAR systems true real-time procedures without having to rely on pre-recorded messages (Catling and McQueen, 1991; Kawashima, 1991, Rillings and Betsold, 1991; Koutsopoulos and Lotan 1990). Similarly, work has been proceeding on evaluating the most appropriate form for the in-car guidance system (e.g., Parkes, Ashby, and Fairclough, 1991; Walker, Alicandri, Sydney and Roberts, 1990; Haselkom, Spyridakis and Barfield, 1991; Allen, Stein, Rosenthal, Ziedman, Torres, and Halati, 1991; Rillings, and Betsold, 1991). Adler et al. (1993) also suggest the concept of multiobjective travel planning as an area that deserves greater focus. Multiobjective travel planning has in part been examined using computational process modeling (CPM). For example, Garling et al., (1994) and Kwan (1995) have presented related

94


models (SCHEDULER and GISICAS) that model the scheduled trip behaviour of household members over specific time-periods (e.g., 24 hours). Using an a priori scheduled set of activities, which throughout the day are influenced by changing traffic conditions such as congestion and travel delays, they show how the initial priorities set on different trip purposes can result in adjustments such as rerouting, activity deletion, activity delay, destination substitution, and activity rescheduling.

PRE-TRIP BEHAVIOUR While many repetitive trips (e.g, the commuting trip to work) are determined after limited experimentation and become more or less habitual in nature, encouraging stereotyped behaviour, other trip purposes may require more activity in the trip preplanning phase— especially things such as the time of day in which the trip should be undertaken, departure times, probable length of trip, path selection criteria, the route to be followed, and expectations associated with this pre-trip travel planning activity (Huff and Hanson, 1989; 1990). Axhausen (1992) and Garling and Golledge (1989) have emphasized the importance of access to relevant information in the pre-trip planning phase. Jou and Mahmassani (1996) have used extensive diary surveys of actual commuter behaviour in two different environments (the North central expressway corridor in Dallas and the Northwest corridor in Austin), to examine the day-to-day dynamics of commuter decisions. These decisions include selection of departure time, and selection of the route to be followed for both the morning and evening commutes. They then relate pre-travel decision making concerning route selection, departure time, and switching patterns, to commuters' socioeconomic characteristics, workplace conditions, and traffic system characteristics. They found remarkable similarity in the general commuting behaviour patterns observed in the two cities. Significant results included evidence of greater switching activity in the evening commute and, relatively speaking, a greater frequency of departure time switching relative to route changes. Although, as they expected, there would be some differences between the two samples because of the larger size of Dallas relative to Austin, results did indicate that behaviours were similar and that model parameters consequently were feasibly transferable from one city to the other. Much of the recent work on the influence of pre-trip activities on travel behaviour has been undertaken at the Texas Transportation Center. For example, Mahmassani and Herman (1990) have reviewed the evolution of these approaches from theoretical microeconomics-based analyses of idealized situations to elaborate simulation studies and observational work based on interactive laboratory experiments. Their own experiments (Mahmassani and Chang, 1985,


95

1987; Jou and Mahmassani, 1996; and Jou, Mahmassani and Joseph, 1992) have examined commuter choices of route and departure time as principal dimensions of commuter response to congestion information and as strategies for mitigating travel behaviour in the face of this information. Their study combines "day-to-day" and "deviation from normal" approaches to look at potential switching behaviour. They found that commuters tend to change departure times, routes, or both, more frequently in the morning than in the evening, possibly a reflection of constrained arrival time at the workplace compared to a flexible arrival time at the home. They also found that departure time switching was a more common phenomenon than route switching for commuters. In particular they found that trip chaining patterns were significant in commuters' decisions to switch either departure times or planned routes and that workplace characteristics (e.g., parking availability) and traffic system conditions (e.g., anticipations or expectations of congestions at certain places) were also significantly influential. An interesting question that Jou and Mahmassani (1996) raise is that, since switching of departure times and routes from those originally pre-planned was common already, how much more switching will take place if a valid and reliable real-time information system was available prior to making these decisions? Much of this work and other research related to departure time and day-to-day dynamics of commuter decision-making are summarized in recent chapters by Mahmassani (1996, 1997).

CHOICE RULES AND STRATEGIES Mahmassani (1996) has suggested that the behaviour of repetitive travelers is guided by simple heuristic strategies and a limited set of mental choice rules. As such, it has been necessary to depart somewhat from the formal utility maximizing paradigm, replacing this rigid constraint with the more flexible one that travelers act in a boundedly rational manner by searching for an "acceptable" outcome. In previous work, (Mahmassani and Chang 1985, 1987) and in work by Supemak (1992) evidence is gathered from psychological and behavioural decision theory literature to show that boundedly rational processes were more realistic and reliable predictors of travel behaviour than were predictions made on the basis of utility maximization. Mahmassani (1996) suggests that the first strategy concerns the willingness of a commuter to change their latest choice of route, departure time, or both, and that these choices are made conditional on each other. A second set of influences concern the degree to which a potential traveler is familiar with route and traffic conditions, and the extent to which exogenous information is likely to be

96


accepted. Important concepts include the traveler's preferred arrival time at a destination which are known to be dependent on attitudes towards risk as well as conditions in the workplace itself (e.g., traffic volume, congestion, parking availability). The boundedly rational character of the decision process is operationalized via a satisficing rule. This specifies that the user does not change departure time if the schedule delay (on day 1) is within a user-specified indifference band or tolerance band. The limits of this tolerance band represent the earliest acceptable arrival times. Information about trip delay and experienced congestion, influence the location of the upper and lower limits of the indifference band. Route selection requires another set of rules. As Garling and Hirtle (1995) have indicated, many trips may be regarded and modeled as a set of choices in which local decisions relating to route switching or selection of route segments to complete a trip chain, may result in behaviours that are other than globally optimal. For example, a traveler's schedule on a particular day may involve a trip to the workplace, a trip from workplace to another destination (e.g., lunch), a return to work trip, a trip to recreation or social activities from work, a trip from recreation/social to shopping and a trip from shopping to home in the evening. On different segments of such a trip, criteria for path selection may change from minimizing time or distance to maximizing aesthetics, maximizing use of freeway segments, or restricting travel to arterial or local streets such that signalization at intersections is minimized. The result may be a trip that is perfectly acceptable and satisfactory for the traveler but which significantly exceeds traditional time, cost or distance minimization rules. In general it is assumed that the purpose of providing exogenous information to travelers en route is to help them optimize route selection and overall to optimize system performance. Experimentation with different choice rules has only recently begun (e.g., Mahmassani, 1996). There appears to be considerable potential, both in the use of driver simulators and the use of Geographical Information Systems (CIS) to explore how different decision rules, when implemented, impact features such as departure time and path selection. As the traffic system evolves throughout the day, travel plans developed prior to initial departure, may have to be modified or altered. These changes invariably are related to the type, amount, time and reliability of information that can be accessed by the potential traveler.


97

RECENT GIS DEVELOPMENTS Traditional travel demand analyses face two obstacles. The first is that the four major travel components (trip generation, trip distribution, modal split and network assignment), are solved in a sequential manner which may result in inconsistencies and nonconvergence. Next, the data required are often complex and difficult to manage. However, recent advances in formal methods for network equilibrium-based travel demand modeling computational platforms with spatial data handling may help to overcome these obstacles. Miller and Storm (1997) offer a prototype Geographic Information System (GIS) design to support network equilibrium-based travel demand models, which includes a) realistic representation of the multimodal transportation network; b) increased likelihood of database integrity after updates; c) effective user interfaces; and d) efficient implementation of network equilibrium solution algorithms. For example Boyce, Zhang, and Lupa (1994) and Sheppard (1995) show that a sequential approach to the modeling of the four dominant components of transportation planning, can generate inconsistencies and nonconvergence amongst the components, even when feedback loops and re-estimation procedures are used. Data required for an urban-scale travel demand analysis is often considerable, and as Shaw (1993) points out, they require an origin and destination zonal system, aggregate travel demand attributes for each zone, disaggregate travel demand data from surveys or diaries, and a separate transportation network representation for each travel mode considered in the analysis. Recently, equilibrium-based travel demand models have extended the theory embedded in the sequential approach to provide consistent estimates with reasonable computational times even for extremely large urban areas. The models provide both static and time-dependent (dynamic) estimation, facilitating their use in infrastructure planning and policy analysis. Within the context of developing intelligent transportation systems and transportation environmental assessments, recent developments in Geographic Information Systems have facilitated model accessibility, database maintenance and updating, and graphic or cartographic display of model results. Perhaps the greatest potential use of GIS in an ITS context is that it provides opportunities for developing different testbeds from which to run ATIS/ATMS models. For example, it has been suggested that object oriented database models and management systems might prove to be the most amenable to testing of ATIS/ATMS because of the feasibility of representing a transport network in a manner similar to the way drivers perceive network systems. Kwan and Hong (1998) suggest that by using a lane-based object oriented database of a transportation network it is possible to use a greater variety of route selection criteria (e.g., both link related "shortest xxx" and node related ("minimize xxx turns") while at the same time allowing an ITS operator the possibility of designating specific lanes at different levels of road hierarchy as specialty

98


traffic lanes for any time period (e.g., HOV, or transit lanes). Kwan and Hong also suggested using GIS functions such as corridoring and buffering to reduce the number of places that may be considered as alternative destinations given that destination substitution was required because of unexpected traffic conditions and delays as might occur through congestion, accidents, or construction. Summers and Southworth (1998) have designed a testbed to assess alternative traveler behaviour models within an ITS system architecture. Their testbed included an ITS architecture simulator, a module for constructing ITS-sensitive travel behaviour models, and a simulation support environment for generating synthetic traveler populations. A specialized, objectoriented development environment provided a framework for the economic development of the testbed's various software components. The ITS testbed they proposed included both a spatial database and the modeling tools to support the analysis of ITS-impacted travel behaviours. Summers and Southworth argue that the application of ITS in the future does not imply that we can build a single or indeed a set of travel behaviour models and try them out in the real world. Like others who have previously focused on driver simulation models, they also suggest that a laboratory modeling system designed specifically to minimize the cost of developing and redeveloping alternative models and performing experiments with them is highly desirable. They suggest that an appropriate modeling framework would include: • a module containing a flexible set of traveler behaviour models • an ITS deployment simulator module capable of representing variously deployed regional transportation system infrastructures • a simulation support environment capable of generating representative traveler populations in different travel scenarios • a system for displaying and summarizing the results of each of the different modeling processes • a software development environment that provides a framework for the rapid and economical development and integration of software components. Such a testbed would provide a simulation support environment to generate different travel scenarios and to assist the collection of environmental results along with data visualization and analysis. It would also include an ITS deployment simulator that would include an ATIS simulator, a traffic control simulator, path-based vehicle movement simulator, incident detection systems, an ATMS simulator, and probe and surveillance system simulator. Linked to these two systems would be the traveler behaviour model clearly differentiating between pretrip behaviour and on-route behaviour. Using an object-oriented approach, they proposed developing a path-based vehicle movement simulator (VMS) as their testbed that would


99

contain different types of objects including transportation network objects, traffic signal objects, vehicle objects, and driver objects. Transportation network objects would include different types of networks, links, lanes, nodes and routes. The authors particularly suggest a number of interesting questions associated with the use of spatial data within an ATIS framework that require immediate attention. These include: a) how is up-to-the-minute traffic information supplied to the ATIS customers? b) how could in-vehicle routing databases be updated efficiently while minimizing both data storage requirements and the time it takes to perform the processing of local information requests? c) what spatial data processing will be required to support map-based or other visual forms of in-vehicle intelligent display systems? d) How will this information be presented to the traveler? e) How can human-friendly path-based representations be generated efficiently from dynamically updated databases containing route attributes and latest traffic conditions? The authors also point out that while much of the literature associated with ATIS focuses on invehicle guidance systems, for those travelers who can neither afford an IVGS or who are loath to use such a device while traveling, traveler information kiosks (e.g., the Smart Information Kiosk recently tested in the Los Angeles area) may prove to be an extremely attractive alternative for those drivers. A question of considerable concern to many researchers concerns the way that information is presented to the potential users. Streeter et al. (1985) had found that voice directions were superior to maps when drivers attempted to follow routes in unfamiliar environments. Deakin (1997) who surveyed route switching behaviour caused by the 1994 Northridge earthquake, suggested that drivers who were relatively familiar with the region through which they passed were able to assimilate a variety of informational sources. These included landmark based and procedural information as well as configurational (layout based) spatial knowledge. Experiments investigating what combination of voice and visual modalities could be used in IVGS are currently being undertaken in many different environments (e.g., Albert, 1997). Although many of the components of the Oakridge testbed still remain to be developed, some progress has been made. Examples include: Summers and Pilai (1996) development of a multiplatform parallel processor for rapidly generating multiple alternative paths through a transportation network database; Kwan and Hong's (1997) feasible opportunity set generator; and the building of object oriented network databases on which to apply different ATIS/ATMS scenarios (Kwan, Spiegle and Golledge, 1997).

100


Although there appears to be an increasing interest in the potential for using GIS in a transportation context, there are also problems and limitations of a GIS that should be noted (Medyckyj-Scott and Henshaw, 1993; Aronoff, 1989). Most of the expense and effort in establishing a GIS is focused in developing and maintaining the database and ensuring quality control of the data. There are also problems caused by different transformation residuals between different coordinate systems from which data may originally have been extracted. Despite these limitations, a survey of state Departments of Transportation (DOTs) and Metropolitan Planning Organizations in 1994 indicated that approximately 75% of them had initiated GIS systems within their planning activities, though many of these had just started the process of using GIS. The growth in GIS usage has been large, substantial and recent. The key components of a GIS, from the point of view of their use in transportation modeling, include: (a) data input (i.e., conversion of data from an existing form into one that can be used by the GIS, including georeferencing, mapping, development of attribute tables and remote sensing and satellite imagery); (b) data management (i.e., functions needed to store and retrieve data from the database); (c) data manipulation and analysis (i.e., the use of specific techniques to manipulate and analyze spatially referenced data); and (d) data output (i.e., visual, graphic, text, tables and reports).

EN-ROUTE DECISIONS AND INFORMATION PROCESSING Included among the temporal and spatial factors involved in en-route driver decision-making and information processing are those such as perception of speed, perception of traffic volume, perception of time lapse associated with completing segments of the designated route, familiarity with the network through which travel takes place, the ease and rate of information acquisition about the driving environment, travel goal specification and destination choice, evaluation of the priority of the goals associated with a specific trip, knowledge of landmarks and waypoints on and off a particular route, and perception of and familiarity with areas through which a route passes (including perceived safety, perceived areas to be traveled, and perceived feasibility for destination substitution). These factors appear to be elemental to the decision-making processes of trip making, goal specification, route choice, information search, and reaction to possible diversion. The information needed to allow the previously specified decision-making processes to work include knowledge of the location of the destination at which a goal can be achieved, plus


101

information about a feasible set of possible alternative destination sites; information that will influence route segment selection and the overall configuration of the most satisfactory route for achieving goals; information about congestion and its applicability to different lanes along which traffic flows; and the availability of information technologies to provide input to each of these decision making strategies. These information technologies, embodied in ATIS, signalization, and radio broadcasts, to date have focused more on system variables, such as network conditions, system disruptions, and warnings of potential time delays. Thus both network information and individualized knowledge structures are combined to provide the basic en-route information that allows an individual to either conform to or change route selection criteria during a trip, or provides the opportunity for driver preferences for handling both network and traffic conditions to be expressed in the ongoing decision-making process. A number of recent studies have been undertaken to examine driver en-route behaviour under the influence of real-time traveler information (e.g., Bonsall and Parry (1991); Allen, Ziedman, Rosenthal, Stein, Torres, and Halati (1991); Ayland and Bright (1991); Ben-Akiva, De Palma and Kaysi, (1991); Khattak, Schofer, and Koppelman, (1992)). The experimental mode chosen for much of this analysis necessarily involves laboratory experiments with traffic and driver simulators. Adler, Recker and McNally (1993) characterize en-route driver behaviour as an iterative process through which drivers assess the current state of the system and adapt travel behaviour in response to the severity of the perceived travel conditions. They further suggest that critical changes to en-route behaviour that might be expected on any particular trip include route diversion, new information acquisition, and revision of travel objectives. Adler et al. (1993) and Bonsall and Parry (1991) suggest that the following factors influence en-route behaviour: estimated delay, expected travel time, congestion levels, perception of the existence of alternative routes, prior knowledge of travel conditions on these routes, risk taking propensity, tolerance thresholds to traffic conditions, expectations of meeting travel goals and objectives, mode of travel, purpose of trip, time of day of the trip, and the potential for deleting an activity from a daily schedule or rescheduling an activity to a later time period. In their simulation model FAST CARS, Adler, McNally and Recker (1991), and Adler, Recker and McNally (1992A, B) have their players perform three actions that include road changing, lane changing and information acquisition. Subjects do not control their cars, and speeds are determined by lane selection (i.e., switching to a lane with a higher speed). They have total control over road changing and their goal is to navigate their car through the network to a given destination. They can actively seek real-time information through a Highway Advisory Radio (HAR) and an In-Vehicle Navigation System (IVNS). The HAR system provides real-time traffic incident

102


and congestion information for the freeways in a network. The FVNS calculates the shortest time path from a player's current position to the chosen destination. The en-route information fed to the players are the calculated times to destination and distance along the chosen path. The players' performance was evaluated after connecting a player profile with trial event data. At this time, studying the effect of real-time information acquisition on driver behaviour has necessarily been limited to simulation conditions, partly because of limited real word implementation of ATIS technologies, and partly because few drivers have ever used or are aware of ATIS technologies (Madenat et al., 1995). The advantage of such simulators is of course to go beyond the case of static choice sets and well-defined traffic scenarios to include those cases where observation and recording of actual driver performance during conditions of normalcy and stress can be evaluated. Real-time changes to traffic and network conditions often are sudden and the drivers' responses reflect their perceptual and cognitive processing ability, both of which are temporally and spatially dependent. It has also been shown that often people respond differently when stating preferences as opposed to the behaviours they exhibit when revealing preferences (Mahmassani and Jayakrishnan (1991). As mentioned earlier, there is an increasing volume of travel diary, panel, and survey data which is providing more and more details of human movement and an increasing number of valuable insights into the process of human wayfinding and travel behaviour. However, detailed testing of how activities are selected and how selections are combined into schedules is still relatively unresearched. In addition, emphasis is only now being placed on how Geographic Information Systems can be expressively incorporated into travel behaviour models (Kwan and Golledge, 1995; Miller and Storm, 1996; Goodchild (1998). There is a need to fiirther examine the potential for different types of GIS in this context. In particular, we need to know which data model that lies at the heart of each different GIS most closely approximates how humans think about and use real world information in their decision making and choice activities. Such research first of all requires matching cognitive processes with characteristics and components of data models and ftmctionalities of GISs. For example, humans generally cognize trips as strings of chunked road systems rather than as a linear sets of links (block faces) and nodes (intersections) (Allen, 1982). Thus, humans tend to think in terms of streets rather than a collection of blocks, and to chunk segments of trips that follow similar levels of the road and street hierarchy (e.g., the interstate chunk, the arterial road chunk, the local street chunk). At this stage, only the object oriented data model appears amenable to using this type of representational mode (Kwan and Hong, 1998).


103

Other data that is being collected includes finding the landmarks that are used as orientation nodes or primers to warn when upcoming decisions are required. For example, one may be traveling along the freeway, and when a particular church steeple comes into view, the driver realizes that two-thirds of the trip has been accomplished (i.e. by using off-route cues), or perhaps it is time to start merging towards an upcoming exit; or in the local street system, viewing a given blue house on the street might signify a need to take the next left turn in order to reach a destination (i.e. using on-route cues).

AXIS AS A DRIVER-DECISION SUPPORT SYSTEM Traffic congestion is a worldwide problem. Reducing congestion and the accidents and hazards that are associated with it is a major goal of ITS. ATIS is a critical part of these activities. It consists of in-vehicle information and guidance systems and pre-trip planning and informational systems that help a driver or potential driver to select feasible routes that will reduce congestion, find parking even when it is sparse, and facilitate activity rescheduling when it becomes necessary. The implication is that by benefiting real and potential travelers, the system in turn will benefit, and its hazardous, congestive, or inhibitive characteristics will be reduced. A primary objective of an in-vehicle guidance system is to assist the driver to select routes, which will help reduce congestion and improve driver satisfaction. Such a move benefits drivers in terms of achieving their scheduled behaviours and activities as well as benefiting the system by improving traffic flow. Before the information given by the en-route guidance system can be effective, it must be perceived as being valuable by the driver. The driver must recognize that information so obtained is valid and reliable, that actions taken in response to receiving the information will produce positive benefits, and that even when forced to drive through unfamiliar territory, the guidance information will provide low stress, safe, and easily negotiated routes. Any student of individual differences will be aware that drivers may respond in markedly different ways to the same in-vehicle information and route guidance material. In other words, although information may be commonly available, it can be used in an extremely personal manner. These manners will range from ignoring it (e.g. let others follow this advice while I stay here and the congestion or hazard will clear more quickly), to immediately accepting it (e.g. exiting as soon as possible and blindly following a suggested route through often unfamiliar territory as a way of bypassing the system problem. To that extent it is best to think

104


of the AXIS as a driver Decision Support System (DSS)—i.e., a basic information supplement to the driver's existing knowledge base. As a decision support system, an AXIS can be used either in the in-vehicle mode or in the pre-trip travel-planning mode (Garling and Golledge, 1993). Decision support systems are integrated sets of tangible and intangible information that are designed to supplement personal knowledge during problem solving activities (Densham and Rushton, 1988). A DSS does not replace individual decision making (e.g. via requiring that a driver slavishly adhere to its informed recommendations), but rather acts in a support mode allowing drivers to keep the prerogative of individual decision making. It is, after all, this freedom that makes the private automobile such an attractive force within the entire transportation schema. A DSS should bring to bear on a problem the strengths of personal expert knowledge and comprehensive exogenous knowledge that may not readily be available to the decision maker (e.g. the exact location of accidents, the amount of backed up traffic that an accident produces, construction places, breakdowns, etc., all of which are part of the dynamics of the operation of a road system). Xhe critical question confronting the driver when faced with such conditions is simply, "What do I do now?" Much of the literature on conflict resolution and decision making in the past several decades has emphasized the importance of offering not just a single solution to a problem but a set of solutions from which can be chosen the one that best suits the constraints under which a decision maker is working (e.g., Adler, et al. 1992 A, B). Xhus, it makes no sense to provide a single message based on helping people minimize time loss in congested traffic if time loss is not one of the critical variables that are used in a person's trip planning, departure time selection, route choice and destination definition process. Xhus, as a DSS, an AXIS should provide a set of alternatives among which drivers can select, depending on whether they wish to wait out the effects of a delay, to change travel such as by rerouting, to change their scheduled activities by rescheduling or deleting an activity, to replace an anticipated activity with another, or to substitute destinations for selected trip purposes. Xhe question is, what are the range of possible options available to drivers, and how can information relevant to these options be accessed (e.g. by a push-button, visual signal, or voice command interface mode)? Again, I stress here that it is important not to impose a priori decided rational or optimal alternatives on a client population, but to give them the freedom to choose alternatives within the context of their own activity scheduling and route selection criteria. Such might be the case, for example, if congestion produces an AXIS response requiring drivers to detour through an unfamiliar neighborhood or through a neighborhood which a driver perceives as being unfriendly or unsafe. In such circumstances, the probability of ignoring the information


105

provided is quite high and may even vary significantly between the sexes and among age groups. Thus, simple network or street system solutions for transmission via ATIS to in-car receivers, must be modified by information about the perceived and actual structure of the environment (i.e. common perceptions as well as new information). An ATIS designed as a DSS must: (i) be easy to use, (ii) have a user friendly in-car interface, (iii) help drivers achieve travel objectives, (iv) not divert them from obtaining those objectives, and (v) enable a user to benefit from the information dispensed. To elaborate briefly: (a) Ease of use: For ease of use an ATIS must have an interpretable language and communicate about the system in a way that is compatible to how humans think about it (e.g., "natural" vs. "technical" language.) (b) User friendliness: This must be friendly in both the in-car and the travel plan mode (Jayakrishnan et al., 1993). For the in-car recipient it must not divert them from their driving experience and must be able to relate to what can be visually experienced from a particular road segment. This includes being able to refer to on- and off-route landmarks as orienting and frame of reference guideposts for the traveler, providing eye-level perspectives rather than flat, planar diagrams—see Albert, 1997, Adler, 1997; Janson and Robles, 1995. (c) Achieving travel objectives: Most travel is motivated and directed towards specific goals or objectives (Golledge, 1992; 1995 a, b). Given that travel is motivated, specific objectives must be known to the driver. Drivers may have multiple objectives with several objectives having approximately equal salience. This can allow for rescheduling or reordering and changing objectives consequent to receipt of pertinent information. We need to continue researching on how driver objectives can be categorized and how they can be served by specific forms of information (e.g. roadside signs, radio broadcasts, in-vehicle displays, etc. Janson and Southworth, 1992). (d) Not diverting: Once objectives have been established, most drivers work out strategies, tactics, and actions in order to achieve them. Information from an ATIS will be most acceptable to a driver if it can be seen as compatible with their existing objectives rather than diversionary. Diversionary information, if given, must be seen to be consistent with previously held objectives. For example, if a substantial travel blockage has occurred on a freeway system, a single diversionary message may be seen as antithetical to some drivers' goals. Choice from a set of relevant messages may obviate this perception. (e) Use of benefits: It is suggested that ATIS information will be most readily adopted if the potential benefits are immediately obvious to the potential user (FHWA, 1995). Benefits may be given on screen or in auditory mode (e.g. in terms of estimated time

106

(f)

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges saved when bypassing a trouble spot). This may provide supplementary local information that allows rescheduling and destination substitution, as might happen if data on the location of local stores in the vicinity of the bypass is also provided. The essence however is for the user to have immediate and easily perceived benefit statements. The usefulness of provided information will thus become immediately obvious, and behaviours designed to alleviate a problem can be prompted. When using an ATIS, the first task is to decide whether the information being dispensed is relevant to solving a general system problem (e.g., clearing a point of congestion) or is providing information to allow drivers to reformulate goals, objectives, path selection criteria and preferred routes. Up to this time, most emphasis has been placed on the former. Alternate bypass routes are the favorite mechanism for clearing such blockages. More recently, as the ATIS is interpreted more as a DSS, there has been a trend to develop driver simulators to observe how people react to different system blockage conditions (for an overview see Koutsopoulos, Lotan, and Yang, 1994). Both revealed behaviour (i.e. that observed) and preferred behaviour (i.e. responses given to test questions) have been collected (Caplice and Mahmassani, 1992; Cyhen and Mahmassani, 1996; Stopher, 1996). The long run goal appears to be providing the driver with access to an information pool that they would not normally consider when used in their pre-trip travel planning or en-route navigation activities alone. Providing information that is not regularly used, or providing it in a form that is not consistent with the normal mode of interaction or communication, inhibits driver acceptance, A realistic time frame: ATIS will be considered useful only if information can be obtained, processed, and used within a realistic time frame (Jayakrishnan et al., 1993, 1994). There are several unresolved problems with respect to this. First, there is insufficient research on whether or not people are more efficient at processing visual or auditory information while concurrently doing other tasks (Albert, 1996). The latter constraint is important. There is little doubt that traditional work in psychology has shown that vision is the spatial sense par excellence and that it far outperforms the other senses in terms of the volume of information perceived in a single exposure (e.g. a glance). Audition involves longer response times, simply because of the grammatical structure of verbal discourse. A blinking cursor on a map can give an immediate idea of current location; a verbal description of where one is may involve several sentences, and it may take a minute or more to present and absorb. However, as it has also been shown that distracting a driver from visually assessing traffic conditions (especially when traffic is moving at speed) is potentially very dangerous. Most drivers are used to receiving auditory information via a radio station, cassette tape, CD, cellular phone, or passenger conversation. More experiments need to be undertaken on the advantages and


(g)

107

disadvantages of different presentation modes, and whether or not paying attention to those modes while in the act of driving in a dynamic travel context, will materially affect the chances of perceiving, absorbing, and reacting to information provided in different modalities. If the driver considers accessing information a normal part of the driving process, it is more likely to be accepted than if it is seen to require attentional processes other than those focused on safe driving, Language use and capabilities: An increasing volume of work is accruing in spatial linguistics, and in geography, on the use of spatial language (Golledge, 1996; Golledge and Stimson, 1997). Distinction is made between naive or natural language (the language of everyday speech) and technical language. Everyday speech is laden with fuzzy spatial prepositions, nouns, adjectives and adverbs, as well as being anchored by spatial verbs that can have several interpretations (e.g. "run" as in a footrace vs. "run" as in turning on a switch). Much of natural language is what Lakoff (1987) called "container" oriented. In other words, spatial information is given with respect to position (e.g. front, back, sides, above, below, inside, outside). Often the container is a human form and egocentric referencing prevails. This introduces perspective error into communication systems. For example, two people facing each other and conversing will have different ideas of what's in front of, behind, and to the left or right as these words appear in common speech. But, most people do not use the precise spatial information of experts. For example, few people will say "The church is northeast by north of you", or that "the church is 70° from your current facing direction" or "the church is at 2:15 o'clock to the way you are heading." The latter are all more technical, more exact and more for expert uses that require precision. An important question within the context of most ATIS is simply "What degree of precision is required, and can this be supplied using common speech or natural language?" Some experiments along these lines for pedestrian travel are currently being undertaken by our research group (Loomis, Klatzky, Golledge, and Tietz, 1997).

SPATIAL LEARNING Research by Gale, Golledge, Pellegrino and Doherty (1990) and Golledge, Ruggles, Pellegrino and Gale (1993) has indicated that people have significant difficulty in integrating information about separate routes, even when they are partially overlapping and experienced via same day travel. In simple matters of sequencing of landmarks, for example, unidirectionally experiencing a route allows almost perfect recall of sequence after a series of trials. Estimating location or interpoint distances between consecutive cues fares almost as well, with rank order

108


correlations being extremely high, but with some variability in interpoint distance estimation. If learned bidirectionally, sequencing and interpoint distance estimation both get worse, as did the sensing of direction and orientation. When two partially overlapping routes are learned unidirectionally, sequencing is again good, distancing is not quite as good, and direction giving is very poor. If layout information is required by integrating one's knowledge of on and off route landmarks experienced on either of the partially overlapping routes, and from directions given to point to landmarks on or near one route from locations on the other route, performance degenerates substantially. This is a clear indication that more experimentation needs to be done on the type of complex or simple geometries that facilitate or inhibit comprehension of information about location, sequencing, and layout of on-route, off-route, and landmark and path information. The same is required in terms of finding how differently configured routes can be integrated to make up some type of configurational understanding. Thus, essentially we need to know how drivers learn about environments, what they learn, and whether they are capable of synthesizing information that could be provided through an ATIS in different formats.

PATH SELECTION CRITERIA One of the more under-researched areas of travel behaviour includes that of examining the different reasons for path or route selection. Traditional econometric models use minimized distance, minimized cost, or minimized time. Psychological evidence on pedestrian and driver behaviour (Saisa, Svenson-Garling, Garling, and Lindberg, 1986) and laboratory and field experiments, (Pas and Koppelman, 1986, 1987) plus an increasing number of driver simulation experiments, appear to indicate that this type of rational or optimizing behaviour is not widespread among individual travelers. The question is then, "What criteria are used?" Golledge (1995) found a variety of criteria being used when subject's undertook laboratory experiments in differently constructed environments. For some, shortest path was the most frequently adopted and regularly chosen activity. Route retrace differed significantly from environment to environment (e.g. 60% retrace in one environment as opposed to 20% retrace in a slightly modified one). Depending on the regularity of the system, path selection criteria might change. For example, in a rectangular system where the origin and destinations were at diagonally opposite comers, many direct routes qualify for the same distance. However, routes that are selected could follow a variety of other criteria such as fastest time, minimizing left turns, minimizing total turns, taking the longest leg first, taking the shortest leg first, trying to approximate a diagonal, always heading in the direction of the objective or destination, and so


109

on. Many people indicated quite readily that the path that they selected on the hypothetical environment used the same criteria as they normally use in their daily activity patterns. Apparently, people use different criteria for different purposes. For the most part it seems that when the home to work trip involves no intermediate stops, it frequently conforms to a minimum time, minimum distance, or minimum cost procedure. But, the trip home is not always a simple reversal of the trip to work. It is more likely to be part of a chained trip, and while parts of the original home to work trip may be repeated, new segments may be added depending on what additional purposes are integrated into the chain (Mahmassani, Hatcher, and Caplice, 1996). As trip purpose changes from shopping, to recreational, to health related, professional consultation, education, and so on, reasons for choosing a particular route can also change substantially. Unless an ATIS can prepare a number of alternate routes to bypass a problem on a network, then people may choose to ignore it. Certainly one cannot assume that all the people on a freeway at 5:15 p.m. on a weekday are going directly home. It makes little sense, therefore, to adopt a strategy of providing information that only satisfies the homeward bound segment of the driver population. Recent research (Golledge 1995 a and b; Kwan, 1994, Mahmassani et al. 1996; Adler, 1996) has indicated that there are a number of feasible route selection criteria embedded in daily activity patterns. The implication of this is that models of travel behaviour should incorporate multiple models or subroutines that allow different path selection criteria to control route selection. For example, in the morning it may well be that minimizing time or distance would be an acceptable criteria to be used for many journeys to work. This criteria may not be invoked again on any other trip during the day (unless a business activity is performed). Activities for other purposes (e.g., a health related visit, dining away from home, recreation, socializing or visiting friends, shopping for clothes, food shopping, servicing vehicles), all may be motivated by (or motivate) different route selection criteria. Such criteria might include minimizing time, minimizing left turns, maximizing freeway travel, minimizing encountered traffic control devices, selecting streets free of truck traffic, bypassing perceived or real congestion points, variety seeking behaviour or exploratory activity, avoiding dangerous neighborhoods, passing by a place of scenic beauty, avoiding hazardous or polluted areas and so on. At this stage, we do not know enough about which criteria are likely to be more popular amongst different groups of people for different trip purposes when specific modes are used. At this time, there is a tendency to adopt single criteria for all trip-making.

110


A second characteristic of route selection comes from recent work in psychology. It has been shown on several occasions that people do not always take the same route in both directions (i.e. that route reversal is not necessarily a common way of traveling) (Golledge, 1996). Much routinized behaviour is very hard to extinguish. This is one reason why people prefer to suffer the effects of congestion and traffic delays rather than go through the experience of learning a new route. This appears to be so when dealing with well-established and long-held habits, as is the case with respect to journey to work. After all, the main purpose of learning a route and repeating it in a relatively invariant way is to minimize the stress of conscious decision-making and allow routinized behaviour to replace problem solving activities. When obstructions occur, conscious decision-making is again activated and problem solving strategies have to be defined. All of this takes effort. Again, an ATIS should be designed to quickly provide enough information such that this new decision process does not become onerous. As with many decisions, without being conscious of it the traveler is likely to do a cost benefit analysis of the information provided. It is in the interests of the ITS planners, therefore, to ensure that just sufficient information is given so that it can be absorbed, and acted upon, with as great a probability of acceptance as possible. Another area of potential research, therefore, is just how much information should be presented from an ATIS source? One research question might be whether or not the volume of information for travel planning can differ significantly from the type and volume of information needed for on-route in-car decision making. There may also be different qualities and quantities of information required for different types of obstacles and barriers (e.g. an accident involving a chemical spill may invoke a different information set from an ATIS source than an accident involving a chain reaction collision, or a long term construction project).

INTRODUCING BEHAVIOURAL CONCERNS INTO THE SUPPLY SIDE: DEFINING AN ATIS-COMPATIBLE NETWORK While much of the recent work on ITS dynamics has been focused on ensuring that real behaviours are considered, rather than hypothetical, rational, or optimizing ones, less attention has been paid to ensuring that those behaviours take place within a real system. A realistic representation of an existing transportation system (i.e. the supply side) would include an exact rendering of the road hierarchy, including traffic directionality, volumes, episodic frequencies, hazards, traffic control devices, speed controls, turning controls, laneage, signage, and traffic related features, such as pedestrian crossings (particularly school crossings), the presence or absence of bicycle lanes, on-street parking, dividing strips, restrictions on turning (such as U-


111

turns), the type of surface, whether through-traffic is possible, and so on. Many AXIS simulations and models have been developed simply as node-link graphs, and while some of the above attributes are included, many of them are not. The need for a more realistic representation of the system on which the behaviour is supposed to take place is needed because many of the system characteristics are consciously or unconsciously taken into consideration in a traveler's decision-making process when forging a route. A major problem, of course, that is part of the central dynamics of the system, is when temporary obstructions or closures occur. These things are not built into one's cognitive map (Sholl, 1987, 1996), and if they are, they are done so only if their existence exceeds some threshold time interval. We know little about the extent to which obstructions will cause people to divert from learned behaviours, how long the diverted behaviour lasts, and whether or not the exact previously used route is resumed after blockages or obstructions have been removed. Since many causes of congestion are temporary (e.g., accidents or hazards) it can be safely assumed that the driver does not encode the location of the accident in any permanent way nor even encode the alternate route chosen (or recommended) to bypass the blockage. A considerable volume of work is being carried out in cognitive science, spatial cognition, and geography on the processes of navigating and wayfmding. Understanding these processes will influence a variety of features such as the reason for choosing path selection criteria, the decision to implement a single-purpose or chained trip, the selection of route segments, and the perception of time taken to complete a trip. (Loomis et al., 1997)

NETWORK AND ROUTE REPRESENTATION Traditionally, transportation systems have been represented as networks (Ran and Boyce, 1994; Ran, Hull and Boyce, 1996). The networks consist of link-node combinations. Links are usually block faces, nodes are generally intersections. A street, therefore, consists of m links and n nodes. This is a simple way of organizing a network such that a computer can easily comprehend it and operate on it (Mahmassani and Peeta, 1993). However, this is not the way that humans cognize systems. More frequently, the human will recognize streets rather than strings of block faces and intersections. In fact, very little if any information may be known about on- or off-route landmarks related to a particular string and certainly for many strings, the names of cross streets or knowledge of the number of intersections is not usually known. Thus, "State Street" is conceived of as a chunk not a link-node string. It is a "whole" or "object" that can be integrated into other chunks of the system to make up a route. Chunking is usually segmented at places where significant choices have to be made. For example, the route

112


from one's home to work may consist of a somewhat complex local street chunk, an arterial road chunk connecting to a freeway, a freeway chunk, and a local street chunk leading to the destination. The transition from one chunk to the next can be facilitated by clear and unambiguous primers at the places where one needs to make specific decisions (e.g., once you come up to the blue house take the next turn right; or once you see the sports arena, move to an exit lane on the freeway). As Golledge et al. (1993) have shown this chunking distorts spatial relations. Spatial relations such as location, sequence, distance and direction within chunks are often maintained in true relation to their existence in reality. Relations between adjacent chunks can maintain sequence but may distort distance, time or other connective measures. Relations between objects in widely separated chunks become extremely generalized and inexact (e.g., the restaurant is "over there"). Transportation systems can be chunked according to many different criteria, such as: volume of traffic flow, number of lanes, types and frequency of traffic control devices, direction, existence within a system hierarchy (e.g., whether alleys, lanes, local streets, arterials, state highways, interstates, etc.), or cognized neighborhood segments. Not all data models are capable of representing street and road systems in such a manner. Current research by Kwan, Speigle, and Golledge (1997) is examining if objectoriented data models are more effective in representing transportation environments in a way more similar to how people perceive and cognize them than other systems do. The essential argument here, is that people conceive of environments as consisting of sets of objects, and in this process of objectification, simplify, label, categorize, cluster, and organize material in such a way as to promote identification and recall. However, this is not an exact imaged reproduction of the environment and in the process of encoding and storing such information, generalizations, simplifications, and errors accrue. It is on this altered and error prone cognitive representation that processing is performed in order to determine features such as connectivity, proximity, sequence, direction, orientation, association, inclusion, and so on. While at this stage it may seem a little unreal to think of an ATIS that is constructed in a manner very similar to the way people think, this possibility is not far off. Egenhofer (1991) in an article on system topology has shown that nonmetric or topological representations of environments appear to be much closer to the traditional human way of organizing and comprehending space than is the more formal and exact method of geometric duplication of a system. Other experiments over the years on features such cognitive distance (Montello, 1991) show that humans often simplify "over the road" distances to be more equal to crow-fly or city block metrics (Minkowskian r=l and r=2 metrics) than to realistic ground truthed representations. The question still remains unsolved, however, as how best to incorporate this information into the preparation of a spatial database and how best to codify the types of operations to be performed on that database, such


113

that a geographically naive human can instantly comprehend the spatial information or directions displayed on or communicated through an ATIS.

USE OF G I S FOR REPRESENTING NETWORKS In many situations (Golledge, 1996) exact transportation systems models have been replaced by accurate geocoded mappings within the context of a geographic information system (GIS). GIS are essentially sets of interacting computer programs that allow a large set of functions (both analytical and manipulative) to be applied to the data. For example, functions such as overlay and dissolve, corridoring and buffering, detecting adjacency, arranging strings, and regionally including or excluding information, are common GIS functions. Using a transportation GIS (GIST), it should be possible to give an instruction such as: "Find the closest arterial road to location x-y (where a system blockage frequency occurs) that will allow travelers to link with highway N in a distance minimizing fashion." In this case if the different levels of the road and street system are stored in the form of a hierarchy or even at separate levels, overlaying can quickly solve the problem by providing a graphic, auditory, or written description of the location of such an arterial and how to access it. But, we have allowed an emphasis on the behavioural component of the supply side of modeling to lag somewhat. It appears to be time for us to consider both the demand and supply side of travel behaviour modeling, as it is being practiced more frequently today. Examining the supply side will parallel the efforts on the demand side that have been stimulated by the activity-based approach and its increased emphasis on understanding human purposes and activities more than just examining data representing the a posteriori results of human decision making. Let us briefly examine how a GIS might assist different driver reactions to an ATIS communication about a system blockage.

Rescheduling Rescheduling may be more of an internal manipulation of destination sequences than requiring external aids to solving order or frequency aspects of travel behaviour. However, if some characteristics of the local environment are not well known (e.g. business hours) then such information should be readily accessible via the GIS. For example, if one was going to work

114


and was informed of a significant time delay on a blocked traffic artery, and one had planned to fill the car with gasoline on the trip home, knowing where the closest gas station was and how it can be accessed might cause a rescheduling change. A GIS should be able to provide a response to the question: "Find me the closest Mobil station within two blocks of the freeway segment representing my current location." When rescheduling, often the driver is forced to move into a different time frame. For example, if a salesman has a customer whom he has to see at 10:00 o'clock, for a thirty-minute interview, and he finds that traffic will delay arrival at the agreed on site by twenty minutes, it may be impossible to reschedule the client for an immediately succeeding time period. Thus, begins the problem of deciding where to fit the client into the rest of the day or following day's schedules. Similarly, if one was planning to go to a bank on the way to work and was delayed by traffic hold up, the visit to the bank may be delayed until lunchtime, after work, or the next day. Rescheduling to lunch time may force a reschedule of planned lunch activities; rescheduling until later in the day may force a previously scheduled activity into a different time interval such as evening, the following day, or a more distant time period; and rescheduling the next day, may involve altering the entire temporal sequence of planned activities (e.g. cutting everything the next morning by ten minutes to get the desired thirty minutes to see the client). Some scheduled activities are very hard to reschedule. For example, if one was limited by congestion from keeping an appointment with a health professional (e.g. a doctor or dentist), it could be weeks before a new appointment was made. Certainly just turning up whenever the system cleared, is not always a viable alternative. This means an irregularity may be thrown into the long-term episodic frequency with which activities occur. Thus, rescheduling has both a short term and a longterm calendar that must be taken into consideration.

Rerouting Given information about a system blockage, a GIS should be able to respond to the question: "Find me an alternate route past the blockage which allows me to rejoin the freeway at the first available exit and that is no further than a quarter of a mile from the freeway itself Also minimize the number of left turns that I will have to make to follow this route." Again, a GIS should be able to do this and because of the second constraint, may eliminate routes that pass under the freeway and proceed along the opposite side. For example, rerouting decisions require information on the nature of the roads that make up the transport network in a problem area. Selection of an alternative path or route requires knowledge of all levels of a road hierarchy of highways to lanes and alleys. Information must also be available on the type of flow, for example (e.g. one or two way) and the directionality of flow (that is, toward or away


115

from the center). Other information may include whether or not there is a central divider, whether U-Turns are possible in a street segment or at intersections, whether the intersections have traffic control devices such as stop signs or traffic lights, whether there are turn lanes, whether there are bike lanes at the side of the road, whether there are curbs, whether parking is allowed, the type of neighborhood through which the street passes, speed controls, and warnings of any commonly known hazards or dangerous areas (e.g. narrow bridges). This information must be embedded in yet another set that indicates the temporal volume of traffic (again probably chunked rather than represented as a linear stream of time).

Destination Substitution Assume that a person had an activity schedule which required going to a grocery store, clothing store, bakery, and a discount store. The GIS should be able to respond to a request to find alternative destinations for each of these. A query might be: "Find me the nearest place that I can go where I can find (list functions) within a half mile of each other." A GIS should be able to perform this activity, and may result in the traveler finding new and previously unknown places at which to shop and interact. Rescheduling may require the traveler to search for an alternate destination than the one originally planned to be visited. Our knowledge of environments outside of well-traveled and well remembered areas, is often sparse and highly distorted. The ATIS/GIS may potentially provide information on the closest opportunities for destination substitution, but even these may not be selected. For example, if one was traveling on a freeway section through an area that had a negative crime image, then one may not wish to exit the freeway in search for a drug store, liquor store, grocery store, or restaurant to which one was currently being directed. Obviously, destination substitution may have only a limited possibility for work trips. But for many other trips it can become a viable alternative—such as substituting a different restaurant from the one originally planned for lunch, a different gas station, a different shopping center, or even a different recreational area. Dealing with congestion invariably involves time loss. Time loss actually may prevent extensive rescheduling, just as the nature of the surrounding road system may make rerouting a less desirable alternative. Destination substitution may be possible if the constraints involved in travel behaviour are not grossly violated.

116


In addition to this basic information users will need other information such as the directionality of any alternate route. For example, a route that is roughly parallel to a previously proposed but now obstructed one provides little difficulty for integrating into one's cognitive map. Landmarks are seen from basically the same perspective, sequences of environmental cues are seen in roughly the same order, and distances and direction should not vary too much. However, a route that is orthogonal or angular to the originally proposed route can change all this. Landmarks may be observed in different sequence or order, sequences of interpoint distances may be distorted, priming cues that signal critical decisions may not be experienced, different perspectives can substantially change geometric understanding of where one is and what the relationship is between environmental features, and so on (Golledge, Ruggles, Pellegrino, and Gale, 1993; Golledge and Stimson, 1997; Lloyd and Cammack, 1995; Kirasic, Allen, and Siegel, 1984; Siegel, 1981; Moar and Bower, 1983; Moar and Carleton, 1982; Sholl, 1987, 1996). Forcing the driver to adopt a completely new frame of reference, a new perspective, and a loss of well known landmarks or experienced cues, can drastically change feelings of well being, increase stress, and potentially reduce the attractiveness of an alternate pathway - even if the pathway is economically more efficient or is optimal for clearing system blockage. Under these circumstances we might expect drivers to begin ignoring the information sent through their AXIS. Obviously, more research needs to be done on the extent to which each of these characteristics, when changed, might alter decision-making processes. It appears this would be an ideal type of problem for a good traffic simulator.

Activity Compression The loss of time associated with congestion and possibly with following alternate paths to avoid congestion, may result in activity compression. In this case, the traveler would simply maintain the original activity schedule but reduce the amount of time allocated to each remaining activity. For example, a two hour planned recreational period might be reduced to 45 minutes; a one-hour shopping trip in a supermarket reduced to a fifteen minute stop at a fast food outlet. Activity compression is often associated with destination substitution. For example congestion on the way home might mean compression of food shopping along with destination substitution such that a quick frozen meal is picked up at a 7-11 or AM/PM rather than shopping at a supermarket for fresh meats or produce.


117

Activity Deletion If time loss associated with traffic blockage and rerouting is significant, yet another possible behavioural response is deletion of a scheduled activity. Earlier we saw that rescheduling at a later date might be feasible. An alternative is to delete the activity from the schedule and not attempt to complete it until its scheduled appearance in the next episodic interval. For example, a traveler may have scheduled a haircut on the way home on a given day, but be forced to delete the activity from an already busy schedule. It may not be possible to resurrect the activity until the schedule for the following week at approximately the same time. Another ahemative is to delete low priority activities. For example, if traffic delay caused one to miss a client and rescheduling was possible, a lower priority activity such as socializing or recreating might be deleted from the episodic package (i.e., for a day, week, month, etc.).

SUMMARY The purpose of this workshop is to open areas for discussion. This chapter was designed to help achieve that goal. The areas I chose to reference can be handled with laboratory, field, real, or simulated conditions. What is required in each case is a thorough experimental design, the appropriate selection of data, and of course, the appropriate forms of analysis and summarization into meaningful models. Many of the questions and problems that I have raised so far have been touched on in part within a different context - the context that behavioural geographic research or research on spatial cognition in psychology, cognitive science, and artificial intelligence. There is a need to bring all this research together and put it in the context of real world application. ITS provides such a context. In this area, basic academic research can be neatly tied to applied needs. Of course, there are innovative and creative actions that must be taken to link the basic and the applied work. But, that is the challenge of workshops like this and indeed conferences such as the lATBR. Along with many others, I hope to get considerable academic stimulation and fresh ideas from this workshop. Having seen some of the abstracts of papers due to be presented in this and other sessions, I am already assured that this must be a necessary outcome of the conference.

118


REFERENCES Adler, J. (1997) Perspectives on modeling routing behaviour under AXIS. Position paper presented to lATBR 97. Adler, J. L., and M. G. McNally (1993). In-laboratory experiments to investigate driver behaviour under Advanced Traveler Information Systems (AXIS). (UCTC Working Paper 184). Institute of Transportation Studies and Department of Civil Engineering, University of California at Irvine, CA. Adler, J. L., T. F. Golob and M. G. McNally (1993). A structural model with discrete-choice variables for predicting en-route behaviour under ATIS. Working Paper, August, UCTC No. 174. The University of California Transportation Center (UCTC), University of California at Berkeley. Adler, J. L., T. F. Golob and M. G. McNally (1996). A structural model with discrete-choice variables for predicting en-route behaviour under ATIS. Working Paper, August, UCTC No. 174. The University of California Transportation Center (UCTC), University of California at Berkeley. Adler, J. L., W. W. Recker and M. G. McNally (1992A). Using interactive simulation to model driver behaviour under ATIS. Proceedings of the ASCE / ^ International Conference on Microcomputers and Transportation, Baltimore, Maryland. Adler, J. L., W. W. Recker and M. G. McNally (1992B). Using interactive simulation to model driver behaviour under ATIS. Proceedings of the 4^^ International Conference on Microcomputers and Transportation, July 1992, Baltimore, Maryland. Adler, J. L., W. W. Recker and M. G. McNally (1992C). A conflict model and interactive simulator (FASTCARS) for predicting en-route assessment and adjustment behaviour in response to real-time traffic condition information.. Presented at the 71^^ Annual Meeting of the Transportation Research Board, Washington, DC. Adler, J. L., W. W. Recker and M. G. McNally (1993). A conflict model and interactive simulator (FASTCARS) for predicting en-route driver behaviour in response to realtime traffic condition information. Transportation, 20, 2, 83-106. Albert, W.S. (1996) The role of attention in learning spatial relationships during navigation. Ph.D. dissertation. Department of Geography, Boston University. Allen, G. L. (1982) The organization of route knowledge. In R. Cohen (Ed), New directions for child development: Children's conceptions of spatial relationships. San Francisco: Jossey-Bass, pp. 31-39. Allen, R. W., A. C. Stein, T.J. Rosenthal, D. Ziedman, J. F. Torres and A. Halati (1991). A human factor simulation investigation of driver route diversion and alternative route selection using in-vehicle navigation systems. Proceedings of the Vehicle Navigation and Information Systems Conference, 1. Society of Automotive Engineers, Inc., Dearborn, Michigan, 9-26. Ampt., E. S., A. J. Richardson and W. Brog, eds. (1985) New Survey Methods in Transportation. VNU Science Press, The Netherlands. Aronoff, S. (1989) GIS: A management perspective.


119

Axhausen, K. W. (1995) The data needs of activity scheduling models. Paper presented at the International Conference on Activity based Approaches: Activity Scheduling and the Analysis of Activity Patterns, May 25-28. Eindhoven University of Technology, The Netherlands. Axhausen, K. W., and T. Garling (1992). Activity-based approaches to travel analysis: Conceptual frameworks, models, and research problems. Transport Reviews, 12, 4: 323-341. Axhausen, K.W. and Herz, R. (1989) Simulating activity chains: a German approach. Journal of Transportation Engineering, 115, 3, 316-325. Ayland, N. and J. Bright (1991). Real-time responses to in-vehicle IVHS technologies—a European evaluation. Presented at the Transportation Research Board lOf^ annual meeting, Washington, DC Beckmann, R. J., K. A. Baggerly and M. D. McKay (1996). Creating synthetic baseline populations. Transportation Research 30A: 415-429. Ben-Akiva, M., A. dePalma and I. Kaysi (1991). Dynamic network models and driver information systems. Transportation Research, 25B, 5 Bonsall, P. W. and T. Parry (1991). Using an interactive route-choice simulator to investigate drivers' compliance with route guidance advice. Transportation Research Record, 1036, 59-68. Bovy, P. H. Y. and E. Stem (1990). Route choice: Wayfinding in transport networks. Dordrecht: Kluwer Academic Publishers, The Netherlands. Boyce, D. E., Y-F. Chang and M. R. Lupa (1994). Introducing feedback into four-step travel forecasting procedure versus equilibrium solution of combined model. Transportation Research Record, 1443, 65-74. Brog, W. and E. Erl (1981). Application of a model of individual behaviour (situational approach) to explain household activity patterns in an urban area to forecast behavioural changes. Paper presented to the International Travel Demand Analysis: Activity Based and other New Approaches. Oxford, July. Brog, W., A. H. Mayburg, P.R. Stopher and M. J. Wermouth (1985). Collection of household travel and activity data: development of a survey instrument. In E.S. Ampt, A.J. Richardson, and W. Brog, eds.. New Survey Methods in Transport, VNU Science Press, The Netherlands, 151-172. Caplice, C. and H. S. Mahmassani (1992). Aspects of commuting behaviour: preferred arrival time, use of information and switching propensity. Transportation Research A, 26A, (5), 409-418. Catling, I. and B. McQueen (1991). Road Transport Informatics in Europe—^Major Programs and Demonstrations. IEEE Transactions on Vehicular Technology, 40, 1, 132-140. Chen, P. S.-T., and H.S. Mahmassani (1993). A dynamic interactive simulator for studying commuter behaviour under real-time traffic information supply strategies. Transportation Research Record, 1413, 12-21. Deakin, A. K. (1997). Procedural knowledge and its potential for enhancing Advanced Traveler Information Systems. Paper presented at the 76^^ Annual Transportation Research Board Meetings, January 12-16, Washington, DC Densham, P. J. and G. Rushton (1988). Decision support systems for locational planning. In R.G. Golledge and H. Timmermans (Eds.), Behavioural modelling in geography and planning. London: Croom Helm, pp. 56-90.

120


Duncan, G. F. (1987) The Role of Panel Studies in Research on Economic Behaviour. Transportation Research, 21A, 4/5, 249-263. Fan, H. S. L., A. P. G. Menon and P. S. Olszewski (1992). Travel demand management in Singapore. ITE Journal, 62, 12, 30-34. FHWA (1995) Driver reaction to unreliable traffic information. Summary report FHWA-RD95-128. Federal Highway Administration, Washington, DC, 20590 Gale, N. D., R. G. Golledge, J. Pellegrino and S. Doherty (1990). The acquisition and integration of neighborhood route knowledge in an unfamiliar neighborhood. Journal of Environmental Psychology, 10, 1: 3-25. Garling, T. and R. G. Golledge (1989). Environmental perception and cognition. In E. H. Zube and G. T. Moore (Eds.), Advances in environment, behaviour, and design. Volume 2. New York: Plenum Press, pp. 203-236. Garling, T. and R. G. Golledge (1993). Understanding behaviour and environment: A joint challenge to psychology and geography. In T. Garling and R. G. Golledge (Eds.), Behaviour and environment: Psychological and geographical approaches. North Holland: Elsevier Science Publishers, pp. 1-15. Garling, T. and R. G. Golledge (Eds.) (1993). Behaviour and environment: Psychological and geographical approaches. North Holland: Elsevier Science Publishers. Garling, T. and S. C. Hirtle (1990). Heuristic rules for sequential spatial decisions: A cognitive approach. Paper presented in the Workshop on Behavioural Models, Commission on Mathematical Models of the International Geographical Union, Boston University, MA, November 8. Garling, T., M-P. Kwan and R. G. Golledge (1991). Computational-process modelling of travel decisions: Review and conceptual analysis. Paper presented in symposium in microsimulation. Commission on Mathematical Models, International Geographical Union (IGU), Stockholm, Sweden, September 2-4. Garling, T., M-P. Kwan and R. G. Golledge (1994). Computational-process modelling of household activity scheduling. Transportation Research B, 28B, 5: 355-364. Garling, T., K. Brannas, J. Garvill, R. G. Golledge, S. Gopal, E. Holm and E. Lindberg (1989). Household activity scheduling. In Transport policy, management and technology towards 2001: Selected proceedings of the fifth world conference on transport research. Volume IV. Ventura, CA: Western Periodicals, pp. 235-248. Golledge, R. G. (1992). Place recognition and wayfmding: making sense of space. Geoforum, 23,2, 199-214. Golledge, R. G. (1995a) Defining the criteria used in path selection. Paper presented at the International Conference on Activity based Approaches: Activity Scheduling and the Analysis of Activity Patterns, May 25-28. Eindhoven University of Technology, The Netherlands. Also printed as a University of California Transportation Center Work Paper, UCTC No. 278. Golledge, R. G. (1995b) Path selection and route preference in human navigation: A progress report. In A. U. Frank and W. Kuhn (Eds.), Spatial information theory: A theoretical basis for GIS. Proceedings, International "Conference on Spatial Information and Theory" (COSIT '95), Semmering, Austria, September. New York: Springer. Also printed as a University of California Transportation Center Work Paper, UCTC No. 277.


121

Golledge, R. G. (1996) "What" and "Where" are geographic primitives: So why should they be so fuzzy? Comments on "Perspective choice in spatial descriptions" by Scott Mainwaring and Barbara Tversky. Presented at the AAAI Conference, Stanford University, March 25-27. Golledge, R. G. and R. J. Stimson (1997). Spatial behaviour: A geographic perspective. New York: Guilford Press. Golledge, R.G., and H. Timmermans (Eds.) (1988). Behavioural modelling in geography and planning. London: Croom Helm. Golledge, R. G., M-P. Kwan and T. Garling (1994). Computational-process modelling of household travel decisions using a Geographical Information System. Papers in Regional Science, 73, 2: 99-117. Golledge, R. G., A. J. Ruggles, J. W. Pellegrino and N. D. Gale (1993). Integrating route knowledge in an unfamiliar neighborhood: Along and across route experiments. Journal of Environmental Psychology, 13, 4: 293-307. Goodchild, M. F. (1998). GIS and Disaggregate Transportation Systems. Geographical Systems (In Press). Hagerstrand, T. (1970). What about People in Regional Science? Papers and Proceedings, European Regional Science Assoc, vol. XXIV, pp 7-24. Hanson, S. and P. Hanson (1981). The impact of married women's employment on household travel patterns. Transportation, 10, 165-183. Hanson, S., and P. Hanson (1981). The travel activity patterns of urban residents: Dimensions and relationships to socio-demographic characteristics. Economic Geography, 57, 332347. Hanson, S., and J. O. Huff (1986). Classification issues in the analysis of complex travel behaviour. Transportation, 13, 211-293. Hasselkom, M., J. Spyridakis and W. Barfield (1991). Surveying commuters to obtain functional requirements for the design of a graphic-based traffic information system. Proceedings, Vehicle Navigation and Information Systems Conference, 2, Society of Automotive Engineers, Inc. Dearborn, MI. 1041-1044. Hatcher, S. G. and Mahmassani, H.S. (1995) Daily variability of route and trip scheduling decisions for the evening commute. Transportation Research Record, 1357, 72-81 Hirtle, S. C. and J. Hudson (1991). Acquisition of spatial knowledge for routes. Journal of Environmental Psychology, 11, 335-345. Hu, T-Y, R. W. Rothery and H. S. Mahmassani (1992). DYNASMART: Dynamic Network Assignment-Simulation Model for Advanced Road Telematics. Center for Transportation Research, The University of Texas at Austin. Huff, J.O., and S. Hanson (1989). Measurement of habitual behaviour: Examining systematic variability in repetitive travel. In P. Jones (Ed.), Ne\s approaches in dynamic and activity based approaches to travel analysis. Gower Publishing, pp. 229-249 Huff, J. O. and S. Hanson (1990). Measurement of habitual behaviour: Examining systematic variability in repetitive travel. In P. Jones (Ed.), Developments in dynamic and activity based approaches to travel analysis. Avebury Aldershot: pp. 229-249. (in bibliography:) Institute of Transportation Engineers (1989). A toolbox for alleviating traffic congestion. Institute of Transportation Engineers, Washington D.C. IVHS America. Strategic Plan for Intelligent Vehicle-Highway Systems. TRB, Washington, DC.

122


Janelle, D. G., B. Klinkenberg and M. F. Goodchild (in press). The temporal ordering of urban space and daily activity patterns for population role groups. Janson, B. N. and J. Robles (1995). Dynamic traffic modeling of the I-25/HOV corridor. Report CTI-UCD-95-11. Prepared by the University of Colorado at Denver for the Federal Highway Administration Washington, DC 20590. Janson, B. N. and F. Southworth (1992). Estimating departure times from traffic counts using dynamic traffic assignment. Transportation Research, 26 B, 3-16. Jayakrishnan, R., H. S. Mahmassani and T-Y. Hugh (1994). An evaluation tool for Advanced Traffic Information and Management Systems in urban networks. Transportation Research C 2, (3) \29'\47. Jayakrishnan, R., H. S. Mahmassani and U. Rathi (1993). User-friendly simulation model for traffic networks with ATIS/ATMS. Proceedings of the 5^^ International Conference on computing in Civil and Building Engineering, Anaheim, CA., June 1993. Jones, P. (Ed.) (1990). Developments in dynamic and activity-based approaches to travel analysis. Avebury lAldershot Joseph, T., H. S. Mahmassani and R-C. Jou (1992). Dynamic framework for the analysis of user responses to traffic system disruptions and control actions. Research report 12162F, Center for Transportation Research, The University of Texas at Austin. Jou, R-C. and H. S. Mahmassani (1996). Comparability and transferability of commuter behaviour characteristics between cities: departure time and switching decisions. Transportation Research Record (Forthcoming). Jou, R. C, H. S. Mahmassani and J. Joseph (1992). Daily variability of commuter decisions: Dallas survey results. Technical report 1216-1, Center for Transportation Research, University of Texas Austin. Kawashima, H. (1991). Two major programs and demonstrations in Japan. IEEE Transactions on Vehicular Technology, 40, 1, 141-146. Khattak, A. G., J. L. Schofer and F. S. Koppelman (1993). Commuters' en-route diversions and return decisions: Analysis and implications for Advanced Traveler Information Systems. Transportation Research A, 27A, 101-111. Kirasic, K. C, G. L. Allen and A. W. Siegel (1984). Expression of configurational knowledge of large-scale environments: Student's performance of cognitive tasks. Environment and Behaviour, 16,687-712. Kitamura, R. (1987) An evaluation of activity-based travel analysis. Transportation, 15,9-34. Kitamura, R. and T. Van der Hoom (1987). Regularity and irreversibility of weekly travel behaviour. Transportation, 15, 227-251. Koutsoupolos, H. N. and T. Lotan (1990). Motorist Information Systems and Recurrent Traffic Congestion: a Sensitivity and Analysis of Expected Benefits. In Transportation Research Record 1281, TRB National Research Council, Washington, DC, 148-159. Koutsopoulos, H. N., T. Lotan and Q. Yang (1994). A driving simulator and its application for modeling route choice in the presence of information. Transportation Research C, 2, (2): 91-107. Koutsopoulos, H. N., A. Polydoropoulou and M. Ben-Akiva (1995). Travel simulators for data collection on driver behaviour in the presence of information. Transportation Research C, 3,(3), 143-159. Kwan, M-P. (1994). A GIS-based model for activity scheduling in intelligent vehicle highway systems (IVHS). Ph.D. dissertation. University of California at Santa Barbara.


123

Kwan, M-P. (1995). GISICAS: An activity-based spatial decision support system for ATIS. Paper presented at the International Conference on Activity based Approaches: Activity Scheduling and the Analysis of Activity Patterns, May 25-28. Eindhoven University of Technology, The Netherlands. Kwan, M-P., and R. G. Golledge (1995). Integration of GIS with activity-based model in ATIS. Paper presented at the Annual Meeting of the Transportation Research Board, Washington, DC, January. Kwan, M-P., J. M. Speigle and R. G. Golledge (1997). Developing an object-oriented testbed for modeling transportation networks. (To be submitted to International Journal of GIS). Lakoff, G. (1987) Women,fire,and dangerous things: What categories reveal about the mind. Chicago, IL: University of Chicago Press. Leiser, D. and A. Zilberschatz (1989). The TRAVELLER: A computational model of spatial network learning. Environment and Behaviour, 21, 4: 435-463. Lenntorp, B. (1976). Paths in time-space environments: A time-geographic study of movement possibilities of individuals. Lund Series in Geography, Series B, Human Geography No. 44, University of Lund, Sweden. Lloyd, R. E. and R. Cammack (1995). Constructing cognitive maps with orientation biases. In J. Portugali (Ed.), The construction of cognitive maps. Dordrecht: Kluwer Academic Publishers. Loomis, J. M., R. L. Klatzky, and R. G. Golledge (1997). Human navigation by path integration. In R.G. Golledge (Ed.), Wayfinding: Cognitive maps and spatial behaviour. Baltimore: The Johns Hopkins University Press (Under Review). Madanat, S. M., C. Y. D. Yang and Y_M. Yen (1995). Analysis of stated route diversion under Advanced Traveler Information Systems using latent variable modeling. Transportation Research Record 1485, 10-17. Mahmassani, H. S. (1996) Dynamics of commuter behaviour: recent research and continuing challenges. In Lee-Gosselin and Stopher (eds.). Understanding Travel Behaviour in an Era of Change. Pergamon Press, New York. Mahmassani, H. S. (1997). Dynamics of commuter behaviour: recent research and continuing challenges. In P. Stopher and M. Lee-Gosselin (eds.). Understanding Travel Behaviour in an Era of Change. Pergamon/Elsevier Science Ltd., Oxford, UK Mahmassani, H. S. and G. L. Chang (1985). Dynamic aspects of departure time choice behaviour in a commuting system: Theoretical framework and experimental analysis. Transportation Research Record, 1037, 88-101. Mahmassani, H. S. and G. L. Chang (1987). On boundedly-rational user equilibrium in transportation systems. Transportation Science, 21 (2), 89-99. Mahmassani, H. S. and R. Herman (1990). Interactive experiments for the study of tripmaker behaviour dynamics in congested commuting systems. In P.M. Jones (Ed.), Developments in dynamic and activity-based approaches to travel analysis. Avebury: Aldershot, pp. 272-298. Mahmassani, H. S. and R. Jayakrishnan (1991). System performance and user response under real-time information in a congested traffic corridor. Transportation Research A, 25A, (5): 293-307. Mahmassani, H. S. and Y-H. Liu (1995). Commuter pre-trip route choice and en-route path selection under real-time information: Experimental results. Proceedings of the Second World Congress on Intelligent Transportation Systems, Yokohama, Japan.

124


Mahmassani, H. S. and S. Peeta (1993). Network performance under system optimal and user optimal assignments: implications for Advanced Traveler Information Systems. Transportation Research Record 1408, 183-193. Mahmassani, H. S., C. Caplice and C. M. Walton (1990). Characteristics of urban commuter behaviour: switching propensity and use of information. Transportation Research Record, 1285, 57-69. Mahmassani, H. S., G.L. Chang and R. Herman (1986). Individual decisions and collective efforts in a simulated traffic system. Transportation Science, 20, 362-384 Mahmassani, H. S., S. G. Hatcher and C. Caplice (1996). Daily variation of trip chaining, scheduling, and path selection behaviour of work commuters. In Lee, Sessler, and P. Stopher (Eds.), Understanding Travel Behaviour in an Era of Change. New York: Pergamon. Mannering, F. and H. Hamed (1990). Occurrence, frequency and duration of commuters workto-home departure delay. Transportation Research, V. 24B, 2. Medyckyj-Scott, D. J. and M. Blades (1990). Users' cognitive representations of space: Their relevance to the design of Geographical Information Systems. Midlands Regional Research Laboratory Report No. 13. Leicester, England: University of Leicester and Loughborough University. Medyckyj-Scott, D. J. and M. Blades (1992). Human spatial cognition: Its relevance to the design and use of spatial information systems. Geoforum, 23, 2: 215-226. Medyckyj-Scott, D. J. and H. M. Hemshaw (1993). Human factors in geographic information systems. London: Belhaven Press. Miller, H. J. (1991). Modelling accessibility using space-time prism concepts within geographical information systems. International Journal of Geographical Information Systems, 5,3:287-301. Miller, H. J. and J. V. Storm (1997). Geographic Information System designed for network equilibrium-based travel demand models. Dept. of Geography, Univ. of Utah, Salt Lake City. Moar, I., and Bower, G.H. (1983) Inconsistency in spatial knowledge. Memory and Cognition, 11,2: 107-113. Moar, I. and L.R. Carleton (1982). Memory for routes. Quarterly Journal of Experimental Psychology, 34A, 381-394. Montello, D. R. (1991). The measurement of cognitive distance: Methods and construct validity. Journal of Environmental Psychology, 11,2: 101-122. Office of Traffic Operations and IVHS (1991). Intelligent Vehicle-Highway Systems (IVHS) projects in the United States: Operational tests and research activities. FHWA, U.S. Department of Transportation. Parkes, A. M., M. C. Ashby and S. H. Fairclough (1991). The effects of different in-vehicle route information displays on driver behaviour. Proceedings of the Vehicle Navigation and Information Systems Conference, 1, Society of Automotive Engineers, Inc., Dearborn, Michigan, 61-70. Pas, E. I. and F. Koppelman (1986). An examination of day-to-day variability in individuals' urban travel behaviour. Transportation, 14, 13-20. Pas, E. I. and F. Koppelman (1987). An examination of the determinants of day-to-day variability in individuals' urban travel behaviour. Transportation, 13, 183-200. Pas, E. I. and F. Koppelman (1986). An examination of the determinants of day-to-day variability in individuals' urban travel behaviour. Transportation, 13, 183-200.


125

Ran, B. and D. E. Boyce (1994). Dynamic urban transportation network models: theory and implications for intelligent vehicle highway systems. Berlin: Springer-Verlag. Ran, B., R. W. Hull and D. E. Boyce (1996). A link-based variational inequality model for dynamic departure time/route choice. Transportation Research B, 30 B, 31-46. Recker, W. W., M. McNally and G. Root (1986a). A model of complex travel behaviour: Part I: Theoretical development. Transportation Research A, 20,4:307-318. Recker, W. W., M. McNally and G. Root (1986b). A model of complex travel behaviour: Part II: An operational model. Transportation Research A, 20, 4: 319-330. Rillings, J. H. and R. J. Betsold (1991). Advanced Driver Information Systems. IEEE Transactions on Vehicular Technology, 40, 1, 31-40. Road Transportation Research (1988) Route guidance and in-car communication systems. Road Transport Research, Organization for Economic Coordination and Development, Paris, France. Root, G. and W. Recker (1983). Towards a dynamic model of individual activity pattern formation. In S. Carpenter and P. Jones (Eds.), Recent advances in travel demand analysis. Aldershot: Gower. Rosenbloom, S. (1978). Peak-period traffic congestion: A state-of-the-art analysis and evaluation of effective solutions. Transportation, 10, 167-191. Saisa, J., A. Svenson-Garling, T. Garling and E. Lindberg (1986). Intra-urban cognitive distance: The relationship between judgments of straight-line distances, travel distances, and travel times. Geographical Analysis, 18, 2: \61-\1A. Shaw, S-L. (1993). GIS for urban travel demand analysis: Requirements and alternatives. Compute. Environ, and Urban Systems, 17, 15-29. Sheppard, E. (1995). Modeling and predicting aggregate flows. In S. Hanson (ed.) The Geography of Urban Transportation, second edition. New York: Guilford Press, 101128. Sholl, M. J. (1987). Cognitive maps as orienting schemata. Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 4: 615-628. Sholl, M. J. (1996). From visual information to cognitive maps. In J. Portugali (Ed.), The construction of cognitive maps (pp. 157-186). Dordrecht: Kluwer Academic Publishers. Siegel, A. W. (1981). The extemalization of cognitive maps by children and adults: In search of better ways to ask better questions. In L.S. Liben, A. Patterson, and N. Newcombe (Eds.), Spatial representation and behaviour across the life span: Theory and application. New York: Academic Press, pp. 167-194. Small, K. (1982). Scheduling of consumer activities: Work trips. American Economic review, 72, 467-479. Stopher, P. (1996). Measurement, Models, and Methods: Recent Applications. In P. Stopher and M. Lee-Gosselin (eds.), Understanding Travel Behaviour in an Era of Change. Elsevier Sciences Press, Tarrytown, New York. Stopher, P. (1996). Measurement, models, and methods: Recent applications. Chi7 in lATBR Selected Proceedings from the Quebec Conference. Stopher, P. and M. Lee-Gosselin (eds.) (1997). Understanding Travel Behaviour in an Era of Change. Elsevier Sciences Press, Tarrytown, N.Y. Streeter, L. A., D. Vitello and S. A. Wonsiewicz (1985). How to tell people where to go: Comparing navigational aids. International Journal of Man - Machine Studies, 22, 549562.

126


Summers, M. and R. Pilai (1996). Generic, object-oriented distributed k-shortest path system. Oakridge National Laboratory, Oakridge, Tennessee, 37831. Working paper. Summers, M. and F. Southworth (1998). Design of a testbed to assess alternative traveler behaviour models within an Intelligent Transportation System architecture. Geographic Systems (In Press). Supemak, J. (1992), Temporal utility profiles of activities and travel: Uncertainty and decision making. Transportation Research B, 26, 61-76. Taylor, M. A. P., W. Young, M. R. Wigam and K. W. Ogden (1992). Designing a large-scale travel demand survey: New challenges and new opportunities. Transportation Research, 26A, 3, 247-262. Timmermans, H., and R. G. Golledge (1990). Applications of behavioural research on spatial problems II: Preference and choice. Progress in Human Geography, 14, 311-354. van Aerde, M., D. Hellinga, M. Baker and H. Rakha (1996). INTEGRATION: An overview of traffic simulation features. Paper presented at the 1996 Transportation Research Board Meetings, January 1996, Washington, DC. Vaughn, K. M., M. A. Abdel-Aty, R. Kitamura and P. P. Jovanis (1993). Experimental analysis and modeling of sequential route choice behaviour under ATIS in a simplistic traffic network. Presented at the 72"^ Annual Meeting of the Transportation Research Board, Washington, DC. Walker, J., E. Alicandri, C. Sedney and K. Roberts (1990). In-vehicle navigation devices: the effects on the safety of driver performance. Proceedings of the Vehicle Navigation and Information Systems Conference, 1, Society of Automotive Engineers, Inc. Dearborn, MI. 499-525. Working Group on ADIS (1990) Advanced Driver Information Systems—Final report of the Working Group on ADIS. Mobility 2000, Dallas, Texas.


RESEARCH INTO AXIS BEHAVIOURAL RESPONSE: AREAS OF INTEREST AND FUTURE PERSPECTIVES

Ennio Cascetta andlsam A. Kaysi

INTRODUCTION Understanding and modeling behavioural response to Advanced Traveler Information Systems (AXIS) represents a significant area of focus for the travel behaviour research community. Early research in this domain simply ignored or oversimplified behavioural response. Evidence has indicated that inappropriate consideration of traveler reaction to information could lead to undesirable results including instability in system performance and inconsistency between information and observed traffic conditions (Ben-Akiva et al. 1991; Kaysi 1991; Cantarella and Cascetta 1995). Such a case would naturally compromise AXIS credibility and result in potential loss of confidence in traveler information. Xhis has led to a recognition that improved understanding and representation of traveler behaviour in the presence of information is needed, and has motivated research into the different areas covered by this workshop and described below. Xhis is an outgrowth of the deliberations of the workshop on "Dynamics and IXS Response" conducted during the Austin lAXBR meeting. Xhe workshop focused on user response to IXS and its implications for design and operation of such systems.

128


UNDERSTANDING AND MODELING BEHAVIOURAL RESPONSE TO

AXIS

Research efforts in understanding and modeling behavioural response to AXIS can be classified into four distinct but related research directions. To start with, research has focussed on identifying the dimensions as well as the dynamics of behavioural response; in other words, traveler behaviour modification due to the availability of AXIS. The second area of focus is on identifying and understanding the factors that affect traveler response and behavioural modification. Moreover, since much of the behavioural response reflects the availability of multiple sources of travel information, a third focus of research was on developing models of information representation and combination. Finally, the complexity of the interactions inherent in modeling traveler response to ATIS encouraged a fourth line of research which investigated innovative methodological approaches to deal with such complexity. Next, an overview of each of the four areas of research focussing on understanding and modeling behavioural response to ATIS is presented, and some of the related research efforts are discussed.

Dimensions and Dynamics of ITS/ATIS Behavioural Response Behaviour response dimensions do not necessarily imply changes in actual choices, but may also be related to perception and attitude. Several basic dimensions of response to ITS were considered in understanding traveler behaviour. Acceptance of ATIS services by travelers has been and continues to be of concern to researchers (Ygnace, 1991 and Van der Laan, 1997). Next, information acquisition and usage comprised additional dimensions of traveler, with Polydoropoulou et al. (1994) proposing modeling framework for the acquisition and processing of pretrip and en-route information. Moreover, research has addressed driving behaviour in terms of lane changing, speed adjustment, and, in general, the extent of aggressive driving. Evidence has indicated that availability of information may reduce the level of stress while driving. In terms of pre-trip and en-route driver choices, work has focused to a large extent on route choice and day-to-day route adjustment as well as departure time choice and day-to-day adjustment. Khattak et al (1991, 1993), Adler et al (1993), Polydoropoulou et al (1996), and Khattak and Khattak (1998) considered en-route driver behaviour in the context of information provision. Moreover, Khattak et al (1996), among others, investigated travelers' travel choices at the pre-trip stage and how they might respond to ATIS. Liu and Mahmassani (1998) developed day-to-day dynamic models of commuters' joint departure time and route switching decisions, and utilized them to investigate the dynamic effects of real-time traffic information

Research into ATIS Behavioural Response: Areas of Interest and Future Perspectives

129

on user decisions. Research in this domain has produced some general conclusions regarding, for instance, the impact of information sources, perceived congestion, expected delays, and travel times on route switching behaviour under different ATIS (Polydoropoulou et al, 1996). More comprehensive and conclusive evidence, however, is still contingent on progress in other areas, namely, understanding factors affecting traveler response, methodological development, dealing with information combination issues, and data availability. While significant research has been conducted on dimensions and dynamics of ATIS behavioural response, a rich research agenda lies ahead. "Second order" choice dimensions, including destination switching and trip cancellation, have received little attention so far, and need to be addressed within the context of activity scheduling. Moreover, consideration of transit rider pre-trip and en-route path decisions has lagged far behind in terms of research. Some models have recently been postulated, but significant empirical work has to be carried out in this area. Finally, there is need to investigate changes in perceptions of network structure and performance as well as modifications in attitudes towards other behavioural response dimensions as the use of ATIS becomes more prevalent. Information related to network performance and network topology, in addition to its direct role, can also influence the traveler's choice set - especially in a transit context.

Understanding Factors Affecting Traveler Response Evidence from research has indicated that there exist at least four classes of factors affecting traveler response to ITS/ATIS. First, ATIS attributes, or the type of information provided hy ATIS plays a significant role in this context. Thakuriah and Sen (1996) considered the impact of information quality on ATIS usage, while Fox and Boehm-Davis (1998) used a driving simulator to identify the effects of inaccurate traffic information on user trust in and compliance with ATIS advice. Kaysi et al (1994) concluded that the level of benefits to be derived from such ATIS depends to a large extent on the quality of the information being provided to travelers. However, information reliability and quality impacts remain underresearched. Second, trip characteristics such as trip purpose, trip chaining, trip length, and travel time reliability are all potentially significant factors in traveler response to ATIS and should be taken into account in the design and modeling of ATIS. The impact of driver familiarity with the choice context on route choice behaviour has been investigated by Lotan (1997), with various trips types (such as work or recreational trips) being associated with differential familiarity measures. Moreover, network characteristics such as network topology (actual and perceived), traffic congestion patterns, and potential pricing systems are also in need of further

130


investigation with respect to their role in influencing AXIS efficiency and traveler response. Finally, traveler characteristics including personality and socio-demographic traveler attributes (such as age and gender) reflect on attitudes and perceptions towards information use and their impacts on traveler behaviour need to be addressed. While research has indicated the significance of these factors in shaping traveler response to information, further work is still required, especially in establishing the relation between personality attributes and travel response. Research has also been targeting specific market segments in an effort to identify their specific information needs (commuters, non-familiar, tourists, etc.). The commuter segment has been the focus of most research, while other segments are clearly relevant and in need of further research. In addition, some researchers have suggested classifying users into categories based on their attitudes and resulting pre-disposition with respect to response to information. For instance, Ng et al (1998), in identifying driver information requirements for an AXIS, used cluster analysis to categorize the driver population based on trip factors. It was concluded that there exist significant differences in trip behaviour and socio-economic characteristics among observed cluster groups. In addition, Srinivasan (1999), using a simulation model of a multimodal pre-trip information system, concluded that different market segments for such information do exist, and that there is need for customized information provision which incorporates context of use considerations.

Information Representation and Combination Information representation relates to defining operational attributes of information such as value and confidence, and is of concern to the travel behaviour research community. Moreover, developing procedures for the combination of various information sources such as experience, en-route observations and perceptions, and multiple AXIS sources remains of interest and centrality. Some work has been done in this respect, primarily in the context of perception updating in day-to-day adjustment of travel behaviour (Jha et al. 1998). While research is advancing on this front, it is still at the early stages and needs to be integrated with modeling approaches and factors discussed above so that it may be of use in operational modeling frameworks.

Research into A TIS Behavioural Response: Areas of Interest and Future Perspectives

131

Figure 1. Structure of Research into ITS Behavioral Response

/IT/S Behavioral Response

Methodological Development Significant research is being conducted in this domain and is expected to pave the way for a more structured handling of issues discussed above in relation to traveler response to AXIS. It should be emphasized that not all models are suitable for all purposes. In that respect, models have been developed along two lines, the first focusing on models to be used in operational frameworks for design and evaluation while the second focus is on models providing detailed insights into behavioural and decision-making mechanisms. Several modeling approaches have been proposed to address traveler response including random utility theory and non-compensatory models such as bounded rationality, proposed by Mahmassani and Stephan (1988). Recent developments have included the integration of latent

132


variable models reflecting perceptions and attitudes in random utility models (Madanat et al., 1995; Polydoropoulou et al, 1997), endogenizing thresholds in satisficing behaviour (Srinivasan and Mahmassani, 1999), incorporating spatial and temporal correlations in traveler switching decisions, and explicit choice set modeling (Ben-Akiva et al, 1997, Cascetta and Papola, 1997, Cascetta and Russo, 1999). Moreover, fuzzy control concepts are finding their way into models of route switching (see for example Lotan, 1992), and cognitive models and approaches (e.g.. Decision Field Theory) are shedding light on localized traveler decision making. Further methodological developments should include comparing modeling approaches using the same data sets. Another area of methodological development relates to the demand-supply interactions under unpredictable (random) supply performances and AXIS information provision. The traditional equilibrium approach to transportation networks is not suited to model effectively these phenomena and different day-to day and within-day dynamic assignment models are being adopted (Mahmassani et al, 1986, Cascetta and Cantarella, 1991, and Cantarella and Cascetta, 1995) either under average (deterministic) or stochastic process framework. This is one area in which demand (i.e. behavioural responses) and supply (i.e. meso or micro simulation of traffic networks) models merge in a system-wide model.

RESEARCH

CONTRIBUTIONS

TO

IDENTIFYING

DATA

NEEDS

AND

IMPROVING DATA QUALITY A preferred mode of data collection should rely on integrated and simultaneous field measurements of different factors potentially affecting behaviour in a travel information context (such as those listed above). This depends on the availability of large-scale fully implemented ATIS, or, in the interim, on field operational tests. Given the limited availability of fully-implemented ATIS, researchers have been resorting to either partial field studies or to Stated Preference (SP) techniques. SP techniques in this context have included attitudinal questionnaires as well as complex travel simulators. Evidence indicates that the degree of realism of the SP experiments has a significant bearing on the reliability of the results. An additional consideration relates to the fact that models embedded in simulators may limit the validity of the results. Early surveys addressed the acquisition of radio traffic reports and the influence of information on drivers' travel behaviour (Khattak et al, 1991 and 1994; Mahmassani et al, 1989 and 1991). Most surveys asked respondents to recall the impact of information on their route


13 3

choice or departure time decisions. For instance, de Palma and Khattak (1994) report a survey conducted in Brussels to explore the impact of various factors on travel decisions, including traffic conditions, unexpected congestion, traffic information, weather conditions, and personal and household attributes. Moreover, Khattak and de Palma (1994) provide a detailed analysis of drivers' propensity to change travel decisions in normal and adverse weather given the availability of guidance provided by radio reports, based on the Brussels survey resuhs. Other types of surveys have obtained detailed diaries of daily trip behaviour and information acquisition and usage by respondents for a designated period of time. A prototypical survey of this type was conducted at MIT in 1991 (see Kaysi 1991 and Lotan 1992 for survey design considerations and Polydoropoulou e/ al., 1994, for data analysis and modeling using survey data). In addition, Abdel-Aty et al. (1997) utilized SP data to investigate the impact of AXIS on route choices of drivers. A combined revealed and stated preference model of traveler response in pretrip and enroute travel contexts was reported by Khattak et al. (1996) and Polydoropoulou et al. (1996), respectively. A number of studies have focused on the use of simulators to analyze driver behaviour under various scenarios of information provision. These studies put drivers in simulated driving circumstances and observe their reactions and travel behaviour. The simulator would "observe" driver decisions and store such information for later analysis. Simulators enable the creation of a wide range of systematically varied driwing situations under laboratory conditions (controlled environment) and the collection of relatively inexpensive data. Travel decision simulators represent an important means through which data on traveler response to ATIS services may be obtained. Mahmassani et al. (1986), Mahmassani and Tong (1986), and Tong et al. (1987) were amongst the first researchers to use an experimental procedure to investigate the effect of information availability on the dynamics of user behaviour in urban commuting systems. Bonsall and Parry (1991) also developed an interactive route-choice simulator to investigate drivers' compliance with route guidance advice. Other early efforts in this domain have been reported by Allen et al. (1991) and lida et al. (1992). Moreover, Halati and Boyce (1992) evaluated the effectiveness of various types of in-vehicle navigation and guidance systems in alleviating traffic congestion through computer simulation. The use of simulators to collect data for estimation of driver behaviour models under the influence of information has also been reported by Adler et al. (1993), Koutsopoulos et al. (1993), Vaughn et al. (1993 and 1995), Chen and Mahmassani (1993), and Bonsall (1997). These experiments have concluded that en-route diversion behaviour is influenced by familiarity of travelers with traffic conditions on potential alternative routes, information

134


provided, and the travelers' risk preferences and that acceptance of advice varied with its quality as well as the quality of recently received advice, the existence of conflicting evidence, and the travelers' knowledge of the network. Finally, Koutsopoulos et al (1995) reviewed existing travel decision simulators and provided an overall evaluation of the use of such simulators to obtain data on traveler response to AXIS. The review indicated that existing simulators have been utilized to collect relatively good data regarding user response to unimplemented AXIS services as far as route choice is concerned, whereas limited data have been obtained for dimensions other than route choice in the overall travel response behaviour. Finally, it was observed that no travel simulator is capable of replicating actual travel conditions exactly; however, quantifying the degree of inconsistency necessitates the availability of RP data which is currently lacking. Driving simulators are different from travel simulators in that they focus on the human factors aspects associated with the provision of information or guidance through AXIS. Driving simulators are used in general for a variety of objectives, including evaluating the mental workload associated with driving, the effect of fatigue, and drivers' comprehension of road signs. A number of driving simulators have considered such driver performance aspects in the context of the provision of information or guidance through AXIS. Examples include simulators developed at the Hughes Aircraft Company (Hein 1993) and the XNO Institute for Human Factors (Van der Mede and Van Berkum, 1991). In the future, it is envisioned that several data collection techniques will have to be adopted to collect different data types (network, traveler, AXIS, and response). While conventional questionnaires can be used to collect basic traveler characteristics, emerging techniques (such as pattern recognition and GPS/GIS technologies) could be adopted to obtain network conditions and to track actual traveler trajectories. Xhis integrated data collection effort represents a priority item on the research agenda of traveler response to AXIS.

RESEARCH CONTRIBUTIONS TO

AXIS DESIGN ASPECTS AND BENEFITS

EVALUATION Behavioural research has made significant contributions to design aspects of AXIS services, in particular with respect to identifying traveler information needs and preferences and the human factors implications and constraints of information provision systems. Moreover, research has been instrumental in evaluating the potential benefits of AXIS schemes form both the network and traveler side, including the development of some willingness to pay measures.


13 5

Contributions of Behavioural Research to AXIS Design Three types of AXIS systems can be identified in relation to design issues, namely, originbased, road-side/area-wide, and in-vehicle systems. Behavioural research contributes to a number of design issues common to the above three types. First, it contributes to the identification of driver information needs. Mannering et al (1995) and Ng et al. (1995) investigated driver preference for navigational information. In a similar manner, Yang et al. (1998) used a driving simulator to identify differences in ATIS information that meet drivers' needs when traveling in familiar and unfamiliar networks, and to recognize desirable formats of transmitting information. Moreover, the need for internal consistency in information provision has been recognized for some time (see Kaysi 1991) and is being addressed by behavioural research. In addition, the incorporation of human factors attributes including workload and information delivery considerations into ATIS design has been identified as a necessity. Work by Dingus and Hulse (1993), among others, has contributed to this aspect. Other issues such as required update frequency (Kaysi et al 1994), liability avoidance, and compatibility with other information sources have been dealt with in behavioural research. Other aspects related specifically to origin-based and in-vehicle systems and which still need to be addressed include customization requirements and the provision of post-trip feedback.

Contribution of Behavioural Research to Evaluation of ATIS Benefits Behavioural response research should contribute to the overall evaluation of ATIS systems. This evaluation should be based on a number of indicators. First, indicators related to network efficiency, such as travel time, waiting time, queue length, fuel consumption need to be developed. Behavioural models can help in simulating these effects in the context of dynamic traffic assignment models. A study by Khattak et al (1994) calculated travel time savings achieved by ATIS-induced route diversion and translated it into monetary benefits. Benefits were determined based on reported and stated behavioural responses to unexpected congestion. Other indicators of value are related to user satisfaction in connection with travel and activityrelated choices as well as reductions in travel-related stress and uncertainty. User satisfaction with traveler information systems usually leads to their adoption and continued use. While some research has considered how user-perceived benefits may be translated into willingnessto-pay for such systems and services (Polydoropoulou et al 1997), more work is needed to develop further such measures. In addition to the design-related issues, research can contribute to evaluation of ATIS based on behavioural measures of satisfaction rather than simply network or system-wide measures of

136


effectiveness. This would represent an interesting and new line of research that may require different data collection tools and analysis approaches.

CONCLUSIONS This overview of the main areas of behavioural response to ATIS has shown the theoretical and operational potential of this field. In particular, it can be concluded that, during the workshop, seven areas of interest and activity have been defined and the relation between them is depicted in Figure 1. In this figure, a structure is proposed whereby various research efforts into ITS behavioural response may be viewed as contributing to one or more of three domains: (i) identifying data needs and improving data quality, (ii) understanding and modeling behavioural response, or (iii) design aspects and benefits evaluation of ATIS. Although this represents a relatively new area of behavioural research, it is certainly one of the most active and challenging ones as new dimensions of traveler behaviour are introduced and a closer connection of research results with actual system design and evaluation is established. In particular, one of the main theoretical contributions to behavioural research as a whole has been the recognition of the central role that information plays in traveler response and the need to better understand and model such phenomena. However, as interesting and innovative as this area may be, its future developments are dependent on the extent to which ATIS will be implemented and operated in the real world. As a matter of fact, only large scale deployments of ATIS will supply enough motivation and empirical evidence to allow this sector to reach the level of maturity of other areas of travel behaviour research.

ACKNOWLEDGEMENTS The authors wish to thank the participants in the workshop on "Dynamics and ITS Response" at the 8^^ Meeting of the International Association for Travel Behavior Research for their valuable contribution to the framework presented in this chapter and which has been the result of a truly collective effort. Participants included: J. Adler, E. Cascetta, P. Chen, T. J. Cherrett, K. Delvert, L. Engelson, R. Golledge, E. Hato, N. Huynh, D. Jasperse, R.-C. Jou, H. A. Katteler, I. Kaysi, M. Namgung, E. Parkany, V. Shah, N. Sobhi, G. H. M. Speulman, E. Stem, Y. Sugie, M. F. A. M. van Maarsevee, H. Wakabayashi, K. Westin, and S. Zhao.


13 7

REFERENCES Abdel-Aty, M., R. Kitamura and P. Jovanis, P. (1997). Using stated preference data for studying the effect of advanced traffic information on drivers' route choice. Transportation Research 5C(1), 39-50. Adler, J.L., W. W. Recker and M. G. McNaily (1993). A conflict model and interactive simulator (FASTCARS) for predicting enroute assessment and adjustment behavior in response to real-time traffic condition information. Transportation 20, 83-106. Allen, R.W., D. Ziedman T. Rosenthal A. Stein J. Torres and A. Halati (1991). Laboratory Assessment of Driver Route Diversion in Response to In-Vehicle Navigation and Motorist Information Systems. Transportation Research Record 1306, 82-91. Ben-Akiva, M., A. de Palma and I. Kaysi (1991). Dynamic network models and driver information systems, Transportation Research 25A(5), 251-266. Ben-Akiva, M., J. Walker, A. Bernardino, D. Gopinath, T. Morikawa and A. Polydoropoulou (1997). Integration of choice and latent variable models. Preprints of the 8^^ Annual Meeting of the International Association for Travel Behavior Research, Austin, Texas, U.S.A. Bonsall, P.W. and T. Parry (1991). Using an Interactive Route-Choice Simulator to Investigate Drivers' Compliance with Route Guidance Advice. Transportation Research Record 1306, 59-68. Bonsall, P., P. Firmin, M. Anderson, I. Palmer and P. Balmforth (1997). Validating the results of a route choice simulator. Transportation Research 5C(6), 371-387. Cantarella G.E., and E. Cascetta (1995). Dynamic Processes and Equilibrium in Transportation Networks: Towards a Unifying Theory. Transportation Science 29, 305-329. Cascetta E., and G. E. Cantarella (1991). A day-to-day and within-day dynamic stochastic assignment model. Transportation Research 25A, 277-291. Cascetta, E., and A. Papola (1997). Random utility models with implicit availability/perception of choice alternatives. Preprints of the 8f^ Annual Meeting of the International Association for Travel Behavior Research, Austin, Texas, U.S.A. Chen, P. and H. Mahmassani (1993). Dynamic interactive simulator for studying commuter behavior under real-time traffic information supply strategies. Transportation Research Record UU,\2-2l. de Palma, A. and A. J. Khattak (1994). Understanding Commuters' Travel Decisions in Brussels: An Overview. Working Paper, COMIN Department, University of Geneva. Dingus, T. and H. Hulse (1993). Some human factors design issues and recommendations for automobile navigation information systems. Transportation Research 1C(2), 119-132. Fox, J. and D. Boehm-Davis (1998). Effects of age and congestion information accuracy of advanced traveler information systems on user trust and compliance. Transportation Research Record 1621, 43-49. Halati, A. H. and D. Boyce (1992). Simulation Evaluation of In-Vehicle Navigation Systems Methodology and Application. Paper presented at the 71st Annual Meeting of the Transportation Research Board, Washington, D.C., January. Hein, C. M. (1993). Driving Simulators: Six Years of Hands-on Experience at Hughes Aircraft Company. Proceedings of the Human Factors and Ergonomics Society 37th Annual Meeting, Seattle, Washington, U.S.A. lida, Y., T. Akiyama and T. Uchida (1992). Experimental Analysis of Dynamic Route Choice Behavior, Transportation Research 26B, 17-32.

138


Jha, M., S. Madanat and S. Peeta (1998). Perception updating and day-to-day travel choice dynamics in traffic networks with information provision. Transportation Research 6C(3), 189-212. Kaysi, I. (1991). Framework and models for the provision of real-time driver information. Ph.D. Dissertation, Department of Civil Engineering, MIT, U.S.A. Kaysi, I., M. Ben-Akiva and A. de Palma (1994). Design aspects of advanced traveler information systems. Urban Traffic Networks, eds. N. Gartner and G. Improta. Khattak, A. J., J. L. Schofer and F. S. Koppelman (1991). Factors influencing commuters' enroute diversion behavior in response to delay. Transportation Research Record 1318, 125-136. Khattak, A. J., J. L. Schofer and F. S. Koppelman (1994). Commuters' enroute diversion and return decisions: analysis and implications for advanced traveler information systems. Transportation Research 27A(2), 101-111. Khattak, A. J. and A. de Palma (1994). Results of a Behavioral Survey in Brussels and the Propensity to Change Travel Decisions in Normal and Adverse Weather. Paper presented at the 7th World Conference on Transport Research. Khattak, A., A. Kanafani and E. Le Colletter (1994). Stated and reported route diversion behavior: implications of benefits of advanced traveler information system. Transportation Research Record 1464, 28-35. Khattak, A., A. Polydoropoulou and M. Ben-Akiva (1996). Modeling revealed and stated pretrip travel response to advanced traveler information systems. Transportation Research Record 1537, 46-54. Khattak, A. and A. Khattak (1998). Comparative analysis of spatial knowledge and en route diversion behavior in Chicago and San Francisco: implications for advanced traveler information systems. Transportation Research Record 1621, 27-35. Koutsopoulos, H., T. Lotan and Q. Yang (1993). A driving simulator and its application for modeling route choice in the presence of information. Paper presented at the 72nd Annual Meeting of the Transportation Research Board, Washington, D.C. Koutsopoulos, H., A. Polydoropoulou and M. Ben-Akiva (1995). Travel simulators for data collection on driver behavior in the presence of information. Transportation Research 3C(3), 143-159. Liu, Y. and H. Mahmassani (1998). Dynamic aspects of commuter decisions under advanced traveler information systems: modeling framework and experimental results. Transportation Research Record 1645, 111-119. Lotan, T. (1992). Modeling route choice behavior in the presence of information using concepts from fuzzy set theory and approximate reasoning. Ph.D. Dissertation, Department of Civil Engineering, MIT, U.S.A. Lotan, T. (1997). Effects of familiarity on route choice behavior in the presence of information. Transportation Research 5C(3-4), 225-243. Madanat, S., C. Yang and Y. Yen (1995). Analysis of stated route diversion intentions under advanced traveler information systems using latent variable modeling. Transportation Research Record 14S5, 10-17. Mahmassani, H. and C. C. Tong (1986). Availability of Information and Dynamics of Departure Time Choice: Experimental Investigation. Transportation Research Record 1085,33-47. Mahmassani, H., G. Chang and R. Herman (1986). Individual Decisions and Collective Effects in a Simulated Traffic System. Transportation Science 20(4), 258-271.


139

Mahmassani, H. and D. Stephan (1988). Experimental investigation of route time and departure time choice dynamics of urban commuters. Transportation Research Record 1203,63-83. Mahmassani, H., C. G. Caplice and C. M. Walton (1989). Characteristics of Urban Commuter Behavior: Switching Propensity and Use of Information. Transportation Research Record 12S5, 57-69. Mahmassani, H., G. S. Hatcher and C. G. Caplice (1991). Daily Variation of Trip Chaining, Scheduling, and Path Selection Behavior of Work Commuters. Proceedings of the 6th International Conference on Travel Behavior, Quebec City, Canada. Mannering, F., S. Kim, L. Ng and W. Barfield (1995). Travelers' preferences for in-vehicle information systems: an exploratory analysis. Transportation Research 3C(6), 339-351. Ng., L., W. Barfield and F. Mannering (1995). A survey-based methodology to determine information requirements for advanced traveler information systems. Transportation Research 3C(2l 113-127. Ng., L., W. Barfield and F. Mannering (1998). Analysis of private drivers' commuting and commercial drivers' work related travel behavior. Transportation Research Record 1621,50-60. Polydoropoulou, A., M. Ben-Akiva and I. Kaysi (1994). Revealed preferences models of the influence of traffic information on drivers' route choice behavior. Transportation Research Record 1453, 56-65. Polydoropoulou, A., M. Ben-Akiva A. Khattak and G. Lauprete (1996). Modeling revealed and stated en-route travel response to advanced traveler information systems. Transportation Research Record 1537, 38-45. Polydoropoulou, A., D. Gopinath and M. Ben-Akiva (1997). Willingness to pay for advanced traveler information systems: Smartraveler case study. Transportation Research Record 1588, 1-9. Srinivasan, K. and H. Mahmassani (1999). Role of congestion and information provision in tripmakers' dynamic decision process: an experimental investigation. Preprints of the 78^ Annual Meeting of the Transportation Research Board. Washington, D.C. Srinivasan, K. (1999). Pre-trip information systems (PTIS): an investigation into users' information acquisition process. Preprints of the 78^^ Annual Meeting of the Transportation Research Board. Washington, D.C. Thakuriah, P. and A. Sen (1996). Quality of information given by an advanced traveler information system. Transportation Research 4C(5), 249-266. Tong, C. C, H. Mahmassani and G. L. Chang (1987). Travel time Prediction and Information Availability in Commuter Behavior Dynamics. Transportation Research Record 1138, 1-7. Van der Laan, J., A. Heino and D. de Waard (1997). A simple procedure for the assessment of acceptance of advanced transport telematics. Transportation Research 5C(1), 1-10. Van der Mede, P.H. and E. C. Van Berkum (1991). Modeling Route Choice, Inertia, and Responses to Variable Message Signs. Proceedings of the 6th International Conference on Travel Behavior, Quebec City, Canada. Vaughn, K. M., M. Abdel-Aty, R. Kitamura, P. Jovanis and H. Yang (1993). Experimental analysis and modeling of sequential route choice behavior under ATIS in a simplistic traffic network. Transportation Research Record 1408, 75-82. Vaughn, K. M., R. Kitamura and P. Jovanis (1995). Experimental analysis and modeling of advice compliance: results from advanced traveler information system simulation experiments. Transportation Research Record 1485, 18-26.

140


Yang, C, J. Fricker and T. Kuczek (1998). Designing advanced traveler information systems from a driver's perspective: results of a driving simulation study. Transportation Research Record 1621, 20-26. Ygnace, J. L. (1991). An example of consumer acceptance of route guidance technologies. Paper presented at the 70th Annual Meeting of the Transportation Research Board, Washington, D.C.

SECTION 3 TELECOMMUNICATIONS-TRAVEL INTERACTIONS



EMERGING TRAVEL PATTERNS: Do TELECOMMUNICATIONS MAKE A DIFFERENCE?

Patricia L Mokhtarian andllan Salomon

ABSTRACT This chapter reviews empirical studies of the relationships between telecommunications and travel. The studies are classified into three approaches: macro-scale, micro-scale application-specific, and micro-scale comprehensive (activity-based). Within the second category we review the literature on the applications of telecommuting, teleconferencing, teleshopping, and the telephone. A diversity of relationships is identified, with some studies finding complementarity and others finding substitution. However, the preponderance of evidence suggests that the net impact is complementarity, and continued growth in both telecommunications and travel should be expected. Hypotheses and directions for future research are discussed, including the need to further develop the comprehensive activity-based approach and to synthesize accounting exercises with behavioral modeling approaches to yield causal forecasts of the impacts of telecommunications on travel.

INTRODUCTION By definition, human interactions depend on communications among individuals and institutions. Such communications have taken place since the early days of civilization, through the use of three basic modes: (1) traveling for the purpose of physical presence; (2) various forms of remote interaction, from smoke signals to modem telematics; and (3) the exchange of a physical object containing information. The last mode combines elements of the

144


other two: a trip to deliver or receive the object is required, but the origination and receipt of the communication are separated in time and (often) space. Given that communications of both forms involve costs but convey different benefits, questions about the relative use of travel versus telecommunications" are warranted. In fact, ever since the invention of the telephone, more than a century ago, the issue of relationships between travel and telecommunications has been addressed (Pool, 1983). Travel patterns are continuously changing. As numerous time series clearly demonstrate, there is a universal increase in the reliance on the automobile, despite its social costs (Schafer and Victor, 1997); there is a widespread reduction in the use of public transport, even in most European cities; and people travel more, to more distant locations, at least during their holidays. Much has been written about the underlying causes for these developments: economic change that makes automobile purchase and use cheaper, social and demographic changes, particularly the changing role of women in society, changes in land use patterns and more. As some of the travel trends are considered to be socially inefficient, there is some interest in ways to mitigate excess travel. Technological changes, and the accompanying social changes, are suggested to offer such remedies. Do they make a dent in the overall current or future trends? Current interest in the relationships between transportation and telecommunications can be attributed to two separate trends. First, there is in recent decades a growing awareness of the full social costs of travel, as congestion and pollution are reaching high levels in some areas. Second, there is a rapid transition into what is often labeled 'the information age', in which information becomes a central asset in the economy, and information technology becomes a popular and inexpensive means for processing and distributing information resources. The intensity of these disparate trends has reached a level where information technology is now suggested to be a (partial) solution for the growing transport costs. This review intends to explore exactly that issue: Do telecommunications make a difference? The difference may be of relevance in two respects. First, does the growing availability of information technology, and the growth in the role of information in society, lead to changes in travel patterns that can be part of the solution for the infamous ills of transportation? This phrasing of the question seeks to establish a degree of substitution of telecommunications for travel. Second, whether or not substitution is a major phenomenon, what other types of changes in travel behaviour can be expected?

Emerging Travel Patterns: Do Telecommunications Make a Difference?

145

To that end, we review the research of the last decade or so. The literature in the field seems to have matured to the point that there are sufficient conceptual and empirical studies worthy of attention, in contrast to previous scholarly work which included many descriptive and expository articles of a purely speculative nature. Telecommunications and information technologies have penetrated many facets of human lives. Previously more limited to professional and employment situations, now such technologies are being used for a variety of domestic purposes, from household maintenance to leisure activities. People are exposed to IT in many different forms and contexts and consequently, their behavioural responses are affected by these multiple encounters with technology. In the remainder of the introduction, we briefly discuss the notions of substitution and complementarity and also briefly review the types of studies found in this field.

A Typology of Relationships A number of authors (ECMT, 1983; Mokhtarian, 1990; Niles, 1994; Salomon, 1985, 1986; USDOE, 1994) have described the different types of relationships between the two spatial technologies (transport and telecommunications) in various ways. We do not review that literature extensively here, but mention four main types of relationships: substitution (elimination, reduction), generation (stimulation, complementarity), modification, and neutrality. The first two are discussed at greater length below. By modification we mean, for example, that telecommunications may alter the time, mode, destination, route, or other characteristics of a trip that would have been made regardless.^ By neutrality we mean, for example, that telecommunications has no impact on travel (as when the predominant effect of e-mail seems to be the generation of more e-mail; Balepur, 1997). Interest in the telecommunications - transport interaction is twofold. Much attention in recent years is based on the underlying question of whether or not the two are substitutive entities^. If they exhibit such a relationship, then identifying its extent is of interest as an input to policymaking processes. This focus is paramount to policy makers, and to suppliers as well. However, the behavioural research perspective is very different. Whether or not the relationship has immediate policy implications is less important. The implications of technology on behaviour are of interest, regardless of whether they have a 'posifive' policy

146


potential. We should attempt to understand the wide variety of factors that facilitate or restrict travel and activity patterns. In this chapter, we primarily focus on the two main types of interactions of interest: substitution 'versus' complementarity (or generation, or stimulation). We place 'versus' in quotes because, although the two concepts are contrasts, they are not mutually exclusive. The growth over the last century in the use of both telecommunications and travel, at almost any level of analysis, hints at a clear complementary relationships On the other hand, the widespread availability and daily use of such sophisticated yet inexpensive communications will inevitably, it is often argued, substitute for increasingly costly trips. Thus, it is the simultaneous presence of both types of impacts that makes the study of this issue so challenging. Whereas substitution seems to be a more straightforward relationship, for which the main question is the extent to which it is occurring, complementarity seems to be a more complex relationship, by nature. For two entities to be substitutive they need to offer similar functions with similar properties. A user will be indifferent if two entities provide essentially the same utility at similar costs and the indifference is interpreted to imply perfect substitution. Indifference is an extreme point in the distribution of preference. Slight differences in the quality of products will result in a removal from indifference to preference of one over the other. Substitution may result from the fact that one option is more attractive, and/or because it provides similar utility but has lower costs. As individuals differ in their preferences for particular attributes, substitution too may take place for some people and some of the time. Cases of technological substitution such as the transition from wood or metal to fiberglass boats, or from turbo-prop to jet engines (Linstone and Sahal, 1976), are often cited as examples of situations where one option dominates the other on all or at least the important determinants of preference. However, in many cases there is no clear dominance. So, the question of substitution between travel and telecommunications must examine the extent to which travel and telecommunication are indeed similar. Complementarity, on the other hand, depends on the degree to which each technology or service elicits the use of the other. The range of possible relationships here is wide. Perfect complementarity exists when one mode cannot be used without the other. This is clearly the case for face-to-face communications, which cannot be accomplished without travel. However, complementarity also relates to situations in which the use of one mode encourages the use of the other. For example, the more one travels, the more one tends to use a mobile phone. The use of both modes need not be simultaneous, however: the discovery of a colleague


147

on the Internet may later generate a trip to meet in person. All of these types of complementary relationships may be considered examples of enhancement, in which the use of one mode of communication directly increases the use of another mode. Another type of complementary relationship deals with situations where one mode makes the use of another more efficient. Real-time or short-term coordination of meetings by use of mobile communications serves as an example of an efficiency effect of telecommunications on face-to-face communication. More generally, face-to-face meetings are almost inevitably accompanied by a variety of electronic communications both before (to set them up and provide supplemental information - a case of efficiency) and after (to continue conversations begun at the meeting - a case of enhancement). Real-time information collected and distributed through telematics is indispensable to realizing the Intelligent Transportation System (ITS) goal of increasing the effective capacity of the transportation system. In the other direction, improved communications (and fewer errors) using telematics-based modes may be attributed to personal acquaintance based on face-to-face meetings between the parties involved. In considering aggregate measures of telecommunications and travel that demonstrate simultaneous increases in both, it is clear that part of the observed relationship is due to thirdparty correlation of each measure with indicators of economic activity (Helling and Mokhtarian, 2000). To some extent these correlations are spurious - reflecting separate relationships of telecommunications and travel with economic activity but not with each other. However, the simultaneous growth in telecommunications and travel also reflects both types of complementarity described above - a direct causal relationship between the two measures with or without the involvement of economic indicators as additional causal factors. The degree to which each type of relationship accounts for the observed trends is unknown, and constitutes a fertile subject for future research. Helling and Mokhtarian also point out that the relative quantities of travel versus communication consumed are partly functions of income effects and of substitution effects. The income effect indicates that both travel and communication increase with real incomes. The substitution effect indicates that communication can increasingly be substituted for travel, as the enabling technology becomes more realistic, affordable, ubiquitous, and easy to use. The net outcome for communication is clearly an increase, since both income and substitution effects favor that result. The outcome for travel is more complex, with an ambiguous net result, since the two effects act in opposite directions.

148


A Typology of Studies Dozens of articles and books addressing, directly or indirectly, the topic at hand have been published in the last fifteen years. The studies of this subject vary in so many dimensions that a formal classification is beyond the scope of our present task (see for example, Salomon, 1986; Salomon, 1998). Briefly, however, some major classes should be described. One important reason for classifying the literature in the field is to distinguish between studies that are based upon some theory or scientific approach and those that are closer to science fiction or commentaries. Writings about the impacts of technology commonly include both types, and much of the widespread expectations with regard to technological fixes emanate from the latter. A classification at this point in time is important, among other reasons, in order to avoid the indiscriminate citation of studies that often lead to premature conclusions. Much of the literature in the field is based on 'armchair' exercises in which ideas regarding possible relationships are exposed for further study. These are usually expectations that draw on the experience with other technologies and on the interpretation and professional judgment of the authors based on available knowledge in relevant fields. At the opposite end of the spectrum from the armchair research lie the empirical studies, which undertake the testing of specific hypotheses on the basis of revealed behaviour. All studies, in fact, fall somewhere on a continuum between the armchair and empirical approaches. Another criterion for classification is the time horizon addressed in the studies. Some studies focus on the short-term while others deal with long-term perspectives. There are two important differences between these types. First, long-term analyses entail much greater uncertainty with regard to technological advances. This is a serious restriction in an era characterized by rapid dynamics. Second, and even more difficult to address, long-term studies face more uncertainty regarding changes in societal values and norms. Social scientists who address such issues may be at risk of dealing with science fiction. Short-term studies can avoid this pitfall, but consequently, may be poor predictors. Studies also differ in the range of relationships they address. Most focus on the direct impacts, namely the extent to which the use of one technology directly affects the relative use of the other. Fewer studies (e.g., Lund and Mokhtarian, 1994; Salomon, 1996) also address indirect impacts like the effects of telecommunications on land use, and through such changes, on travel behaviour.^


149

Yet another distinction can be made between studies, in terms of the breadth of the question addressed. Some studies are comprehensive, trying to unveil the relationship between transportation and telecommunications in a broad context. In this category, for example, a focus on substitution of a single trip type (e.g., work or shopping) would be 'unjustified', as it is possible that changes in travel with respect to one trip type will change not only the entire communication and trip-making patterns of the individual, but also the activity patterns of other household members. Comprehensive studies explicitly recognize the interrelationships among the various modes (and submodes) of communication - face-to-face (involving passenger travel), exchange of a physical object (involving goods movement), and telecommunications - and focus on the combined, system wide effects of such relationships. Other studies are limited to the analysis of particular situations, with implied or explicit recognition (or ignorance) of the broader context within which the limited case is relevant. Finally, the distinction can be made between macro-scale (aggregated to regional, national, or international levels) and micro-scale (disaggregate, individual-level) approaches. For the current context we categorize the empirical studies reviewed into three research approaches, presented in Table 1, which shows a cross classification of Limited vs. Comprehensive studies and Macro-scale vs. Micro-scale studies. This results in three research approaches, discussed in the subsequent sections of the chapter. One cell of Table 1 is empty, since there are, to the authors' knowledge, no empirical studies that examine a limited scope on a macro scale (there are several such studies of a hypothetical nature in the areas of telecommuting and teleconferencing, e.g. USDOT, 1993; USDOE, 1994; Harkness, 1977). To maintain a manageable scope for this chapter, we have limited our review specifically to studies which have either provided a theoretical or conceptual frameworkybr the relationships between telecommunications and travel, developed testable hypotheses on those relationships, or in fact tested such hypotheses in an empirical or quasi-empirical context. We do not extensively review studies that confine themselves to modeling the adoption of telecommunications-related activities (such as telecommuting or teleshopping), although we discuss the role of such studies in a later section. The analogy between the evolution of the research in this area and in travel behaviour is illustrative. Early travel modeling focused on aggregate studies and evolved into disaggregate, micro-level analyses. Initially, behavioural models focused on specific choice situations (e.g., mode choice, destination and so on), and later the focus shifted to activity-based approaches. In studying the relationship of travel to telecommunications, it seems that we have now reached the point at which we need to progress from specific applications to activity-based approaches. We will return to this in the concluding section.

150

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Table 1 A Classification of Research Approaches Scope of Coverage Limited

Scale

Macro Micro

1

(not used)

Comprehensive 1

Industrial and consumer contexts

Application-specific

Activity analysis

Further sections address each of the three research approaches shown in Table 1. In each case, we provide a brief description of the approach, summarize any empirical results to date, and assess the advantages and disadvantages of the approach. We also examine the importance of behavioural modeling of activity mode selection to understanding the relationship between telecommunications and travel. There is also discussion of the implications of the literature reviewed here and suggestion of directions for further research.

THE MACRO-SCALE COMPREHENSIVE APPROACH Description of approach: The macro-scale comprehensive approach to studying telecommunications B transportation relationships analyzes transportation and communication sectors of the economy in the aggregate to determine the net impact of each sector on the other(s). To date, the macro-scale analyses undertaken have focused on the national and international scales, but the same methods could be applied to state, regional, or even metropolitan economies if those were of particular interest and if the appropriate data were available at those levels. To the authors' knowledge, only three studies using the macro-scale approach have been published, and they are complementary both in methodology and in results. Selvanathan and Selvanathan (1994) used the Rotterdam demand system, a set of equations simultaneously modeling the demand for multiple commodities, to analyze relationships in consumer demand for the private transportation, public transportation, and communication sectors of the economy. They compared the United Kingdom and Australia, using time-series data for the period 1960-1986. The Netherlands Organization for Applied Scientific Research (1989, cited in Button and Maggi, 1994) has taken a similar approach. Plaut (1997), on the other hand, argues that industrial uses account for half to two-thirds of all expenditures on transportation and communication in the U.S. and Europe. Accordingly, she used cross-sectional input/output


151

analysis to analyze industrial demand for transportation and communication in nine member countries of the European Union in 1985. Results: The results of the studies are intriguing. Selvanathan and Selvanathan found that, at the consumer level, private transportation, public transportation, and communication were pairwise substitutes, but with relatively small price elasticities (e.g., in the UK, the price elasticity of communication with respect to private transport was 0.57). They further found exponential growth in communication, at the expense of the two types of transportation. Similar results have also been put forward by NOASR (1989), suggesting relatively low elasticities and a reduction of travel by only 8% over the next 35 years. Plaut, by contrast, found that at the industrial level, transportation and communications were complements. Both results are plausible, although replication studies are essential.^ The consumer-oriented finding of net substitution is consistent with the nearly unanimous empirical results of numerous micro-scale studies (presented in the section below), whereas the industrial-oriented finding of complementarity is consistent with historically-observed simultaneous increases in both transportation and communication in the aggregate. The divergent findings are not only empirically substantiated, but are also conceptually reasonable (Plaut, 1997). As indicated in the Introduction, complementarity can arise both through an enhancement effect (in which use of one mode of communication directly stimulates use of other modes) and through an efficiency effect (in which use of one mode in conjunction with another improves the efficiency of the latter). It is quite possible that both effects are obtained more strongly in an industrial context than in a consumer one. For example, the expansion of personal contacts through electronic means is more likely to lead to increased travel (enhancement) in a business context than in a social one. The use of electronic data interchange and global positioning systems (efficiency) have benefited goods movement more than, say, automobile drivers. On the other hand, ITS approaches may begin to shift that balance as efficiency-improving technologies such as in-vehicle navigation systems permeate the consumer sector more deeply. Hence, it is possible that, over time, the net substitution effect now seen for consumer demand may weaken and even reverse into a complementary effect. This again calls quite urgently for studies replicating the Selvanathan and Selvanathan methodology at other times and places. The latest year in their time-series data was 1986; a shift may already be detectable in the intervening one-and-a-half decades. Advantages and disadvantages: The macro-scale comprehensive approach has the obvious advantage of offering a 'big picture' view. It illuminates the net sectoral impacts of telecommunications and transportation in a way that neither of the micro-scale approaches

152


described below can possibly do. It offers the potential for developing aggregate forecasts of the impacts of telecommunications on travel (and conversely) more readily than the other two approaches. On the other hand, the macro-scale approach does not completely dominate the others in conceptual superiority and usefulness. The macro-scale approach offers no insight into behavioural or other causal mechanisms driving the observed results. Its findings are based on the temporal or cross-sectional relationships exhibited by the data analyzed, but the focus on net impacts may conceal various counteracting relationships. If the underlying structure of those relationships changes over time or space - as, for example, may be the case for the impact of ITS and other telecommunications technologies on consumer demand for communication and travel - then its findings will not be robust. Further, the focus of this approach on monetary value may obscure some relationships. In some contexts (for example, understanding the impacts of telecommunications on urban traffic congestion), volume is more important than value. The number of person-trips made, and the number and length of messages flowing over an electronic link, are of legitimate interest in assessing the impacts of one mode of communication on another. The relationship between value and volume will probably differ by mode and over time: in brief, one might expect that, over time, volume of activity per unit of monetary value has been rising more rapidly for telecommunications modes than for transportation (Webber, 1991). If that is true, then a finding that, say, expenditures on communication and travel are positively correlated may or may not mean that volumes are rising together as well, and conversely.

THE APPLICATION-SPECIFIC APPROACH Description of approach: The application-specific approach analyzes one telecommunications application at a time. It is by far the approach most often taken in empirically assessing the impacts of telecommunications on travel. The evaluation is generally performed by collecting survey data from individual users about the transportation impacts of the application. The survey may be prospective (e.g., stated preference), contemporaneous (e.g., travel diaries), or retrospective. Advantages and disadvantages: The application-specific approach has the advantage of manageability. It offers the opportunity for a detailed look at the impacts of a single type of communication, for which the boundaries around the activities or process being studied can be relatively easily drawn. Studying individual behaviour brings the analyst closer to the decisionmaking unit than is the case for the macro-scale approach. On the other hand, this approach has clear disadvantages as well. By narrowing the focus to a single application context, it is easy to


153

lose sight of the big picture. The short-term nature of most studies in this category may give rise to findings that will change considerably over the long term. In particular, this approach seems likely to underestimate any stimulation effects of telecommunications, which tend to be longer-term and more indirect (occurring outside the boundaries of the process being studied), in favor of the shorter-term and more direct substitution effects. Results: As noted above, the key question in studying the degree of potential substitution between competing technologies or services is the extent to which they fulfill similar needs, at similar costs. Three activities and their telematics-based alternatives are discussed below: work, conferencing and shopping. We would argue, for each one of them, that the telematicsbased alternative is a substantively different activity or experience. Consequently, the degree of substitution depends not only on how well the tele-activity fulfills the 'basic' function (work, convey information, or shop), but also on the distribution of preference with regard to the other aspects of the activity. It is also necessary to recall that advanced telecommunications are not supplied in a vacuum, in which individuals (and more so, policy makers) are acting to reduce travel to work, conferences, and shopping. The other parties involved in these industries may have a different agenda, acting quite vigorously to encourage visitation to shopping malls and to convention centers, and airline travel. Thus, tele-activities may offer new opportunities, but their relative attractiveness also depends on changes in a very dynamic environment (Albertson, 1977). In the sections below, we discuss key results relating to the applications of telecommuting, teleconferencing, and teleshopping. We also review recent studies of the impact of the telephone (conventional and mobile) on travel.

Telecommuting Commuting constitutes the dominant single trip purpose, in terms of its share both of trips and of distance traveled (Hu and Young, 1992; US DOT, 1997). Further, it is the most routine type of trip, performed in very well-defined time slots and serving as an anchor to which other trips are chained. It is therefore the trip purpose contributing most heavily to peak-period congestion in urbanized areas. Further, it may be that a higher number of commute trips would be amenable to substitution by telecommunications than would be the case for trips for other purposes such as shopping (which are both less frequent and more diverse). For these reasons, the potential of telecommuting for reducing congestion holds particular appeal for policy

154


makers and planners, and hence telecommuting is doubtless the most commonly-researched telecommunications application in the context of understanding its travel impacts. Technically speaking, work entails the performance of particular tasks, usually at defined times, in return for some (financial) compensation. However, it is clear that the quality of this activity extends widely beyond the time, task and compensation. Under the purely technical view, the performance of many work situations can easily be carried out through telecommunications. But, for many people, work is a series of tasks requiring face-to-face communications, it is an opportunity to socially interact with others, it serves as an opportunity to exit the home environment for some time, it is an opportunity to see and be seen by others, and so on. In a nutshell, work often involves, in addition to the financial gains, social and psychological gratification that may not be explicitly stated or recognized, even by the individual. Although net substitution is obviously the most (socially) desired and perhaps the most expected effect of telecommuting, it does not mean that the same holds true for the individual. To the extent that other employment-related benefits and costs are important to the individual, the likelihood of substitution will be affected. Furthermore, travel stimulation is certainly possible as well, due to non-commute trip generation, changes in mode choice from ridesharing or transit to driving alone on regular commuting days, induced demand caused by the same telecommunications technology that supports telecommuting, latent demand^ realized if telecommuting perceptibly reduced congestion, and long-term changes in residential location that increase commute lengths (Salomon, 1985; Mokhtarian, 1991, 1998). Numerous individual studies of telecommuting have been conducted (Hamer et al., 1991, 1992; Henderson et al, 1996; Henderson and Mokhtarian, 1996; Kitamura et al, 1990; Koenig et al, 1996; Pendyala et al, 1991; RTA, 1995; Mokhtarian and Varma, 1998), and reviews of empirical results have periodically appeared (Mokhtarian, 1991, 1997, 1998; Mokhtarian et al, 1995; Nilles, 1988). The studies generally involve the collection of multi-day travel diary data before and some months after telecommuting began, often from a control group of nontelecommuters as well. These data are then analyzed to ascertain the impact of telecommuting on travel indicators such as number of trips and distance traveled. To date, there is some empirical evidence on non-commute trip and mode choice impacts, little evidence on residential relocation impacts, and virtually none on induced and latent demand impacts.


155

All the empirical studies are unanimous in finding that total distance traveled by telecommuters decreased markedly on telecommuting days. The change in non-commute trips and distance was sometimes positive and sometimes negative, but essentially statistically negligible. Gould et al (1997) found a similarly insignificant result for home-based workers in general. Interestingly, one study of telecommuting centers found a small increase in commute trips on telecommuting days, mostly due to trips home for lunch and back to the center in the afternoon, but again, the net reduction in distance traveled remained substantial. Little actual shift in mode choice has been found, although there was some evidence that trips eliminated by telecommuting tended to be disproportionately by transit or rideshare modes. That is, the more difficult-to-use modes were the ones more readily given up. No significant impact on residential relocation has been measured to date. Hence, the empirical results so far indicate a net impact of substitution. Given the unknown extent of the induced demand, latent demand, and long-term residential relocation effects, however (as well as the likely unrepresentativeness of the early adopters of telecommuting analyzed in these studies), it is certain that the long-term, systemwide effects of telecommuting will be less positive than is suggested by the results from the short-term, small-scale studies conducted to date. It is not certain how much less positive, but it is possible for the generation effects to nearly equal or even exceed the substitution effects. Two studies of the latent demand issue, one in the context of telecommuting (USDOE, 1994) and the other in a general context (Hansen and Huang, 1997) suggest that the realization of latent demand could amount to anywhere from 30 to 90% of the newly available capacity. Some initial evidence on the induced demand issue is offered below.

Teleconferencing Teleconferencing, increasingly in the form of videoconferencing, enables individuals or groups to conduct information transfers while spatially separated. The information typically is verbal (with visual cues in the case of videoconferencing), often supplemented by written or graphical material. Teleconferencing systems vary in sophistication and costs, as well as in accessibility (some require specialized studios, thus imposing greater need for pre-conference coordination, while others are readily accessible). Generally, the market for teleconferencing is in the institutional (private and public) sector and not in the domestic sector, although simple desktop videoconferencing systems being developed for personal computer users may eventually find a niche there too. Button and Maggi (1994) have looked into adoption patterns in Switzerland and the United Kingdom and found that large, often multi-national corporations are likely to be

156


early adopters, primarily for intra-organizational communications. This is consistent with product life cycle theory. The ability of teleconferencing to reduce travel is a very popular notion in the promotion of these services as well as in the popular literature (e.g., Arvai, 1994). In many cases, promotion material is very explicit about the trade-off and the expected cost savings due to the eliminated travel. However, some providers seem to agree that the substitution effect is only used for promotion to budget-conscious decision makers, whereas for the most part, the benefits of videoconferencing are simply the increase in inter-organizational communications, and not the trip savings (see, e.g., Egido, 1990; Mette, 1995). Further, as Salomon et a/. (1991) point out, even straightforward cost considerations do not always favor teleconferencing. They illustrate that, under then-prevailing price structures, travel costs could be lower than telecommunications costs for meetings involving short distances, long duration, and/or few participants. While the costs of teleconferencing have fallen since that study was conducted, there are likely still to be instances in which the trip can be justified in purely economic terms. Applications of the technology include four basic types of'electronic meetings': formal group meetings, informal group discussions, single presentations and repetitive presentations. Each seems to have different travel implications. For all types, the technology is increasingly userfriendly, costs are falling, and availability is increasing. Formal group meetings are electronic conferences, intra- or inter-organizational, in which people convene without the need to travel for face-to-face meetings (although some short travel may be necessary to a studio). Periodic conferences incur high costs for organizations that have to pay the travel and accommodation costs, in addition to the costs of time. Thus, using videoconferencing as a cost reduction strategy seems attractive. But organizations, and individuals within them, have recognized that travel to conferences entails external benefits (Button and Maggi, 1994). These include the value of face-to-face interaction in terms of the richness of information exchanged and opportunities for personal acquaintances (which may improve future mediated communications), as well as perks such as the break in routine and the enjoyment of visiting appealing places. (Especially the latter benefits are independent of any technological advances that increase the realism of the teleconference experience). All these aspects are missing from videoconferences, and in this respect the two types of meetings are far from similar. Consequently, they are also less amenable to substitution of teleconferencing for travel. Again, the importance of such attributes is not uniformly distributed among employees. For example, those who travel a great deal may do so in part because of the higher average value they place on the advantages of face-to-face communication, but at the same time, they may have a lower marginal utility for a single trip compared to those who travel less.


157

Informal meetings usually take place among team members within or across organizations^. These are more routine meetings involving people who usually know each other and share an organizational culture. These meetings may be substituting some travel between facilities of an organization, but in many cases they do not, as the users may all be in the same facility, with little or no physical separation. Arguably, this category has the most potential for substitution. But even here, new uses of the videoconferencing facility are likely to arise which do not replace regularly-scheduled in-person meetings, but rather which represent communication that would not have occurred otherwise. Single presentations of new products seem to be among the most popular applications of videoconferencing. This is to a great extent similar to broadcast systems where a new product or service is announced and potential clients gather in locations across the country or the world and can view the announcement and present (usually audio-only) questions to the originating agent. This is increasingly also used for religious activities. We hypothesize that this form of teleconferencing does not replace travel (in fact it probably generates local travel to the videoconference location) and that it is more of a marketing tool. In its absence, some alternative, possibly a travel-based mode of promotion, would be used but it is highly unlikely that significant differences in travel would take place. Repetitive presentations relate especially to education and training applications of teleconferencing. Here the technology is used to distribute information to students who may be spatially scattered in remote classrooms or even in their homes. Of course tele-education differs in many respects from the traditional classroom environment. It is widely applied in contexts where either the students or the teachers experience a mobility constraint (permanent due to distances, such as in Australia and Canada, or temporary, due to illness). Although the technology facilitates engagement in activities that formerly were inaccessible, it may not always result in substitution, that is, in being used in lieu of travel. A common theme for all four of these types of electronic meetings is that not every teleconference substitutes for a trip to the in-person version of the same meeting (Albertson, 1977). In many cases, the alternative to 'teleconferencing' is not 'traveling to the meeting', but rather 'not attending the meeting at all'. Hence, the primary impact in those cases is the generation of new communication, not the replacement of travel. Even when travel is substituted, as, say, in the case of routine informal meetings, the time thereby saved may be partially spent in making non-routine trips which are not readily replaced by telecommunications, such as those to the desirable conference locations, or those involving establishing a new business relationship.

158


Most of the literature dealing with teleconferencing (e.g. Egido, 1990; Johansen and Bullen, 1984; Green and Hansell, 1984) focuses on the quality of the communications process, in all of the above types. Much attention is paid to the applications for educational purposes. Given a lack of data due to the novelty of the technology and the fact that many organizations do not maintain sufficiently detailed data on teleconferencing and travel (Button and Maggi, 1994), very few studies have empirically addressed the impacts of teleconferencing on travel. Bennison (1988) analyzed the effectiveness of videoconferencing for a UK trial in the early 1980s. Among other findings, 87% of the 31 users responding to that part of the study reported a decrease in travel (and a smaller decrease in other modes of telecommunications) due to videoconferencing. However, there are a number of concerns with such a number, aside from the small sample on which it is based. Respondents are reporting a general perception, not a careful measurement of actual travel before and after the trial began. They may be more aware of the direct effect on trips eliminated (especially if those were the less pleasant, more routine trips), than of indirect effects involving 'in-filF travel. The promotion around the trial may have sensitized them to an expected outcome of travel savings, thereby biasing their responses. And finally, they may be reporting a short-term outcome before new stable patterns of travel are achieved. The author of the study, examining all the evidence, concludes that '...clearly substitution of the former [face-to-face meetings] by the latter [videoconferences] was at best partial. Indeed, the pattern that emerged was essentially that of videoconferences complementing conventional meetings rather than supplanting them' (Bennison, 1988, p. 293). Mokhtarian (1988) studied an experiment in which a regular monthly meeting of the Southern California Association of Governments was held by videoconferencing instead of conventionally. While participating individuals reduced their trip length to the meeting on a per capita basis (by 24% on average), thus indicating substitution, more participants took part in the meeting, increasing total distance traveled (by 29%) and thus indicating a complementary relationship (Albertson, 1977 makes a similar observation). Erdal and Hallingby (1992) studied the impact of the 1991 Persian Gulf war on travel and telecommunications to and from Norway. They found that while air travel declined noticeably due to the fear of terrorism (at least 40,000 fewer passengers in each of the three months January - March 1991, compared to the same period in 1990), the increase in telecommunications (teleconferencing as well as ordinary voice and data traffic) was negligible by comparison. This occurred despite considerable media attention being devoted to the use of telecommunications as a substitute. In fact, the primary impact on travel seemed to be the postponement or cancellation of planned trips rather than fulfillment of the same purpose through telecommunications.


159

Our understanding of teleconferencing allows us to speculate about its future impacts on travel. It is not likely that the demand for business travel will be reduced to a noticeable extent. The other contributing motivations for travel (the external benefits) will simply carry more weight. However, organizations (which develop travel-telecommunications policies) will be able to sustain some reductions in travel when conditions require them to do so, while still maintaining contact with their counterparts. Thus for teleconferencing (and other telecommunications used in the business sector), complementarity is probably the dominant effect, with the potential of some limited substitution.

Teleshopping Shopping refers to a service (or activity) that takes a variety of forms, but common to all is the ability to obtain information about products and services and to perform a transaction by which ownership is transferred and a product or service is spatially relocated to its new owner. We focus here only on active shopping, namely an activity for which the individual engages in a search for a product and may in fact generate a purchase. The 'shopping revolution' (Batty, 1997) offers a growing range of ways to do it. The conventional store environment alone has now diversified itself from street comer stores, to neighborhood supermarkets or specialty stores, to shopping centers and shopping malls, to factory outlets, warehouse stores or clubs and variants thereof. Likewise, non-store shopping comes in a variety of forms, most of which involve telematics to obtain information about and/or purchase consumer goods, and hence can be considered teleshopping. This category includes pre-World Wide Web services such as home shopping channels on cable television, specialized early systems such as the Minitel in France, and even telephone or fax orders from a catalog mailed to the home. It now includes the use of the Web for obtaining information, comparison shopping, placing orders, and even downloading digital products (typically news, music, software, and books) electronically. As commonly used, the terms teleshopping and electronic commerce represent intersecting but not coincident concepts. While teleshopping takes the consumer perspective, electronic commerce takes the vendor perspective. The latter term encompasses business-to-business and business-toconsumer transactions that take place through telecommunications networks, whether private systems or, increasingly, the Web. It includes virtual supply chain management activities (production and distribution), as well as demand chain activities. Electronic retailing is the subset of e-commerce that caters to an individual consumer (who participates by teleshopping). This segment is literally growing exponentially, with consumer spending via the Internet doubling annually {The Economist, 1999). The ubiquitous coverage and instantaneous information

160


capabilities of the Internet are generating entirely new forms of transactions, such as last-minute sales of surplus goods and services, and electronic auctions. Airlines are now using these techniques to sell otherwise-unused airplane capacity. Understanding the impact of teleshopping on conventional shopping activities requires an analysis of the nature of shopping activities, and through that, an examination of the extent to which teleshopping modes offer substitutable activities. As in other choice situations, the decision on whether to use one of the many travel-based shopping modes or to use a homebased mode, depends on the extent to which these activities are in fact similar enough to be substitutes, and on the characteristics of the decision-maker. Store shopping is presently (still) very different from teleshopping in terms of such attributes as the information provided, the sensual stimulation, the ability to compare prices and to attain immediate ownership. Beyond the functional attributes related to shopping and purchasing, store shopping also offers numerous other experiences, which to varying degrees are less amenable to electronic environments. These include, for example, the ability to interact with real salespeople and to bargain, the opportunity to be outside the home or work environment, and so on. Shopping, for many but not all, is a combined maintenance - leisure activity. Shopping modes differ in numerous attributes, thus entertaining different tastes, time and money budgets, and 'activity integration'. This term refers to the degree to which each shopping option allows or facilitates other activities to be intertwined with the shopping activity. For example, shopping at a mall offers high activity integration, as opposed to shopping at a street-comer grocery. One can (and should) assume that teleshopping services will be increasingly user-friendly, especially with respect to the quantity and quality of product information supplied to users. With the increasing similarity to the 'real' shopping experience, it is plausible to assume that more substitution will take place. But, the choice of shopping mode also depends on the individual's preferences and attitudes (Handy and Yantis, 1996). Individuals who prefer out-of home activities, especially if they are confined to home due to work (e.g. telecommuters) or household responsibilities, are likely not to forego the store shopping opportunities. Conversely, individuals who have a very busy schedule and desire some quiet time when off work, may prefer to use home-based shopping options, as reported by Gould et al. (1997). While shopping alternatives are changing in character and becoming more diverse, one could assume that human attitudes toward in-home and out-of-home activities do not change at the


161

same pace. We would speculate that basic attitudes toward the two types of activities change rather slowly. Thus, segmentation of the market on the basis of attitudes and preferences will provide some understanding as to the relative potential of substitution. Koppelman et al. (1991), using models of hypothetical choices, have shown that teleshopping is probably perceived as an electronic catalog, and at least in the context of shopping for appliances, does not seem to be substitutive to store shopping for those who would not purchase from a catalog. On the other hand, some shopping activities appear to lend themselves quite well to substitution, at least for some people. Balepur (1997) reports anecdotal evidence that computer-based information acquisition regarding an automobile purchase saved the respondent trips to multiple dealerships for the same purpose, and that another respondent reported saving trips to computer stores by finding and downloading a desired piece of software online. Such activities, comparatively rare at the time his data were collected (1994-1995), are now exceedingly commonplace. Handy and Yantis (1996) point to the complexity of the relationship, suggesting that some substitution may occur as systems develop, but again supporting the case that store shopping is a different adventure, and thus not easily substitutable. Gould et al. (1997), using structural equations to estimate time allocation between activities, point to the fact that busy women are more likely than others to take advantage of home-based shopping modes. Tacken (1990) found that users of a grocery teleshopping service in the Netherlands were predominantly (a) older people who chose it because of their limited mobility, and (b) dual-income households who chose it to save time. As the development and adoption of teleshopping modes are still in their infancy, the authors are not aware of any empirical studies explicitly examining the impact of teleshopping on travel patterns. In that regard, however, several different trends, operating in different directions, can be postulated: • To the extent teleshopping replaces store shopping, travel by the consumer will directly decrease. However, shopping trips are often chained to other trips. To the extent that those other trips still occur when the shopping trip is eliminated, the calculated travel distance reduction must be adjusted accordingly (Handy and Yantis, 1996). • Replacing store shopping by teleshopping shifts the travel required to deliver the purchased goods from the consumer to the provider, with an uncertain net impact. Provider-side delivery trips may be more efficiently organized than consumer-supplied deliveries - or they may not be, depending on both the extent to which the consumer trip was chained to other activities, and the tradeoffs made between efficiency and timeliness

162

•

•

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges of delivery for provider-side trips. The demand for fast delivery (which is related to the rise in standard of living) is clearly in conflict with efficiency. To the extent teleshopping for physical products supplements store shopping, travel (for manufacture and delivery) will increase. For example, many on-line purchases of products such as compact disks (while they are still demanded in physical form rather than downloaded directly) may not replace trips to a store, but may represent new purchases by consumers who otherwise would not have shopped for them in a conventional store (at least as often). Television home shopping networks, and their newer Internet counterparts, are prime examples of impulse teleshopping for which an actual trip to purchase the same items would not have been made.

Teleshopping may change the frequency of the shopping activity. If the convenience of shopping on-line prompts consumers to place orders more frequently than they would have by traveling to a store, delivery travel will increase. The extent to which it will increase depends in part on whether the total volume of goods purchased is greater, or whether the same volume is simply divided among multiple orders. In the latter case, consolidation of deliveries with other customers will reduce but not eliminate the increase in delivery travel. On the other hand, delivery charges will mitigate this effect, and may even lead to lower shopping frequencies for some people than if they made the trips themselves. • Teleshopping may alter the 'destination' of the shopping trip: with the Internet offering global reach even to small providers, manufacturing and delivery travel may increase as consumers and businesses readily obtain information about and order products and services from distant providers. • More widespread dissemination of information about physical stores using telematics, e.g. through the Internet or through sophisticated in-vehicle navigation devices, may prompt new trips to stores. • The time saved by not traveling to shop, and the increased time spent on the in-home activity of teleshopping, may be partially compensated for by additional out-of-home activities requiring travel, or by longer travel to more attractive destinations. Thus, it is clear that a complete accounting of the travel impacts of teleshopping must take place from a system-wide perspective, not just the individual's; must analyze the supply side as well as the demand side; and must consider second-order as well as first-order effects (Marker and Goulias, 2000; Gould et al., 1997; Gould, 1998). The bottom line, it is suggested, is that teleshopping is not likely to have a noticeable effect on travel reduction or enhancement, as processes cancel each other and there is no overriding effect that clearly dominates.


163

Conventional and Mobile Phones The telephone is clearly the most accessible form of telecommunications, with penetration levels exceeding 90% of the households and most business establishments in developed countries. As noted earlier, interest in its impact on travel dates back to its invention. The demand for telephones can be separated into the demand for access and the demand for use. In a series of studies on the telephone's impacts during the first half of the century, Fischer (1987) and Fischer and Carrol (1988) show that in the choice between a car and a telephone, many American households during the Depression preferred the former, which had a greater perceived promise for economic productivity. In recent years the plain old telephone has assumed numerous innovative uses. Aside from a series of smart services such as call waiting, call forwarding, and caller identification, telephone lines are now commonly performing facsimile and data transmission services in businesses and in homes. All these make the telephone an elaborate telecommunications medium, which potentially increases the options for substituting travel-based activities. Probably the most significant innovation in telephony in recent years which has direct ramifications for spatial behaviour is the introduction of relatively cheap and accessible mobile telephones. Nevertheless, relatively few studies have empirically examined the impacts of the telephone itself on mobility. One key study was reported by Claisse and Rowe (1993), who surveyed the residential telephone use of 663 people living in the Lyon, France metropolitan area in 1984, using a one-week diary of all calls made and received while at home. Among other questions, respondents were asked whether each call led to an unplanned trip (generated travel), and what they would have done if the telephone network had been down for an extended period (to which responses of 'made a trip' or 'sent someone' imply that the phone call replaced a trip). Depending on their focus (local calls only or all calls), Claisse and Rowe estimated that residential phone use generated trips 3-5% of the time, and replaced trips 21-27% of the time, for a net substitution impact of 17-22%. On the other hand, Massot (1997) compared those who used the telephone (most often a pay or stationary phone) during trips (4.6%) to those who did not, for a sample of 14,000 French respondents to a 1994 national survey of travel and communications behaviour. She found evidence of (efficiency-related) complementarity in that those who used a phone during trips were considerably more mobile (longer trips, greater travel time) than those who did not. She concluded that 'there is much more complementarity than substitution between modes' [telephone

164


and travel], and that the telephone plays an important role in the management of the lives of busy people. Finally, Yim (1994) studied the role of the cellular telephone in daily travel, through a 1991 mail survey of 7,347 cell phone subscribers in the San Francisco Bay Area. Regarding the present context, the general conclusion was drawn that 'the effect of the cellular calls on trip reduction was more significant than trip generation.' For example, 14.8% of the respondents reported driving less often after getting a cell phone than before, compared to 8.0% reporting driving more often. Similarly, 11.5% reported driving shorter distances afterwards, compared to 5.8% driving longer distances. On the other hand, in a later, smaller study of Bay Area cell phone users, Yim (2000) concluded that 'cellular communication generated additional trips rather than substituting for them.' However, these results should be interpreted with caution. As with the teleconferencing results mentioned previously, the numbers in the earlier Yim study represent the respondents' general impression, not a rigorous measurement. Further, the proper attribution of causality is not clear. The question was phrased in terms of, 'after getting a cell phone', but that is not necessarily 'because of the cell phone'. Conversely, when the respondents reported making an unscheduled trip 'because of a cell phone call, they may actually have just meant, 'using' a cell phone. If, without the cell phone, a pay phone would have been used, or the call would have taken place at another time, then it was not the cell phone/?er se but rather any phone that caused the trip. None of these results permit a mile-for-mile calculation of net impacts. In the Claisse and Rowe study, for example, it is not known whether the generated trips are longer or shorter on average than the replaced trips. Similarly, in Yim's studies, without knowing how much shorter or longer the distances were after obtaining the cell phone compared to before, or how long the generated trips were compared to the substituted trips, the net impact on vehicledistance traveled cannot be computed. Even the impact on number of trips cannot be measured in the earlier study, since the increase in trip-making for those who reported driving more often might exceed the decrease for those reporting driving less often. THE ACTIVITY-BASED APPROACH Description of approach: The activity-based approach is the newest of the analysis methods presented here. To date, empirical applications are only partial at best. But the approach seems to have been spontaneously and independently generated among several researchers, as a logical next stage in our analysis capabilities. Conceptually, the activity-based approach falls


165

between the other two approaches discussed. The activity-based approach is micro-scale, with measurement taken at the disaggregate level, but in theory it takes a comprehensive look at an individual's communications activity rather than focusing on a single application. As Claisse and Rowe (1993, p. 277) indicate, a key reason for the neglect of the stimulation effect (both of new travel and of new telecommunications that would not have occurred anyway) is that 'this work is often based on an in-depth analysis of the demand for transport; few studies start off from an analysis of the demand for and the use of telecommunications in order to evaluate their influence on transport.' The activity-based approach is intended to remedy this deficiency precisely by making the holistic study of communications the focus, rather than the demand for transportation in a specific context. The methodology is expected to involve a specialized activity or time use diary, through which measures of the amount of engagement in each mode of communication, over some period of time, can be obtained. Those measures should permit the analysis, perhaps through the use of techniques such as (time-dependent) structural equations modeling, of the impacts of each communication medium on itself and the others over time. Socioeconomic and other explanatory variables can be controlled for in such an analysis. Results: As indicated, this approach has not yet been applied in its entirety. Zumkeller (1996) describes a study in which 166 employees of the University of Karlsruhe, Germany completed diaries recording information on all trips and contacts (communication activities) they made for one day in 1994. He concludes (p. 79) that 'the complementary factor of the interrelationship between travel and communication is much stronger than the substitutional one' since high levels of trip-making were found to be associated with high levels of communication activity (an observation made more than a quarter-century ago by Day (1973, citing an unpublished research proposal by James KoUen)). Using time-use data, Harvey and Taylor (2000) analyze the contact and travel behaviour of nationally representative samples collected from Canada, Norway, and Sweden in 1990-92 (total N = 17,496). They conclude (p. 53) that '[t]here is a tendency for persons with low social interaction [specifically including those who work at home] to travel more. It is argued that individuals need, or want, social contact and if they cannot find it at the workplace they will seek it elsewhere thus generating travel... [This suggests] that working in isolation at home will not necessarily diminish travel but rather may simply change its purpose.' Hjorthol (2000) studied the relationship between travel and home use of information and communications technology for a sample of several thousand Norwegians in 1997-98. Although the measures of ICT activity used in this study are general and not based on a

166


comprehensive diary of particular communication episodes, the results are of interest. Low (0.07 - 0.12) but significant positive correlations were found between the use of a home computer for paid work and vehicle travel (N = 2,834), suggesting a complementary effect. As with the aggregate trends discussed in this chapter, however, the positive relationship is confounded by third-party variables such as income, occupation, and gender, which act in the same directions on both computer use and travel. When Hjorthol controlled for income, no significant differences in total daily car driver trips or distance traveled were found between groups having home computers and using them for work and those who did not have them or did not use them for work (N = 2,822). Thus, while there is not strong support for complementarity in a causal sense in this study, there is no support at all for a net impact of substitution. A Dutch study (KMPG, 1997) also looked at travel behaviour by three categories of people: heavy IT users (not specifically defined), a reference group of other people with sociodemographic characteristics similar to those of the heavy IT users, and the Dutch population as a whole. It was found that although the heavy IT users traveled more than the population as a whole, most of the difference was explained by sociodemographic distinctions since their overall trip frequency and distance traveled were similar to (although slightly higher than) that of the reference group. However, the heavy IT users traveled considerably more frequently (47% more trips) and farther (53% more kilometers covered) for business than did the reference group. Research conducted at the University of California, Davis also represents an early attempt to partially implement the activity-based approach. The study involves evaluating the communication and travel impacts of the Davis Community Network (DCN). DCN was launched slowly, beginning in January 1994; it is still in operation today. At the time collection of the evaluation data was completed in June 1995, the primary features of the system were electronic mail, newsgroup-reading, and web-browsing capabilities. In view of that, the evaluation constitutes primarily an assessment of the impact of Internet access on communication and transportation. That impact, especially on transportation, may not be expected to be sizable - particularly not as sizable as might be expected if more information about community activities had been posted and if more transaction opportunities had been available at the time the evaluation data were collected. Notwithstanding that, the methodology used in the study is of broader applicability, and even the empirical findings themselves are of some intrinsic interest. Multiple data collection instruments were developed for the evaluation. In the present context, two instruments are most relevant: an Activity Diary, collecting data on the antecedents and


167

likely consequences (for communication and travel) of a sample of DCN uses; and a Communications/ Travel (C/T) Log, in which respondents tallied each instance of communication in each of numerous categories (phone, e-mail, fax, document, in-person, and so on), over a four-consecutive-day period, once before and once several months after beginning to use DCN. Analysis of the Activity Diary data (148 respondents; 636 uses of DCN) suggested that the net impacts of DCN were to: greatly increase the number of electronic communications; leave the number of in-person communications essentially unchanged (generating some communications, but eliminating or substituting just as many); decrease the number of communications through physical objects (such as a book or diskette); and decrease the number of trips (Balepur, 1997). Hence, the overall impact of DCN as far as travel is concerned appears to be one of substitution. However, Balepur points to several limitations in the data. Again, it is primarily respondents' impressions (in this case, of hypothetical consequences of real behaviour) which are being obtained. When they reported the likely consequences of the DCN use in question (e.g., generating a trip), they could have been thinking of just the immediate consequences or of the chain of consequences extending into the indefinite future. The number of times a consequence was expected to occur was not reported, so 'generating a trip' may have meant one trip in one case and five in another. The same issue of the proper attribution of causality which was discussed previously applies here as well: it was very easy for respondents to confuse 'using' DCN with 'caused by' DCN, when the two are not necessarily the same. Further, this part of the analysis is essentially an example of the application-specific approach where the application is Internet access. DCN, viewed alone, may reduce travel and increase electronic communication, but the net impacts of all modes on each other cannot be ascertained by taking only DCN uses as the focal point. A different picture emerges in the analysis of the C/T Logs, which more closely embodies the spirit of the activity-based approach. The latter analysis identifies changes in communications patterns, not solely due to DCN, but any changes that occurred over the approximately six months (on average) between the before and after measurements. Before and after logs for 91 respondents provided data for the estimation of a six-equation system, in which the endogenous variables were transformations of the daily average numbers of communications (sent and received by the respondent) involving each of five modes (phone, fax, e-mail, physical object, and personal meeting), and the number of trips made by the respondent (Mokhtarian and Meenakshisundaram, 1999). Exogenous variables included

168


elapsed time between the before and after measurements (which varied across the sample), seasonal dummies, and socioeconomic characteristics. Results were that: the elapsed time variable was positive in all equations and generally significant, meaning that each form of communication is generally increasing over time, all else equal; the amount of communication by each mode in the after wave was positively and (except for the physical object equation) significantly related to the amount by the same mode in the before wave; and significant 'cross-mode' relationships (impacts of the amount of communication by one mode on amounts by other modes) were mostly positive, indicating the presence of complementary effects across modes. Taken together, these results suggest that self-generation and complementarity rather than substitution are the predominant impacts. The fact that an apparently different result is obtained from a broader look at (nearly) all communication activity than when the focus is on a particular type of communication is provocative, lending support to the supposition that the application-specific results discussed are necessarily incomplete. Advantages and disadvantages: The activity-based approach theoretically combines some of the strengths of the other two methods - comprehensive coverage with behavioural insight, at the level of the individual decision maker. On the other hand, it presents a number of measurement difficulties that make it an imperfect solution in practice. For example, typical activity or time-use diaries would need to be modified to focus on the desired modes of communication, and the level of detail needed to perform the desired analyses might be tedious for the respondent. Further, in such data collection instruments the unit of measurement is normally time spent on each activity (as well as simply number of activities). Other units of measurement are important, however. One such unit is the quantity of communication involved: 15 minutes spent reading the newspaper transfers considerably more information than 15 minutes spent writing an e-mail message. But transforming disparate modes of communication into a common denominator of quantity is problematic to say the least. The quality and value of the communication are also important dimensions, as the same example illustrates (the smaller quantity of information transferred through the e-mail message may have a higher value). These dimensions present even greater measurement challenges. Also, to properly analyze impacts on travel (trips, distance, mode, time of day, and so on) requires that, for each trip made, the diary collect information which is at a higher level of detail than is found in most activity or time-use diaries.


169

Finally, this approach is likely still to be most commonly applied on a short-term (months or a year) rather than on a long-term basis, although obtaining panel data over a longer period of time (several years) is certainly possible in theory.

MODELING THE CHOICE OF AN ACTIVITY MODE This chapter has focused on empirical studies directly addressing the relationship between telecommunications and travel. To date, most of those studies have evaluated the impact of telecommunications, given an activity pattern. Regardless of which of the three approaches discussed in this chapter are taken, all of the studies that permit an assessment of complementarity or substitution to be made represent essentially accounting exercises. The micro-scale approaches are oriented toward comparing the number of trips (or distance traveled) generated to those reduced to obtain a net impact, whereas the macro-scale approaches compare aggregate expenditures on travel to those on communications, whether cross-sectionally or over time. These calculations are only indirectly undergirded with conceptual models of choice among communication alternatives. On the other hand, the literature also contains a number of behavioural models of telecommunications-based choices, which have not been thoroughly reviewed here. In telecommuting, there are behavioural models of preference for the home-based form (Bernardino and BenAkiva, 1996; Bernardino et al., 1993; Mokhtarian and Salomon, 1997) and between the homeand center-based forms (Bagley and Mokhtarian, 1997; Stanek and Mokhtarian, 1998; Mokhtarian and Bagley, 2000), choice of home-based telecommuting (Mokhtarian and Salomon, 1996b), frequency of home-based (Mannering and Mokhtarian, 1995; Sullivan et aL, 1993; Yen and Mahmassani, 1997) and center-based (Ho, 1997) telecommuting, and duration of center-based telecommuting (Ho and Mokhtarian, 1999). There are models of choice between various forms of communication in a business context (Carlson and Davis, 1998; Fischer et al., 1992; Hauser, 1978; Moore and Jovanis, 1988; Webster and Trevino, 1995). And there are models of teleshopping behaviour (Koppelman et al., 1991; Manski and Salomon, 1987; Timmermans et al., undated). The demand for information generates communications activities. Analogously to trip generation, distribution and modal choice, the demand for information can also be satisfied by different quantities, at different destinations and by different modes. However, this comprehensive conceptualization and the development of (mathematical) models that explain the broader communications behaviour have so far received only scant attention.

170


Some suggestions have been presented, but, unfortunately, not pursued sufficiently yet. Moore and Jovanis (1988) have suggested an integrated framework that includes the generation and type of communications activities, which then leads to the choice of either travel or telecommunications. Ben-Akiva et al. (1996) have also suggested that IT options be integrated into the spatial choices structure, both at the level of offering alternative activities and in supplying information for short term decisions. Starting from this literature and the foregoing discussion, a prototypical model of activity 'mode' choice begins to emerge. Given the demand for a particular type of activity (such as work, or shopping, or conducting a business communication), the individual evaluates altemative ways or modes of performing that activity. Generically, we can refer to a location-based altemative L (requiring travel) and a telecommunications-based altemative C (potentially not requiring travel, or requiring less), but of course in specific applications there may be several variations on these categories. We have frequently pointed out that a given activity (such as shopping) may fulfill a number of purposes in addition to the primary or most apparent one (making a purchase). This means that the individual will choose between altemative activity modes based on a variety of relevant dimensions, and thus that the analyst should characterize each mode in terms of those dimensions and measure the individual's evaluation of each mode on each dimension. Generically, the utility of an individual for an activity mode could be viewed as a function of the following variables or dimensions (where each is both individual- and mode-specific unless otherwise indicated): • the quantity, quality, and timeliness of information obtained by the individual (quality is likely to be superior for L compared to C, whereas the winner on quantity and timeliness may depend on the situation); • the quantity, quality, and/or timeliness of the activity completed by the individual (being more productive working from home is an example in which C is superior here, but examples in the opposite direction can also be constructed); • the social/psychological content (possibly superior for L); • the physical exertion required (higher for L, but that may be a positive trait for some individuals some of the time); • the aesthetic content (may be higher for L, e.g. when travel to a conference is chosen because of its scenic venue; or higher for C, e.g. when one has a nicer office and/or view at home than at the regular workplace); • other positive qualities specific to the context; • travel cost/time/stress (potentially zero for C, but anyway presumably favoring it); • telecommunications costs (potentially zero for L, but anyway presumably favoring it); • other situation-specific costs and constraints;

Emerging Travel Patterns: Do Telecommunications Make a Difference? • • •

171

personal characteristics of the individual; mode-specific constant(s); and unmeasured variables (error).

Using this framework, it is easy to see how the individual's utility-maximizing choice between C and L for conducting the activity depends on the relative advantages of each alternative on the relevant dimensions (i.e. the values of the explanatory variables), and on how those dimensions are weighted or traded off by the individual (i.e. the coefficients of those variables as estimated from the data through maximum likelihood or some other means). The travel impacts of the collective choices made by a given sample (or population) can then be calculated. Assuming that distance (or travel time, cost, or some other measure of spatial separation) is a dimension relevant to the choice and hence is measured for each alternative (often zero for C), the expected travel impacts can be obtained by multiplying the travel outcome (distance, time, cost, etc.) for each alternative by the estimated probability of choosing that alternative, and summing across the sample (or population). As initially described here, the alternatives apply to a single activity and hence are application-specific. However, it may be possible to design more complex alternatives involving the choice of a pattern of activities across, say, a day, and hence to model the transportation impacts more comprehensively.

SUMMARY AIVD DIRECTIONS FOR FUTURE RESEARCH To the simple question: Do telecommunications make a difference? we can probably answer affirmatively with a high level of confidence. However, if the underlying assumption is that 'difference' implies substitution, the answer must be qualified. The differences telecommunications make are diverse, as shown in Table 2, which summarizes the results discussed here. As noted throughout this chapter, the relationship between telecommunications and travel can be of several basic types: substitution, stimulation, modification, and neutrality. Despite the widespread expectations of a substitution effect of telecommunications for travel, it may be just part of a more complex relationship. In fact, it is very likely that much of the impact is in the form of modifications in travel patterns (Salomon, 1985), such as trip timing, destination change, coupling with other users or a change of mode of travel. It may also be the case that some constraints are relaxed (or vice versa) and that travel-based activities are changed, as a result. Furthermore, as noted above, telecommunications may change land use patterns, and as a result, modify travel. So, there seems to be a variety of differences introduced by the availability of telecommunications-based activities.

172

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Table 2 Summary of Empirical Results to Date

Approach

Dominant Empirical Result

Macro-scale comprehensive Consumer

substitution

Industry

complementarity

Micro-scale limited Telecommuting

substitution

Teleconferencing

complementarity

Teleshopping

no results to date; approximately neutral impact expected

Telephone

mixed, ambiguous

Micro-scale comprehensive

complementarity

While various 'gross' impacts are observed depending on the specific focus of interest, these different impacts combine and counteract to result in an overall 'net' impact. We find the evidence for a net relationship of mutual complementarity between telecommunications and travel to be compelling. The evidence is both conceptual - with logical expectations of enhancement and efficiency effects, and empirical - with aggregate measures of the use of both modes rising simultaneously, and disaggregate studies of comprehensive communications finding own- and cross-mode generation effects. We see no reason for the historical relationship of net complementarity to weaken substantially in the future. Nevertheless, a great deal of work remains to better understand the complex relationships we are observing. As indicated in the previous section, we have on the one hand the 'accounting' studies, which take a given set of telecommunications-based activities and attempt to calculate the net impact on travel, and on the other hand the 'modeling' studies, which attempt to understand the individual decision process and inform a demand forecast. Improvements are possible in both cases. For the studies of net impact on travel, it is clear that research will benefit from the further accumulation of empirical evidence and the growing availability of


173

data. Data collection efforts must be carefully designed to produce the input necessary for rigorous research. That is true for micro-level as well as for macro-level analyses, which may complement each other. But, the more significant gains are to be expected from further development of conceptual and theoretical models of communication choice behaviour. Much of the modeling research done to date relies on stated preference (SP) approaches, which are very suitable to situations in which revealed preference is difficult to observe, or nonexistent, as may be the case with new technologies. However, despite considerable advancement in applying SP methods in recent years, their reliability is inherently attenuated when the behaviour considered is unfamiliar to the respondents. There is much evidence that telecommuting, for example, is far more often preferred than chosen (Mokhtarian and Salomon, 1996a). While doubtless some of this gap is due to genuine constraints on a genuine desire to telecommute, it also appears that in many cases the expressed preference is a weak one that may disappear entirely as the disadvantages of telecommuting are made more apparent. In general, much of what people believe about the impacts of information technology is derived from futuristic notions, which in turn, tend to overemphasize substitution effects (Salomon, 1998). Thus, future modeling research will also benefit from the growing availability of revealed preference data. Further, the prevailing modeling research approach, which has focused on specific applications, resembles the early and now almost obsolete interest in isolated trips, classified by trip purpose. Future research should broaden the view. Spatial behaviour, in our current understanding, assumes that individuals (or households or firms) generate a demand for activities. This demand is translated into travel which can, for practical purposes, be simplified into a set of behavioural choices, modeled as a sequential or simultaneous process of trip generation, trip distribution, modal choice and route choice. The demand for activities is conventionally assuming a spatial distribution of opportunities, which is a reflection of the land-use map. The emergence of aspatial opportunities for performing various activities calls for more attention to the nature of the activities, and their subsequent implications for behaviour. Thus, the currently-prevailing micro-scale application-specific models, which combine a conceptual model of choice behaviour with an empirical context, are necessary building blocks for further development of causal models which explain communications and spatial activity at the disaggregate level. If the past experience in activity modeling and in telecommunication travel modeling is relevant, then the complexity of the tasks ahead is enormous, but, nevertheless, necessary.

174


While some of the attributes of the demand for physical activities are quite well recognized (more so in the case of work, less so in the case of shopping and leisure), the attributes of the demand for information, and consequent communications activities, are presently less apparent. Their temporal, spatial and other attributes need to be better understood before comprehensive models can be developed. For example, the transportation-telecommunications issue has focused to a great extent on the impacts of the latter on distance or space, culminating in The Economist cover story on the 'Death of Distance' (Sept. 29, 1995). However, the research reviewed above may in fact warrant a shift in the focus from space to time. As activities are inherently performed in both time and space, focusing primarily on space may be too narrow a view of the potential impacts of telecommunications. We suggest that much of the impact on travel patterns is moderated through the time-saving (rather than space-saving) capability of telematics and the consequent reorganization of activities along the temporal dimension. One particularly troublesome problem in this general type of research is the role of values and norms. It is reasonable to assume that norms can change as younger generations adopt and assimilate information technology. It is not clear, however, how this will affect their behaviour in a few years' time. The very least that current students of the field need to do is to identify the often implicit value- and norm-laden assumptions in current models. This requires a critical assessment of the current models. Two interesting avenues of research, among others, emerge from the current state of the art. First, it would be interesting to study the adoption of activity modes in a cross-cultural context. One can hypothesize that in some cultures there will be greater acceptance than in others of electronic forms in lieu of face-to-face communications. Such differences may be attributed to the importance associated with personal contact and/or attitudes toward technology. (Similar differences may also be found across sectors in the economy). Second, there is a growing body of literature suggesting hypotheses about the positive utility of travel (e.g., Maggi et al., 1995; Salomon and Mokhtarian, 1998; Mokhtarian and Salomon, forthcoming). If people differ significantly in their attitudes towards travel then they are likely to differ in their propensities for substitution and complementarity. If this is the case then the understanding of the impacts of telematics on travel really depends on travel attitudes, which should be the focus of detailed hypothesis testing. Ultimately, however, it will not be sufficient simply to develop better models of activity choices. It is important to remember that, from the perspective of the transportation profes-


175

sional, the demand for the activity-based approach is a derived demand - derived from our need for more accurate transportation modeling. This means that, for transportation planners, it is the trip that counts: activities don't congest or pollute - trips do. The activity-focused approach is simply the means of getting better data about, and insight into, the trip. Thus, achieving an understanding of the activity decision process for its own sake is commendable, but beyond that (where the rubber meets the road, so to speak), there will still be a real transportation network for which it will still be important to forecast link volumes by mode and time of day. Any approach that does not allow that ultimate outcome, or that stops short of obtaining it, will be of limited use in a regional planning context. Hence, what is needed is a combination of the accounting and modeling approaches. In the short term, it may be possible to synthesize insights gained from both methods to develop aggregate, application-specific forecasts. This is the approach taken by Mokhtarian (1998), in which she used both behavioural modeling results and empirical evidence on net impacts to forecast the future systemwide effects of telecommuting on travel. In the long term, however, the desired goal is a comprehensive, integrated, completely behavioural model of communication and travel choice, leading to causally-based aggregate forecasts of the transportation outcomes. The generic approach outlined in the previous section may be a fruitful direction forward. We do not know at this point in time how much substitution and how much stimulation of travel is taking place. What we are uncovering is mostly the complexity of the interactions of telecommunications with the already complex phenomena of activity and travel behaviour. Will we know in the future? That depends very much on how successful we are at developing the comprehensive analysis approach suggested above. None of the approaches taken so far achieve the desired ideal, but they are nevertheless still of considerable value. It's just that we should not fail to keep the forecast in mind while working on the trees.

ACKNOWLEDGEMENTS An earlier draft of this chapter has benefited from comments of Piet Bovy, Elizabeth Raney, an anonymous reviewer, and participants in the Telecommunications - Travel Interactions workshop of the 8th Conference of the International Association for Travel Behaviour Research.

176


REFERENCES Albertson, L. A. (1977). Telecommunications as a travel substitute: Some psychological, organizational, and social aspects. Journal of Communication 27(2), 32-43. Arvai, E. S. (1994). Telecommunications and business travel: The revolution has begun. Transportation Research Circular 425, 28-31. TRB, Washington DC, May. Bagley, M. N. and P. L. Mokhtarian (1997). Analyzing the preference for non-exclusive forms of telecommuting: Modeling and policy implications. Transportation 24(3), 203-226. Balepur, P. (1997). Impacts of Computer-Mediated Communication on Travel and Communication Patterns: The Davis Community Network Study, Master's thesis. Department of Civil and Environmental Engineering, University of California, Davis, December. Batten, D. F. (1989). The future of transport and interface communication: Debating the scope for substitution growth. In David F. Batten and Roland Thord, eds. Transportation for the Future. Berlin: Springer-Verlag, 67-86. Batty, M. (1997). The retail revolution. Environment and Planning B: Planning and Design 24, 1-2. Ben-Akiva, M., J. Bowman and D. Gopinath (1996). Travel demand model system for the information era. Transportation 23, 241-266. Bennison, D. J. (1988). Transport/telecommunication interactions: Empirical evidence from a videoconferencing field trial in the United Kingdom. Transportation Research A 22A(4), 291-300. Bernardino, A. and M. Ben-Akiva (1996). Modeling the process of adoption of telecommuting: A comprehensive framework. Transportation Research Record 1552, 161 -170. Bernardino, A., M. Ben-Akiva, and I. Salomon (1993). Stated preference approach to modeling the adoption of telecommuting. Transportation Research Record 1413, 22-30. Brotchie, J., P. New1:on, P. Hall and P. Nijkamp (eds.) (1985). Technological Change and Urban Form. London: Croom Helm. Button, K. and R. Maggi (1994) Videoconferencing and its implications for transport: An Anglo-Swiss perspective. Transport Reviews 15(1), 59-75. Claisse, G. and F. Rowe (1993). Domestic telephone habits and daily mobility. Transportation Research A 27A(4), 277-290. Carlson, P. J. and G. B. Davis (1998). An investigation of media selection among directors and managers: From "self to "other" orientation. MIS Quarterly (September), 335-362. Day, L. H. (1973). An assessment of travel/communications substitutability. Futures 5(6), 559-572. r/2e£'co«owz.s'/(1999). Business and the Internet: The net imperative. June 26. 5-44. Egido, C. (1990). Teleconferencing as a technology to support cooperative work: Its possibilities and limitations. In Jolene Galegher, Robert Kraut, and Carmen Egido, eds., Intellectual Teamwork: Social and Technological Foundations of Cooperative Work. Hillsdale, NJ: Lawrence Erlbaum Associates, 351-371. Erdal, O. B. and H. S. Hallingby (1992). Business as usual during the Gulf War - the Norwegian case. Paper presented at the 71st Annual Meeting of the Transportation Research Board, Washington, DC, January. European Conference of Ministers of Transport (ECMT) (1983). Transport and Telecommunications. ECMT Round Table 59, Paris, France.


177

Fischer, C. (1987). Technology's retreat: The decline of rural telephony in the United States, 1920-1940. Social Science History 11 (3), 295-327. Fischer, C. and G. Carrol (1988). Telephone and automobile diffusion in the United States, 19021937. American Journal of Sociology 93(5), 1153-1178. Fischer, M. M., R. Maggi and C. Rammer (1992). Stated preference models of contact decision behavior in academia. The Journal of the Regional Science Association 71(4), 359371. Gottmann, J. (1983). Urban settlements and telecommunications. Ekistics 50, 411-416. Gould, J. (1998). Driven to shop? Role of transportation in future home shopping. Transportation Research Record 1617, 149-156. Gould, J., T. Golob, and P. Barwise (1997). Why do people drive to shop? Future travel and telecommunication trade-offs. Presentation at the Annual Meeting of the Transportation Research Board, January, 1998, Washington, DC. Graham, S. and S. Marvin (1996). Telecommunications and the City: Electronic Spaces, Urban Places. New York: Routledge. Green, D. and K. Hansell (1984). Videoconference. Business Horizons 27, 57-61. Griibler, A. (1989). The Rise and Fall of Infrastructures: Dynamics of Evolution and Technological Change in Transport. Heidelberg: Physica Verlag. Hamer, R., E. Kroes and H. van Ooststroom (1991). Teleworking in the Netherlands: An evaluation of changes in travel behaviour. Transportation 18, 365-382. Hamer, R., E. Kroes, H. van Ooststroom and M. Kockelkoren (1992). Teleworking in the Netherlands: Evaluation of changes in travel behaviour - Further results. Transportation Research Record 1357, 82-89. Handy, S. and T. Yantis (1997). The Impacts of Telecommunications Technologies on Nonwork Travel Behavior. Research Report SWUTC/97/721927-lF, Southwest Region University Transportation Center, Center for Transportation Research, The University of Texas at Austin, January. Hansen, M. and Y. Huang (1997). Road supply and traffic in California urban areas. Transportation Research A 31A(3), 205-218. Harkness, R. C. (1977). Technology Assessment of Telecommunications/Transportation Interactions. Report No. NSF/RA-770157, performed by the Stanford Research Institute, Menlo Park, CA, for the National Science Foundation, May. Harvey, A. S. and M. E. Taylor (2000). Activity settings and travel behaviour: A social contact perspective. Transportation 27(1), 53-73. Hauser, J. R. (1978). Forecasting and influencing the adoption of technological innovations: Applications to telecommunications innovations. Unpublished working paper. Transportation Center, Northwestern University, Evanston, IL, October. Helling, A. and P. L. Mokhtarian (2000). Changing worker telecommunication and mobility: Consequences for planning. Submitted to Journal of Planning Literature. Henderson, D. K., B. E. Koenig and P. L. Mokhtarian (1996). Using travel diary data to estimate the emissions impacts of transportation strategies: The Puget Sound Telecommuting Demonstration Project. Journal of the Air and Waste Management Association 46(1), 47-57. Henderson, D. K. and P. L. Mokhtarian (1996). Impacts of center-based telecommuting on travel and emissions: Analysis of the Puget Sound Demonstration Project. Transportation Research D 1 (1), 29-45. Hjorthol, R. J. (2000). The relation between daily travel and use of the home computer. Submitted to Transportation Research.

178


Ho, C-I. (1997). Modeling the Engagement in Center-Based Telecommuting. PhD dissertation, Department of Civil and Environmental Engineering, University of California, Davis, September. Ho, C-I. and P. L. Mokhtarian (1999). Accelerated failure time modeling of the duration of engagement in center-based telecommuting. Unpublished manuscript, available from the authors. Hojer, M. and L—G. Mattsson (2000). Determinism and backcasting in future studies. Futures 32,613-634. Hu, P. S. and J. Young (1992). Nationwide Personal Transportation Survey: Summary of Travel Trends. Report No. FHWA-PL-92-027. Federal Highway Administration Office of Highway Information Management, Washington, DC, March. Johansen, R. and C. Bullen (1984). Thinking ahead: What to expect of teleconferencing. Harvard Business Review (March-April), 166-174. Kitamura, R., J. Nilles, D. Fleming and P. Conroy (1990). Telecommuting as a transportation planning measure: Initial results of State of California pilot project. Transportation Research Record 1285, 98-104. KMPG Bureau for Economic Research and Documentation (1997). The Influence of the Information Society on Traffic and Transportation. Final report, commissioned by Ministry of Transport, Public Works, and Water Management of the Netherlands, Transport Research Centre (AW), Strategic Studies Division (VMV), Rotterdam, the Netherlands, October. ISBN 903693603. Koenig, B. E., D. K. Henderson and P. L. Mokhtarian (1996). The travel and emissions impacts of telecommuting for the State of California Telecommuting Pilot Project. Transportation Research C 4(1), 13-32. Koppelman, F., I. Salomon, and K. Proussaloglou (1991). Teleshopping or store shopping? A choice model for forecasting the use of new telecommunications-based services. Environment and Planning B: Planning and Design 18, 473-489. Linstone, H. and D. Sahal (1976). Technological Substitution: Forecasting Techniques and Applications. NY: Elsevier. Lund, J. R. and P. L. Mokhtarian (1994). Telecommuting and residential location: Theory and implications for commute travel in the monocentric metropolis. Transportation Research Record 1463, 10-14. Maggi, R., I. Salomon, and J. Orfeuil (1995). Why do people travel? Presented at the Euroconference on European Transport and Communications Networks, Espinho, Portugal. Mandeville, T. (1983). The spatial effects of information technology. Futures, 65-70. Mannering, J. S. and P. L. Mokhtarian (1995). Modeling the choice of telecommuting frequency in California: An exploratory analysis. Technological Forecasting and Social Change 49(1), 49-73. Manski, C. F. and I. Salomon (1987). The demand for teleshopping: An application of discrete choice models. Regional Science and Urban Economics 17, 109-121. Marchetti, C. (1994). Anthropological invariants in travel behavior. Technological Forecasting and Social Change 47, 75-88. Marker, J. T. and K. Goulias (2000). A review and framework for the analysis of household replenishment and its impacts on travel behavior and goods movement. Paper No. 000566 presented at the Annual Transportation Research Board Meeting, Washington, DC, January.


179

Massot, M-H. (1997). Telephone uses and daily mobility. Paper presented at the 8th meeting of the International Association for Travel Behaviour Research, Austin, Texas, September 21-25. Mette, M. (1995). Impact of Videoconferencing on the Demandfor Air Travel. Master's thesis, Department of Aeronautics and Astronautics, Flight Transportation Laboratory Report 95-1, Massachusetts Institute of Technology, February. Mokhtarian, P L. (1988). An empirical evaluation of the travel impacts of teleconferencing. Transportation Research A 22A(4), 283-289. Mokhtarian, P. L. (1990). A typology of relationships between telecommunications and transportation. Transportation Research A 24A(3), 231-242. Mokhtarian, P. L. (1991). Telecommuting and travel: State of the practice, state of the art. Transportation 18(4), 319-342. Mokhtarian, P. L. (1997). The transportation impacts of telecommuting: Recent empirical findings. Understanding Travel Behaviour in an Era of Change, eds. P. R. Stopher and M. Lee-Gosselin. Oxford, Great Britain: Pergamon (Elsevier Science Ltd.), 91-106. Mokhtarian, P. L. (1998). A synthetic approach to estimating the impacts of telecommuting on travel. Urban Studies 3 5(2). Mokhtarian, P. L. and M. N. Bagley (2000). Modeling employees' perceptions and proportional preference of work locations: The regular workplace and telecommuting alternatives. Transportation Research A 34(4), 223-242. Mokhtarian, P. L., S. L. Handy and I. Salomon (1995). Methodological issues in the estimation of travel, energy, and air quality impacts of telecommuting. Transportation Research A 29A(4), 283-302. Mokhtarian, P. L. and R. Meenakshisundaram (1999). Beyond tele-substitution: Disaggregate longitudinal structural equations modeling of communication impacts. Transportation Research C 7(1% 33-52. Mokhtarian, P. L. and I. Salomon (1996a). Modeling the choice of telecommuting 2: A case of the preferred impossible alternative. Environment and Planning A 28, 1859-1876. Mokhtarian, P. L. and I. Salomon (1996b). Modeling the choice of telecommuting 3: Identifying the choice set and estimating binary choice models for technology-based alternatives. Environment and Planning A2S,\ 877-1894. Mokhtarian, P. L. and I. Salomon (1997). Modeling the desire to telecommute: The importance of attitudinal factors in behavioral models. Transportation Research A 31(1), 35-50. Mokhtarian, P. L. and I. Salomon (forthcoming). How derived is the demand for travel? Some conceptual and measurement considerations. Transportation Research A. Mokhtarian, P. L. and K. V. Varma (1998). The tradeoff between trips and distance traveled in analyzing the emissions impacts of center-based telecommuting. Transportation Research D 3(61 4\9-42S. Moore, A. and P. P. Jovanis (1988). Modelling media choices in business organizations: Implications for analyzing telecommunications - transportation interactions. Transportation Research A 22A(4), 257-273. Netherlands Organization for Applied Scientific Research (1989). De invloed van telecommunicatie op verker: Gevlgen voor energie en millieu, TNO, Delft. Nijkamp, P. and I. Salomon (1989). Future spatial impacts of telecommunications. Transportation Planning and Technology 13(4), 275-287.

180


Niles, J. (1994). Beyond Telecommuting: A New Paradigm for the Effect of Telecommunications on Travel. Prepared for the US Department of Energy, Offices of Energy Research and Scientific Computing, Washington, DC 20585, Report No. DOE/ER0626, September. Available from NTIS, and on the Web at http://www.lbl.gov/ICSD/Niles. Nilles, J. M. (1988). Traffic reduction by telecommuting: A status review and selected bibliography. Transportation Research 22A, 301-317. Nilles, J. M. (1991). Telecommuting and urban sprawl: Mitigator or inciter? Transportation 18(4), 411-432. Organisation for Economic Co-operation and Development (OECD) (1992). Cities and New Technologies. Paris: OECD. Pendyala, R., K. G. Goulias and R. Kitamura (1991). Impact of telecommuting on spatial and temporal patterns of household travel. Transportation 18, 383-409. Plaut, P. O. (1997). Transportation - communication relationships in industry. Transportation Research A 31 A(6), 419-429. Pool, I. (1983). Forecasting the Telephone: A Retrospective Technology Assessment. Norwood, Mass.: Ablex. Roads and Traffic Authority (RTA) of NSW (1995). Report on Travel Impacts: RTA Teleworking Pilot Project. Network Development Branch, RTA, Surry Hills, New South Wales, Australia. Salomon, I. (1985). Telecommunications and travel: Substitution or modified mobility? Journal of Transport Economics and Policy 19(3) (September), 219-235. Salomon, I. (1986). Telecommunications and travel relationships: A review. Transportation Research A 20A(3l 223-238. Salomon, I. (1996). Telecommunications, cities, and technological opportunism. Annals of Regional Science 30(1), 75-90. Salomon, I. (1998). Technological change and social forecasting: The case of telecommuting as a travel substitute. Transportation Research C 6(1/2), 17-45. Salomon, I. and P. L. Mokhtarian (1998). What happens when mobility-inclined market segments face accessibility-enhancing policies? Transportation Research D 3(3), 129140. Salomon, I., H. N. Schneider and J. L. Schofer (1991). Is telecommunicating cheaper than travel? An examination of interaction costs in a business setting. Transportation 18, 291-318. Schafer, A. and D. Victor (1997). The past and future of global mobility. Scientific American (October), 58-61. Selvanathan, E. A. and S. Selvanathan (1994). The demand for transport and communication in the United Kingdom and Australia. Transportation Research B 28B(1), 1-9. Stanek, D. M. and P. L. Mokhtarian (1998). Developing models of preference for home-based and center-based telecommuting: Findings and forecasts. Technological Forecasting and Social Change 57(1/2), 53-74. Sullivan, M. A., H. S. Mahmassani and J-R. Yen (1993). A choice model of employee participation in telecommuting under a cost-neutral scenario. Transportation Research Record 1413,42-48. Tacken, M. (1990). Effects of teleshopping on the use of time and space. Transportation Research Record 1285, 89-91.


181

Timmermans, H., A. Borgers and M. Gunsing (undated). The potential adoption of teleshopping technologies in a spatial context: A decompositional choice experiment. Unpublished manuscript. United States Department of Energy (1994) Energy, Emissions, and Social Consequences of Telecommuting. Report DOE/PO-0026, Office of Policy, Planning, and Program Evaluation, Washington, DC, June. Available from NTIS, Springfield, VA 22161, (703) 487-4650. United States Department of Transportation (1993). Transportation Implications of Telecommuting. US DOT, Washington, April. United States Department of Transportation (1997). Our Nation's Travel: 1995 NPTS Early Results Report. US DOT, Federal Highway Administration, Office of Highway Information Management, Publication No. FHWA-PL-97-028, Washington DC, September. Webber, M. M. (1991). The joys of automobility. The Car and the City, eds. Martin Wachs and Margaret Crawford. Ann Arbor, Mich.: University of Michigan Press, 274-284. Webster, J. and L. K. Trevino (1995). Rational and social theories as complementary explanations of communication media choices: Two policy-capturing studies. Academy of Management Journal 38(6), 1544-1572. Yen, J-R and H. S. Mahmassani (1997). Telecommuting adoption: Conceptual framework and model estimation. Transportation Research Record 1606, 95-102. Yim, Y. (1994). The effects of mobile telephones on transportation and urban form. Paper presented at the 33rd Annual Meeting of the Western Regional Science Association, Tucson, Arizona, February 23-27. Yim, Y. (2000). Telecommunications and travel behavior: Would cellular communications generate more trips? Paper No. 00-0625 presented at the Annual Transportation Research Board Meeting, Washington, DC, January. Zumkeller, D. (1996). Communication as an element of the overall transport context - an empirical study. Proceedings, 4th International Conference on Survey Methods in Transport, pp. 66-83. 'The term telecommunications is used here interchangeably with information technology (IT) and telematics. 'Telecommunications' refers to a broad set of technologies and services provided by such technologies. In the present context we choose to narrow the term to refer mainly to point-to-point interactive telecommunications networks. This excludes broadcast systems, which are 'one-to-many' and commonly not interactive. The definition used here implies that the user must take action in order to engage in a communications session, much like the action taken to engage in travel. Telematics are the systems and services which use telecommunications and computers to provide information and communications services. ^ Note however that while the number of person trips is by definition unchanged in this case, the number of vehicle trips, or the person- or vehicle-distance traveled, or the number of peak period trips, and so on, might change. Hence, the net impact of a particular modification effect might in fact be generation or substitution, in terms of one of those alternate indicators of the level of travel (e.g., an increase or decrease, respectively, in vehicle distance traveled). ^ The term 'entity' is used to emphasize that both transportation and telecommunications are complex systems which entail technology, infrastructure and services. ^ See, for example, the figure first prepared by Gr^bler (1989), and reproduced in Batten (1989), Marchetti (1994), and Graham and Marvin (1996), which illustrates the simultaneous growth of passenger transport and communications in France since the early 1800s. Although HHjer and Mattsson (2000) rightly point out that the drama of that figure changes materially with the choice of base point and scale, the key trait that both transport and communications are generally rising is not altered. Batten (1989) and Niles (1994) present data from Sweden

182

In Perpetual

Motion: Travel Behaviour

Research

Opportunities

and

Challenges

and the United States, respectively, that also illustrate the simultaneous growth in both transport and communications. ^ Numerous researchers (e.g. Brotchie, et al., 1985; Gottmann, 1983; Graham and Marvin, 1996; Mandeville, 1983; Nijkamp and Salomon, 1989; OECD, 1992) have examined the impacts of telecommunications on land use. The focus here is on the subsequent effects of those land use changes on travel, which are seldom addressed with any specificity. ^ It would be of particular interest to apply the time-series approach to industrial data and the cross-sectional approach to consumer data (or better yet, both approaches to both kinds of data, collected within the same time frame), since the use of those two different approaches may be a confounding factor in the difference between the two outcomes. ^ Although the terms 'induced demand' and 'latent demand' are often used interchangeably, we distinguish them. By induced demand we mean (in this context) travel generated directly by telecommunications, such as finding out about an activity through a community network and then traveling to that activity. By latent demand we mean the phenomenon that increasing the transportation system capacity, or reducing the costs (whether through providing new infrastructure or, in this context, through telecommunications reducing demand), attracts new vehicle trips, whether through changes of mode or route, new development along the corridor, and so on. Neither necessarily implies the other: although the capacity may be freed up through tele-substitution, the latent demand could be realized by anyone for any reason. And induced demand can be generated by telecommunications even if no travel were substituted (and hence no capacity were freed to attract latent demand). * The distinction between inter- and intra-organizational communications is important but sometimes blurry. Some very big organizations, like NASA or IBM, which operate in multiple locations, can have systems that are officially intra-organizational but in effect may be used inter-organizationally.


8

TRANSPORTATION AND TELECOMMUNICATION: FIRST COMPREHENSIVE SURVEYS AND SIMULATION APPROACHES

Dirk Zumkeller

ABSTRACT The chapter examines the interrelationship between physical and virtual traffic (telecommunication) and deals with the difficulty of developing a concept to measure this interrelationship. The scope of spatial behaviour patterns is expanded by including the virtual dimension of overcoming space. The analysis is supported by empirical data from household surveys of both physical travel and telecommunication activities in Germany and Seoul. The next section checks if the sociological formation of person groups is confirmed by intra-homogeneity and inter-heterogeneity in their traffic and telecommunication behaviour. This is verified by the observation that person groups with a large physical activity radius also show the greatest virtual mobility. Thus, complementary effects dominate substitutional impacts. Following this result, the concept of a model reflecting the interdependent travel and telecommunication behaviour is presented. Finally, some aspects of modeling are examined with the help of empirical data from Seoul. This modeling exercise demonstrates which fields necessitate further analysis and research.

BACKGROUND The development of today's society towards an increased usage of telecommunication services will affect people's traveling and telecommunication habits. There are theories about what these effects may look like, but they have not been tested nor quantified (Garrison and Deakin, 1988).

184


On the other hand, innovations in information technology enable massive real-time information transactions between individuals, economic institutions and their public counterparts, affecting life-styles and behavioural patterns of all members of the community (Cerwenka, 1992; Heinze, 1985; Heldman, 1995; Salomon, 1986). The occurrence of new services such as teleworking, -shopping, -conferencing etc. is observed without quantitative indications of their impacts. They may lower the impedance of space and thus contribute to a decentralization of private residences and public organizations. Undoubtedly, they will lead to changes in leisure activities and inevitably result in changes in traffic phenomena - but in which direction and to which extent? Therefore, transportation planners need to identify the interactions between telecommunication and transport. How do individuals choose between various types of contact media? When do they choose to travel and in which types of situations do they prefer to make a telephone call? Which kinds of topics are dealt with in e-mails and when do people rely on ordinary letters? What are the differences and similarities in travel behaviour between various groups of people? Is there potential for substitution and/or complementarity to take into account in future planning of infrastructure and investment? Accordingly, transport planners have to analyze the interrelation between traveling and other types of communication in order to develop a model capable of simulating travel and communication behaviour as a response to new information services.

THE EMPIRICAL BASIS Aggregate traffic flows are always the result of the superposition of individual behavioural patterns. Individual behaviour and observation of the individual are therefore given priority. Consequently, the individual has to form the elementary sampling unit in survey work. Due to these considerations, it is necessary to obtain as much information as possible about the intrapersonal context of travel and communication. For that reason, an empirical investigation of the daily context of telecommunication and trips was carried out on a microscopic individual level, aiming at deeper insight into the intrapersonal frequency of using different modes and media (including cars, public transport, walking, cycling, telephone, fax, letters, mobile phones, e-mail etc.) and their interrelationships.

Transport and Telecommunication: First Comprehensive Surveys

185

Development of the Questionnaire One major problem in developing the questionnaire was the formulation of a common sense context to measure the travel and telecommunication events. The two following questions had to be settled first: • what should be measured? (see Figure 1) • how should it be measured? Conventional travel surveys aim for private households to report both the private and the business activities of all household members. This type of survey had to be extended: • in the household part, by including the access to telecommunication facilities (phone, cordless phone, mobile phone, answering machine, fax (private or office), broad band TV, satellite dish, TV-set, videotext, btx-system, internet, etc.) and • in the personal part, by the reporting contacts performed. The resulting problem was: which information should be recorded for each trip and contact? For trips, the usual characteristic features like mode, trip length, etc. (see Figure 1, left side) are already defined. Some of these features are quite similar to those of a contact. However, the comprehensive description of a contact requires additional features, as shown in Figure 1. The next and most important question refers to the way to obtain this information. This step was represented by a so-called notebook (see Figure 2). Each person had to enter all trips and contacts performed on a given day. Care was exercised so that the procedure would not cause the respondent unnecessary trouble. Since the sequence of the trips and contacts was not known in advance, the book was structured in such a way that any possible order of trips and contacts (e.g. trip-trip-contact-trip) could be filled in without the need to turn pages other than to proceed in the given order. For that reason each page of the notebook was divided into two parts: the first part to fill in a trip, and the second part to fill in a contact (see Figure 2).

Samples and Data-Sets The Institute for Transport Studies carried out a pilot survey in summer 1994. 261 households of employees of the university were selected at random. After a first contact by phone, 168 households agreed to participate in the survey, to which 94 households or 166 persons actually responded. Within the framework of a pilot sample such a low response rate and a possibly

186


non-representative segment of the population are acceptable.

What has to be measured ? Travel

Telecommunication

transport behaviour trips

communication behaviour phone, fax, letters

for each trip

for each contact

departure time

^

time of contact

mode:

walk bike car public transport

medium:

trip purpose:

work education shopping leisure etc.

contact purpose: organizing activities exchanging information social conversation other purposes

trip length

phone letter fax

distance between the two locations of the contact additionally:

performance:

active? passive?

performance:

active? passive?

Social segment: private business semi-private organization Figure 1 Information to be Gathered

187

Transport and Telecommunication: First Comprehensive Surveys

— •

trip

contact

1

^

i phone

D

or

letter

D

or

fax

D

• - When did the contact happen ?

time

- Did you n call, write or fax or D receive a call/letter/fax - What was the purpose of the contact ? n personal (e.g. leisure etc.) D business D other - What was the aim of your contact ? (more than one possibility) n organize an activity (visit, meeting, etc.) n exchange information n social conversation D other as follow s Distance to the location of the contact

When did you start your trip?

km

time

What was the trip purpose ? n to work / education D leisure

D business

D shopping

D other as follows __

D return home _____

Which means of transport did you use? n walking

D bicycle

to a location D car

n public transport

different to your current location 1 next page!

• oth er as follows Distance of y

V

p2-g

s

1 V3

]«

a •2

rs

la

5« .2

F 21 Or;

w 1 '^ 1 ^

o

o

>-> [ o CT 0)

c= 1 o '55

U, 1/3 t - - tS 1> «^

o

13

§ ^

ON

3 .2 c tS

Qe E

.22 1 ea

ON

O

ON

ON

«

rt

PQ

J

1

S ^ 3 J Sj .S § "2 o -S c .2

O 3 o c/2 2i a. o c .o 3 -o

OS

1

c o 3 -o

1 3 -o T3 O c«

'o, W

O

'5,

13 O

13 T3 O

e

•o -^

e ^ •So

ON

2 E

c .2

3

1

1

-o a> Si T3

ON

5

s .^

c

00

i

316 In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges

-4-*

u o

a 2 H o

2

s:

a,

o -^ C3

Li -^ S Mr 1=:

x:

tu J= is

o

ho oSig

u

pi 1

i 1 o

1 p

B

s \%

s

-T3

||g|

-a

o

•So S s S DO

zi p

0\

on

"^ S

of g 1 1 E? 1 ti o .S U PQ H

o

on

a

O

.2

1 >

1 a, >

.

"on O

o — c« S C3

•2 ^

^

o

1 d^ •r-

^

j=

f

r.

«

a

^

o

II

1

Time Use and Travel Demand Modeling

•c

o

&dO a

«-> o

O a> D

£ £ . ^ "^ 13

^2 ro |0

ON ON

=3

3

cr

.2

a

13 -o

Q 13

.£ .2

Q

•— ex •—

^ :§ 1

"^ i s

^ .2 i^

.2

V5

II i 't^ S
"i -^ ^ C 3

c ^ — "O

1l o

11 s (/5

5 J 1 f*^ ON ON

00

5 s

N

ON E S ON

NO

•1 ^

8 ^

ll

o

E

13 -o o

3

3

2

ON

=:

E ^ fE 3 o u Q Xi

SI

s i

§ .£ c 8

on X 3 «J

ll

^ o

Q 13

.3 .2

;>^ CO

II ^ § 1

8 ^ -S X.

o

fe O '-' o

S § o 2^

"S1 f §

«-g 1 s

•S "> o .«

2^ ^ "^

-§ fe §

•i|2

.S *^ o « -a ^

2 ^ ^

318 In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges

^

3 5 ^^

S Ô '••3

e«

s

s T3

-2 Z

p S

.2

H

•S

y

=^

r-.S

> E2 i 1 •21 -8 i a ^

c O

"^ ^

C

o 5 = t> ^ -s

i

13

'^

0^

ON

'S

'-3

to

c .2

•5^ c3 ^ ^5 ^ ^ ."" rj c« ^ Z3 D . 5w

1 1 i >• o E

^

.2 a

"51 .2 c « 2

.E s e i

03

£

"N

%1

s s ON

s p0

2 0 '>

.3 -d "§ 1

1/3

-a c

•5 ^ T3 :S :S 3 '

on

0

•^ .5 3

on

-o

3

•2

£

13 -o

Q1

"3 T3

13 TD O

S S cd

cd 3 T3

.£ £ c c id ^

00.-

ON

-6

J^

3

£ c

-o

Q 13

13 -o o £ c TD

3 O 3

3

5o

13 •o £ c .2 "3

3 O 3 k«

ca

5o

ON ON

a*

ON ON

W5

3

CO

-o c

CU

00

3 CT

C

1

1

1


^

on ^

^ -o >

£

re 1 . 1 -M o

•5 g ^ oj «^ . S g O •— ' 3

'^ .2 .2 a. S "S "S '5 -o -2. o

ON

o

£

e

'C tn

.2

"a.

00

T3

(-< =*

1 o

1 \0 o

^ '"5

'••3

ig S'c/e?7ce, 10, 1, 1-23. Jain, D. C, N. J. Vilcassim and P. K. Chintagunta (1994). A random-coefficients logit brandchoice model applied to panel data. Journal of Business and Economic Statistics, 12, 3, 317-328. Johnson, N. and S. Kotz (1970). Distributions in Statistics: Continuous Univariate Distributions, John Wiley, New York, Chapter 21. Kamakura, W. A., B-D. Kim, and J. Lee (1996). Modeling preference and structural hQtQrogenQity in consumQY choicQ, Marketing Science, 15, 152-172. Kiefer, N. M. (1988). Economic duration data and hazard functions. Journal of Economic Literature, 27, June, 646-679. Killingsworth, M. R. (1983). Labor Supply, Cambridge University Press, Cambridge. Kitamura, R. and S. Fujii (1996). Two computational process models of activity-travel behavior, paper presented at the Theoretical Foundations of Travel Choice Modeling Conference held in Stockholm, August 7-11.

Recent Methodological Advances Relevant to Activity and Travel Behaviour

411

Koppelman, F. S., C. R. Bhat and J. L. Schofer (1993). Market research evaluations of actions to reduce suburban traffic congestion: commuter travel behavior and response to demand reduction actions, Transportation Research, 27A, 5, 383-393. Koppelman, F. S. and C-H Wen (1996). The paired combinatorial logit model: properties, estimation and application. Technical Paper, Department of Civil Engineering, Northwestern University, Evanston, Illinois. Kurani, K. S. and R. Kitamura (1996). Recent developments and the prospects for modeling household activity schedules, report prepared for the Los Alamos National Laboratory, Institute of Transportation Studies, University of California, Davis, CA. Kwan, M-P. (1994). A GIS-based computational process model of activity scheduling for intelligent vehicle highway systems (IVHS), Unpublished Ph.D. Dissertation, Department of Geography, University of California at Santa Barbara Lam, S-H (1991). Multinomial probit model estimation: computational procedures and applications. Unpublished Ph.D. dissertation. Department of Civil Engineering, The University of Texas at Austin. Lam, S-H. and H. S. Mahmassani (1991). Multinomial probit model estimation: computational procedures and applications, in Methods for Understanding Travel Behavior in the 1990's, Proceedings of the International Association of Travel Behavior, 229-242. Lancaster, T. (1985). Generalized residuals and heterogenous duration models with applications to the weibull model. Journal of Econometrics, 28, 1, 155-169. Lee, L. F. (1983). Generalized econometric models with selectivity, Econometrica, 51, 2, 507512. Lee, M. (1996). Analysis of accessibility and travel behavior using GIS, Unpublished Master of Science Thesis, Department of Civil and Environmental Engineering, The Pennsylvania State University, May. Lillard, L., J. P. Smith and F. Welch (1986). What do we really know about wages? the importance of nonreporting and census information. Journal of Political Economy, 94, 31,489-506. Maddala, G. S. (1983). Limited-Dependent and Qualitative Variables in Econometrics, Cambridge University Press, Cambridge. Mahmassani, H. S. (1997). Dynamics in commuter behavior: recent research and continuing challenges, in P. Stopher and M. Lee-Gosselin (editors). Understanding Travel Behavior in an Era of Change, 315-338, Elsevier Science Ltd., Oxford. Mahmassani, H. S. and R-C. Jou (1996) Bounded rationality in commuter decision dynamics: Incorporating trip chaining in departure time and route switching decisions, presented at the Conference on Theoretical Foundations of Travel Choice Modeling, Stockholm. Mannering, F. L. and D. A. Hensher (1987). Discrete/continuous econometric models and their application to transport analysis. Transport Reviews, 7, 3, 227-244. Mannering, F., E. Murakami and S. G. Kim (1992). Models of traveler's activity choice and home-stay duration: analysis of functional form and temporal stability, submitted to Transportation. Mannering, F. L. and C. Winston (1985). A dynamic empirical analysis of household vehicle ownership and utilization. Rand Journal of Economics, 16, 215-236. Manston, K. G., E. Stallard, and J. W. Vaupel (1986). Alternative models for the heterogeneity of mortality risks among the aged. Journal of the American Statistical Association, 81, 395, 635-644. McCall, B. P. (1994). Testing the proportional hazards assumption in the presence of unmeasured heterogeneity, Journal ofApplied Econometrics, 9, 321-334.

412

In Perpetual Motion: Travel Behaviour Research and Opportunities

McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior, in Zaremmbka, P. (ed.) Frontiers in Econometrics, Academic press, New York. McFadden, D. and K. Train (1996). Mixed MNL models for discrete response, working paper, Department of economics. University of California, Berkeley. McMillen, D. P. (1995). Spatial effects in probit models: a monte carlo investigation, in L. Anselin and R.J.G.M. Florax (editors) New Directions in Spatial Econometrics, Springer-Verlag, New York. Mehndiratta, S. (1996). Time-of-day effects in intercity business travel, Ph.D. thesis. Department of Civil Engineering, University of California, Berkeley. Meyer, B. D. (1987). Semiparametric estimation of duration models, Ph.D. Thesis, MIT, Cambridge, Massachusetts. Meyer, B. D. (1990). Unemployment insurance and unemployment spells, Econometrica, 58, 4, 757-782. Miller, E. J. (1996). Microsimulation and activity-based forecasting, paper prepared for publication in the proceedings of the Conference on Activity-Based Travel Forecasting held in New Orleans, June 2-5. Miller, E. J. and P. Salvini (2001). Activity-based travel behavior modeling in a microsimulation framework. Chapter 26 in this volume. Niemeier, D. A. and J. G. Morita (1996). Duration of trip-making activities by men and women. Transportation, 23, 353-371. Pas, E. I. (1996). Recent advances in activity-based travel demand modeling, paper prepared for publication in the proceedings of the Conference on Activity-Based Travel Forecasting held in New Orleans, June 2-5. Prentice, R. and L. Gloeckler (1978). Regression analysis of grouped survival data with application to breast cancer data. Biometrics, 34, 57-67. Recker, W. W. (1995). Discrete choice with an oddball alternative. Transportation Research, 29B, 201-212. Revelt, D. and K. Train (1996). Incentives for appliance efficiency: Random-parameters logit models of households' choices, technical report, Department of Economics, University of California, Berkeley. Sakomoto, A. and M. D. Chen (1991). Inequality and attainment in the dual labor market, American Sociological Review, 56, 295-308. Small, K. A. (1987). A discrete choice model for ordered alternatives, Econometrica, 55(2), 409-424. Steckel, J. H. and W. R. Vanhonacker (1988). A heterogenous conditional logit model of choice, Journal of Business and Economic Statistics, 6, 391-398. Stopher, P. R. (1993). Deficiencies of travel-forecasting methods relative to mobile emissions. Journal of Transportation Engineering, 119, 5, 723-741. Stopher, P. R. (1996). Household travel surveys: new perspectives and old problems, paper presented at the Theoretical Foundations of Travel Choice Modeling Conference held in Stockholm, August 7-11. Sueyoshi, G. T. (1992). Semiparametric proportional hazards estimation of competing risks models with time varying covariates. Journal of Econometrics, 51, 25-58. Swait, J. and J. Sweeney (1996). Perceived value and its impact on choice behavior in a retail setting, working paper. Department of Marketing, College of Business Administration, University of Florida.

Recent Methodological Advances Relevant to Activity and Travel Behaviour

413

Swait, J. and E. C. Stacey (1996). Consumer brand assessment and assessment confidence in models of longitudinal choice behavior, presented at the 1996 INFORMS Marketing Science Conference, Gainesville, FL, March 7-10. Swait, J. and W. Adamowicz (1996). The effect of choice environment and task demands on consumer behavior: discriminating between contribution and confusion, working paper. Department of Rural Economy, University of Alberta. Swait, J. (1994). A structural equation model of latent segmentation and product choice for cross-sectional revealed preference choice data. Journal of Retailing and Consumer Services, 1,77-89. Train, K. (1986). Qualitative Choice Analysis: Theory, Econometrics, and an Application to Automobile Demand, The MIT Press, Cambridge, Massachusetts. Train, K. (1995). Simulation methods for probit and related models based on convenient error partitioning, working paper, Department of economics. University of California, Berkeley. Train, K. (1996). Unobserved taste variation in recreation demand models, working paper, Department of Economics, University of California, Berkeley. Vilcassim, N. J. and D. C. Jain (1991). Modeling purchase-timing and brand-switching behavior incorporating explanatory variables and unobserved heterogeneity. Journal of Marketing Research, 28, 29-41. Vovsha, P. (1997). The cross-nested logit model: application to mode choice in the Tel-Aviv metropolitan area, presented at the 1997 Annual Transportation Research Board Meeting, Washington, D.C. Waddell, P. (1993). Exogenous workplace choice in residential location models; is the assumption valid?, Geographical Analysis, 25, 65-82. Wedel, M., W. A. Kamakura, W.S. Desarbo, and F.T. Hofstede (1995). Implications for asymmetry, nonproportionality, and heterogeneity in brand switching from piece-wise exponential mixture hazard models. Journal ofMarketing Research, 32, 457-462. Williams, H. C. W. L (1977). On the formation of travel demand models and economic evaluation measures of user benefit, Environment and Planning, 9A, 285-344. Winship, C. and R. D. Mare (1992). Models for sample selection bias. Annual Reviews of Sociology, 18,327-350. ' Travel demand literature, in general, attributes the IIA property to the IID assumption of the error covariance structure and does not discuss the assumptions of response homogeneity and error variance-covariance homogeneity. ^ A related model is the cross-correlated logit (CCL) model of Williams (1977). The CCL model allows correlation among alternatives across both dimensions in a two-dimensional choice model by specifying the error covariance matrix to include variance terms specific to each dimension (the error terms specific to each dimension and to the combination of dimensions are assumed to be gumbel distributed). Vovsha's CNL model, on the other hand, enables a flexible correlation structure by allowing the same alternative to appear in multiple nests. The CCL model is not consistent with random utility maximization, while the CNL model is. ^ The reader will note that the nested logit model cannot accommodate such a correlation structure because it requires alternatives to be grouped into mutually exclusive nests. ^ It is useful only in instances where there is a clear bound to the perceived attractiveness of an alternative, such as "in route choice models where it may not be unreasonable to assume that the perceived attractiveness of a route cannot be positive, since perceived travel time cannot be reasonably expected to be negative" (Daganzo, 1979; pl6).

414

In Perpetual Motion: Travel Behaviour Research and Opportunities

^ The reader is referred to Hensher (1996a; 1996b) and Hensher et al. (1996) for applications of the HEV model to estimation from revealed and stated preference data. The HEV model has also been applied in a marketing context by Allenby and Ginter (1995). ^ Appropriate identification conditions will have to be imposed in this structure. In the most general case, each group can represent a pair of alternatives. If there are / alternatives, the number of pairs of alternatives is /(/-l)/2. However, we cannot accommodate a covariance term for each pair; one of the pairs should be normalized to have a covariance of zero, so the covariance of the other pairs is relative to that of the base pair. Unfortunately, the covariances are generated by variance terms and so are pre-specified to be positive. Thus, the normalization of which pair to select as the base is not innocuous; the base pair should be the one with least covariance, which of course we do not know a priori (see also Ben-Akiva and Bolduc, 1996 for a related discussion). Thus, in general, we have to impose a restrictive structure for the covariance patterns based on a priori theoretical considerations. ^ In concept, the endogenous segmentation approach is equivalent to a random-coefficients approach with nonparametric discrete probability distributions for the heterogeneity specification (see Jain et al, 1994 and Chintagunta and Honore, 1996). ^ Sueyoshi (1992) has also extended the Han and Hausman framework to the multivariate case. However, like all earlier competing risk models, he characterizes the exit state implicitly based on the duration spells. Further, the Sueyoshi approach becomes cumbersome when dealing with muhivariate competing risks since it requires computation of multivariate integrals. In contrast, Bhat's approach requires only the computation of bivariate integrals independent of the number of competing risks. 1995 also formulate a competing risk model to model activity duration with the termination states being any one of several activity types such as in-home leisure, work/education, shopping, etc. They use an accelerated lifetime model to include the effect of covariates so that the covariates rescale time directly. Unfortunately, with such a specification, they are unable to capture unobserved heterogeneity and also they have to impose the assumption of independence among risks.


20

THE GOODS/ACTIVITIES FRAMEWORK FOR DISCRETE TRAVEL CHOICES: INDIRECT UTILITY AND VALUE OF TIME

Sergio R. Jara-Diaz

ABSTRACT We present the properties of the conditional indirect utility function corresponding to an expanded version of the goods/leisure trade-off model, which includes work and travel time as direct sources of utility. The analysis is focused on the role of the marginal utilities in the formation of a general interpretation of the subjective value of travel time. We show that this analysis depends on the exogeneity of income, and on the relation between goods consumption and consumption time. The marginal utility of work is shown to be particularly important.

INTRODUCTION The microeconomics of travel choices is presently sustained by two powerful frameworks. One is the stream of consumer behaviour theories that include explicitly the time dimension in various forms, and the other is the theory of discrete choices, which are those decisions related with the acquisition of one unit of a general type of good within a specific set of finite alternatives. Time related consumer theories have an evident relation with travel, as travel means relocation in space-time. On the other hand, travel choices have been for long understood and modelled as a series of discrete decisions, i.e. to travel or not, where to go, and how to do it; hence the role of discrete choice theory.

416


When discrete choices are seen from a microeconomic viewpoint, the most important single theoretical construct is the Conditional Indirect Utility Function, which represents the maximum utility that could be obtained if a particular (discrete) choice was made. The arguments and properties of this function depend on the particular manner in which consumer behaviour is understood and modeled. In this article I want to explore the properties of the (conditional) indirect utility function that corresponds to an expanded version of the goods/leisure framework used to model travel choices: the goods/activities framework. I will show that, in its simplest form, this framework generates neat interpretations of the value of travel time, which further complicates when some necessary relations are added, regarding a usually neglected relation between goods and leisure. The important role of the marginal utility of work is particularly highlighted.

FROM GOODS/LEISURE TO GOODS/ACTIVITIES The specific approach to address the presence of time in transport models within the discrete choice framework, has been the goods/leisure trade-off model with mode choice for a specific trip. The approach rests upon a utility function that increases both with the general consumption of goods (G) and with time available out of work (leisure L). Two versions of this model can be built: the original one (Train and McFadden, 1978) in which the individual decides how many hours ^ t o work at a pre-specified wage rate w, and an alternative one (JaraDiaz and Farah, 1987) in which ând /are fixed within the relevant period. The goods /leisure trade-off arises because of the inverse effect of ^ o n G and L : a high value of ^ makes goods consumption large and makes leisure small. A small value of ^reverses the effects. Within this framework, the individual has potentially two choices: how many hours to work and which mode to use. Each mode / has an associated cost c, and travel time tf. If fastexpensive modes compete against slow-cheap ones, the trade-off will be present even if ^(and 7) are fixed, because mode choice translates into a goods-leisure choice through the income and time constraints. On important property of the goods/leisure framework in its original version (self decision on Wand I=wW) is that the subjective value of travel time SK calculated from the corresponding discrete travel choice model is equal to the wage rate. This result should not be a surprise, as the individual adjusts his/her hours of work such that utility is maximized. As the only reward from work time is w, the level of ^will be adjusted until the value of leisure time is w as well.

The Goods/Activities Framework for Discrete Travel Choices

417

as if it was larger (smaller) than w the individual would work less (more). Evidently, this property vanishes in the exogenous income version. The goods/leisure trade-off model corresponds directly to the approach by Becker (1965), who introduced time in the individual utility function in addition to market goods and thus expanded consumer's theory. The time vector T in Becker explicitly accounted for the preparation and consumption of the so-called final commodities, therefore leaving aside working time and other activities (as travel to work) as direct sources of (dis)utility. This omission was criticized by various authors (Johnson, 1966; Oort, 1969; De Serpa, 1971; Evans, 1972) through a series of articles that ended up in the pioneering proposition by Evans (curiously ignored in the literature), who postulated that utility depends primarily on what the individual does ; the goods X would play the role of a necessary input. This activity framework has appeared again in the literature (e.g Gronau, 1986; Winston, 1987 ; Juster, 1990), probably due to the never-ending pressure (socially induced) on the individual's committed time. I have recently proposed a general model of transport users' behaviours that rescues Evans' contribution (Jara-Diaz, 1994). By introducing only leisure in addition to goods in the utility function, the Train and McFadden (1978) and Jara-Diaz and Farah (1987) models suffer from the same limitation as Becker's. In what follows we extend the framework to explore its implications regarding the subjective value of time. The main idea in the activity framework is that all activities have a potential impact on direct utility. In fact, a U(T) utility function as in Evans (1972) or Jara-Diaz (1994) has limitations because the marginal utility of an additional time unit assigned to a specific activity certainly depends on the type and amount of goods used to actually perform the activity. Thus, a U(X ,T) function seems general enough (although we would like to stress the fact that it is T the basic source of satisfaction). When using the goods/leisure framework within an activity approach, we need to introduce both ând travel time using mode z, //, as potential sources of direct (dis) utility. Therefore, the most general expanded version of the model with endogenous income is (A) subject to

G+c.=wW L+W+t.=T ieM

418


where ris the reference period and Mis the set of alternatives (modes). Using the discrete choice procedure, the conditional continuous problem in W is obtained assuming / given and replacing G and L'mU from the constraints. This yields Max

U[{wW-cMT-W-t,),W,t,]

•

(1)

W

The first order condition is

dU

dU =

dW

dU w

dG

dU +

dl

=0

(2)

dW

from which the conditional optimal amount of work W (-c, ,w, T - ti , ti) can be obtained. Result (2) shows that the individual will choose working hours such that the marginal utility of leisure equals the marginal utility of labour, but this latter now has two components : the wage rate times the marginal utility of goods, and the marginal utility of pure labor (which was nil in the original goods/leisure model). The intuitive explanation is straightforward : those who like their jobs would be willing to work not only for the reward in terms of purchasing power, but also for the pleasure of it; everything else constant, this would make such individuals to work more, having less leisure with a higher marginal utility. The reverse occurs if work provokes disutility. From this, the conditional indirect utility function flows as

V. =U (wW* -c-),{T-W*

- / . ) , ^ * , / . ] = F ( c . ,/.)

I

,

(3)

which is the generic version of the so-called modal utility that commands mode choice. From this, the subjective value of travel time SVt can be obtained in the usual manner, as the ratio between d Vj/ dtt and d Vi/ dci First we trivially get

_ ^ _ ^ dt, ~ dG^

dW dt.

i

dU_ dW ~ dl V

/

dW

Û

But W* fulfills condition (2), which combined with eq. (4) yields

^^

The Goods/Activities Framework for Discrete Travel Choices dV, = dt.

dU dG

dU dW

w

+

dU dt.

.

419

f5) ^^

A similar procedure can be used to obtain

^F dc,

Û dG

f

w

dW dc.

I

*

\

1

\

^

dV dW * dJJ dW * + = dl dc. dW dc. I

I

dU dG

.

(6) '^

'

^'^^

From eqs. (5) and (6) we finally get

_ dV, I dt, ^^' " dV, I dc, ~^

dU I dW dU I dt, "^ dU I dG ~ dU I dG

Equation (7) is in fact a very general result for the case of endogenous income. It says that the subjective value of travel time is equal to the wage rate (which is the value in goods units of a unit time saved in travel) plus the subjective value of pure work (which is the goods equivalent of one additional unit time at work) minus the subjective value of pure travel (which is the goods equivalent of one less unit time traveling). In other words, a reduction in travel time is (individually) important because of more work (more (dis) pleasure, more money) and less travel. Note that the result is general, as it holds for positive, negative and null values for the marginal utilities of work and travel. Thus, if an individual likes the job and dislikes travelling, the SVt is definitely higher than the wage rate as saving one minute would mean more money, more pleasure from work, and less displeasure from travel. The case of exogenous income is simple but interesting. It presents, though, an asymmetric condition in this model with work as a source of direct (dis) satisfaction: the individual can not diminish working hours even if he or she dislikes work, but nothing prevents that person from working more if work is pleasurable, in spite of no additional money reward. The general formulation is identical to that of problem (A), but now wW=I and ^ h a s to be at least equal to the fixed amount WF . Replacing the equality constraints we get

420

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Max

u\{l-c,),{T-W-t,),Wj,]

W,i&M

/w-w,>0

^

(B)

/

If problem (B) is solved for W conditional on z, only leisure and work contribute to the variation in U after a change in W. Thus, the first order conditions are

Û

dU

juiW-Wp,)=0

,

// > 0

(8)

If the individual chooses to work more than required, (W* > Wf) then the multiplier //will be nil and the marginal utility of leisure (positive) equals the marginal utility of pure labor, which has to be positive in this case. Then we have W [ (I - ct), (r - tj), tj and ^ / (W positive. Then the conditional indirect utility ftinction for W>Wf is

(9) Therefore ^V, dt,

dU dl

(

dW

*

\

su aw du ^ dw Û ^ Û

(10)

Applying condition (8) for // = 0,

+ ^',

A similar procedure for dVi/dc, to

SV. =

'

gw

(II)

^'z

yields the usual result - ^/6G,

avja.

SU / aw

av, lac,

au i ao

dUldt,

au lac

which makes the SV, equal

(12)


421

Before making an interpretation of this result, note that the case of /J>0 (or W*=Wf) makes ^ W / dti in eq. (10) equal to zero which yields

dVJdt. SV = • -= ' dV; Idc,

dU I dL dU I dG

dU Idt^ dV IdG

*

03) ' ^

Therefore, in the case of exogenous income the subjective value of travel time is always equal to the money value of leisure minus the money value of pure travel; if the individual works more than strictly required, then the money value of leisure is equal to the subjective value of pure work (which has to be positive). A general result for the fixed income case is

dU I dW u SV = + ' dU I dG dV I dG

dUldt, dUldG

'

(14) ^ ^

Results (7) and (14) synthesize the endogenous and exogenous income cases respectively in terms of the subjective value of time, within the "goods/activities" framework. It is interesting to note that there is no reason a priori to expect an SVt equal to either the wage rate w in the first case or the ratio I/W in the second case. This last property also holds for the goods/leisure model with exogenous income, as shown by Jara-Diaz and Farah (1987).

THE MISSING LINK BETWEEN GOODS AND LEISURE The extended version of the goods/leisure framework is richer than its predecessor, as it allows for a direct effect of both work and travel on utility. This is not only appealing intuitively, but also generates analytical results, which contain the previous ones as particular cases. Nevertheless, the new framework still lacks an important relation among its elements, which we will now discuss. Although Becker (1965) established an implicit relation between goods Xand consumptionpreparation times r, it was DeSerpa (1971) who made it explicit that there were minimum consumption times, adding a set of constraints that account for this. Later, Evans (1972) imposed a matrix that turned goods into activities. This idea was rescued by Jara-Diaz (1994)

422


and Jara-Diaz et. al. (1994) in the form of a transformation function that represented a (potentially large) series of technical relations that turned goods into time and time into goods. Within the goods/leisure and goods/activities frameworks, as presented here, the relations between physical consumption and its time counterpart become relations between G and the time aggregates. In what follows, I will make the simplifying assumptions that goods are consumed during L and that work and travel do not require the acquisition of goods. As goods consumption require time, let me define a as the rate of consumption in time units per unit G. According to the first simplifying assumption, G and L have to fulfil

L-aG>0

,

(15)

which means that the resulting leisure time should be large enough to permit the consumption of the resulting amount of goods. If this framework was stated in terms of detailed consumption X and activities T, a second type of condition should be imposed, as activities require certain minimum combination of goods to be developed (think of sports, entertainment, social life, domestic life, etc.) But, in this aggregated conceptual framework, the minimum goods requirement looses meaning as leisure activities have been fully aggregated. Thus, let us concentrate on the conditional analysis of problem A with the addition of constraint (15) and its associated multiplier 0. Again, assuming / given, the problem can be solved in W by replacing G and L as functions of W from the income and time constraints, in both the utility function and the new constraint (15). Thus, we get

Max

U[{wW

-C.),{T-W

-t.),W,t.]

w

subject to

W -t.

- a (wW - c.) > 0

The first order conditions for problem C are

•

(C)

The Goods/Activities Frameworkfor Discrete Travel Choices

û

Iz-t.

du

du

+ a Ci - W

î

^

X ^

{l + a w)\ = 0

,

6> > 0

423

(16)

.

(17)

Equations (16) and (17) yield generic solutions for ând 0. In general we will get W (w, Cj, t^) which, once replaced in the utility of problem C, yields an indirect utility function formally equal to (3) and a marginal utility of time that looks exactly as (4). In this case, by equation (16) we get ^F,

^W*

dU dL

^/.

dU dt.

(18)

I

Similarly

dc,

dc,

{\ + aw)0

(19)

dG

As expected, the case of ^ = ^yields a value of 5*^; given by equation (7). The novel case is 0 > 0, for which

W

T - ^ + a c.

{c,,t,)^—^

(20)

a w

and I+a w

dW d c. I

1+ a w

If these are replaced in the general expressions for d Vi /dti and d Fi /dci, we get

(21)

424


= - 6> dt.

dL

a c.

+— dt.

(22) ^ ^

aG

I

and a more general result for SVt is obtained, namely

The interpretation of this general result follows. If the multiplier îs non-zero (positive), then constraint (15) is active, which means that goods consumption is limited by leisure time. Thus, travel time is more important and travel cost is less important than in the usual case. Accordingly, the marginal utility of travel time is (in absolute terms) larger than the sum of the gain in (more) leisure and the gain in (less) travel (equation 22). And the marginal utility of cost is, in absolute terms, smaller than the marginal utility of goods consumption (equation 23). Regarding the relation between leisure and work, equation (16) confirms that the subjective value of leisure is less than the subjective value of work plus the money value of the goods equivalent. The subjective value of travel time somehow synthesizes the effects of leisure as an active constraint for goods consumption. The denominator of the second and third terms in equation (24) is less than the marginal utility of goods consumption and, therefore, the second term is larger than the subjective value of pure work and the third expression is larger than the direct value of travel time. Thus, if a person likes working and dislikes traveling, his/her subjective value of travel time will exceed the wage rate by a larger amount than in the non-binding leisure time case. This has a clear intuitive interpretation, as travel time not only reduces leisure and (pleasurable) working time, but also diminishes goods consumption through leisure availability. The case we have seen in this section is one in which the individual might run out of free time. In a model with more detail than the one presented here (goods and activities), constraints


425

regarding goods requirement for leisure activities might cause the opposite resuh, in which the individual runs out of money and still have free time left. This case was in fact identified by Evans (1972) and rescued by Jara-Diaz (1997). The fixed income case is, again, simple but interesting. As stated earlier, now the individual has to work at least Wp hours, but he/she can work more if desired. The problem in ^ i s similar to B plus constraint (15) written as in problem C above, i.e.

Max

-t,),W,t,]

U[{I-C,),{T-W

(D)

w

subject to T -W

-t.

a {I - c.)

-

> 0

F

The first order conditions are

\T-W

- t . - a{I-cMe=()

(w - W^Y^

=0

0>O

ju > 0

(26)

(27)

from which we get W*(ci, U) and the conditional indirect utility ftmction Vt (see equation 9). The expression for dVi/dtt is similar to equation (10), and replacing from eq. (25) we get

426


dt.

dt.

^

^^

dL

dt,

^

'

' ^

On the other hand, after a few manipulations,

ac-

oG

oc.

Out of the many possible combinations on ând /^ the most illustrative is the case with 0> 0 and /u = 0, which means that at the optimum the individual works more than the strictly necessary Wf, but is limited by time availability to consume G. Analytically this means

W* = T-t^ - a(I -c^)>Wj,

(9W*

,

=1

dt:

,

^PF*

=a

(30)

dc.I

and dU dW

dU +e dL

.

(SI)

Note that eq. (31) proves that this case is possible only if work is pleasurable, as both ând the marginal utility of leisure are positive. Intuitively, the marginal utility of work should be large enough for the unpaid extra work to overweight the loss in leisure and the loss in consumption because of limited leisure. From equations (28), (29), (30) and (31) we obtain dU dG

IdW

dU

Idt. I

dG

which is equal to the endogenous income case, except for the wage rate. This was to be expected, as work is freely adjusted but the marginal money reword is nil.

The Goods/Activities Frameworkfor Discrete Travel Choices

All

Finally, note that both cases with 6 = 0 m fact represent conditions which we have seen previously, as this means a non-binding leisure regarding goods consumption.

SYNTHESIS AND CONCLUSIONS We have expanded the goods/leisure framework to account for all activities as potential sources of utility, keeping the analysis at an aggregated level. Postulating goods, leisure, work and travel as direct sources of utility is not new, but a strict analysis through a general conditional indirect utility function (CIUF) within the framework of discrete travel choices, is a novel treatment which improves our understanding of what is behind time perception, the role of income and the subjective value of travel time. Two families of models appear as extensions of the goods/leisure framework: those where income is endogenous (i.e. the individual decides how many hours to work at a pre-specified wage rate), and those with exogenous income. In the former case, the subjective value of travel time SVt has three components, namely the wage rate, the direct subjective value of work and the direct subjective value of travel time. In the latter case, the individual can decide to work more (unpaid), provided the marginal utility of work is positive. The corresponding SVt has two components if the individual works more than required. These components are equivalent to those in the endogenous income case with zero wage rate. If the individual works strictly according to contract, there is a third term corresponding to the difference between the direct subjective values of leisure and work. Important results do arise after a new constraint is introduced, namely the necessary relation between goods and leisure, as goods consumption require consumption time. The first important result is that the marginal utility of travel cost, obtained from the CIUF, is not necessarily equal to the marginal utility of goods consumption. Moreover, if goods consumption happens to be limited by leisure, then its marginal utility is in fact larger than the marginal utility of income. This is not only interesting but also intuitively attractive, in line with the discussion by Evans (1972). We must consider, though, that we are assuming that all income is spent, and this needs fiirther exploration. If travel is seen as part of what individuals do, and utility depends primarily on activities as a result of time and money assignment, then travel decisions should be studied, modeled and understood, within the context of human activities. This has many implications like, for example, the need to understand the nature and perception of both work and leisure (e.g.

428


alienated or not, in the sense defined in Fromm, 1965), or the need to take into account the socially induced necessities, as highlighted by Marcuse (1968). This seems to be a very relevant point, as living in a hurry or "having no time" is becoming a new syndrome (and symbol of status) in both developed and developing societies. As stated by Lasch, "a profound shift in our sense of time has transformed work habits, values, and the definition of success" (pp.107). Thus, the social study of behaviour should become part of the effort towards meaningfiil microeconomic interpretations of travelers' decisions. Linking sociology, psychology and economics is not a sophisticated step towards meaningfiil travel choice models. It is a must.

ACKNOWLEDGMENTS This research was partially fijnded by FONDECYT Chile.

REFERENCES Becker, G. (1965). A theory of the allocation of time. The Economic Journal 75, 493-517. DeSerpa, A. (1971). A theory of the economics of time. The Economic Journal 81, 828-846. Evans, A. (1972). On the theory of the valuation and allocation of time. Scottish Journal of Political Economy, February, 1-17. Fromm, E. (1965). Humanismo Socialista^ 2"^. Edition. Paidos, Barcelona, 1984. Gronau, R. (1986). Home production - a survey. In Handbook of Labor Economics, Vol. 1, O. Ashenfelter and R. Layard, eds. North Holland, 273 - 304. Jara-Diaz, S. R. (1997). Time and income in travel demand : towards a microeconomic activity fi*amework. En Theoretical Foundations of Travel Choice Modelling, T. Garling, T. Laitia y K. Westin, eds. Elsevier, forthcoming. Jara-Diaz, S. R. (1994). A general micro-model of users' behavior: the basic issues. Seventh International Conference on Travel Behavior, Conference Preprints, 91-103. Jara-Diaz, S. R. and M. Farah (1987). Transport demand and user's benefits with fixed income : the goods / leisure trade - off revisited. Transportation Research 2 IB, 165-170. Jara-Diaz, S. R., F. Martinez and I. Zurita (1994). A microeconomic framework to understand residential location. 22nd European Transport Forum, Proceedings Seminar K, 115 128. Johnson, B. (1966). Travel time and the price of leisure. Western Economic Journal, Spring, 135-145. Juster, F. T. (1990). Rethinking ufility theory. Journal of Behavioral Economics 19, 155 - 179. Lasch, C. (1979). The Culture of Narcissism: american life in an age of diminishing expectations, Warner, New York. Marcuse, H. (1986). An Essay on Liberation. Beacon Press, MA.


429

Oort, C. (1969). The evaluation of travelling time. Journal of Transport Economics and Policy, September, 279 - 286. Train, K. and D. McFadden (1978). The goods/leisure trade-off and disaggregate work trip mode choice models, Transportation Research 12, 349-353. Winston, G.C. (1987). Activity choice : a new approach to economic behavior. Journal of Economic Behavior and Organization 8, 567 - 585.



21

INTEGRATION OF CHOICE AND LATENT VARIABLE MODELS

Moshe Ben-Akiva, Joan Walker, Adriana T. Bernardino, Dinesh A. Gopinath, Taka Morikawa, andAmalia Polydoropoulou

ABSTRACT This chapter presents a general methodology and framework for including latent variables—in particular, attitudes and perceptions—in choice models. This is something that has long been deemed necessary by behavioural researchers, but is often either ignored in statistical models, introduced in less than optimal ways, or introduced for a narrowly defined model structure. The chapter is focused on the use of psychometric data to explicitly model attitudes and perceptions and their influences on choices. The methodology requires the estimation of an integrated multi-equation model consisting of a discrete choice model and the latent variable model's structural and measurement equations. The integrated model is estimated simultaneously using a maximum likelihood estimator, in which the likelihood function includes complex multi-dimensional integrals. The methodology is applicable to any situation in which one is modeling choice behaviour (with any type and combination of choice data) where (1) there are important latent variables that are hypothesized to influence the choice and (2) there exist indicators (e.g., responses to survey questions) for the latent variables. Three applications of the methodology provide examples and demonstrate the flexibility of the approach, the resulting gain in explanatory power, and the improved specification of discrete choice models.

INTRODUCTION Recent work in discrete choice models has emphasized the importance of the explicit treatment of psychological factors affecting decision-making. (See, for example, Koppelman and Hauser, 1979; McFadden, 1986a; Ben-Akiva and Boccara, 1987; Ben-Akiva, 1992; Ben-Akiva et al., 1994; Morikawa et al., 1996.) A guiding philosophy in these developments is that the

432

In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities

incorporation of psychological factors leads to a more behaviourally realistic representation of the choice process, and consequently, better explanatory power. This chapter presents conceptual and methodological frameworks for the incorporation of latent factors as explanatory variables in choice models. The method described provides for explicit treatment of the psychological factors affecting the decision-making process by modeling them as latent variables. Psychometric data, such as responses to attitudinal and perceptual survey questions, are used as indicators of the latent psychological factors. The resulting approach integrates choice models with latent variable models, in which the system of equations is estimated simultaneously. The simultaneous estimation of the model structure represents an improvement over sequential methods, because it produces consistent and efficient estimates of the parameters. (See Everitt, 1984 and Bollen, 1989 for an introduction to latent variable models and Ben-Akiva and Lerman, 1985 for a textbook on discrete choice models.) Three applications of the methodology are presented to provide conceptual examples as well as sample equations and estimation results. The applications illustrate how psychometric data can be used in choice models to improve the definition of attributes and to better capture taste heterogeneity. They also demonstrate the flexibility and practicality of the methodology, as well as the potential gain in explanatory power and improved specifications of discrete choice models.

SUPPORTING RESEARCH Discrete choice models have traditionally presented an individual's choice process as a black box, in which the inputs are the attributes of available alternatives and individual characteristics, and the output is the observed choice. The resulting models directly link the observed inputs to the observed output, thereby assuming that the inner workings of the black box are implicitly captured by the model. For example, discrete choice models derived from the random utility theory do not model explicitly the formation of attitudes and perceptions. The framework for the random utility choice model is shown in Figure 1. This figure, as well as the remaining figures in the chapter, follows the convention of depicting a path diagram where the terms in ellipses represent unobservable (i.e. latent) constructs, while those in rectangles represent observable vsinablQs. Solid arrows represent structural equations (cause-and-effect relationships) and dashed arrows represent measurement equations (relationships between observable indicators and the underlying latent variables).

Integration of Choice and Latent Variable Models

433

There has been much debate in the behavioural science and economics communities on the validity of the assumptions of utility theory. Behavioural researchers have stressed the importance of the cognitive workings inside the black box on choice behaviour (see, for example, Abelson and Levy, 1985 and Olson and Zanna, 1993), and a great deal of research has been conducted to uncover cognitive anomalies that appear to violate the basic axioms of utility theory (see, for example, Garling, 1998 and Rabin, 1998). McFadden (1997) summarizes these anomalies and argues that "most cognitive anomalies operate through errors in perception that arise from the way information is stored, retrieved, and processed" and that "empirical study of economic behaviour would benefit from closer attention to how perceptions are formed and how they influence decision-making." To address such issues, researchers have worked to enrich choice models by modeling the cognitive workings inside the black box, including the explicit incorporation of factors such as attitudes and perceptions. A general approach to synthesizing models with latent variables and psychometric-type measurement models has been advanced by a number of researchers including Keesling (1972), Joreskog (1973), Wiley (1973), and Bentler (1980), who developed the structural and measurement equation framework and methodology for specifying and estimating latent variable models. Such models are widely used to measure unobservable factors. Estimation is performed by minimizing the discrepancy between (a) the covariance matrix of observed variables and (b) the theoretical covariance matrix predicted by the model structure, which is a function of the unknown parameters. Much of this work focuses on continuous latent constructs and continuous indicators. When discrete indicators are involved, direct application of the approach used for continuous indicators results in inconsistent estimates. For the case of discrete indicators, various corrective procedures can be applied. Olsson (1979), Muthen (1979, 1983, and 1984), and others developed procedures based on the application of polychoric correlations (rather than the Pearson correlations used for continuous indicators) to estimate the covariance matrix of the latent continuous indicators from the discrete indicators. Consistent estimates of the parameters can then be obtained by minimizing the discrepancy between this estimated covariance matrix and the theoretical covariance matrix. (See Bollen, 1989, for more discussion of discrete indicators.) Estimation methods for the case of discrete latent variables and discrete indicators was developed by Goodman (1974)—see McCutcheon (1987) for a discussion. In the area of choice modeling, researchers have used various techniques in an effort to explicitly capture psychological factors in choice models. One approach applied is to include indicators of psychological factors (such as responses to survey questions regarding

434


individuals' attitudes or perceptions) directly in the utility function as depicted in Figure 2 (see, for example, Koppelman and Hauser, 1979; Green, 1984; Harris and Keane, 1998). Another frequently used approach is to first perform factor analysis on the indicators, and then use the fitted latent variables in the utility, as shown in Figure 3. (See, for example, Prashker, 1979a,b; and Madanat et al., 1995). Note that these fitted variables contain measurement error, and so to obtain consistent estimates, the choice probability must be integrated over the distribution of the latent variables, where the distribution of the factors is obtained from the factor analysis model (See, for example, Morikawa, 1989).

Figure 1: Random Utility Choice Model

Figure 2: Choice Model with Indicators Directly Included in Utility

Factor Analysis Explanatory Variables

-M

Indicators

Figure 3: Sequential Estimation: Factor Analysis followed by a Choice Model

Figure 4: Choice Model with Latent Attributes


435

Other approaches have been developed in market research (in an area called internal market analysis), in which both latent attributes of the alternatives and consumer preferences are inferred from preference or choice data. For a review of such methods, see Elrod (1991) and Elrod and Keane (1995.) For example, Elrod (1988 and 1998) Elrod and Keane (1995) and Keane (1997) develop random utility choice models (multinomial logit and probit) that contain latent attributes. In estimating these models, they do not use any indicators other than the observed choices. Therefore, the latent attributes are alternative specific and do not vary among individuals in a market segment. However they do use perceptual indicators post-estimation to aid in interpretation of the latent variables. The framework for their model is shown in Figure 4. Wedel and DeSarbo (1996) and Sinha and DeSarbo (1997) describe a related method based on multidimensional scaling. This research extends the above-described methods by formulating a general treatment of the inclusion of latent variables in discrete choice models. The formulation incorporates psychometric data as indicators of the latent variables. We employ a simultaneous maximum likelihood estimation method for integrated latent variable and discrete choice models, which results in consistent and efficient estimates of the model parameters. The formulation of the integrated model and the simultaneous estimator are described in the following sections. Our work on this methodology began during the mid-1980s with the objective of making the connection between econometric choice models and the extensive market research literature on the study of consumer preferences (Cambridge Systematics, 1986; McFadden, 1986a; and BenAkiva and Boccara, 1987). We first developed a unifying framework for the incorporation of subjective psychometric data in individual choice models. We then proceeded to undertake a number of empirical case studies, some of which are described in this chapter. Finally, in this chapter, we present a general specification and estimation method for the integrated model.

THE METHODOLOGY The objective of this research is the integration of latent variable models, which aim to operationalize and quantify unobservable concepts, with discrete choice models. The integrated model is employed to include latent variables in choice models. The methodology incorporates indicators of the latent variables provided by responses to survey questions to aid in estimating the model. A simultaneous estimator is used, which results in latent variables that provide the best fit to both the choice and the latent variable indicators.

436


Notation The following notation, corresponding to choice model notation, is used: X

observed variables, including S characteristics of the individual Z. attributes of alternative/

X*

latent (unobservable) variables, including S* latent characteristics of the individual Z* latent attributes of alternative / as perceived by the individual

/

indicators of X* (e.g., responses to survey questions related to attitudes, perceptions, etc.) I^ indicators of S* I2 indicators of Z*

U^ U

utility of alternative / vector of utilities

y.

choice indicator; equal to 1 if alternative / is chosen and 0 otherwise

y

vector of choice indicators

a, p, y

unknown parameters

Tj, £, V

random disturbance terms

E, G

covariances of random disturbance terms

D

generic distribution

Framework and Definitions The integrated modeling framework, shown in Figure 5, consists of two components, a choice model and a latent variable model. As with any random utility choice model, the individual's utility JJ for each alternative is assumed to be a latent variable, and the observable choices y are manifestations of the underlying utility. Such observable variables that are manifestations of latent constructs are called indicators. A dashed arrow representing a measurement equation links the unobservable U to its observable indicator y. Solid arrows representing structural equations (i.e., the causeand-effect relationships that govern the decision making process) link the observable and latent variables (X, X*) to the utility V.


437

It is possible to identify a choice model with limited latent variables using only observed choices and no additional indicators (see, e.g., Elrod, 1998). However, it is quite likely that the information content from the choice indicators will not be sufficient to empirically identify the effects of individual-specific latent variables. Therefore, indicators of the latent variables are used for identification, and are introduced in the form of a latent variable model. The top portion of Figure 5 is a latent variable model. Latent variable models are used when we have available indicators for the latent variables X*. Indicators could be responses to survey questions regarding, for example, the level of satisfaction with or importance of attributes. The figure depicts such indicators I as manifestations of the underlying latent variable X*, and the associated measurement equation is represented by a dashed arrow. A structural relationship links the observable causal variables X (and potentially other latent causal variables X*) to the latent variable X*. The integrated choice and latent variable model explicitly models the latent variables that influence the choice process. Structural equations relating the observable explanatory variables X to the latent variables X* model the behavioural process by which the latent variables are formed. While the latent constructs are not observable, their effects on indicators are observable. The indicators allow identification of the latent constructs. They also contain information and thus potentially provide for increased efficiency in model estimation. Note that the indicators do not have a causal relationship that influences the behaviour. That is, the arrow goes from the latent variable to the indicator, and the indicators are only used to aid in measuring the underlying causal relationships (the solid arrows). Because the indicators are not part of the causal relationships, they are typically used only in the model estimation stage and not in model application.

General Specification of the Model As described above, the integrated model is composed of two parts: a discrete choice model and a latent variable model. Each part consists of one or more structural equations and one or more measurement equations. Specification of these equations and the likelihood function follow. Structural Equations.

For the latent variable model, we need the distribution of the latent

variables given the observed variables, f{X* \ X;/,Z^). For example:

438

In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities r=h(X;r)^r?

and 7]-D(0,Z^)

(1)

This results in one equation for each latent variable.

Explanatory Variables X

Indicators /

}

Latent Variable Model

Choice Model Figure 5 Integrated Choice and Latent Variable Model For the choice model, we need the distribution of the utilities, f2(U\X,X*;j3,I.^).

For

example: U = ViX,X*;j3) + s

and

f~Z)(0,SJ

(2)

Note that the random utility is decomposed into systematic utility and a random disturbance, and the systematic utility is a function of both observable and latent variables. Measurement Equations.

For the latent variable model, we need the distribution of the

indicators conditional on the values of the latent variables, f^(I \X,X*;a,ll^). I = g{X,X'';a)-hu

and

u-^D(0,\)

For example:

(3)


439

This results in one equation for each indicator (i.e. each survey question). These measurement equations usually contain only the latent variables on the right-hand-side. However, they may also contain individual characteristics or any other variable determined within the model system such as the choice indicator. In principle, such parameterizations can be allowed to capture systematic response biases when the individual is providing indicators. For example, in a brand choice model with latent product quality (Z*), one may include the indicator j^, for the chosen brand, for example, I^ = a^^Z* + a2,yj + u^, where /^ is an indicator of the perceived quality of alternative /. This would capture any exaggerated responses in reporting the perceived quality of the chosen brand, perhaps caused by justification bias. For the choice model, we need to express the choice as a function of the utilities. For example, assuming utility maximization: fl, y^=\

ifU^=max{UJ ' [0, otherwise

(4)

Note that /?(•), F(), andg()are functions, which are currently undefined. Typically, as in our case studies, the functions are specified to be linear in the parameters, but this is not necessary. Also note that the distribution of the error terms must be specified, leading to additional unknown parameters (the covariances, 2). The covariances often include numerous restrictions and normalizations to both simplify the model and provide identification. Integrated Model. The integrated model consists of equations (1) to (4). Equations (1) and (3) comprise the latent variable model, and equations (2) and (4) comprise the choice model. From equations (2) and (4) and an assumption about the distribution of the disturbance, s, denoted as f2{U\X,X*;fi,Z^),

we derive P{y\X,X*;j3,llJ,

the choice probability conditional on

both observable and latent explanatory variables. Likelihood Function. We use maximum likelihood techniques to estimate the unknown parameters. The most intuitive way to create the likelihood function for the integrated model is to start with the likelihood of a choice model without latent variables: Piy\X;fi,ZJ

(5)

The choice model can be any number of forms, e.g., logit, nested logit, probit, ordered probit, and can include the combination of different choice indicators such as stated and revealed preferences.

440


Now we add the latent variables to the choice model. Once we hypothesize an unknown latent construct, X*, its associated distribution, and independent error components (;;, s), the likelihood function is then the integral of the choice model over the distribution of the latent constructs:

/>(j;|Z;Ar,2,,S,)= \p{y\X,X';P,I.,)f,{X'\X;y,I.^)dX'

(6)

x'

We introduce indicators to improve the accuracy of estimates of the structural parameters. Assuming the error components (//, s, o) are independent, the joint probability of the observable variables >' and /, conditional on the exogenous variables X, is:

UyJ\X;a,p,Y,Z,,Y.,,i:^)=

(7)

|p(jl^,r;A2,)/3(/|x,r;a,s„)y;(r|Z;r,i:,)^' x'

Note that the first term of the integrand corresponds to the choice model, the second term corresponds to the measurement equation from the latent variable model, and the third term corresponds to the structural equation from the latent variable model. The latent variable is only known to its distribution, and so the joint probability of >^, /, and X* is integrated over the vector of latent constructs X*. Functional Forms. The forms of the variables (e.g. discrete or continuous) and assumptions about the disturbances of the measurement and structural equations determine the functional forms in the likelihood equation. Frequently we assume linear in the parameter functional forms, and disturbances that have normal (or extreme value for the choice model) distributions. The choice model portion of the likelihood function is a standard choice model, except that the utility is a function of latent constructs. The form of the probability function is derived from equations (2) and (4) and an assumption about the distribution of the disturbance, s. For example, for a choice of alternative /: U^^V.+s. and V.^V.{X,X*\p)

, / G C, C is the choice set

P ( 7 , = l | X , r ; > ^ , S J =?([/, >^^,y/GC)


441

If the disturbances, s, are iid standard Gumbel, then: P{y^=\\X,Xr^P)

=- ; ^

[Logit Model]

Or, in a binary choice situation with normally distributed disturbances: P{y, =\\X,X*\P) = ^{V.- Vj)

[Binary Probit Model]

where O is the standard normal cumulative distribution function The choice model can take on other forms. For example, ordered categorical choice indicators would result in either ordered probit or ordered logistic form (e.g.. Case Study 3 in this chapter). The form of the distribution of the latent variables is derived from equation (1); the form of the distribution of the indicators is derived from equation (3). The disturbances of the structural and measurement equations of the latent variable model are often assumed to be normally and independently distributed. The latent variables are assumed to be orthogonal and the indicators are assumed to be conditionally (on X* a n d ^ independent. In this case, the resulting densities are:

^x;-h(X;r;)] f(r\X;r,cT^) = Yl— V

Î^-giXXia,)" where: ^ is the standard normal density ftinction (Tj^ and cr^^ are the standard deviations of the error terms of v^ and //^, respectively R is the number of indicators L is the number of latent variables Both the indicators and the latent variables may be either discrete or continuous. See Gopinath (1995) and Ben-Akiva and Boccara (1995) for details on the specification and estimation of models with various combinations of discrete and continuous indicators and latent constructs.

442


Theoretical Analysis The methodology presented here improves upon the techniques described by Figures 1 through 4. Figure 1 - Omitting important latent variables may lead to mis-specification and inconsistent estimates of all parameters. Figure 2 - We a priori reject the use of the indicators directly in the choice model - they are not causal, they are highly dependent on the phrasing of the survey question, and, furthermore, they are not available for forecasting. Figure 3a - The two-stage sequential approach without integration leads to measurement errors and results in inconsistent estimates. Figure 3b - The two-stage sequential approach with integration results in consistent, but inefficient estimates. As long as one is integrating (and therefore, by necessity, not using a canned estimation procedure) one may as well estimate the model simultaneously. Figure 4 - The choice and latent variable model without indicators is restrictive in that the latent variables are alternative specific and cannot vary among individuals. In summary, the approach we present is theoretically superior: it is a generalization of Figures 1 and 4 (so cannot be inferior) and it is statistically superior to sequential methods 3a and 3b. How much better is the methodology in a practical sense? The answer will vary based on the model and application at hand: in some cases, it will not make a difference and, presumably, there are cases in which the difference will be substantial.

Identification As with all latent variable models, identification is certainly an issue in these integrated choice and latent variable models. While identification has been thoroughly examined for special cases of the integrated framework presented here (see, e.g, Elrod, 1988 and Keane, 1997), necessary and sufficient conditions for the general integrated model have not been developed. Identification of the integrated models needs to be analyzed on a case-by-case basis. In general, all of the identification rules that apply to a traditional latent variable model are applicable to the latent variable model portion of the integrated model. See Bollen (1989) for a detailed discussion of these rules. Similarly, the normalizations and restrictions that apply to a standard choice model would also apply here. See Ben-Akiva and Lerman (1985) for further information.


443

For the integrated model, a sufficient, but not necessary, condition for identification can be obtained by extending the Two-step Rule used for latent variable models to a Three-step Rule for the integrated model: 1. Confirm that the measurement equations for the latent variable model are identified (using, for example, standard identification rules for factor analysis models). 2. Confirm that, given the latent variables, the structural equations of the latent variable model are identified (using, for example, standard rules for a system of simultaneous equations). 3. Confirm that, given the distribution of the latent variables, the choice model is identified (using, for example, standard rules for a discrete choice model). An ad-hoc method for checking identification is to conduct Monte Carlo experiments by generating synthetic data from the specified model structure (with given parameter values), and then attempt to reproduce the parameters using the maximum likelihood estimator. If the parameters cannot be reproduced to some degree of accuracy, then this is an indication that the model is not identified (or there is a coding error). Another useful heuristic is to use the Hessian of the log-likelihood function to check for local identification. If the model is locally identified at a particular point, then the Hessian will be positive definite at this point. The inverse Hessian is usually computed at the solution point of the maximum likelihood estimator to generate estimates of the standard errors of estimated parameters, and so in this case the test is performed automatically.

Estimation Maximum likelihood techniques are used to estimate the unknown parameters of the integrated model. The model estimation process maximizes the logarithm of the sample likelihood function over the unknown parameters:

maxthMy„,I„\X„;a,fi,r,J:)

(8)

The likelihood function includes complex multi-dimensional integrals, with dimensionality equal to that of the integral of the underlying choice model plus the number of latent variables. There are three basic ways of estimating the model: a sequential numerical approach, a simultaneous numerical approach, and a simulation approach.

444


The sequential estimation method involves first estimating the latent variable model (equations 1 and 3) using standard latent variable estimators. The second step is to use fitted latent variables and their distributions to estimate the choice model, in which the choice probability is integrated over the distribution of the latent variables. The two step estimation method results in consistent, but inefficient estimates. See McFadden (1986a), Train et al. (1986), and Morikawa et al. (1996) for more details on the sequential approach. An important point is that a sequential estimation procedure that treats the fitted latent variables as non-stochastic variables in the utility function introduces measurement error and results in inconsistent estimates of the parameters. If the variance of the latent variable's random error (;;) is small, then increasing the sample size may sufficiently reduce the measurement error and result in acceptable parameter estimates. Increasing the sample size results in a more precise estimate of the expected value of the latent variable, and a small variance means that an individual's true value of the latent variable will not be too far off from the expected value. Train et al. (1986) found that for a particular model (choice of electricity rate schedule) the impact of the inconsistency on parameter estimates was negligible in a 3000 observation sample. However, this result cannot be generalized; the required size of the dataset is highly dependent on the model specification, and it requires that the variance of the latent variable's error (/;) be sufficiently small. Note that the sample size has no effect on the variance of ;;. In other words, the measurement errors in the fitted latent variables do not vanish as the sample size becomes very large. Therefore, without running tests on the degree of inconsistency, it is a questionable practice to estimate these integrated choice and latent variable models by chaining a canned latent variable model software package with a canned choice model package. Performing these tests requires integration of the choice model. The first case study presented in this chapter uses the sequential estimation approach. This case involved a small choice set, and it was necessary to integrate the choice probability over the latent variables. The inconsistency issue already makes application of the sequential estimation approach quite complex, and it produces inefficient estimates. Alternatively, a fully efficient estimator can be obtained by jointly estimating equations (1) through (4). This involves programming the joint likelihood function (equation 8) directly in a flexible estimation package, which, ideally, has built in numerical integration procedures. This is the method that is used in the second and third case studies presented in this chapter. The dimensionalities of the likelihoods are such that numerical integration is feasible and preferred.


445

As the number of latent variables increases, numerical integration methods quickly become infeasible and simulation methods must be employed. Typical estimation approaches used are Method of Simulated Moments or Simulated Maximum Likelihood Estimation, which employ random draws of the latent variables from their probability distributions. For illustration purposes, consider the use of simulated MLE for the model that we later present as Case Study 1. This is a binary choice (probit) model with 2 latent variables and six indicators (see the Case Study for further details). The likelihood function is as follows:

fSyJ\X;a,p,r,Y.)=\[My{xp,+rp,)Y 6

1

L-Z'a,

Z, -Xy, /=1

\iZ*

^rj,

Note that since this is only a double integral, it is actually more efficient to estimate the model using numerical integration (as we do in the case study). However, the model serves well for illustration purposes. Typically, the random draws are taken from a N(0,I) distribution, so we transform the likelihood by substituting:

Z] = XYI +r]i , / = 1,2 , Tj- N{0, I^ diagonal) {the structural LV equation}

which leads to: /,(3;,/|X;a,Ar,2)=JJ[WÂ+(^ri+^..î)A2+(^r2 + ^,,^2)A2)}* ^

1

cr.. To simulate the likelihood, we take D random draws from 7, and fj^ for each observation in the sample, denoted ff^ and fj^, d=l,...,D. The following is then an unbiased simulator for f{yJ\X;a,j3,y,i:): liyJl

X;a,p,y,i:)

= ^ E { O M ^ A +(^^1 +^,,/7f )A2 +(^^2 + ^,,^2 )A2)}

446


*n—^ 6

1

=rcr,.

r=l

The parameters are estimated by maximizing the simulated likelihood:

Note that, by Jensen's Inequality, In f^ is a biased estimator of In / . When a small number of draws is employed, this results in a non-negligible bias in the parameter estimates. Therefore, one has to verify that a sufficient number of draws is used to reduce this bias. This is usually done by estimating the model using various number of draws, and showing empirically that the parameter estimates are stable over a certain number of draws. For more information on simulation methods for estimating discrete choice models, see McFadden (1986b and 1989) and Gourieroux and Monfort (1996).

Model Application The measurement equations are used in estimation to provide identification of the latent constructs and further precision in the parameters estimates for the structural equations. For forecasting, we are interested in predicting the probability of the choice indicator, P{y\X;a,j3,y,I.). Furthermore, we do not have forecasts of the indicators, /. Therefore, the likelihood must be integrated over the indicators, and the model structure used for application is:

P{y\X;a,/3,r,^)=

jP(y\X,X';/3,^M(X'\^;r,^nydX'

(6)

X'

So once the model is estimated, equation (6) can be used for forecasting and there is no need for latent variable measurement models nor the indicators.


447

BEHAVIOURAL FRAMEWORK FOR CHOICE MODELS WITH LATENT VARIABLES The behavioural framework for choice models with latent variables is presented in Figure 6 [Ben-Akiva and Boccara, 1987]. The modeling framework presented here attempts to analyze explicitly latent psychological factors in order to gain information on aspects of individual behaviour that cannot be inferred from market behaviour or revealed preferences. In this framework, three types of latent factors are identified: attitudes, perceptions, and preferences.

The Cause-Effect Behavioural Relationships Attitudes and perceptions of individuals are hypothesized to be key factors that characterize the underlying behaviour. The observable explanatory variables, including characteristics of the individual (e.g., socio-economics, demographics, experience, expertise, etc.) and the attributes of alternatives (e.g., price) are linked to the individual's attitudes and perceptions through a causal mapping. Since attitudes and perceptions are unobservable to the analyst, they are represented by latent constructs. These latent attitudes and perceptions, as well as the observable explanatory variables, affect individuals' preferences toward different alternatives and their decision-making process. Characteristics of the Individual 5 and Attributes of the Alternatives Z

Attitudinal Indicators

I—-H

Perceptual Indicators

Stated Preferences /p

Revealed Preferences

Figure 6 Behavioural Framework for Choice Models with Latent Variables

448


Perceptions are the individuals' beliefs or estimates of the levels of attributes of the alternatives. The choice process is expected to be based on perceived levels of attributes. Perceptions explain part of the random component of the utility function through individualspecific unobserved attributes. Examples of perceptions in a travel mode choice context for the transit alternative are safety, convenience, reliability, and environmentalfriendliness.Examples of perceptions for toothpaste are health benefit and cosmetic benefit (Elrod, 1998). Attitudes are latent variables corresponding to the characteristics of the decision-maker. Attitudes reflect individuals' needs, values, tastes, and capabilities. They are formed over time and are affected by experience and external factors that include socioeconomic characteristics. Attitudes explain unobserved individual heterogeneity, such as taste variations, choice set heterogeneity and decision protocol heterogeneity. Examples of attitudes in a travel mode choice context are the importance of reliability ov preferences for a specific mode. Examples of attitudes about toothpaste are the importance of health benefits, cosmetic benefits, and price. In this framework, as in traditional random utility models, the individual's preferences are assumed to be latent variables. Preferences represent the desirability of alternative choices. These preferences are translated to decisions via a decision-making process. The process by which one makes a decision may vary across different decision problems or tasks, and is impacted by type of task, context, and socioeconomic factors (Garling and Friman, 1998). Frequently, choice models assume a utility maximization decision process (as we do in our case studies). However, numerous other decision processes may be appropriate given the context, for example habitual, dominant attribute, or a series of decisions each with a different decision-making process. This framework is flexible and can incorporate various types of decision processes.

The Measurement Relationships The actual market behaviour or revealed preference (RP) and the preferences elicited in stated preference (SP) experiments are manifestations of the underlying preferences, and thus serve as indicators. Similarly, we may also have indicators for attitudes and perceptions such as responses to attitudinal and perceptual questions in surveys. For example, one could use rankings of the importance of attributes or levels of satisfaction on a semantic scale. As stated earlier, indicators are helpful in model identification and increase the efficiency of the estimated choice model parameters.


449

Benefits of the Framework The integrated choice and latent variable modeling framework allows us to explicitly model the cognitive processes enclosed by the dashed lines in Figure 6. Incorporating such latent qualitative variables in choice models requires a hypothesis of the type and the role of the latent variables, as well as indicators of the latent variables. The simple framework shown in Figure 6 is a bit deceiving. Attitudes can in fact be any latent characteristic of a decision-maker and thus incorporate concepts such as memory, awareness, tastes, goals, etc. Attitudes can be specified to have a causal relationship with other attitudes and perceptions, and vice-versa. Temporal variables can also be introduced in the specification, and different processes by which people make decisions could be included, such as those described in the section above. There is still a tremendous gap between descriptive behavioural theory and the ability of statistical models to reflect these behavioural hypotheses. Examining the choice process within this framework of latent characteristics and perceptions opens the door in terms of the types of behavioural complexities we can hope to capture, and can work to close the gap between these fields. As with all statistical models, the consequences of mis-specification can be severe. Measurement error and/or exclusion of important explanatory variables in a choice model may result in inconsistent estimates of all parameters. As with an observable explanatory variable, excluding an important attitude or perception will also result in inconsistent estimates. The severity depends highly on the model at hand and the particular specification error, and it is not possible to make generalizations. Before applying the integrated choice and latent variable methodology, the decision process of the choice of interest must also be considered. For more information on behavioural decision theory, see Engel et al. (1973), Olson (1993), and other references listed in the "Supporting Research" section of this chapter.

CASE STUDIES The unique features of the integrated choice modeling framework are demonstrated in three case studies. For each case study, the problem context, a problem-specific modeling framework, survey questions, model equations, and results are presented. The Role of the Case Studies. These case studies have been assembled from a decade of research investigating the incorporation of attitudes and perceptions in choice modeling. The case studies provide conceptual examples of modelfi*ameworks,along with some specific

450


equations, estimation results, and comparison of these models with standard choice models. The aim is to show that the methodology is practical, and to provide concrete examples. The case studies emphasize the general nature of the approach by providing likelihood functions for a variety of model structures, including the use of both SP and RP data, the introduction of an agent effect, and the use of logit, probit, and ordered probit. Model Estimation. The dimensionalities of the likelihoods in each of the three case studies were small enough such that numerical integration was feasible and preferred over simultaneous estimation techniques. Therefore, numerical integration was used in all three studies. The first case study was estimated sequentially, where the choice probability was integrated over the latent variables in the second stage, resulting in consistent, inefficient estimates of the parameters. In the second and third case studies, the latent variable and choice models were estimated jointly (by programming the likelihood function in GAUSS and employing its numerical integration routines), resulting in consistent, efficient estimates. Identification was determined via application of the Three-step Rule as described earlier, as well as using the inverse Hessian to check for local identification at the solution point. Further References. Additional applications of the integrated approach can be found in Boccara (1989), Morikawa (1989), Gopinath (1994), Bernardino (1996), Borsch-Supan et al. (1996), Morikawa et al. (1996), and Polydoropoulou (1997).

Case Study 1: Mode Choice with Latent Attributes The first case study (Morikawa, Ben-Akiva, and McFadden, 1996) presents the incorporation of the latent constructs of convenience and comfort in a mode choice model. The model uses data collected in 1987 for the Netherlands Railways to assess factors that influence the choice between rail and car for intercity travel. The data contain revealed choices between rail and auto for an intercity trip. In addition to revealed choices, the data also include subjective evaluation of trip attributes for both the chosen and unchosen modes, which were obtained by asking questions such as those shown in Table 1. The resulting subjective ratings are used as indicators for latent attributes. It is presumed that relatively few latent variables may underlie the resulting ratings data, and two latent variables, ride comfort and convenience, were identified through exploratory factor analysis. Figure 7 presents the framework for the mode choice model. The revealed choice is used as an indicator of utility, and the attribute rafings are used as indicators for the two latent variables. Characteristics of the individual and observed attributes of the alternative modes are exogenous


451

explanatory variables. Figure 8 provides a full path diagram of the model, noting the relationships between each variable. Charact. of the Traveler S and Attrib. of the Modes Z

- - "H

Indicators of Ride Comfort and Convenience /^

Revealed Preference y (Chosen Mode)

Figure 7 Modeling Framework for Mode Choice with Latent Attributes

- ^x:^^^.- - u

^{T>-».

Figure 8 Full Path Diagram for Mode Choice Model with Latent Attributes (See Table 2 and the model equations for notation.)

452

In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities Table! Indicators for Ride Comfort and Convenience

Please rate the following aspects for the auto trip: very poor Relaxation during the trip 1 2 Reliability of the arrival time 1 Flexibility of choosing departure time 1 Ease of traveling with children and/or heavy baggage 1 Safety during the trip 1 Overall rating of the mode 1

3 3 3 3 3 3

very good 4 5

The mode choice model with latent attributes is specified by the following equations. All variables, including the latent variables, are measured in terms of the difference between rail and auto. This was done to reduce the dimensionality of the integral (from 4 to 2), and was not necessary for identification of the joint choice/latent variable model. Structural Model Z;=Xri-^T7i , 1 = 12 , 77-N(0,Y

diagonal)

{2 equations}

(1X1) (IXIOXIOXI) (1X1)

U = Xp,+Z*p^ + s , f ~A^(0,1)

{1 equation}

(1X1) (1X10)(10X1) (1X2)(2X1) (1X1)

Measurement Model. I^=Z*a^+u^ , r = l,...,6 , L>~A^(0,Z^ diagonal)

{6 equations}

(1X1) (1X2)(2X1) (1X1)

y= (1X1)

1, i f ^ > 0 -1, if^45 vears

K 0.15 0.64 0.18 0.69 0.63 0.16 0.13 0.25 0.06 0.82

Practical Findings from the Case Studies In the case studies reported here, and in our other applications of the methodology, we generally find that implementation of the integrated choice and latent variable model framework results in:

466


•

improvement in goodness of fit over choice models without latent variables.

•

latent variables that are statistically significant in the choice model, with correct parameter signs

•

a more satisfying behavioural representation

Several practical lessons were learned from our application of the methodology. First, in terms of the measurement equations {eq. 3), we found that a sufficient number of indicators relevant to the latent variable under consideration, as well as variability among the indicators, are critical success factors. Second, for the structural equations {eq. I), we found that it can be difficult to find solid causal variables (X) for the latent variables. In some cases, it is difficult to even conceptually define good causal variables, that is, cases in which there are no good socioeconomic characteristics or observable attributes of the alternatives that sufficiently explain the latent attitudes and/or perceptions. However, quite frequently, even if one can define good causal variables, these types of data have not been collected and are not included in the dataset. To address both of these issues, it is critical for the successful application of this methodology, to first think clearly about the behavioural hypotheses behind the choices, then develop the framework, and then design a survey to support the model. The final major lesson is that these integrated models require both customized programs and fast computers for estimation. The estimation programs and models tend to be complex, and therefore the use of synthetic data to confirm the program's ability to reproduce the parameters should be done as a matter of routine. Such a test provides assurance that the model is identified and that the likelihood is programmed correctly, but does not otherwise validate the model specification.

CONCLUSION In this chapter, we present a general methodology and framework for including latent variables—in particular, attitudes and perceptions—in choice models. The methodology provides a framework for the use of psychometric data to explicitly model attitudes and perceptions and their influences on choices. The methodology requires the estimation of an integrated multi-equation model consisting of a discrete choice model and the latent variable model's structural and measurement equations. The approach uses maximum likelihood techniques to estimate the integrated model, in which the likelihood function for the integrated model includes complex multi-dimensional integrals (one integral per latent construct). Estimation is performed either by numerical integration or simulation (MSM or SMLE), and requires customized programs and fast computers.


467

Three applications of the methodology are presented. The findings from the case studies are that implementation of the integrated choice and latent variable model framework results in: improvements in goodness of fit over choice models without latent variables, latent variables that are statistically significant in the choice model, and a more satisfying behavioural representation. Application of these methods require careflil consideration of the behavioural framework, and then design of the data collection phase to generate good indicators and causal variables that support the framework. To conclude, we note that the methodology presented here and the empirical case studies have merely brought to the surface the potential for the integrated modeling framework. Further work is needed to assess ramifications and to transcribe the methodological developments from an academic setting to practical applications, including investigation in the following areas: Behavioural Framework: By integrating latent variable models and choice models, we can begin to reflect behavioural theory that has here-to-for primarily existed in descriptive flowtype models. The behavioural framework and the methodology we present needs to be extended to further bridge the gap between behavioural theory and statistical models. For example, including memory, awareness, process, feedback, temporal variables, tastes, goals, context, etc. in the framework. Validation: The early signs indicate that the methodology is promising: the goodness of fit improves, the latent variables are significant, and the behavioural representation is more satisfying. For specific applications it would also be useful to conduct validation tests, including tests of forecasting ability, consequences of misspecifications (e.g., excluding latent variables that should be present), and performance comparisons with models of simpler formulations. Identification: Other than the methods we present for identification (the Three-step Rule, the use of synthetic data, and the evaluation of the Hessian), there are no additional rules for identification of the general formulation of the integrated choice and latent variable models. Similar to the way that necessary and sufficient rules were developed for LISREL, the knowledge base of identification issues for the integrated model must be expanded. Computation: Application of this methods is computationally intensive. Investigation of techniques such as parallel computing, particularly for estimation by simulation, would greatly ease the application of such models.

468


The approach presented in this chapter is a flexible, powerful, and theoretically grounded methodology that will allow the modeling of complex behavioural processes. Now we need to further explore its potential.

ACKNOWLEDGMENTS The seed for the research described in this chapter was planted by Dan McFadden around 1985 when he invited one of us (Ben-Akiva) to join him in a research project on the use of discrete choice models for energy market research. His 1986 Marketing Science paper presented the key ideas that we have been pursuing. We have also benefited from discussions of the paper at the lATBR '97 Conference in Austin, Texas; the 1998 AMA ART Forum in Keystone, Colorado; and the 1998 HEC Choice Symposium in France. In particular, input from Axel Borsch-Supan, Terry Elrod, Tommy Garling, Michael Keane, Frank Koppelman, and Ken Small was helpful. In addition, we received useful feedback from Brian Ratchford and several anonymous reviewers.

REFERENCES Abelson, R. P., and A. Levy (1985). Decision Making and Decision Theory. Handbook of Social Psychology 1. G. Lindzey and E. Aronsom, Eds. Random House, New York. Ben-Akiva, M. (1992). Incorporation of Psychometric Data in Individual Choice Models. The American Marketing Association Advanced Research Techniques Forum, Lake Tahoe, Nevada. Ben-Akiva, M. and B. Boccara (1987). Integrated Framework for Travel Behavior Analysis. LATBR Conference, Aix-en-Provence, France. Ben-Akiva, M. and B. Boccara (1995). Discrete Choice Models with Latent Choice Sets. InternationalJournal of Research in Marketing 12: pp. 9-24. Ben-Akiva, M., M. Bradley, T. Morikawa, J. Benjamin, T. Novak, H. Oppewal, and V. Rao (1994). Combining Revealed and Stated Preferences Data. Marketing Letters 5, 4: pp. 335-350. Ben-Akiva, M. and S. Lerman (1985). Discrete Choice Analysis: Theory and Application to Travel Demand. The MIT Press, Cambridge, MA. Bentler, P. M. (1980). Multivariate Analysis with Latent Variables. Annual Review of Psychology 31: pp. 419-456. Bernardino, A. T. (1996). Telecommuting: Modeling the Employer's and the Employee's Decision-Making Process. Garland Publishing, New York. Boccara, B. (1989). Modeling Choice Set Formation in Discrete Choice Models. Ph.D. Thesis, Department of Civil Engineering, Massachusetts Institute of Technology. Bollen, K. A. (1989). Structural Equations with Latent Variables. Wiley Series in Probability and Mathematical Statistics, John Wiley & Sons.


469

Borsch-Supan, A., D. L. McFadden, and R. Schnabel (1996). Living Arrangements: Health and Wealth Effects. Advances in the Economics of Aging. D.A. Wise ed. The University of Chicago Press. Cambridge Systematics, Inc. (1986). Customer Preference and Behavior Project Report. Prepared for the Electric Power Research Institute. Elrod, T. (1988). Choice Map: Inferring a Product-Market Map from Panel Data. Marketing Science 7 1: pp. 21-40. Elrod, T. (1991). Internal Analysis of Market Structure: Recent Developments and Future Prospects: Recent Developments and Future Prospects. Marketing Letters 2 3: pp. 253266. Elrod, T. and M. P. Keane (1995). A Factor-Analytic Probit Model for Representing the Market Structure in Panel Data. Journal of Marketing Research 32 1: pp. 1-16. Elrod, T. (1998). Obtaining Product-Market Maps from Preference Data. American Marketing Association Advanced Research Techniques Forum, Keystone, Colorado. Engel, J. F., D. T. KoUat, and R. D. Blackwell (1973). Consumer Behavior: Second Edition. Holt, Rinehart and Winston, Inc. Everitt, B. S. (1984). ^« Introduction to Latent Variable Models. Monographs on Statistical and Applied Probability. Chapman and Hall. Garling, T. (1998). Theoretical Framework. Working paper, Goteborg University. Garling, T. and M. Friman (1998). Psychological Principles of Residential Choice. Draft chapter prepared for Residential Environments: Choice, Satisfaction and Behavior, J. Aragones, G. Francescato and T. Garling eds. Goodman, L. A. (1974). The Analysis of Systems of Qualitative Variables When Some of the Variables are Unobservable. Part 1-A: Modified Latent Structure Approach. American Journal of Sociology 79: pp. 1179-1259. Gopinath, A. D. (1995). Modeling Heterogeneity in Discrete Choice Processes: Application to Travel Demand. Ph.D. Thesis, Department of Civil and Environmental Engineering, Massachusetts Institute of Technology. Green, P. (1984). Hybrid Models for Conjoint Analysis: An Expository Review. Journal of Marketing Research 21: pp. 155-169. Greene, W. H. (1997). Econometric Analysis Third Edition. Prentice-Hall, Inc. Harris, K. M. and M. P. Keane (1998). A Model of Health Plan Choice: Inferring Preferences and Perceptions from a Combination of Revealed Preference and Attitudinal Data. Forthcoming in Journal of Econometrics. Joreskog, K. G. (1973). A General Method for Estimating a Linear Structural Equation System. Structural Models in the Social Sciences. A.S. Goldberger and O.D. Duncan, Eds. Academic Press, New York. Keane, M. P. (1997). Modeling Heterogeneity and State Dependence in Consumer Choice Behavior. Journal of Business and Economic Statistics 15 3: pp. 310-327. Koppelman, F. and J. Hauser (1979). Destination Choice for Non-Grocery-Shopping Trips. Transportation Research Record 611)'. pp. 157-165. Keesling, J. W. (1972). Maximum Likelihood Approaches to Causal Analysis. Ph.D. Thesis, University of Chicago. Madanat, S. M., C. Y. D. Yang and Y-M. Yen (1995). Analysis of Stated Route Diversion Intentions Under Advanced Traveler Information Systems Using Latent Variable Modeling. Transportation Research Record 1485: pp. 10-17.

470


McFadden, D. (1986a). The Choice Theory Approach to Marketing Research. Marketing Science 5, 4: pp. 275-297. McFadden, D. (1986b). Discrete Response to Latent Variables for Which There are Multiple Indicators. Working paper, Massachusetts Institute of Technology. McFadden, D. (1989). A Method of Simulated Moments for Estimation of Discrete Response Models without Numerical Integration. Econometrica 57 5: pp. 995-1026. McFadden, D. (1997). Rationality for Economists. Presented at the NSF Symposium on Eliciting Preferences. Berkeley, California, July. McCutcheon, A. L. (1987). Latent Class Analysis. Sage Publications, Newbury Park. Morikawa, T. (1989). Incorporating Stated Preference Data in Travel Demand Analysis. Ph.D. Thesis, Massachusetts Institute of Technology. Morikawa, T., M. Ben-Akiva, and D. McFadden (1996). Incorporating Psychometric Data in Econometric Choice Models. Working paper, Massachusetts Institute of Technology. Muthen, B. (1979). A Structural Probit Model with Latent Variables. Journal of the American Statistical Association 74: pp. 807-811. Muthen, B. (1983). Latent Variable Structural Equation Modeling with Categorical Data. Journal of Econometrics 22: pp. 43-65. Muthen B. (1984). A General Structural Equation Model with Dichotomous, Ordered Categorical and Continuous Latent Variable Indicators. Psychometrika 49: pp. 115-132. Olson, J. M., and M. P. Zanna (1993). Attitudes and Attitude Change. Annual Review of Psychology 44: pp. 117-154. Olson, P. (1993). Consumer Behavior and Marketing Strategy: Third Edition. Irwin, Inc. Olsson, U. (1979). Maximum Likelihood Estimation of the Polychoric Correlation Coefficient. Psychometrika 44: 443-460. Polydoropoulou, A. (1997). Modeling User Response to Advanced Traveler Information Systems (ATIS). Ph.D. Thesis, Department of Civil and Environmental Engineering, Massachusetts Institute of Technology. Prashker, J. A. (1979a). Mode Choice Models with Perceived Reliability Measures. Transportation Engineering Journal 105 TE3: pp. 251-262. Prashker, J. A. (1979b). Scaling Perceptions of Reliability of Urban Travel Modes Using Indscal and Factor Analysis Methods. Transportation Research A 13: pp. 203-212. Rabin, M. (1998). Psychology and Economics. Journal of Economic Literature XXXVI 1: pp. 11-46. Train, K., D. McFadden and A. Goett (1986). The Incorporation of Attitudes in Econometric Models of Consumer Choice. Cambridge Systematics working paper. Sinha, I. and W. S. DeSarbo (1997). An Integrated Approach Toward the Spatial Modeling of Perceived Customer Value. The American Marketing Association Advanced Research Techniques Forum, Monterey, California. Wedel, M. and W. S. DeSarbo (1996). An Exponential-Family Multidimensional Scaling Mixture Methodology. Journal of Business & Economic Statistics 14 4: pp. 447-459. Wiley, D. E. (1973). The Identification Problem for Structural Equation Models with Unmeasured Variables. Structural Models in the Social Sciences. A. S. Goldberger and O. D. Duncan, Eds. Academic Press, New York.


22

METHODOLOGICAL DEVELOPMENTS: WORKSHOP REPORT Juan de Dios Ortuzar and Rodrigo Garrido

INTRODUCTION This chapter attempts to summarize the discussions and presents the main conclusions of a workshop devoted to examine recent developments and applications of econometric and psychometric methods, and statistical modeling techniques. The workshop was actually designed as a direct extension of the work presented at the 1996 Stockholm Conference on Theoretical Foundations of Travel Demand Modelling (Garling et. al., 1998). The workshop's resource paper presented an overview of some considerable methodological advances in travel and activity modelling that have been made in recent years and which are of direct relevance to improved transport policy analysis and travel demand forecasting (Bhat, 2001). Due to size limitations it did not touch on topics such as methodological advances in data collection techniques, use of stated preference (SP) in combination with revealed preference (RP) information, and longitudinal (panel) data. Complementing the resource paper, several other papers setting up the tone of the discussions were presented dealing with the following important themes: •

Formulation of models integrating choice and latent variables (Ben-Akiva et. al., 2001), models allowing for fuzzy choice set formation (Cascetta and Papola, 2001) and dynamic models to handle attrition and non-response biases in panel samples (Sasaki andMorikawa, 1997).

•

Formulation and estimation of hybrid logit-probit models (better known as mixed logii) with SP and RP data (Brownstone et. al, 2000).

472


•

Estimation and validation of probit models for freight demand forecasting in space and time (Garrido and Mahmassani, 2000) and for modelling choice situations with similar options and structured covariance (Yai, 1997).

•

Definition of a microeconomic framework for the goods/activities trade-off in discrete travel choices (Jara-Diaz, 2001) and formulation of a SP approach to develop models of activity behaviour (Wang et. al, 1997).

The rest of the chapter is organized as follows: first we present the major messages arising from the discussions. Then we summarize an illuminating debate concerning the roles of theory and data in the specification and estimation of appropriate model forms. Finally we put together and comment upon a series of other issues that were identified as sources for new research efforts in the area. The workshop participants, whose knowledge was instrumental in arriving at the conclusions discussed below, are identified in the acknowledgements.

MAJOR MESSAGES There was complete agreement that the main workshop's message should be the following: • •

For well defined (i.e. simple) choice problems with a small number of options, there is a wide range of sophisticated modelling techniques available. However, for complex problems (i.e. dependent variable not well defined, or with a large number of options), we are only able to use rather simple models.

The reasons are that in the case of behavioural models of relatively simple short-term travel decisions (such as mode choice), there are no technical limitations nowadays to the use of an appropriate model form (i.e. allowing for heteroscedasticity, correlation and taste variations). It has also become feasible to incorporate data of different nature or obtained from various sources in the modelling effort. So, the simple logit (MNL, see Domencich and McFadden, 1975) or hierarchical logit (HL, see Williams, 1977; Daly, 1987) models, which prevail in practice, should only be justified when there is evidence that the situation under study does not violate severely their restrictive assumptions. In this sense, other lesser know and hardly ever-used forms (most of them reviewed by Bhat, 2001) may turn out to be completely unnecessary. The multinomial probit (see Daganzo, 1979), the faster mixed logit approximation or the probit with a logit kernel (see Ben-Akiva and Bolduc, 1996) are now possible to estimate and use in reasonable time spans. Thus, in these cases the rationale for model formulation and estimation is as follows:

Methodological Developments: Workshop Report • •

•

Alii

First, anticipate the expected error structure on the basis of theory, knowledge or experience, trying to avoid overtly compHcated error structures. Second, use theory and/or intuition to postulate what variables (and in what form) should enter the systematic utility function. Recall the difference between understanding and forecasting. Measure all the attributes with precision. Third, estimate simple model forms and then test for generalisations using the general probit or mixed logit frameworks.

It is important to note that although these advanced model frameworks can accommodate latent variables and use data from different sources, they may fail to converge in certain cases (in fact, convergence is only guaranteed for the simplest structure, the MNL). Also, it might well be that neither of these could be integrated into supply-demand equilibration mechanisms in practice, thus furthering the gap between the states of theory and practice. On the other hand, in the case of more complexes travel choice situations (i.e. destination choice, trip chains) with a large number of options, or for models of daily activity patterns, there seems to be no recourse but to use the very same restrictive forms mentioned above. Although in principle the treatment is the same, the need to deal with complex dependent variables makes the difference, as in these kinds of settings the problems associated to estimate and use the more general model forms are still insurmountable. Not only are there convergence issues at stake here, but there is also a need to learn how to integrate models in practice for each choice dimension involved. Finally, another important recommendation supported by all participants, was the need to have available hold out samples (for model validation) and to develop other validation tests when estimating behavioural models. This is not only to check on problems like the equi-fmality issue (see for example, Williams and Ortuzar, 1982). Perhaps more importantly is the need to avoid the dangers of over-fitting, which may turn out to be a severe problem in models allowing for sophisticated error structures, even in the case of relatively simple choice situations.

A MAJOR DEBATE Workshop participants included a mixture of specialists from several of the sub-disciplines our profession is now divided into: theoreticians, data gatherers and model fitters. It was interesting to contrast the cool and collected debate between model fitters (in this case, econometric experts) and developers of theory (i.e. microeconomic foundations of models), with the hot

474


(and to some extent bitter) debate between the former and data collection experts. The reason for the latter was the apparent disregard of model fitters to the need for collecting data at a high level of precision in order to understand travel behaviour and estimate appropriate model structures (see the discussion by Ortuzar and Willumsen, 1994). The former debate was extremely reminiscent of the description by Leamer (1978) of the discussions at econometric meetings between a celibate priesthood of statistical theorists on the one hand, and a legion of inveterate sinner data analysts on the other. The present debate was perhaps more interesting precisely because it differed from Leamer's picture, where: "...a calm equilibrium permeates the meetings as priests draw up list of sins ... and sinners need only confess their errors openiy\ Here there were no equilibrium an apparent disagreement about the roles of theory and empirical data in model specification and forecasting. At data collection conferences everything revolves around how to obtain representative samples and to avoid (non-response and other) errors, and how to weigh and correct the information in order to achieve this goal. Essential aspects are also questionnaire design, interviewer training and so on. The only point of agreement relating to these issues here, was the unanimous call for validation studies leading to multiple imputation techniques and non-response modeling methods.

OTHER IMPORTANT TOPICS OF DISCUSSION Although we were not capable of answering them in full, given the time available for discussion, from the beginning we tried to keep in mind the following questions: •

What is the state of the art?

•

Where do we have a consensus of opinion about what we know how to do right?

•

What is not yet known or well understood?

AREAS FOR FURTHER RESEARCH Parts of the answers to the questions above have already been touched upon in the chapter. In particular, we believe that further research is needed on the whole area of dealing with large or complex problems. In the case of modeling activity-based patterns, the problem starts with the need to know how to define the dependent variables. The time assigned to activities has implications in terms of the demand functions for goods and activities. In principle the same

Methodological Developments: Workshop Report

475

framework used for modeling trips in classical studies may be used, as the choices are also discrete although they involve not only time and money, but also quality. Another example is the treatment of space/time interactions (e.g. travel demand is temporally and/or spatially correlated). In both cases the complex structures needed may not converge and there is a need to learn how to combine demand with supply in order to reach equilibrium in practice. Incidentally, there was a small discussion on the meaning of equilibrium, which concluded with the notion that it actually meant internal consistency. Another interesting area for further research has to do with combining SP and RP data in discrete choice modeling. In particular what to do with variables that are common to both types of data sets if their coefficients are widely different (e.g. even with a different sign)? For example, are we certain that the variables are the same? Is it possible and at what cost, to leave certain variables out of one the data sets?

OTHER ISSUES DISCUSSED •

Choice set formation and/or modelling: Although choice set generation is not a compensatory process, this is an area where great improvements can be made due to recent advances in modelling methodology. In particular, the fuzzy set approach proposed by Cascetta and Papola (2001) can be modelled by means of a further index in the framework incorporating attitudes and perceptions of Ben-Akiva et. al. (2001) and solved using latent variable models. One interesting problem relates to the perception of alternatives, and it was noted that it might be not enough to ask which is the choice set, particularly if interested in forecasting.

•

Magnitude of errors in explanatory variables: This concerns the covariance matrix structure. When variables are measured not all variances would be the same (i.e. gender v/s income); the question is,.what happens when you combine? It was agreed that this problem is similar to that of combining information of a different nature and there is a need for validation studies (including more precise measurements).

•

Treatment of continuous/discrete variables: In particular time, which is simple to deal with in the simple MNL but it is not so clear in more complex structures. It was proposed that the same treatment given to spatial fields (where continuous models are transformed into discrete versions) could be used in the case of time.

•

Role and treatment of omitted variables: These may lead to problems with the temporal transferability of models, but it was noted that in some cases variables are omitted purposely (e.g. use of public transport by old pensioners is perceived as derogative). Another kind of omitted variable is the treatment of seasonally in dynamic models, but

476

In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges solving for these may lead to over-fitting. In general the problem is more complicated if the variable is endogenous, and it may be preferable to have reduced form models.

•

Non-response modelling: In general this is an issue of growing importance because non-response is getting higher as times go by. Validation studies allow finding out more about non-respondents. Also, more flexible models can be obtained if validation data is available. The approach is similar to the latent variable method. However, it is not possible to model item non-response error separately from behavioural error, so different non-response models may be obtained using the same data when modelling different choice situations. For unit non-response it is even preferably to impute an entire observation than re-weighting. However, imputation models have errors and it is preferable to use multiple imputation methods (i.e. impute five values, estimate five choices and from these estimate parameter means and variances), in order to get an idea about how much of the error comes from the imputation.

•

Use of external information to improve models: There is no doubt that incorporating outside data should improve model estimation (such as in the case of using traffic counts and aggregate choice data in mode and destination choice estimation). There is evidence suggesting that Bayesian methods could be used in this case.

Finally, it was suggested that information should be circulated about programs available for estimating various types of models, for example in Web sites. Many codes are available for public use but information about them is not sufficiently disseminated.

ACKNOWLDGEMENTS The workshop chairman was Juan de Dios Ortuzar (Chile), the resource paper author was Chandra Bhat (USA) and the reporter Rodrigo Garrido (Chile). The rest of the workshop members were (in alphabetical order): Staffan Algers (Sweden), Moshe Ben-Akiva (USA), Julian Benjamin (USA), Muriel Beser (Sweden), David Brownstone (USA), Sergio Jara-Diaz (Chile), Ying Kang (USA), Frank Koppelman (USA), Kuniaki Sasaki (Japan), Andrea Papola (Italy), Jans Rekdal (Norway), Francesco Russo (Italy), Vaneet Sethi (USA), Ken Small (USA), Frode Voldmo (Norway), Douggen Wang (Holland), Chie-Hua Wen (USA) and Tetsuo Yai (Japan). Ennio Cascetta (Italy), Hani Mahmassani (USA) and Taka Morikawa (Japan) also made useful guest appearances.

Methodological Developments: Workshop Report

All

REFERENCES Ben-Akiva, M. and D. Bolduc (1996). Multinomial probit with a logit kernel and a general parametric specification of the covariance structure. Working Paper, Department of Civil and Environmental Enginnering, MIT. Ben-Akiva, M., J. Walker, A. T. Bernardino, D. A. Gopinath, T. Morikawa and Polydoropoulou (2001). Integration of choice and latent variable models. Chapter 21 in this volume. Bhat, C. (2001). Recent methodological advances relevant to activity and travel behaviour analysis. Chapter 19 in this volume. Brownstone, D., D. S. Bunch and K. Train (2000). Joint mixed logit models of stated and revealed preferences for alternative-fuel vehicles. Transportation Research 34B (5), 315-338. Cascetta, E. and A. Papola (2001). Random utility models with implicit availability and perception of choice alternatives. Transportation Research 9C (4), 249-263. Daganzo, C. F. (1979). Multinomial Probit: The Theory and its Applications to Demand Forecasting. Academic Press, New York. Daly, A. J. (1987). Estimating "tree" logit models. Transportation Research 21B, 251-268. Daly, A. J. and J. de D. Ortuzar (1990). Forecasting and data aggregation: theory and practice. Traffic Engineering and Control 31, 632-643. Domencich, T. and D. McFadden (1975). Urban Travel Demand: A Behavioural Analysis. North Holland, Amsterdam. Garling, T., T. Laitila and K. Westin (1998). (eds.) Theoretical Foundations of Travel Choice Modelling. Elsevier, Amsterdam. Garrido, R. A. and H. S. Mahmassani (2000). Forecasting freight transportation demand with the space-time multinomial probit model. Transportation Research 34B (5), 403-418. Jara-Diaz, S. R. (2001). The goods/activities framework for discrete travel choices: indirect utility and value of time. Chapter 20 in this volume. Leamer, E. E. (1978). Specification Searches: Ad Hoc Inference with Non-Experimental Data. John Wiley & Sons, New York. Sasaki, K. And T. Morikawa (1997). Dynamic choice model revising attrition and nonresponse biases of panel sample. Preprints 8^ Meeting of the International Association for Travel Behaviour Research, Austin, USA, 21-25 September 1997. Wang, D., A. Borgers, H. Oppewal and H. Timmermans (1997). A stated choice approach to developing multi-faceted models of activity behaviour. Preprints 8* Meeting of the International Association for Travel Behaviour Research, Austin, USA, 21-25 September 1997. Williams, H. C. W. L. (1977). On the formation of travel demand models and economic evaluation measures of user benefit. Environment and Planning 9"", 285-344. Williams, H. C. W. L. and J. de D. Ortuzar (1982). Behavioural theories of dispersion and the mis-specification of travel demand models. Transportation Research 16B, 167-219. Yai, T. (1997). Multinomial probit with structured covariance for several choice situations with similar alternatives. Preprints 8^ Meeting of the International Association for Travel Behaviour Research, Austin, USA, 21-25 September 1997.


SECTION 8 FORECASTING



23

FORECASTING THE INPUTS TO DYNAMIC MODEL SYSTEMS

Konstadinos G. Goulias

ABSTRACT Travel behaviour modeling is increasingly moving toward more disaggregate approaches using a variety of dynamic quantitative methods. In parallel, travel demand management, transportation system management, and intelligent transportation system impacts evaluation requires increased resolution in land use and strategy description that also calls for a move at finer disaggregate levels (e.g., employment sites and employees). This methodological movement from zonal to person-based, household-based, and site-based forecasting (called microanalytic approach here) has amplified the need for finer detail in forecasts of the social and economic circumstances for each person and/or household used in the travel behaviour equations. The premier tool used to provide modelers with such data is called sociodemographic microsimulation, promising potential for higher predictive power and flexibility when compared to other approaches. This approach, however, provides only partial coverage in the data batteries needed by the newly developed dynamic travel demand systems and there remain many issues yet to be resolved. In this chapter a brief review of policies that need detailed data is provided first with a list of data needs. Then, specific areas that need additional research are illustrated. This review is address data needs and models/methods that are in their infancy and lag behind policy analysis needs.

INTRODUCTION Models for activity-based and dynamic travel forecasting methods are increasingly developed by researchers mainly in Europe and the United States in support of policy actions that cannot be addressed by existing modeling methods and forecasting applications (see Hofman et al., 1995, Mierzejewski, 1996, Kitamura et al., 1994, Kitamura, 1988, Jones, 1990, and papers in

482


this conference). Significant progress has been accomplished in the three areas of data collection, modeling, and simulation (see Bradley, 1997 on hypothetical scenarios, the review on a variety of survey methods with papers edited by Bonnel et al., 1997, the two recent collections of papers edited by Stopher and Lee-Gosselin, 1996, and Ettema and Timmermans, 1997, and the review by Bowman and Ben-Akiva, 1996, representing a sample of work in this field). The majority of these approaches use dynamic (longitudinal) models of travel behaviour to predict policy impacts on future travel demand. The dynamics included in the models are within-a-day, day-to-day, and from one year to the next while they control for observed and unobserved human heterogeneity. The models, on one hand, aid in examining a region's evolution under alternative travel policies (e.g., market penetration of telecommuting over time) and they do not focus on a single time point in the future (e.g., as it is very often done with the overused UTPS procedures), thus, the new approaches recognize more explicitly the interaction between policy decisions and alternative future paths of change moving along the direction of policy needs identified in van der Hoom (1997). The scope of this workshop is to explore the forecasting capabilities of these new approaches and it includes the input forecasts required by these new travel forecasting models, methods by which to develop such forecasts, as they relate to variables that are not provided externally on a routine basis. The striking majority of these recent developments share the need for detailed person-byperson and household-by-household data for a baseline period (a single period database when cross-sectional models are used or a multi-period database when longitudinal models are used). In addition, either for short, medium, or long term forecasting one needs disaggregate forecasts of all the social, economic, and demographic information used by these relatively new models (e.g., when creating synthetic population time-use schedules we will need to also provide in a synthetic way household composition, resources, and household members' roles and task allocation). Surprisingly, and in a stubborn consistent way in travel forecasting history, work in this area is proceeding at a much slower pace than is needed to support the types of policies governments are proposing and/or exploring and their associated model building efforts. As discussed in the review here, however, there are a few research examples in the U.S. and Europe that have provided experiments with social, economic, and demographic forecasting that have been designed to fill this gap. These experiments have been successful within their design domain but they do not satisfy the increased need for better, more detailed, tailored to policy initiatives, and validated exogenous data. This chapter presents, first, examples of policies and associated new modeling frameworks followed by a list of emerging data needs. Then, the data needs are presented in three groups, i.e., aggregate, disaggregate, and based on sociodemographic forecasting methods. The

Uncertainties in Forecasting: The Role of Strategic Modeling following section provides a discussion on data availability and location. concludes with a summary.

483

The chapter

POLICY INITIATIVES AND N E W MODELING FRAMEWORKS Dissatisfaction with trip-based forecasting tools and attempts to move practice toward activitybased approaches predates the milestone legislation of the 1990s in the U.S. (Allaman et al., 1982). Indeed, issues such as forecasting the inputs to travel demand equations emerged with the first development and application of disaggregate choice models (Tye et al., 1982), which need detailed sociodemographic information at the level of a trip, an individual, and/or a household. Similarly, when aggregate approaches are used (e.g., at the traffic analysis zone), forecasts of sociodemographic information of the residents need to also be provided and many methods used in practice are gross approximations that produce many errors throughout the forecasting exercise (Hamburg et al., 1983). The 1980s research on this subject was partially in response to legislation such as the Federal-aid Urban System and the requirement for Metropolitan Planning Organizations to produce long range transportation plans, transportation system management plans, and a list of transportation projects (the transportation improvement program-TIP). Public agency support (by Urban Mass Transit Administration, UMTA, today called Federal Transit Administration, FTA) for the Urban Transportation Planning System (UTPS) made the four-step procedure - trip generation, trip distribution, (trip-based) modal split, traffic assignment-the standard forecasting tools for evaluating large scale urban facility building in the 10- to 20-year horizon. The development of this tool took more than 30 years to mature (for example compare the 1950s applications in Detroit, Chicago, and Pittsburgh to the later UTPS-like systems in Seattle, Portland, and San Francisco among many others). Over time, however, the need for more accurate forecasting tools that contain richer analytical and forecasting instruments to address policy actions has been identified and documented (e.g. Bajpai, 1991) and has yet to be safisfied (Lawrence and Tegenfeldt, 1997). Indeed, emphasis was given more to the development of operational traffic engineering tools to study short-term improvements instead of critical revisions and improvements of the travel model. Over the past decade, the need to examine new and more complex policy initiatives has become more pressing in the U.S. due to the passage of a series of Acts and associated Federal and State regulations on transportation planning. The intermodal character of the new legislation, its congestion management systems and the taxing air quality requirements for selected U.S. regions have motivated many new forecasting applications that have been predominantly based on the four-step process. Air quality mandates, however, motivated

484


impact assessments of transportation control measures and the creation of statewide mobile source air pollution inventories (Stopher, 1994, Loudon and Dagang, 1994, Goulias et al., 1993) that require more detailed and better forecasting tools than in the past. In addition, lack of funding for transportation improvement projects also motivates the need for impact fees' assessment for individual private developments, which in turn necessitates higher resolution for regional forecasting models and interfacing with traffic engineering tools that are recognized in state and local impact fee legislation (Paaswell et al., 1992). This urgency for new forecasting tools is further amplified by the technology "push" under the general name of Intelligent Transportation Systems (i.e., bundles of technological solutions in the form of user services attempt to solve chronic problems such as congestion, safety, and air pollution). Under these initiatives, forecasting models, in addition to long-term land use trends and air quality impacts, need to also address issues related to technology use and information provision to travelers in the short and medium terms in a temporal continuity. Similarly, van der Hoom (1997) provides a European perspective in policy trends that includes increasing citizen participation, intraEuropean integration, decentralization, deregulation, privatization, environmental concern, mobility costs, congestion management by population segment, and private infrastructure finance. These new policy initiatives place more complex issues in the domain of regional policy analysis and forecasting and amplify the need for better methods that produce forecasts at the individual decision maker level instead of the regional aggregate level. Since analysts and researchers in planning need to evaluate the impacts of new technologies, information provision, and pricing/financing strategies (e.g., tolls), transportation management actions, and assessment of environmental policies, their forecasting capabilities need to be more accurate and detailed in space. This can be done by increasing the level of resolution in the current traffic analysis zones to capture much smaller geographic units. When one considers the use of new technologies, for example, and needs to provide scenarios of possible adoption over time the need arises not only for disaggregate sociodemographic forecasts at a given distant point in time but for the entire trace of sociodemographic change in its evolutionary path. This is due to the differential speed of adoption by specific population segments, which is more relevant to policy than the mythical "equilibrium" point in a twenty-year span. This is also particularly useful when we consider the effects of staging in project development, which needs to be incorporated into the usual long range planning process and the submission of TIP projects with related impacts and comparisons in terms of costs, benefits, and cost effectiveness. In addition, accuracy is particularly important for projects that attempt to influence travel demand and transportation supply at the same time.

Uncertainties in Forecasting: The Role of Strategic Modeling

485

Activity-based models, more than any other travel modeling approach, exemplify this need clearly. An activity-based travel forecasting system is a system that uses as inputs sociodemographic information of potential travelers and land use information to create (synthetic) schedules followed by people in their everyday life. The output, for a given day, is a detailed list of activities pursued, time(s) spent in each activity, and travel information from activity to activity (including travel time, mode used, and so forth). This output is very much like a "DayTimer" for each person in a given region. Given the infancy of this approach, a complete operational activity-based forecasting system does not exist yet. However, given the rapid progress and the advanced state of research, we can envision a hypothetical activity-based forecasting system with ingredients developed by several researchers (Hamed and Mannering, 1990; Kitamura et al., 1994; Kwan, 1995; Recker, 1995; Ben-Akiva and Bowman, 1995; Ettema et al., 1995; Pendyala et al., 1995; Bowman and Ben-Akiva, 1996; Ma and Goulias, 1996; Golob, 1996; Golob and McNally, 1996; Golob et al., 1996; and Vaughn et al, 1997). The types of time use data needs, data collection methods, and data quality needed for three types of these new models have been extensively reviewed recently in Arentze et al. (1997). The simplest way to present the type of "external data" needed for activity-travel demand models is to use as an example or case study an improved regional forecasting model (e.g., the vision provided by the Travel Model Improvement Program in the U.S. DOT). This includes the more traditional travel model (known as the four-stage approach) that has been the most popular application of travel demand forecasting throughout the two American continents, many regions and Nations in Europe, and Australia. A typical regional model entails a spatial and temporal representation of the region under study. This includes the region's geography in terms of land use (e.g., location and intensity of developments and residences), transportation facilities (e.g., highways, bus routes, ridesharing facilities, etc.), economic and social circumstances of the region such as current and future prospects for employment, industry plans-trends and potential, and the social evolution of the region (e.g., households by type, car ownership, income, residence type, etc.). This type of information has been the usual external input to the four-stage approach that was developed mainly to answer what-if type of questions when building new highways that are expected to have major regional impacts. An improved regional model needs to do this but also to assess the impacts of new technologies, provide impact comparisons of transportation system and demand management, and assess the use of information by travelers. In addition, staging of investment (see for example the typical tool of highway financing via bonds) requires the provision of streams of forecasts that can depict a temporal profile of demand that is tailored to the financing stream. The need arrises, then, for detailed social, economic, and demographic forecasts of a region over time that are able to

486


produce longitudinal histories to feed the travel demand models that in turn should be capable of assessing the equity and distributional aspects of the policy initiative at hand.

DATA NEEDS The data needed to simply describe the region in a disaggregate and longitudinal way that is consistent with the new model frameworks currently under development can be grouped into disaggregate and aggregate data. Clearly, the disaggregation needed is at a much finer level than the traffic analysis zone, which is most often of a size similar to a census block group. For example, evidence exists of ecologic fallacy when person and/or household data are used with infrastructure characteristics such as travel time computed at the zonal levels (for an example on accessibility see Lee and Goulias, 1997). In addition, when we attempt to describe what may happen to travel demand in a twenty-year period and we want to observe possible evolutions in demand by scenarios of change (e.g., to describe cyclicalities in behaviour, habitual phenomena, adoption of activity and travel patterns induced by infrastructure improvements) the need also arises for complete personal, household, land use, and economic "biographies" of the people, their households (or other social groups), and the region where they reside. Before presenting the data and models needed to create regional biographies, a brief reference is made to some key types of models needed for a regional forecasting system to produce trip making descriptions over time. The first regards the relationship between residence-workplace location/relocation and travel behaviour of household members. This is particularly important in the U.S. where changing jobs and/or residence is a frequent phenomenon and it is becoming even more important in Europe as unification progresses. One such illustration is provided in Southworth (1995) in which regional economic changes determine population and land use changes, which in turn cause a change in demand for services. This, on one hand, causes new business starts and closures that directly affect land use, which again affect the demand for new transportation services and the creation of these services. This in turn affects travel times and costs that determine a change in travel patterns, which in turn affects local economic changes, which eventually affect employment and population. All these changes are not instantaneous. For example, during change in workplace and/or residence people go through stages of "cognitive disengagement" from the previous workplace and/or residence and phases of "cognitive engagement" with the new workplace and/or residence. As a result their activity and travel patterns go through stages of adjustments and adaptation that should be captured by the regional (activity-based or trip-based) travel forecasting system because they may take longer than the planning horizon used. Integrated land use-transportation models exist and have


487

been used for some time. In addition, comparative studies have provided ample evidence on the need for integration with travel demand, and the relationship with recent policy initiatives has been clearly illustrated (Southworth, 1995, TMIP; 1994). The second area regards telecommunications-information and travel. Telecommunications are used today either intentionally or unintentionally to affect the ways people spend their time. For example, telecommuting has been proposed as a method to mitigate traffic congestion. In this forecasting system, models that represent the use of telecommunications and information by people to participate in activities and travel will also need to be included (for an example of the complex relationships among telecommunications, population changes, and travel see U.S. DOE, 1994, p. 53, and for some model frameworks see Sullivan et al., 1993). The third area addresses lifecycle-lifestyle changes and travel. Lifecycle and associated lifestyle are important determinants of travel and time allocation by individuals and their households. A policy maker should then be interested in separating out the effects of a policy, the effects of "natural" population evolution going through lifecycles, and the effects of the interaction between the two on travel change. This is one of the reasons social-economic-demographic forecasting needs to be an integral part of the transportation policy analysis models. Indeed, van der Hoom (1997) has identified decrease in household size, increase in dwellings, increase in elderly households, increase in workers and multi-worker households, increase in education levels, increase in service employment, and decrease in production employment as key sociodemographic trends affecting and, sometimes, affected by activity and travel patterns over time. In addition, current data collection efforts are neglecting the social networks in which people operate. These are within and outside the household networks that determine the roles, relationships, and tasks of household members. For example, the study of these relationships may provide input to understand and predict long term trends such as the change in labor force participation of women, decrease in in-home activities by women, the slow decrease in labor force participation by men, and the slow increase in in-home activity participation by men.

Disaggregate Data The basic groups of data needed are on person and household evolution, spatial and temporal distributions of activity opportunities, and the most neglected, social networks, depicting human interaction.

488


Person evolution. The size and composition of a population from a socio-demographic perspective is a function of births, deaths, marriages, divorces, formation of extended households, labor force entries and exits, educational attainment, shifts in jobs and vsôrkplace, and income generation. From a strict transportation planning perspective the population composition is also a function of driver's licence attainment and loss (e.g., to assess the path of older driver mobility over time), wôrkplace organization and behavioural incentives (e.g., to assess the potential for transportation demand management strategies' adoption). The targeted questions are what differentiates people with respect to the probability of experiencing each event of the above (e.g., finding a new job), how is this probability affected by the transportation system, and how is it affecting travel demand? Household evolution. The first social structure, formed by people, is a household and each decision they make, in particular travel, is a fiinction of their household (e.g., the presence of children and/or elderly in the household). Data are then needed on fertility decisions (the number of children to have), residence of children and formation of extended families, residence location and relocation and, car ownership decisions (for examples see Wilson, 1979; Rudzitis, 1982; Anas, 1982; Loikanen, 1982). Activity Opportunities. Similarly to formation and dissolution of households, businesses open, close, change in their service offered, and relocate. Many of these decisions affect people's travel but they are also affected by the transportation system. For the travel demand equations data are needed on location of shops, opening and closing hours, variety of goods and services offered, and opportunity clustering in the region (e.g., a Mall). For the locationrelocation decisions of businesses examples of data needed include presence of competing firms, rent levels, parking availability and charge to customers (Hunt, 1997). Ultimately these data should be capable of providing a spatio-temporal description and prediction of activity opportunities. In addition, for modeling the impacts of TDM strategies at the regional level each employment site should also contain information on organization and TDM-related incentives (see TMIP, 1994). Social Networks. The relationships among people can be depicted by using indicators of linkages across people to depict the exchange of resources in terms of a network of relationships. In this way social networks provide the structural environment within which opportunities and constraints to individual action (e.g., activity participation and travel) are depicted clearly. While this conceptualization has not been used neither in data collection nor in travel demand modeling yet, the interaction among household members has been recognized as an important factor affecting behaviour (Golob and McNally, 1996).


489

Aggregate Data Adoption of a disaggregate approach that is driven by disaggregate inputs to produce disaggregate regional forecasts, however, does not in any way eliminate the need for aggregate forecasts. This is mainly due to the need for "control totals." For example, when forecasting the supply of housing, in a given region, one needs to consider the capacity of housing based on zoning regulations. Similarly, employment supply needs to consider general economic conditions in the region and the region's position within a broader National or super-regional economic framework. Land use. Traditional land use information includes indicators of the use intensity in the horizontal and vertical directions. For example, intensity of industry, retail, and service development, existing industrial and retail developments, new developments and their characteristics (e.g., clustering and variety, prices and values, and so forth). In residential development we need existing housing developments and their characteristics, new developments and improvements/changes in existing developments (e.g., infill). Infrastructure. The typical information available is on highways, public transportation systems, bikeways, and walkways. In addition, data are collected to derive measures of accessibility and level of service and related congestion indices. Recent evidence argues for the need to collect data that characterize developments based on the connection of streets, separations of non-motorized paths from the motorized paths, presence of on-street parking, access points to neighborhoods and businesses, size of lots, access to recreational facilities, presence and condition of sidewalks, setback design, and so forth (see Loudon et al., 1997). However, McNally and Kulkami (1997) show that sociodemographic factors are stronger in explaining travel behaviour than land use and highway design factors. Independently of this debate, however, the traditional land use indicators are needed to describe the spatial organization of a given region. In addition, when transportation supply controls are explored, such as access management along an arterial, the need arises to describe the highway network in finer detail and to simulate impacts on the street network at an equally fine detail (e.g. Chung and Goulias, 1996). Infostructure. Initiatives such as the National Information Infrastructure in the U.S. (NRC, 1994) and the Global Information Society (Karamitsos, 1997) are expected to have major impacts on a person's everyday life. Data on the existence and use of and levels of service offered by the existing information infrastructure are absent.

490


Methods and Models Forecasting is essentially an accounting method (Shen, 1994). Each individual unit or groups of units evolve over time in a disaggregate and/or aggregate manner. At regular computing times the analyst checks this spatial and temporal evolution not to exceed given and/or predicted control totals. For example, each household purchases a new home in a given neighborhood, based on its probability of new home purchase, provided a new home is available or an older home is vacant. For regional agencies this has been practice for many years (e.g., in transportation see the survey of methods in Hamburg et al. (1983) in demography see Murdock and Ellis, 1991). The functions that are built into these forecasting systems can be classified into five categories that are mathematical extrapolatory models, comparative methods-models, cohort-survival models, migration models, and regression methods. Mathematical extrapolation (for an example see Newell, 1988) entails the projection of a given variable of interest (e.g., number of people, birth rate, death rate, in- and out- migration, etc.) based on the past trends using a suitable equation to fit past data (e.g., linear, exponential, quadratic, etc.). The major advantages of this method are its simplicity and ease of use. However, the method lacks theoretical underpinnings, does not incorporate structural change in the variable depicted, and it is based on subjective analyst judgment. In addition, given its inability to reflect structural change it is unable to also depict the relationship between a policy and its impacts over time (e.g., slow adoption at the onset, possible accelerated penetration, or possible segment-bysegment bifurcations over time). Comparative methods focus on the relationship of the population under scrutiny to another parent or similar population that has experienced a given policy. Essentially, these are ratios of change that are built on assumptions about the sizes of the population under study and a reference population. Any errors in projecting changes in behaviour of the reference population are directly reflected, then, in the method. In a similar way as in the mathematical extrapolation easiness and simplicity are its advantages, whereas, it does not provide any insights regarding population composition and paths of change. The cohort-survival models are a somewhat disaggregate approach (Turner et al., 1980). All projections using these models are based on "cell" values by cross-classifying the population by uniform age and sex. A cohort model provides some detail because it is a dynamic model of population change and accounts for the two most influential determinants of behaviour. Demographers advocate this method as the most suitable for birth, death, and migration rates


491

(Davis, 1995). The method has also found its way into transportation forecasting (van den Broecke, 1988). Migration models have a place of their own because they are usually accounting systems by themselves. In the model a variety of administrative records are used to estimate the inmigrants and the out-migrants. The direct migration models use, within their accounting framework, survey data of driver's license records, property tax payments, and registration of voters. Indirect migration models are based on the enumeration done in CENSUS by subtracting the natural increase in population from the population change, which yields net migration estimates. One particular sample of the census data, called PUMS, can also be used to derive disaggregate models of migration for transportation planning (Chung and Goulias, 1997). Regression methods for socio-demographic forecasting can be unidirectional (single equation) where the dependent variable is the entity of interest (e.g., probability to migrate) and the independent variables are considered exogenously given (e.g., employment and earnings opportunities, fiscal policy, housing costs, distance from origin, crime at origin or destination). The most interesting regression models are bidirectional or multidirectional (Halli and Rao, 1992). These models reflect more complex causalities among the dependent variables (e.g., migration and labor force participation, fertility and income) by treating them as endogenous variables and functions of each other and other explanatory factors. The glue that brings all models together and produces a population forecast is the population accounting system. This can take the form of a simple accounting method that starts at a given year in which population characteristics are known. The rates/models above are applied to this initial population and projections can thus be produced for the region, by social segment, or by geographical subdivision. The system works in much the same way as a savings account. A simple accounting system example of this sort is provided for Harris County, TX, in Rives and Serow, 1984, and the UTPS procedure, which includes the four-step travel model, is one such accounting method (Ortuzar and Willumsen, 1994, p. 24). A second type of accounting system, which is needed when "integrated" approaches are preferred because of the possible endogeneity in the policy variables, is provided by Rima and van Wissen, 1987. This is a complex accounting system that models population, household, and housing market processes using a dynamic version model system for Amsterdam, NL. The model system was validated and calibrated using an observed time-path. The third type of accounting system is a stochastic microsimulator of the type used in Mackett (1985 and 1990),

492


Miller et al. (1987), Wolfe (1989), Goulias and Kitamura (1992), Miller (1996), and Chung and Goulias (1997). This type of system does not produce a single outcome for a given observational unit but a series of likely paths of change over time. As discussed above a forecasting system needs a routine that uses the data as inputs and in which the models are embedded to produce forecasts. In practice, these are a series of logical statements that given an input population in a region create evolutionary paths of change from a given time point to the next using computer software. We can call this a micro simulator because it operates at the level of a single microscopic unit (e.g., a person, a household, or a vehicle). It is a simulation because we numerically exercise a set of models for a given set of inputs to produce forecasts (as opposed to the use of a closed form and mathematically exact solution to predict the future). Lack of knowledge and the inherent randomness of human behaviour dictates the need to design these systems with at least randomness in input components making the evolutionary engine a stochastic micro simulator (see Law and Kelton, 1991, and for a more complete and focused exposition see Miller, 1996). Given the importance of this method for the current directions in travel demand forecasting a brief overview is also provided here. Orcutt's (1957) "brainchild" microsimulation became accepted only in the late 60's and 70's. Attempting to bridge measurement and theory, Bergmann et al. (1980) note that "micro simulation is a potentially efficient device for organizing scattered versions of theorizing in a consistent manner and in a format that makes efficient confrontation with measurement necessary as not before, albeit in a somewhat new and unconventional garb." The first practical application of microsimulation appeared in 1969 (this is the Transfer Income Model designed to assess alternative family income maintenance programs for the Heineman Commission under President Johnson, see Webb et al., 1990). The method is now used routinely for welfare and taxation analysis in the U.S. (Michel and Lewis, 1990 and Kasten and Sammartino, 1990), health and welfare analysis in Canada (Morrison, 1990), economic and social policies' determination in Germany (Galler and Wagner, 1986 and Brennecke, 1980), as a theoretical-empirical tool in Sweden (Eliasson, 1985, and 1986), for detailed description of the economy in terms of individuals' actions/transactions in the U.S. (Bergmann, 1980), for research on labor force participation and income in France (Ballot, 1982), and for tax reforms in Israel (Habib, 1986), among other uses. Arrow (1980) provides in a succinct way the four main sources of the increasing demand for microanalytic simulation models (or microsimulation models). The first arises from a need to assess impacts of policies at the micro-level. Decision making in policy formation requires


493

information about the costs and benefits of proposed policies and the gainers and losers among those experiencing policy impacts. For example, what happens to solo driving of a specific group of people such as the elderly population when dial-a-ride facilities are provided in a region? This question concerns the distributional aspects or distributional impacts of a policy action (e.g., welfare) and the answer must be at the individual or household level. The second "strength" of microsimulation stems from the transparency of its disaggregate models. Aggregate relations conceal the variability across individual decision making units (persons, households, or firms). Unavoidably aggregate relations assume identical individuals and the error resulting from this assumption is exacerbated when the true underlying relations at the disaggregate level are nonlinear (the striking majority of the activity-based models illustrate exactly this). To the contrary, microsimulation follows the basic tenet of microeconomics D a complex entity composed of many components can best be explained and predicted through an analysis of its constituent parts. The third source results from the accuracy offered from direct observation. Estimates of parameters may be obscured when the system as a whole is studied while, if the system is studied at its elementary unit, parameters may be clearer. Finally, microanalytic (not necessarily microeconomic) models provide richness of information (similar to statistical efficiency), i.e., more information is used by the system and therefore it should yield better predictions of aggregate responses. A fifth source may be cited which concerns the treatment of the temporal dimension. Time can be explicitly included in the system and events can be ordered according to an assumed or derived temporal causation. Persons, families, and firms can be made to evolve over time in the same fashion as they evolve and change stages in the real world (Orcutt et al., 1980). Orcutt et al. (1986) classified microanalytic simulation models in two categories: static microsimulation and dynamic microsimulation. In the static approach the initial sample, used for the first year forecasts, is weighted to reflect changes occurring in the population and is then used in subsequent years. This is needed because static models are used in the simulation. Static models assume that individuals do not change their attributes in response to policy changes or other circumstances and events. Therefore, the need arises to weight the initial sample according to changes occurring in the real population (Michel and Lewis, 1990). In the dynaniic approach the initial sample of persons and families is made to evolve over time with simulated births, deaths, marriages, divorces, immigration & emigration, residential mobility, purchases, etc. The models depicting the behaviour of each unit are dynamic. A dynamic model attempts to capture part of the behavioural responses of individuals over time such as adaptation and resistance to change. The development of weights, if the initial sample follows closely the population changes, is not needed. Clearly, the dynamic approach involving a

494


relationship that is a function of time, is superior to the static approach despite its added complexity. In brief, then, an evolutionary engine called microsimulation attempts to replicate the relationship among sociodemographics, land use, time use, and travel. The causal links among these groups of entities can be extremely complex and in many instances unknown or incompletely specified. This is the reason that no closed form solution can be created for such a forecasting model system. In terms of capabilities, however, the engine needs to provide a realistic representation of person and household life evolution (e.g., birth, death, marriages, divorces, birth of children, etc.) and spatio-temporal activity opportunity evolution while at the same time it accounts for uncertainties in data, models, and behavioural variation. Feedback mechanisms of the type shown in Anas (1982) and TMIP (1994) can be built into the system. The last two accounting methods use ratios of change or probability of change associated with an observation or a group of observations and regression models to provide us with the detailed data needed in travel demand models. The usual aggregate forecasts, provided they are of adequate quality, can then be used as control totals.

WHERE A R E T H E D A T A ?

The usual techniques providing socio-demographic input to travel demand models can be summarized as follows. Forecasts on regional economic and demographic changes are obtained from agencies that routinely perform this type of forecast (e.g.. Bureau of the Census and National Bureau of Economic Research) at a geographically aggregate level, i.e., a region. These, in turn are converted into simple projections of change (change in household size, labor force participation, labor force participation of women, car ownership, type of employment, etc.) at the tract level, or more elaborate macro-econometric models of change at the regional level (e.g. Prastacos 1986a, 1986b). While the techniques, models and procedures used to obtain this input are quite disparate, they share one common characteristic: they are only very rarely at the same level of disaggregation as travel demand models (Hamburg et al., 1983). Most agencies "transform" regional information to the district level, and then from the district level to the traffic zone level. These allocation methods do not provide all the information required by the travel demand models. Additional detailed information is obtained using approximate post-processing procedures. Techniques of this type, allocation methods and postprocessing procedures, have been called disaggregation procedures. The provision of input at the zone level necessitates the application of travel demand forecasting models, designed for households and persons, at the traffic zone level, too. As expected, both the conversion of


495

aggregate socio-demographic forecasts to "zonal" forecasts and the conversion of individual travel demands to "zonal" demands produce many errors throughout the process (Tye et al., 1982; Hamburg et al, 1983; Bajpai, 1991). Although the degree of error decreases as the size of geographical aggregation increases (i.e., traffic assignment on freeways is less sensitive to error in sociodemographic inputs than traffic assignment on arterials and collector roads), the methods fail completely for areas in which rapid growth is observed (Bajpai, 1990). Ironically, it is precisely for these areas that more accurate forecasts from UTPP are needed (Lowry, 1988). Bajpai (1990) observed that "techniques to project automobile ownership, household income, and household size from population and employment are highly recommended for future research." The post-processing techniques that Bajpai (1990) and Hamburg et al. (1983) reviewed were found to produce significant errors at any level of disaggregation. These techniques fail to capture the relationships among population, households, labor force participation and mobility effectively because they do not capture the correlation that exists among variables typically used in travel demand models. Travel demand forecasts based on these techniques are questionable (Lawrence and Tegenfeldt, 1997). However, such "top-down approaches" to providing inputs to travel demand models, may be justified when the decision unit is the region, the state, or the developer and they may be the only available method. This is probably the case with some land-use related decisions (e.g., relocation of large public agencies; Rivkin, 1989) or the creation of statewide emissions inventories (Goulias et al., 1993). However, the use of a traffic analysis zone as the decision making unit is never justified. In fact, most of the sociodemographic variables describe and attempt to replicate decisions made by individuals and households so that the need arises for models that predict these variables at that elementary level of decision making. In addition, when the spatial and temporal distributions of activity opportunities and their evolution over time are needed models at the individual business and firm levels are also needed. Aggregate responses to policy changes can be obtained by grouping households, individuals, and businesses into traffic zones or following any other aggregation scheme desired (e.g., some sort of market segmentation). This approach can be called a "bottom-up" procedure. It is well known that bottom-up approaches lead to more accurate results (Bajpai, 1990). Building such a stochastic microsimulation-based accounting system requires the use of the same external information that is currently used in the "post-processing" procedures to feed the UTPS model system. This is necessary for pragmatic reasons (e.g., to assure the disaggregate models reproduce the population at hand faithfully as it is done in TRANSIMS; Barrett, et al..

496


1995). It is also imperative to use a variety of aggregates because decisions are made at that level and the relationship across levels allows one to unveil system dynamics at differing scales (Eliasson, 1985). To do this, most aggregate information is widely available in the U.S. and in Europe. For example, the CENSUS provides most if these data at the levels of SMS A, County, Census tract, block groups, and partially on blocks. In addition, the availability of PUMS and PUMA allows to create disaggregate models at the level of a person or household. Complementary information in between census years is also provided by the Current Population Survey and the Annual Housing Survey (Rives and Serow, 1984) and a wealth of data is in the public domain (Stewart and Kamins, 1993). Disaggregate data, however, are not available for the population in each region of the U.S.. In Europe a review of data and models reveals this information is not even available at aggregate levels, except at the national level for a few countries (Delavelle et al., 1995). In addition, business data are not widely available and the few databases in the public domain require considerable post-processing to become suitable longitudinal sources of information for the new modeling frameworks. This is another major motivation to include social, economic, and demographic forecasting in travel demand forecasting applications. In addition, policy-specific data such as information that describes telecommunication networks has not been assembled in a widely available database. Finally, social network data are totally absent in spite of their importance in determining the ways people interact with each other and their potential to explain long term trends in travel demand.

SUMMARY Dynamic activity-based approaches are a necessity that emerged from recent legislation, unsatisfied technical needs accumulating for the past two decades, and technology applications in the U.S. and Europe. Current proposed approaches attempt to address new policy questions (van der Hoom, 1997) and chronic problems and frustration with the lack of integration between the aging UTPS-based forecasting methods and land use (Southworth, 1995). The need to integrate these new transportation applications with the social-economic-demographic determinants of travel behaviour is also emerging. On one hand, this is the right time for dynamic activity-based forecasting systems for many reasons. Knowledge about activity-based data collection is at a mature stage (Richardson et al., 1995; Stopher, 1996; Stecher et al.; 1996). Activity model estimation/calibration and related frameworks exist and have been implemented in various contexts, and long-term frameworks to be used for activity-based travel forecasting have been designed (Morrison and Loose, 1995; Barrett et al., 1995, Spear, 1994; Kitamura et al., 1994). In addition, evolutionary engines to perform long-term detailed forecasting based on stochastic micro simulation are available and developing rapidly in


497

transportation research (Miller, 1996). On the other hand, practical issues through demonstration and illustration of the methods remain unresolved largely due to lack of specific "field-tests." In spite of the substantial availability of data, models, and computational algorithms in the social sciences early applications of complete and integrated dynamic activity-based tools are incomplete. In terms of data, aggregate information about a region is widely available and in the public domain with exception data on telecommunication and social networks. To the contrary, disaggregate information (in space, time, and observational units) is absent making it necessary to also include in the forecasting exercise social, economic, and demographic forecasting.

REFERENCES Allaman P.M., T. J. Tardiff, and F. C. Dunbar (1982). New approaches to understanding travel behavior. National Cooperative Research Program Report 250. TRB, Washington D.C. Anas, A. (1982). Residential Location Markets and Urban Transportation. Academic Press, New York, NY. Arentze T., F. Hofman, N. Kalfs and H. Timmermans (1997). Data needs, data collection and data quality requirements of activity-based transport demand models. Paper presented at the conference "Transport Surveys, Raising the Standard." Grainau, Germany, May 24-30. Arrow, K. (1980). Microdata simulation: Current status, problems, prospects. In R.Haveman and K. Hollenbeck (Eds.), Microeconomic Simulation Models for Public Policy Analysis, Vol. 2, (253-265), Academic Press, New York, NY. Bajpai J. N. (1990). Forecasting the basic inputs to transportation planning at the zonal level. National Cooperative Highway Research Program Report 328, Washington, D.C. Ballot, G. (1982) Micro-Macroeconomic simulation of the labour market and system dynamics: Model ARTEMIS. Paper presented for the Vllth International Conference of System Dynamics, Brussels, June 16-18, 1982. Barret C, K. Berkbigler, L. Smith, V. Loose, R. Beckman, J. Davis, D. Roberts and M. Williams (1995). An operational description of TRANSIMS. LA-UR-95-2393, Los Alamos, NM. Ben-Akiva M. E. and J. L. Bowman (1995). Activity based disaggregate travel demand model system with daily activity schedules. Paper presented at the conference "Activity Based Approaches: Activity Scheduling and the Analysis of Activity Patterns." Eindhoven, The Netherlands, May 25-29, 1995. Bergmann, B. (1980). A short term description of a microsimulated macroeconomic model: The transactions model of the U.S. economy. In Bergmann, B., G. Eliasson and G. Orcutt (Eds.) Micro Simulation - Models, Methods and Applications, (23-54), The Industrial Institute for Economic and Social Research, Stockholm, Sweden.

498


Bergmann, B., G. Eliasson and G. Orcutt (1980a). Integrating theory with measurement. In Bergmann, B., G. Eliasson and G. Orcutt (Eds.) Micro Simulation - Models, Methods and Applications, (11-19), The Industrial Institute for Economic and Social Research, Stockholm, Sweden. Bowman J. L. and M. Ben-Akiva (1996). Activity based travel forecasting. Tutorial transcript from the TMIP Conference on Activity-Based Travel Forecasting, New Orleans, LA. Bradley M. A. (1997). The future role of hypothetical survey contexts in transport policy research. Resource paper for The International Conference on Transport Survey Quality and Innovation, May 1997, Grainau, Germany. Brennecke, R. (1980). A sociopolitical decision and indicator system for the Federal Republic of Germany - An overview of SPES. In Bergmann, B., G. Eliasson and G. Orcutt (Eds.) Micro Simulation - Models, Methods and Applications, (377-384), The Industrial Institute for Economic and Social Research, Stockholm, Sweden. Chung J. and K. G. Goulias (1996). Access management using GIS and traffic management tools in Pennsylvania. Transportation Research Record (forthcoming). Chung, J. and K. G. Goulias (1997). Travel demand forecasting using microsimulation: Initial results from a case study in Pennsylvania. Transportation Research Record (forthcoming). Davis H.C. (1995) Demographic Projection Techniques for Regions and Smaller Areas. UBC press, Vancouver, BC, Canada. Delavelle C, A. Michon, and P. Roger (1995). Databases and scenarios for European transport. Final report. DGVII - A4/94/Strategic/two. Eliasson, G. (1985). The Firm and Financial Markets in the Swedish Micro-to-Macro Model: Theory, Model, and Verification. The Industrial Institute for Economic and Social Research, Stockholm. Eliasson, G. (1986). The Swedish micro-to-macro model: Idea, design and application. In Orcutt, G., J. Merz and H. Quinke (Eds.)Microanalytic Simulation Models to Support Social and Financial Policy, (179-259), Elsevier Science Publishers B.V., North Holland, Amsterdam. Ettema D. F., A. Borgers, and H. J. P. Timmermans (1995). SMASH (Simulation Model of Activity Scheduling Heuristics): Empirical test and simulation issues. Paper presented at the conference "Activity Based Approaches: Activity Scheduling and the Analysis of Activity Patterns." Eindhoven, The Netherlands, May 25-29, 1995. Ettema D. F.. and H. J. P. Timmermans, Editors (1997). Activity-based Approaches to Travel Analysis. Pergamon. Elsevier Science LTD, The Netherlands. Galler, H.P. and G. Wagner (1986) The microsimulation model of the Sfb3 for the analysis of economic and social policies. In Orcutt, G., J. Merz and H. Quinke (Eds.) Microanalytic Simulation Models to Support Social and Financial Policy, (227-249), Elsevier Science Publishers B.V., North Holland, Amsterdam. Golob T. F. (1996). A model of household demand for activity participation and mobility. UCIITS-WP-96-5. Institute of Transportation Studies, University of California at Irvine, Irvine, CA. Golob T. F., M. A. Bradley, and J. Polak (1996). Travel and activity participation as influenced by car availability and use. Paper presented at the 75th Annual Transportation Research Board Meeting, Washington, D.C.


499

Golob T. F. and M. G. McNally (1996). A model of household interactions in activity participation and the derived demand for travel. Paper presented at the 75th Annual Transportation Research Board Meeting, Washington, D.C. Goulias, K. G. and R. Kitamura (1992). Travel Demand Analysis with Dynamic Microsimulation. Transportation Research Record, 1357, pp. 8-18. Goulias K. G., T. Litzinger, J. Nelson and V. Chalamgari (1993). A study of emission control strategies for Pennsylvania: Emission reductions from mobile Sources, cost effectiveness, and economic impacts. Final report to the Low Emissions Vehicle Commission. PTI 9403. The Pennsylvania Transportation Institute, University Park, PA. Habib, J. (1986). Microanalytic simulation models for the evaluation of integrated changes in taxes and transfers: Their role in tax and transfer reform in Israel. In Orcutt, G., J. Merz and H. Quinke (Eds.) Microanalytic Simulation Models to Support Social and Financial Policy., (117-134),Elsevier Science Publishers B.V., North Holland, Amsterdam. Halli S. S. and K. V. Rao (1992). Advanced Techniques of Population Analysis. Plenum, New York, NY. Hamburg J. R., E. J. Kaiser , and G. T. Lathrop (1983). Forecasting inputs to transportation planning. National Cooperative Highway Research Program Report 266, Washington D.C. Hamed M. M. and F. L. Mannering (1990). Modeling travelers' post-work activity involvement: Toward a new methodology. Department of Civil Engineering, University of Washington, Seattle, WA. (Mimeo). Hofman F., A. W. J. Borgers, and H. J. P. Timmermans (1995). The necessity of activity based modelling. Paper presented at the conference "Activity Based Approaches: Activity Scheduling and the Analysis of Activity Patterns." Eindhoven, The Netherlands, May 25-29, 1995. Hunt J.D. (1997). A stated preference examination of the location choice behaviour of small retail firms. Paper presented at the 76th Transportation Research Board meeting, Washington, D.C. Jones P. (1990). Developments in Dynamic and Activity-Based Approaches to Travel Analysis. A compendium of papers from the 1989 Oxford Conference. Avebury, UK. Karamitsos F. (1997). The information society and mobility: Telematics applications in transport. In M. Papageorgiou and A Pouliezos (Eds.) Preprints of Transportation Systems IFAC/IFIP/IFORS Symposium, 16-18 June, Chania, Greece. Kasten, R. A. and F. J. Sammartino (1990). A method for simulating the distribution of combined federal taxes using census, tax return, and expenditure microdata. In G.H. Lewis and R.C. Michel (Eds.) Microsimulation Techniques for Tax and Transfer Analysis, (173-186), The Urban Institute Press, Washington, D.C. Kitamura R. (1988). An evaluation of activity-based travel analysis. Transportation 15. Pp. 934. Kitamura R., E. I. Pas, C.V. Lula, T. K. Lawton, P. E. Benson (1994). The sequenced activity mobility simulator (SAMS): An integrated approach to modeling transportation, land use and air quality. Mimeo. Kwan M. (1995). GISICAS: An activity-based spatial decision support system for ATIS. Paper presented at the conference "Activity Based Approaches: Activity Scheduling and the Analysis of Activity Patterns." Eindhoven, The Netherlands, May 25-29, 1995.

500


Law A. M. and W. D. Kelton (1991). Simulation Modeling and Analysis. McGraw Hill, New York, NY. Lawrence M. F. and M. Tegenfeldt (1997). The use of economic and demographic forecasts by metropolitan planning organizations. Paper presented at the 76th Annual Transportation Research Board Meeting. Washington D.C. Lee, M. and K. G. Goulias (1997). Accessibility indicators for transportation planning using GIS. Paper presented at the 76th Annual Transportation Research Board Meeting. Washington D.C. Loikanen H.A. (1982). Housing demand and intra-urban mobility decisions: a search approach. Mimeo. Loudon W. R. and D. A. Dagang (1994). Evaluating the effects of transportation control measures. In Transportation Planning and Air Quality II (eds. T.F. WhoUey). American Society of Civil Engineers, New York, NY. Loudon W.R., J. L. Henneman, L. I. Hartnett, and M. J. Lawlor (1997). Integrating transportation and land use planning: Addressing the requirements of federal legislation and rule making. Paper presented at the 76th Transportation Research Board Meeting, Washington, D.C. Lowry, I. S. (1988). Planning the urban sprawl, hi A Look Ahead: Year 2020., (275-312), Transportation Research Board, Washington D.C. Ma J. and K. G. Goulias (1996). Multivariate marginal frequency analysis of activity and travel patterns in the first four waves of the Puget Sound Transportation Panel. Transportation Research Record (forthcoming). Mackett, R. L. (1985). Micro analytic simulation of locational and travel behaviour. In Proceedings of PTRC Summer Annual Meeting, Seminar L, Transportation Planning Methods, PTRC, London, England, 175-188. Mackett R. L. (1990). Exploratory analysis of long term travel demand using micro-analytical simulation. In P.M. Jones (ed.). New Developments in Dynamic and Activity-Based Approaches to Travel Analysis, (384-405), Gower Publishing, Aldershot, England. McNally, M. G. and A. Kulkami (1997). An assessment of the land use - transportation system and travel behavior. Paper presented at the 76th Transportation Research Board meeting, Washington, D.C. Michel, R. C. and G. H. Lewis (1990). Introduction and overview: Issues and problems in microsimulation modelling. In G. H. Lewis and R. C. Michel (Eds.) Microsimulation Techniques for Tax and Transfer Analysis, (1-31), The Urban Institute Press, Washington, D.C. Mierzejewski E. A. (1996). An Assessment of Uncertainty and Bias: Recommended Modifications to the Urban Transportation Planning Process. Unpublished Ph.D. Dissertation, Department of Civil and Environmental Engineering, University of South Florida, Tampa, FL. Miller E. J. (1996). Microsimulation and activity-based forecasting. Paper presented at the TMIP Conference on Activity-Based Travel Forecasting, New Orleans, LA. Miller, E. J., P. J. Noehammer and D. R. Ross (1987). A micro-simulation model of residential mobility. In W. Young (Ed.) Proceedings for the International Symposium on Transport, Communication and Urban Form: 2, Analytical Techniques and Case Studies , (217-234), Clayton, Monash University, Victoria, Australia. Morrison, J. and V. Loose (1995). TRANSIMS Model Design Criteria as Derived from Federal Legislation. LAUR 95-1909. Los Alamos National Laboratory.


501

Morrison, R. J. (1990) Microsimulation as a policy input: Experience at Health and Wellfare Canada. In G. H. Lewis and R. C. Michel (Eds.) Microsimulation Techniques for Tax and Transfer Analysis, (77-108), The Urban Institute Press, Washington, D.C. Murdock S. H. and D. R. Ellis (1991). Applied Demography. Westview Press, San Francisco, CA. National Research Council (1994). Research recommendations to facilitate distributed work. National Academy Press, Washington D.C. Newell C. (1988) Methods and Models in Demography. Belhaven press, London, U.K. Orcutt, G. (1957). A new type of socio-economic system. Review of Economics and Statistics, 58, 773-797. Orcutt, G., A. Glazer, R. Harris and R. Wertheimer II (1980). Microanalytic modeling and the analysis of public transfer policies. In R. Haveman and K. Hollenbeck (Eds.), Microeconomic Simulation Models for Public Policy Analysis, Vol. 1, (81-113), Academic Press, New York, N.Y. Orcutt, G., J. Merz and H. Quinke Eds. (1986). Microanalytic Simulation Models to Support Social and Financial Policy. Elsevier Science Publishers B.V., North Holland, Amsterdam. Ortuzar J. de D. and L. G. Willumsen (1994). Modelling Transport (Second Edition). Wiley, New York, NY. Paaswell R. E., N. Rouphail, and T. C. Sutaria, Editors (1992). Site impact traffic assessment. Problems and solutions. ASCE, New York, NY. Pendyala R., R. Kitamura, and D. V. G. P. Reddy (1995). A rule-based activity-travel scheduling algorithm integrating neural networks of behavioral adaptation. Paper presented at the conference "Activity Based Approaches: Activity Scheduling and the Analysis of Activity Patterns." Eindhoven, The Netherlands, May 25-29, 1995. Prastacos, P. (1986a). An integrated land-use-transportation model for the San Francisco Region: 1. Design and mathematical structure. Environment and Planning, 18A, 307322. Prastacos, P. (1986b). An integrated land-use-transportation model for the San Francisco Region: 2. Empirical estimation and results. Environment and Planning, 18A, 511-528. Recker W. W. (1995). The household activity pattern problem: General formulation and solution. Transportation Research, Vol 29B, pp.61-77. Richardson A. J., E. S. Ampt, and A. H. Meyburg (1995). Survey Methods for Transport Planning. Eucalyptus Press, Parkville, Victoria, AUS. Rima A. and L. J. G. van Wissen (1987). A Dynamic Model of Household Relocation. Ph.D. Dissertation, Free University Press, Amsterdam, The Netherlands. Rives N. W. and Serow W. J. (1984). Introduction to Applied Demography. Sage, Newbury Park, CA. Rivkin, M. D. (1989). Can transportation management reduce traffic in the suburbs? Ask the Nuclear Regulatory Commission. In Transportation Research News No. 141. Rudzitis, G. (1982). Residential location determinants of the older population. The University of Chicago, Department of Geography, Research Paper 202, Chicago, IL. Shen, J. (1994). Dynamic Analysis of Spatial Population Systems. Concepts and Techniques in Modern Geography, London School of Economics, Concepts and Techniques in Modern Geography (CATMOG) 57, London, U.K.

502


Southworth, F. (1995). A technical review of urban land use-transportation models as tools for evaluating vehicle travel reduction strategies. Oak Ridge National Laboratory, TM6881, Oak Ridge, TN. Spear B. D. (1994). New Approaches to Travel Forecasting Models: A Synthesis of Four Research Proposals. DOT-T-94-15. TMIP. Stecher, C. C, S. Bricka and L. Goldenberg (1996). Travel behavior survey data collection instruments, hi Conference on Household Travel Surveys: New Concepts and Research Needs. Conference Proceedings 10. Transportation Research Board, Washington, D.C.. Stewart, D. W. and M. A Kamins (1993). Secondary Research. Sage Publications, Newbury Park, CA. Stopher, P. R. (1994) Predicting TCM responses with urban travel demand models. In Transportation Planning and Air Quality II (eds. T.F. Wholley). American Society of Civil Engineers, New York, NY. Stopher, P. R. (1996). Household travel surveys: New concepts and research needs. In Conference on Household Travel Surveys: New Concepts and Research Needs. Conference Proceedings 10. Transportation Research Board, Washington, D.C.. Stopher, P. R. and M. Lee-Gosselin, Editors (1996). Understanding Travel Behaviour in an Era of Change. Pergamon, Elsevier Science, Ltd., Oxford, U.K. Sullivan M. A., H. S. Mahmassani, and J. Yen (1993). Choice model of employee participation in telecommuting under a cost-neutral scenario. Transportation Research Record 1413, 42-48. TMIP (1994). The effects of land use and travel demand management strategies on commuting behavior. U.S. DOT, DOT-T-95-06, Washington, D.C. Turner C. G., M. D. Smith, and F. A. Lenk (1980). Forecasting economic and demographic change in slow-growth economy. Transportation Research Record, 820, pp. 23-29. Tye, W. B., L. Sherman, M. Kinnugan, D. Nelson and T. Tardiff (1982). Application of Disaggregate Travel Demand Models. National Cooperative Highway Research Program Report 253, Washington, D.C. U.S. DOE (1994). Energy, Emissions, and Social Consequences of Telecommuting. DOE/PO0026S, Washington, D.C. van den Broecke, A. A. J. (1988). Long term forecasting of car ownership with the cohort processing model. PTRC 16th Summer Annual Meeting, Proceedings of Seminar D, PTRC Education and Research Services, Ltd., London, England, pp. 183-191. van der Hoom T. (1997). Practitioner's Future Needs. Paper presented at the conference "Transport Surveys, Raising the Standard." Grainau, Germany, May 24-30. Vaughn K.M., P. Speckman, and E.I. Pas (1997) Generating household activity-travel patterns (HATPs) for synthetic populations. Paper presented at the 76th Transportation Research Board meeting, Washington, D.C. Webb, R. L., R. C. Michel and A. B. Bergsman (1990). The historical development of the Transfer Income Model (TRIM2). In G.H. Lewis and R.C. Michel (Eds.) Microsimulation Techniques for Tax and Transfer Analysis, (33-75), The Urban Institute Press, Washington, D.C. Wilson F. D. (1979). Residential consumption, economic opportunity, and race. Academic Press, New York, NY.


503

Wolfe, R. A. (1989). Comparing three forecasting methods using multiple time period data: Dynamic methods, Bayesian updating and microanalytic updating. Paper presented at the International Conference on Dynamic Travel Behavior Analysis, Kyoto University, 18-19 July, Kyoto, Japan.



24

UNCERTAINTIES IN FORECASTING: THE ROLE OF STRATEGIC MODELING TO CONTROL THEM

Charles Rata

ABSTRACT The growing concern about environmental degradation from transport activity at short-range and long-range horizon calls for policies aiming at reorientation of travel demand trends. However, every transport policy is subject to risks, environmental or financial ones, and has often long-range effects. This explains the renewed interest in tools which allow detection of these risks and their consequences. There is however a methodological challenge in the elaboration of these simulation tools because we have to take into account many different uncertainties. This study analyzes the uncertainties associated with transport forecasts using a strategic model recently developed for Lyon's conurbation. Different sources of error and uncertainty are tested and compared by means of the model. It is argued that a strategy of systematic exploration of uncertainty is the preferred way to cope with it and to detect long-term risks associated with transport policy.

INTRODUCTION There is a growing concern about environmental degradation from transport activity at shortrange horizon (regional pollution, daily life surroundings) and at long-range horizon (global climate change). This concern motivates the search for policies aiming at reorientation of travel demand towards modes less harmful for the environment, or even at reducing in some ways the

506

In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities

vehicle-miles of travel. However, these policies encounter long-range trends in the social and spatial context: these are for instance several decades of urban sprawl, density decrease and car-ownership growth. These long-range trends make the desired changes more costly, and even make in some cases the current situation irreversible. These growing costs are for instance those of providing mass transit in less and less dense areas or managing car-pool programs and so on. Moreover, there is a risk of greater social and economic costs related to the clean up of the environment in the future. This is in conflict with public expenditure shortage, for instance to provide basic transport infrastructure (public transport or roads), yielding the search for social and economic efficiency of transport systems. It clearly appears that every transport policy measure is subject to risks, environmental or financial ones, and that these measures have often long-range effects. This explains why there is a renewed interest from transport agencies in several countries in tools that allow detection of these risks and their consequences. Such tools should contribute to the development of strategies to reduce the occurrence of these risks or to minimize their consequences. This supports the need for tools allowing long-range (ten years) simulation of the potential effects of transport policies in an evolving and not controlled context. These tools should also be flexible, allowing the test of alternative policies under contrasted hypotheses of socioeconomic context evolution (for instance economic growth, incomes, etc.). Of course, the results should not be detailed forecasts of an inescapable future but rather ideas of the size of different development trends connected to the contrasted hypotheses and to different policy actions. These tools should play a pedagogical role, helping transport agencies and local authorities to confront the possible results of their action. There is, however, a methodological challenge in the development of these simulation tools. We have to take into account many different uncertainties but also the evolutionary path in the simulation range. The majority of work on the errors of transport demand models and on the uncertainty attached to the forecasts made using these models date from the beginning of the Eighties. MacKinder and Evans (1981) showed that the main part of the prediction errors made with conventional aggregate models, came from prediction errors of the exogenous context, i.e. of the urban and economic growth. A workshop at the 4th lATBR conference was also devoted to this problem (cf in particular contributions of Horowitz (1981), Talvitie (1981) and Ashley (1981)). The conclusions of this workshop (Koppelman, 1981) were that the structure of interaction errors

Uncertainties in Forecasting : The Role of Strategic Modeling

507

between the various sub-models, as well as the propagation of the errors, were poorly understood. The overall estimation error can be viewed as the outcome of (a) the measurement and sampling errors in the surveys, (b) the errors resulting from incorrect behavioural theories and specifications, and (c) biased estimation procedures. To this overall estimation error are added the errors of context forecast to form the overall prediction error. In this study, we analyze the uncertainties associated with the forecast according to three categories: •

•

•

The first covers the overall estimation error referred to above: there are of course errors in data measurement and sampling on the one hand, in behavioural theory, model specification and calibration on the other hand; The second encompasses uncertainties regarding the exogenous context: economic and income growth yielding for instance car-ownership development; housing development; employment growth and location; demographic trends yielding a population less captive to public transport; time constraints evolution (work, school and facilities schedules); The third arises from projection into the future of the current behavioural mechanisms: these uncertainties reflect the potential long-range instability of travel behaviour models. For instance, the coupling between car-ownership and car-mobility developments could be disrupted in the future; or elasticities which we know to be different in the short versus long range, may significantly evolve.

The purpose of the study is to discuss the relative importance of the different uncertainties listed above, and the strategies to cope with such uncertainties. The discussion relies upon a strategic model recently developed for Lyon's conurbation. In the first section we present the strategic model developed in Lyon. The second section is devoted to the measures of various sources of uncertainty. These results are synthesized and discussed in the conclusion.

THE STRATEGIC M O D E L DEVELOPED IN LYON

Why a Strategic Model? In transport economics as in any scientific discipline, each model can be defined only relative to the modeled phenomenon and the paradigms to which the model refers: this makes it possible to define the application field of the model^^. Each model thus has specific capabilities

508


that match the range of needs for the evaluation needs of urban transport policies. This range of needs can be ordered according to two dimensions, spatial and temporal. The spatial dimension ranges from the neighborhood level up to the level of the agglomeration or conurbation. The temporal dimension goes from the short term (1 to 3 years) to the long term (10 years and beyond). These two dimensions make it possible to order the various types of studies needed to articulate a transport policy. It is clear that there is no all-purpose model and that each type of study requires specific tools. It can be recognized that the analyst is better equipped with short or medium term models, as regards network assignment models or discrete choice models, than with long term models. The long term is accompanied by inherent uncertainty, with multiple dimensions. The quantitative evaluations resulting from a long term model are generally applicable only at a relatively aggregate spatial level. In brief, the strategic model must rely on regularities detected at the agglomeration level, spatially, and a temporal scale of tens of years. The field of application of what we call a strategic model, is to be able to simulate on an agglomeration scale the consequences of various transport policies under alternative urban, socioeconomic and demographic development contexts. It is not a matter of providing detailed forecasts of a unique, inescapable ftiture, but of providing evaluations, in the form of orders of magnitude, corresponding to these alternative scenarios. As such the strategic models have a pedagogical role to inform the of the transport system actors, while making it possible to contrast and confront their respective visions of the agglomeration's evolution. Compared to these objectives the existing operational tools offer only few answers. The toolboxes are very well stocked but do not lend themselves to simulations of strategies. Indeed, they suffer in this respect two major limitations, which are slowness and "heaviness." The slowness of these tools rises from the objective which is assigned to them, namely the calculation of the road and public transport networks loading, which requires a necessarily fine spatial division (more than 100 areas for a million-person agglomeration). This implies considerable time and effort to search for information and interpret results. This level of detail weighs down the treatment of feedbacks considerably. However, the degree of road congestion obviously has important impacts on modal split, households residential location and job location, and challenges the mobility level or the car-ownership level. Explicit representation of these interactions is theoretically possible but is often ignored (Stopher et al, 1996).


509

The concept of strategic model is not new, since for example the QuinQuin model developed in the middle of the Eighties (Bonnafous, 1985; Bouf, 1989; Tabourin, 1989; Bonnafous and Tabourin, 1995), takes into account this strategic long term dimension to evaluate the consequences of transport policies, in the realms of public transport financing and road network congestion (Raux and Tabourin, 1992). However, that model is highly aggregate at the agglomeration level, whereas the model proposed here is intended to be more sensitive to the spatial dimensions of the urban system. Strategic models were also developed in the United Kingdom, within the framework of a series of integrated studies of transport, in particular for London (Oldfield, 1993), Birmingham (Jones et al, 1990) and Edinburgh (Bates et al, 1991). These models operate with a reduced number of area (zones), a representation of the transport supply by speed-flow relations between areas, and the taking into account of the feedback of the supply state on the demand. Other relevant models include the simplified transport demand models developed by Ortuzar (1992).

The Principle of the Strategic Model Developed in Lyon The strategic model developed within the framework of the Lyon agglomeration tries to bring answers to the requirements identified earlier. The search for regularities in behaviour on the scale often years was performed on the basis of analysis of the three household travel surveys carried out in Lyon in 1976-77, 1985-86 and 1994-95. This 20 year retrospective depth is expected to result in more robust invariants (model specifications and parameter values) developed on the basis of the data analysis. The spatial resolution is the result of a compromise that takes account of the acceptable degree of error, which is a function of the sampling errors of the surveys. The Lyon agglomeration includes 1,200,000 inhabitants in an area of approximately 1,000 km^. A spatial division into 25 areas was established, which is ten times less than for a conventional short-term model. It is a subdivision for projection of the results, which is different from a subdivision for estimation of the sub-models. This subdivision is also different from one intended for presentation of the results, with even fewer areas (zones). This limited subdivision is compatible with the municipal zonal structure for which the socioeconomic and demographic data are available. The household travel surveys rely on the trip paradigm. It is not a question of not being aware of the paradigm shift taking place over nearly 20 years, towards activity-based approaches. We think that the activity-based approaches are very relevant instruments when they are used in exploratory studies, in hypothetical situations: HATS (Jones, 1979), CUPIG (Lee-Gosselin, 1990) and other stated adaptations techniques (Raux et al, 1994; Faivre d' Arcier et al, 1997).

510


These approaches are on the other hand far from providing direct operational tools for modeling travel choices. Furthermore, we believe that their role is to contribute to the development and testing of new behavioural hypotheses, as well as guiding the modification of the existing models. Thus, we introduced modifications, which may initially appear to be marginal, in the conventional structure of modeling. Indeed the strategic model rests on a conventional fourstage architecture where the stages of generation, distribution and modal split, are carried out at the daily level, while a transition to the peak hour is introduced before the assignment of trips into the networks. The first modification is to model trips in the form of trip chains. Indeed, the preoccupation with a search for behavioural invariants led us to reconstitute, on the basis of the survey data, the trip chains, which were attached to a main purpose: for example the trip sequence home/ shopping/work is regarded as a home-work chain. The survey analysis underlined a considerable regularity in the generation of these trip chains by individuals. In particular, for work, we note the progressive passage to "the continuous day" (no more return home for lunch) between 1976 and 1995: from then on, on average, one home-work trip chain is carried out daily by each working individual. This stability of the trip chain generation by main purpose and area, is used as a basis for the generation model, while the variability of non home based trips is taken into account by the relation between overall mobility and income growth. The second modification is the step by step operation, which makes it possible to simulate and test the behaviour of the modeled system over the course of time, as well as dealing with feedbacks. The third modification is that of an architecture modulated (specified) according to travel purpose. This modulation consists in dealing with feedbacks between stages in a different way according to the travel purpose. This allows consideration for inertia and rigidities in spatial and temporal behaviour according to the activity type. Step by Step Operation Feed-Back Effects. Instead of directly projecting the calculated state of the system over the horizon year (2005), the model calculates successively the state of the system year after year, starting from a balanced situation between supply and demand for the base year (1995). At each annual step, the travel flows (generation, distribution and modal split) are determined by the socioeconomic and demographic context of the current year, and by the transport conditions - time and costs - of previous years. The network loadings which


511

follow determine the transport conditions of the current year, and are used to calculate the state of the system for the following years. Three reasons underscore this choice: • The first relates to the internal coherence of the model: coherence between supply and demand is ensured in a process of quasi-dynamic equilibrium. It is thus not a static equilibrium of the transport system which is required, but rather a coherent evolution year after year; •

•

The second is to take into account inertia in behavioural adaptation and change (Goodwin, 1988); thus a degradation of the travel conditions by car on a given link, will not produce immediately and for all the activities, a change of mode or destination. The inertia of the response depends on the activity type and the location or schedule constraints. Thus the architecture of the model makes it possible to differentiate the temporal pace of feedback on the distribution and the modal split, according to trip/activity purpose; The third reason rests on the interest to consider the interactions between the temporal development pace of the socio-economic context and the effective implementation of transport policies over several years.

Breakdown by Purpose. This breakdown by purpose arises from the need to take into consideration on the one hand the determinants, and on the other hand the inertia and rigidities in behaviour, which are different according to activity type: • the home-work chains: these are a ftmction of the working population and employment locations, characterized by strong rigidity; • the home-education chains, fiirther categorized into primary, secondary and higher education: these are a fiinction of the population and school establishment locations, as well as of differentiated transport conditions; they are attributed to a degree of rigidity similar to that of work; • the home-shopping or personal, business chains: these are a fiinction of the population and commercial facility locations, with a degree of rigidity quite less than for work; the individual choice of a shopping place is relatively freer and is conditioned sometimes on the modal split (one chooses initially the car then the shopping place). The model architecture (Figure 1) allows calibration of the different sub-models according to purpose, and implementation of different feedback effects according to purpose.

512


r Work J Generation

i5

i

T Distribution

D

T

1

1

i M

Modal split

•

^

Assignment

Figure 1 The Model Architecture The Sub-Models. The model architecture and the flexibility of its implementation in a spreadsheet allow subsequent modification of the various sub-models. The short description of these sub-models in their current version (SEMALY-LET, 1997a, 1997b) makes it possible to distinguish the main parameters which are sources of error and of potential uncertainties. The generation stage consists in calculating first the total trips, for all purposes and modes combined, based on a relation between average daily individual mobility and average income. This relation was established and validated over ten years for the Lyon agglomeration (Bonnafous and Tabourin, 1995). Concurrently to this volume of trips, the productions and attractions of trip chains by area are calculated for each of the main purposes: work, education according to three levels, shopping, other purposes. The calculation relies on generation coefficients. Through analysis of the three surveys, these coefficients exhibited convergence over time, and remarkable stability from one area to another. The production and attraction factors vary according to purpose: total population and working one, school populations, employment by type, places of education, etc. The difference between total trips and trip chains for work, education and shopping, makes it possible to correct the production and attractions of trips for the other purposes and non home based trips. These non home based trips amount to 21% of total trips in 1995.


513

In the stages of distribution, modal split and assignment, generalized travel times used are based on the following description of the supply, at the area level. The public transport supply is represented through the existing direct connections from area to area: it takes account of the average access time to the network in the origin area, estimated according to the network density, the waiting time, the time of travel and the average egress time to the final destination. Travel times for all the origin-destinations are obtained by a shortest path algorithm. The various components of the travel time are weighted to reflect their perception by the user, resulting in a generalized time. Coefficients of perceived disutility of access to the network, and disutility of waiting, as well as preference constants, expressed as differentiated access times added to the various modes. The values of these parameters are based on the calibration of a public transport network assignment model, validated for twenty years on the Lyon agglomeration. The road supply is also represented in the form of an overall supply from area to area. The supply is described by the overall hourly capacity from area to area, the free-flow speed and the distance between areas. In the absence of finer data, the parking difficulty is represented by an indicator of urban density (population and employment) which makes it possible to improve the modal split models. This indicator represents at the same time the average search time for a parking place and the disutility associated with it, and is transformed into a generalized time. On the whole the generalized travel time by car includes the travel time, the search time for parking, its possible cost and the operating cost related to the distance covered. The costs are transformed into time by means of a standard value of time (70 FF/h 1995). The distribution stage is carried out for each purpose separately. It relies on a conventional gravity model where flows from area to area are initially calculated by m^

r>m

jm

I

ij

Tf = E;"AJ expl - y^\

where

TjJ' is the number of trips from area / to areay for purpose m E^ is the number of trips produced by area / for purpose m AJ is the number of trips attracted by areay for purpose m tîs the generalized multimode time (better time) to go from / to J T^ is the impedance coefficient of for purpose m. T^ are calculated from T]J' at the end of an iterative Fratar process, so as to adjust itself with the margins in productions and attractions. This adjustment is carried out at the time of

514


prediction, so as to avoid the use of pre-established balancing factors: we know that such factors are probably not stable over time (Dufftis et al, 1987). For the purposes of work, higher education, shopping and other purposes, the impedances increased regularly over the three surveys. To take account of these developments, these impedances were connected to the development of overall car-ownership level during the three surveys, according to a linear relation. This translates the fact that under unchanged activity location and transport conditions, longer trips are taken when car-ownership level increase. The trip distribution is left unconstrained, in the sense that a change in the transport conditions between areas will cause changes in the flow between areas, under the constraint of the total productions and attractions by area. However, to take account of inertia in the change of behaviour, the average trip duration of the two previous years is taken into account in the calculation of T^J". For travel purposes that have essentially fixed locations, like work and education, the average trip duration of the five previous years is used in the calculation. The modal split is broken up into two hierarchical steps: the first one consists in determining the "light modes" share (walking primarily and bicycle), before proceeding to the division between car and public transport. These trips using light modes proceed almost exclusively inside each of the 25 areas. Their volume has strongly decreased during the last twenty years. The share of these light modes decreases with the surface of the area and the car-ownership level. This share is calibrated by the fiinction ^ ^ r = -^{^^Vi-Kôtor.

) + c^)

where

MZf is the modal share of light modes for the purpose m in area /, a^,b^, câre parameters estimated for the purpose m, Sj is the surface of area /, motor^ is the car-ownership level of area /. Once trips by light modes are obtained from the total number of trips inside each area, a split is operated between the motorized modes (car and public transport) on the T.J' matrix. For the primary and secondary education purposes, which involve short distance trips and the school transport supply, a constant split between public transport and car (as passenger) is applied. For the other purposes (work, higher education, shopping and other purposes) the model used is of the logit form:


ttc.j

Tc; = 1 + exp K+- Ttc/

tvp.j

dj

where

Tvp^motor. 5^

^ motor.

515

Jj

TC^. is the modal share of public transport between areas / andy for purpose m, k^, Ttc^, zv/7^, ^âre parameters of the model for purpose m, ttc-j is the generalized travel time by public transport from / to J, tvpjj is the generalized travel time by car from / toy, dj is a density parameter representing the parking difficulty, motor^ is the car-ownership level of area /. Zonal average data are used in the above model, contributing an additional source of error if one is interested in the disaggregated choice probabilities (Horowitz, 1981). The conversion to peak hour values entails converting daily flows of individuals by car, into vehicle flows during the morning peak hour, according to implicit vehicle occupancy rates. Specific coefficients, calculated in 1995, are implemented for the home-work chains: these corresponding to a spatial scheme with four concentric areas. For all other purposes, two different coefficients are implemented, one for trips inside the area, the other for trips coming out of the area. The assignment is finally carried out by combining trips for all the purposes, in addition to the external exchange and through traffic for the agglomeration. The latter are estimated exogenously, based on screen-line surveys conductd in 1979 and 1990. These traffics grew of 5% a year on average at the peak hour between 1979 and 1990. The assignment procedure follows a conventional iterative approach with previously calibrated link performance ftmctions.

EXPLORATION OF THE SOURCES OF UNCERTAINTIES IN THE FORECAST This model can thus be implemented to evaluate the various sources of uncertainties noted in the introduction. After describing the various categories of uncertainties and the associated method of measurement, we give the results of the various tests (experiments).

516


Various Categories of Errors and Uncertainties Tested We again make a distinction according to three sources of uncertainties presented in the introduction: • • •

Source of uncertainty of type I: estimation errors of the model; Source of uncertainty of type II: prediction errors of the exogenous context (the input variables); Source of uncertainty of type III: errors of projection in the future of the current behavioural mechanisms;

The experimental approach consists of changing in a controlled way the parameters or the exogenous inputs to the model, and evaluating the impacts of these changes relative to two reference points: • The situation calibrated by the model in 1995, named "zero point" or base case: in this case this situation is identical to the situation into 2005 as all parameters of the model are held constant; • A reference situation in 2005 corresponding to a "do-nothing" scenario: this reference scenario is a trend line projection of all the exogenous entries (population, employment, car-ownership level, income), without corresponding action in the transport supply. As the impact of the changes is measured in terms of the output produced by the entire modeling procedure, the resulting error is an error propagated by the model and not only the error attached to any particular sub-model. The following sources of type II uncertainty are analyzed: • The reference or "do-nothing" scenario, already noted; • The exchange (external-internal) and through traffic which reflects the general economic situation; • The population and employment, which reflect the agglomeration's own demographic and urban dynamics; • The incomes, the car-ownership level and the value of time (at the same pace as income), which reflect the development of the households' wealth; • The road supply programmed over the next ten years, i.e. primarily a tolled by-pass of 11km length, brought into service gradually in 1997. In practice, the differences between the three types of uncertainty sources are not so strongly contrasted, as discussed hereafter.


517

The Measurement of these Errors or Uncertainties The principal indicators considered during the evaluation of the model results are primarily the impacts on the modal share of public transport (effectiveness of the transport policy), the total distance traveled by car (environmental aspect), and the flows from area to area (spatial and social balance of the agglomeration). Spatially, the results are aggregated into three concentric areas: Lyon-Villeurbanne (center of the agglomeration, 500.000 inhabitants approximately), the 1st suburb and the 2nd outer suburb, each corresponding to very different degrees of urbanization. Measurement of the gap between the zero point (situation 1995) and the state of the system resulting from each test is made using a Chi2 distance. This is a robust criterion that makes it possible to take account of the difference between two flows in a given origin-destination pair, while normalizing each one of these difference by initial flow (zero point). The measure is computed to compare the OD matrix obtained under a given scenario to the OD matrix corresponding to the zero point (base case). The measure is computed for the matrix with all modes combined, and for each of the* three modesseperately: public transport, car and light modes. We give as an example the values of Chi2 calculated for the reference scenario (Table 1). Table 1 Chi2 Distances from Zero Point for Reference Scenario

Reference

All modes

public transport

car

light modes

total veh-km in morning peak hour (basis 100 in 1995)

41,084

2,684

57,505

12,525

128

|

Results To facilitate comparison across scenarios, the Chi2 values are given as a percentage relative to the reference scenario, for each mode. The distance covered in vehicle-kilometers is expressed as a fraction of the gap between the zero point (100) and the reference scenario(128). Results are given in Table 2. The modal shares are established, in 1995, at 57% for the car, 14% for public transport and 29%) for the light modes. The Exogenous Context. The impacts of the exogenous context are analyzed by carrying out separate tests of the impact of each exogenous entry, under the "do-nothing" scenario.

518


The exchange and through traffic increase under the "do-nothing" scenario (test El) resulting in a gap value that is about 50% of the gap observed between the zero point and the reference scenario, in terms of flow of all modes (46) and of vehicle-kilometers traveled (57). Population and employment (test E2) also produce significant gaps relative to the zero point, but rather small in terms of vehicle-kilometers traveled (11). The impact of incomes, value of time, and car-ownership level (test E3) produces important gaps for flows in public transport (116) and light modes (148). Car-ownership level by itself (test E5) also produces very important gaps for public transport (187) and the light modes (160). On the other hand, the road supply programmed from here 2005 does not generate significant gaps relative to the zero point (test E4). Finally decreasing the value of time by 50% (test E6) produces important gaps for public transport (258), while increasing it by 30% approximately (test E7) produces marginal gaps relative to the zero point. On the whole, one will note the particularly important impacts of (a) exchange and through traffic on the flows for all modes and the vehicle-kilometers traveled, and (b) car-ownership level, income and value of time, on public transport and the light modes. The Generation. The introduction of the higher value of the prediction error (with 95% confidence) on average mobility (i.e. +0.27, test G2), yields a gap to the zero point that is nearly half of that of the reference scenario (all modes, 52). The combination of this prediction error with the "do-nothing" growth of the income, car-ownership level and value of time (test Gl), leads to a gap to the zero point that is greater than that of the reference scenario. This is true for the OD matrices of all modes (103) but particularly for OD matrix for car (167). The lower value of the prediction error on average mobility (-0.27, test G3) leads to results similar to those of G2. Uncertainty on the forecast of average mobility can be interpreted equally as a type I uncertainty (estimation error in the prediction) or as a type III uncertainty (future shift of the relation mobility-income). On the whole, it plays an important part in the uncertainty of the final result. In contrast, the introduction of variation in the generation coefficients, according to high and low values of the prediction confidence intervals of these coefficients, produces only negligible variations compared to the zero point (tests G4 to Gl 1).


519

The Distribution. As the impedances of the distribution models are related to the carownership level, the impact of a de-coupling between the car-ownership level and the impedances was tested: the impedances fixed at their 1995 values, while the car-ownership level is assumed to evolve as the "do-nothing" (test Dl). The gap for flows in public transport is important (340) if one compares it with test E5 where only the car-ownership level evolves according to the "do-nothing" scenario (187). By maintaining the impedances constant, one cancels the direct impact of the car-ownership level on the spatial distribution of trips, which results in important distortions in public transport flows. The Modal Split. The tests of the modal split are concerned with, on the one hand the estimate of the light modes share, on the other hand the modal split between car and public transport. The tests of the light modes share consisted in varying the values of the parameters a, b, c between the extreme values estimated for each trip purpose. The choice of average identical values for all purposes (test Ml) produces small gaps for flows in all modes and for the vehicle-kilometers traveled, but high for public transport (103) and the light modes (106). The choice of high identical values for all purposes (test M2) produces even more significant gaps, in particularly for the light modes (414). The estimate of the light mode share can thus be prone to important errors. The tests of the modal split between car and public transport consisted in fixing the modal shares from area to area at their 1995 values instead of letting them evolve as in the logit model. Compared to the reference scenario, test M3 primarily produces differences for public transport (233) compared to the car (70). Compared to the test E5 of the car-ownership level, test M4 produces gaps twice less significant for public transport (93 instead of 187). The car-ownership level thus plays a critical part in the modal split between car and public transport. All indicates that this role must be maintained and that its intervention in the modal split specification requires considerable vigilance. Conversion to the Peak Hour. The tests of the peak coefficients consisted in exploiting the extreme values measured for the work purpose and the other purposes: in 1995 these coefficients varied between 0.24 and 0.44 for work, by origin-destination, and are worth 0.07 in intra-zone and 0.04 in extra-zone for the other purposes. The first test (PI) corresponds to a stronger concentration of trips in the morning peak hour, and produces significant gaps to the zero point for all the modes. It also implies a significant increase in the vehicle-kilometers traveled: the congestion of the roadway system involves route reassignment on the network.

520


therefore larger distances traveled. The second test (P2) corresponds to a spreading out of the peak and also produces significant gaps to the zero point. The variation of these peak coefficients is thus likely to produce significant impacts. A spreading out of the peak seems however most probable, if recent trends persist. These peak coefficients are to be considered here as a source of uncertainty in the exogenous context (category 11), through the flexibility of work schedules in particular. The Assignment and the Transport Supply. Two series of tests were carried out, one relating to the generalized travel times in public transport, the other to the assignment on the road network. The coefficient of difficulty of access time to the public transport network was set to 1 instead of 2 (test Al), resulting in large gaps for the light modes (183). The coefficient of disutility of the waiting time was reset to 1 instead of 1.8 (test A2), support and the preference constants were standardized to 0 minutes (test A3). In both cases public transport improved somewhat but the impact on the other modes was quite limited. These experiments illustrate the sensitivity of flows by public transport to the values of these parameters; their importance is crucial to evaluate long-term strategies aimed at improving service quality of public transport (access, waiting, connections), rather than their pure speed. The second series of tests addressed the effect of the number of iterations in the assignment procedure (impact of convergence) and the form of the speed-flow curve. The iteration count (with 5 being the default value) does not have any impact on the flows (tests A5 and A6). The form of speed-flow curve which relates the ratio of speed to the freeflow speed to the load (volume) to capacity (V/C) ratio, was changed in various ways. The curve is initially insensitive to congestion (flat), and then starts to increase when the V/C ration reaches 0.5. Test A7 consists in beginning the sensitivity to flow at a V/C of 0.25. This modification produces significant gaps for each of the modes, but not for the vehiclekilometers traveled, reflecting a shift between modes. Test A8 consists in lowering the minimum speed, attained for a V/C of 1.0, from two tenths of the free-flow speed to one tenth of that value. This test did not produce significant gaps. Test A9 consists of a combination of the A7 and A8 tests and produces very important gaps for flows for all modes (187) as well as for each mode. It also produces a reduction in the vehicle-kilometers traveled (-39), reflecting the impact of congestion on the modal shift.


521

These tests show that while convergence of the assignment algorithm does not seem to have important impacts, the speed-flow curve plays a critical part.

CONCLUSION The balance-sheet of the most important uncertainty sources is as follows: • Type I uncertainties (estimation errors): in the road assignment, the form of the speedflow curve can have a particularly strong impact on flows by car, and, indirectly, on the modal split and the distribution; • Type II uncertainties (prediction errors of the context): these uncertainties relate to the majority of the exogenous parameters, i.e. the exchange and through traffic, the working and school populations, employment, the incomes and the car-ownership level, and also the temporal rhythm of activities (peak hours); • Type III uncertainties (errors of projection in the ftiture of current behavioural mechanisms): these are primarily the coefficients of the generalized travel time by public transport. Although the adopted standard values were validated, one cannot rule out changes in the future values of these coefficients, which introduces an important uncertainty on flows by public transport; • Uncertainties of type I and III: these affect the link between mobility and income, the relation between independence and car-ownership level, as well as the estimate of the light modes share and the relation between car-ownership level and modal split. The main uncertainty of type I, here the form of the speed-flow curve, requires more thorough investigation to assess its robustness. Uncertainties exclusively of types II or III, can be controlled only by a strategy of "pragmatic control of uncertainty." By this expression we understand the implementation of multiple simulations, and testing a given transport policy under several alternative exogenous contexts. It is the only means of evaluating the extent of the risks which must be assumed if a given transport policy is implemented. Uncertainties of types I and III can be the target of concurrent approaches, either by trying to refine the current models, or by adopting the strategy of controlling of uncertainty mentioned above. This alternative can be illustrated with the example of the coupling between carownership level, mobility and car use. The past trend, reflected in the specification and the calibration of the various sub-models, is that of a direct impact of the car-ownership level on

522


car use and travel intensity. However, one cannot exclude, in an immediate future, that transport demand management policies could cause a de-coupling between car-ownership level and intensity of car use, by analogy with the de-coupling observed between GDP and energy consumption. However nothing indicates that the current models founded on the revealed preferences or on the surveys of stated adaptations, are ready to provide precise estimates of the future behavioural mechanisms. The only plausible strategy thus seems to us one of exploration of these uncertainties, by testing several possible specifications of the coupling or de-coupling between car-ownership level and car use, along the lines described in the previous tests.

H

^

P!5

.d 'u

s

"S

k) p .s

'o

B

V w

« B •7 .B

>

S 0

a .2 >-3 o 'C CQ c« ;.

u N

u u II

^ «2

«« H i ^ ^ u

V S C9

15 0

s •**

4>

CI.

03 U

U

en

03

.0

,2

"^ ^0

^(» s ^

®

ô •3 fS

is *33

IS S •s 1 B

s >>es 4^

s

1 '^^ U

^ s M)

bx)

-^ a •S

0

0 0 1—1

0 0

0^
1 >

0

II

0 0

Pi

CM CD

l^ (N

Tf Tf

jo_

0

0

0

1

0

0

1

0

0

1

0

0

1

•0 'O

0

'^

CM

r j 1 04 04

f^

^

1 1

0

0 ^ 1

'—1 1• r j

rf

0

y—*

m

0

0

0

0

CM

CO ra 0 '—1 1 t n

»n

,—1

00

*n «n

p

Op

's

ti

0

. ^0 0

t^

0

rf

^ ^

,_H

T3

-. ^^

11

T3

^

1

0 03

-a

tn

'-^ s -0 «U

— ' *S T3 ^ ^ ;U->

.2 ^ 0 'S "S

0 0

11

U

0\ 0

t03i

0

0

0

0

^.^^^ ^ .2

cx

^ ^ 0

CD

c 0

t:

t!. 0

00

0 sZ

.2 0

op

CM

0-

r4

0

+ +

0

XI

1 _s_

x0

>^ >. >-» .2 "S jd 0

S

>

U+H 0

Is

(LT

03 0

0

c

a.

k
^

e

•a

^

W5

fi ,W)

Ti-

"—I

00

m

"^ 'O

O

— ' ' — ' ' (N

OO (N

m TT

O

o

CO

'^

gap + 1), then speed is reduced to gap 3) Randomization (which is applied after rule 1 or 2): With probability p, the velocity of each vehicle (if greater than zero) is decreased by one. • Mixed micro- and macroscopic models, which maintain individual driver-vehicle-units, but use aggregate relationships for the calculation of speeds, such as DYNEMO (Schwerdtfeger, 1984) or Integration (Bacon, Lovell, May and Van-Aerde, 1994). A further alternative is discrete event frameworks, which calculate variable time-steps until the next change in behaviour (e.g. Taale and Middelham, 1997). It should be noted, that the last two alternatives do not allow meaningful disaggregate analyses of driver behaviour or interactions, as their base models are either too crude (cellular automata) or not disaggregate (mixed models). Certain other aspects, for example route choice if included in the model, can be meaningfully traced and analyzed on an individual level. The models of lane-choice and overtaking are rule-bases of differing complexity, depending on context and degree-of-realism sought. Further rule-bases to describe the interaction with traffic control and traffic information are easily implemented, but generally difficult to validate. The available commercial models (see Algers et al., 1997) are intended for smallish networks with a concurrent number of simulated vehicles in the low thousands, which is completely suitable for the normal tasks to which they are put: analysis of specific freeway sections or parts of signalized urban networks. In both cases the work involved in the aggregate validation of the model and the acquisition of the detailed description of networks and flows prohibits larger networks. Due to the expensive specialized data required a new validation of the disaggregate rules and assumptions is hardly ever undertaken during an application. The default parameters are accepted and may be fine tuned. The inclusion of complex rule-bases for interactions with control and information systems makes a disaggregate validation even less likely in a typical application.

Microsimulation: Workshop Report

587

The commercial models respond in their development to the pressures of their market place, which currently is mostly interested in the simulation of various high-tech driver support, traffic control or traveller information systems. These development enrich the range of applicability of the models, but do not challenge the basic framework set out above. The rather non-commercial work on pedestrians and bicyclist and traffic flow in less-developed countries has shown that the lane- and immediate vicinity orientation of nearly all current models also limits the description of the driver-vehicle-unit. The incorporation of more explicit models of the exact choice of lateral position and of the strategic and tactical driving decisions can expand the behavioural core of the models, while requiring a much richer description of the drivers and their plans in comparison to the currently blindly responding simulated drivers. Still, this expansion of behavioural complexity could until now not be justified for real-life applications, although congested urban networks with interactions between car, trucks, trams, busses, cyclists and pedestrians might do so in the future. The second major development direction of the current model generation is their transfer to large networks with 100,000's of concurrently simulated driver-vehicle units for both planning and real-time control applications: simulation of a whole urban area including detailed simulation of emissions production and distribution (e.g. TRANSIMS (Barret et al., 1995) or DYNEMO (SIMTRAP, 1997)), simulation of complete motorways networks of large-scale regions (e.g. the Ruhr-Region or Scottish Midlands) (PLANSIM-T (in Algers et al., 1997) or PARAMICS (McArthur, 1995), or the optimization of traffic control for large networks using simulation to evaluate different control strategies in real-time. A variety of directions are currently pursued to achieve the necessary computational speeds and behavioural realism. Parallel computers are one approach of speeding up the calculation. The more important approaches concern the reformulation of the model to simplify the conceptualisation of the driver-vehicle units and their interactions: e.g. cellular automata or mixed microscopic/macroscopic models. The relative success of this microsimulation tradition is based on the relative simplicity of modelling task at hand, the lack of credible alternative tools, which could address the same questions and the relatively simple starting conditions and relatively low validation requirements'. The progress in this area is important, as fully fledged microsimulation models of travel behaviour require a model of traffic flow to ensure consistency between the time-space regime and the experiences of the individual travellers and to allow for time-space specific interactions, such as parking search or response to traveller information systems.

588


ISSUES AND CHALLENGES While the implementation of microsimulation models of travel behaviour is making progress on the well established basis of consumer choice theory and its extension to dynamic processes, as well as on the new results about network learning, there are a range of issues, as identified by the workshop, which need attention. Microsimulation models of travel behaviour have ambitious scales in time and space and scope in their coverage of human choices. Next to the practical problems of the large amounts of computing time required and file storage required for the intermediate outputs, the main conceptual problem for the user is to maintain the understanding of how the model reacts, in particular for the various nested rule structures, which make this rather difficult. In the face of this challenge the workshop suggested to separate the definition of the "agents" and of their "processes" as strictly as possible and to document the model and the application as comprehensively as possible. A single run of a microsimulation model of travel behaviour is a challenge in itself, especially as the run needs to be prefaced with the creation of a consistent starting solution against which any policy might be tested. A bigger challenge is the necessary and proper experimentation to remove the effects of the random-number seeds, which are known to have potentially substantial impacts in microsimulation models of all kinds. Researchers in the field will have to adopt best practises in experimental design, if they want to keep the effort manageable and if they want to obtain valid results. Given these difficulties the workshop concluded that microsimulation models should be applied only, if their specific strength are called for: description of system evolution over time, modelling of interactions in time and space, complex non-linear decision rules, nonequilibrium situations.

RESEARCH DIRECTIONS The workshop was not able to cover the whole range of possible issues in its definition of research directions, but could only focus on a small number. The following were identified: Development of tools for the construction of consistent scenarios out of individual - not necessarily - internally consistent building blocks

Micwsimulation: Workshop Report

• •

589

Execution of proper experiments with the scenarios, i.e. construction of experimental designs, conduct of sufficient simulation runs, extractions of required results from each run and proper statistical meta-analysis (response surface regressions etc.) Correct integration of flow simulation and higher level traveller choices in the context of traffic control, roadside information and in-vehicle and pre-trip information systems. Formulation of models of environmental learning (updating of mental maps) and of the process of abstracting by which travellers generalize from the specific to the general, e.g. how experiences become general expectations about the performance of types of environments and services. Experimentation with the formulation of discrete rule systems and their comparison with choice models or non-linear classifiers, such as neural networks or fuzzy neural networks. Analysis of the meaning and identification of equilibrium and steady state for such models and of the paths they describe (true system evolution vs. iterative convergence) Analysis of the general stability of such systems and of the effects of the dimensionality of the choices considered and described.

The number and difficulty of the research problems indicates that large-scale application of microsimulation models as a matter of fact is some way in the future. Still, the workshop felt that even now microsimulation models offered the best tool available to test and explore research hypotheses through their formulation as rules and their animation in simulation models (see also Axhausen, 1991).

ACKNOWLEDGEMENTS Participants included: Akmal Abdelfatah, M. C. J. Bliemer, J. Bates, R. Chapleau, U. Crisalli, Y. Hawas, T.-Y. Hu, P. Koskenoja, R. Machemehl, E. Miller, O. Nielsen, R. Pendyala, M. Rrepanier, and D. Watling.

REFERENCES Algers, S., E. Bemauer, M. Boero, L. Breherent, C. Di Taranto, M. Dougherty, K. Fox and J-F. Gabard (1997). Review of micro-simulation models, SMARTEST Deliverable, D3, Institute for Transport Studies, University of Leeds, Leeds. Axhausen, K.W. (1990). A Simultaneous Simulation of Activity Chains, in P.M. Jones (Ed), New Approaches in Dynamic and Activity-based Approaches to Travel Analysis, 206225, Avebury, Aldershot.

590


Axhausen, K.W. (1991). The role of computer-generated role-playing in travel behaviour analysis, paper presented at the 70^^ Transportation Research Board Meeting, Washington, D.C. Axhausen, K.W. and T. Garling (1992). Activity based approaches to travel analysis: Conceptual frameworks, models and research problems, Transport Reviews, 12 (4) 323-341. Axhausen, K.W. and P.B. Goodwin (1991). EUROTOPP: Towards a dynamic and activitybased modelling framework, in Advanced Telematics in Road Transport, 1020-1039, Elsevier, Amsterdam. Axhausen, K.W. and R. Herz (1989). Simulating Activity Chains: A German Approach, ASCE Journal of Transportation Engineering, 115 (3) 316-325. Bacon, V. W. Jr., D. J. Lovell, A. D. May and M. Van-Aerde (1994). Use of Integration model to study high-occupancy-vehicle facilities. Transportation Research Record, 1446, 8-13. Barcelo, J. and J. L. Ferrer (1994) Microscopic simulation of vehicle guidance systems with AIMSIJN2, paper presented at the 23rd Euro-Conference, Glasgow. Barrett, C, K. Berkbigler, L. Smith, V. Loose, R. Beckmann. J. Davis, D. Roberts and M. Williams (1995). An operational description of TRANSIMS, Working Paper, LA-UR-952393, Los Alamos National Laboratories, Los Alamos. Benz, T. (1994). Traffic flow effects of intelligent vehicles, in Traffic Technology International 1994, 118-120, UK and International Press, Dorking. Botha, J. L., X. Zeng and E. C. Sullivan (1993). Comparison of performance of TWOPAS and TRARR models when simulating traffic on two-lane highways with low design speeds, Transportation Research Record, 1398, 7-16. Brilon, W. (1976). Warteschlangenmodell des Verkehrsablaufs auf zweispurigen LandstraBen, Schriftenreihe Strafienbau und Verkehrstechnik, 201, Bundesministerium fur Verkehr, Bonn. Buck, A. (1992). Simulation von Fundamentaldiagrammen bei Eisenbahnen, dissertation, Universitat Karlsruhe, Karlsruhe. Dixon, K. K., A. Lorschneider and J. E. Hummer (1995). Computer-simulation of 1-95 lane closures using FRESIM, Proceedings of the 65th Annual Meeting, ITE, Washington, D.C. Emmerink, R. H. M., K. W. Axhausen, P. Nijkamp and P. Rietveld (1995). Effects of information in road transport networks with recurrent congestion. Transportation, 22 (1) 2153. Fellendorf, M. (1993). Beurteilung des innerstadtischen Verkehrsgeschehens mittels Simulation, Fortschritte in der Simulationstechnik, 6, Vieweg-Verlag, Wiesbaden. Fellendorf, M. (1997). Public Transport Priority within SCATS - a simulation case study in Dublin, In: Compedium of the 67 Annual ITE Meeting, Boston, August 1997 (CD-ROM) Fellendorf, M., T. Haupt, U. Heidi and W. Scherr (1997). PTV Vision: Activity-based demand forecasting in daily practice, in D. Ettema and H. Timmermans (eds.) Activity-based approaches to travel analysis, 55-72, Pergamon, Oxford. Fellendorf, M., C. Mac Aongusa and M. Pierre (1997). LRT priority within the SCATS environment in Dublin - a traffic flow simulation study, in L. Sucharov and G. Bidini (eds.) Urban Transport and the Environment for the 21st Century, 53-62, Computational Mechanics Publications, Southampton. Gipps, P. G. (1981). A behavioural car-following model for computer simulation. Transportation Research, 15B (2) 105-111.

Microsimulation: Workshop Report

591

Gipps, P. G. (1987). Simulation of pedestrian traffic in buildings, Schriftenreihe, 35, Institut flir Verkehrswesen, Universitat Karlsruhe, Karlsruhe. Gunn, H., T. van der Hoom and A. Daly (1989). Long-range, country-wide travel demand forecasts from models of individual choice, in lATBR (ed.) Travel Behaviour Research - Proceedings of the Fifth International Conference on Travel Behaviour, 119-137, Gower, Aldershot. Hesselfeld, M., V. Kahn and K. Weisbrich (1997). Computer-aided planning of rural train runs in Lower Saxony, in L Sucharov and G. Bidini (eds.) Urban Transport and the Environment for the 21st Century, I'^-'il, Computational Mechanics Publications, Southampton. Hu, T. Y. and H. S. Mahmassani (1995). Evolution of network flows under real-time information: Day-to-day dynamic simulation assignment framework. Transportation Research Record, 1493 46-56. Khasnabis, S., R. R. Kamati and R. K. Rudraraju (1996). NETSIM-based approach to evaluation of bus preemption strategies. Transportation Research Record, 1554, 80-89 Leutzbach, W. (1988). Introduction to the Theory of Traffic Flow, Springer, Heidelberg. Leutzbach, W., U. Brannolte and M. Schmidt (1989). Untersuchung des Verkehrsablaufs auf einbahnigen Strassen unter besonderer Berucksichtigung langsamer Fahrzeuge, Schriftenreihe Forschung Strassenbau und Strassenverkehrstechnik, 551, Bundesministerium ftir Verkehr, Bonn. Leutzbach, W. and R. Wiedemann (1986). Development and applications of traffic simulation models at the Karlsruhe Institut ftir Verkehrswesen, Traffic Engineering and Control, 27(5)270-278. Liu, R., D. van Vliet and D. P. Watling (1995). DRACULA: Dynamic route assignment combining user learning and micro-simulation. Proceedings of the 23rd European Transport Forum, E, 143-152, PTRC, London. Mahmassani, H. S., T. Y. Hu and S. Peeta (1994). Microsimulation-based procedures for dynamic network traffic assignment". Proceedings of the 22nd European Transport Forum, H2, 53-64, PTRC, London. May, A. D. (1990). Traffic Flow Fundamentals, Prentice-Hall, Englewood Cliffs. McArthur, D. (1995). The PARAMICS-CM behavioural model. Proceedings of the 23rd European Transport Forum, E, 219-231, PTRC, London. Miller, E. J and P. A. Salvini (2001). Activity-based travel behavior modeling in a microsimulation framework. Chapter 26 in this volume. Nagel, K. and M. Schreckenberg (1992). A cellular automaton model for freeway traffic. Journal of Physical Institute of France, 2, 2221. Nagel, K. (1995). High-speed microsimulation of traffic flow, dissertation, Universitat zu Koln, Koln. Nokel, K. (1997). Specification of the Traffic Model, Deliverable D6 of the SIMTRAP-Project, PTV System, Karlsruhe. Nokel, K. and A. Rehkopf (1996). Betriebsfuehrung von Nahverkehrssystemen mit dem Simulationswerkzeug TRANSIT, SignaRDraht, 88 (3) 15-18. Okura, I. and K. Matsumoto (1990). Traffic simulation study for two-lane rural highway overtaking improvements. Proceedings of the 15th ARRB Conference, 15 (5) 43-56, Darwin, August. Pursula, M. and H. Siimes (1993). A simulation study of a high-class three-lane rural road, in Compendium of Technical Papers from ITE 63rd Annual Meeting, 16-20, ITE, Washington, D.C.

592


Reid, J. D., D. L. Sicking and G. W. Paulsen (1996). Design and analysis of approach terminal sections using simulation, Journal of Transportation Engineering, 122 (5) 399-405. Reiter, U. (1994). Simulation des Verkehrsablaufs mit individuellen Fahrbeeinflussungssystemen, dissertation, Universitat Karlsruhe, Karlsruhe. Reiter, U. (1997). Kapitel E: Verkehrsflussmodelle, in U. Brannolte and C. Theis (eds.) Situationsanalyse iiber den Stand der Simulationsmodelle im Verkehrswesen, report to the Bundesministerium fiir Bildung, Forschung, Technologic und Wissenschaft, Fakultat fiir Bauingenieurwesen, Bauhaus-Universitat, Weimar. Rickert, M., K. Nagel, M. Schreckenberg and A. Latour (1996). Two lane traffic simulations using cellular automata, PhysicaA, 231, 534. Schwerdtfeger, T. (1984). DYNEMO - A model for the simulation of traffic flow in motorway networks, in Proceedings of the 9th Symposium of Transportation and Traffic Flow Theory, 65-87, VNU Science Press, Utrecht. Taale, H. and Middelham, F. (1997). FLEXSYT II - A validated microscopic simulation tool. Proceedings of the 8th IFAC/IFIP/IFORS Symposium on Transportation Systems, 923928, Chania, June 1997. Van der Hoom, T. and H. van Hoek (1988). Translation of results from the Dutch national model into practical information for policy recommendations. Proceedings of the 16th PTRC Summer Annual Meeting, D, 165-182, PTRC, London. Venglar, S. P., D. B. Fambro and T. Bauer (1995). Validation of simulation software for modeling light rail transit, Transportation Research Record, 1494, 161-166. Wegener, M. (1994). Operational urban models: State of the art. Journal of the American Planning Association, 60 (1) 17-28. Wiedemann, R. (1991). Modelling RTI-elements on multi-lane roads, \n Advanced Telematics in Road Transport, 2, 1007-1019, Elsevir, Amsterdam. Wiedemann, R. and X. Zhang (1989). Verkehrssimulation mit Radfahrern, Final report to the Deutsche Forschungsgemeinschaft, Institut fiir Verkehrswesen, Universitat Karlsruhe, Karlsruhe. Zumkeller, D. (1989). Ein sozialokologisches Verkehrsmodell zur Simulation von MaBnahmewirkungen, Verqffentlichungen des Instituts fiir Stadtbauwesen, 46, Technische Universitat Braunsschweig, Braunsschweig.

' In many cases only the reproduction of known macroscopic speed-flow relationships

In Perpetual Motion

The Dream of Perpetual Motion

The Dream of Perpetual Motion

Perpetual War for Perpetual Peace

Perpetual war for perpetual peace

Perpetual War

E. coli in Motion

Blood in Motion

Storyboards: Motion in Art

Gender in Motion

The war over perpetual peace

Wave motion in elastic solids

Вечное движение: История одной навязчивой идеи. (Perpetual Motion: The History of an Obsession, 1977) . Сокращенный

Brownian motion

Brownian Motion

The Human Figure in Motion

Israeli Cinema: Identities in Motion

The Human Figure in Motion

Brownian motion

Illeagal Motion

Motion Mountain

Self-Motion

Perpetual Motion: Transforming Shapes in the Renaissance from da Vinci to Montaigne (Parallax: Re-visions of Culture and Society)

Wave Motion

Motion Mountain

Stefan Heym: The Perpetual Dissident (Cambridge Studies in German)

Large Amplitude Motion in Molecules II

In Perpetual Motion

The Dream of Perpetual Motion

The Dream of Perpetual Motion

Perpetual War for Perpetual Peace

Perpetual war for perpetual peace

Perpetual War

E. coli in Motion

Blood in Motion

Storyboards: Motion in Art

Gender in Motion

The war over perpetual peace

Wave motion in elastic solids

Вечное движение: История одной навязчивой идеи. (Perpetual Motion: The History of an Obsession, 1977) . Сокращенный

Brownian motion

Brownian Motion

The Human Figure in Motion

Israeli Cinema: Identities in Motion

The Human Figure in Motion

Brownian motion

Illeagal Motion

Motion Mountain

Self-Motion

Perpetual Motion: Transforming Shapes in the Renaissance from da Vinci to Montaigne (Parallax: Re-visions of Culture and Society)

Wave Motion

Motion Mountain

Stefan Heym: The Perpetual Dissident (Cambridge Studies in German)

Large Amplitude Motion in Molecules II

Recommend Documents