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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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 Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
This Page Intentionally Left Blank
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 Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Setting the Research Agenda: Response to New Transport Alternatives and Policies
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Setting the Research Agenda: Response to New Transport Alternatives and Policies
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Setting the Research Agenda: Response to New Transport Alternatives and Policies
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Setting the Research Agenda: Response to New Transport Alternatives and Policies
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Setting the Research Agenda: Response to New Transport Alternatives and Policies
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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).
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
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').
Living Models for Continuous Planning
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
Living Models for Continuous Planning
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
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
Living Models for Continuous Planning
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
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
Living Models for Continuous Planning
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
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.
Living Models for Continuous Planning
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
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
Living Models for Continuous Planning
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
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
Living Models for Continuous Planning
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
In Perpetual Motion: Travel Behaviour ResearchOpportunities and Challenges
•
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.
Living Models for Continuous Planning
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.
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Household Adaptations to New Personal Transport Options
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Household Adaptations to New Personal Transport Options
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.
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Workshop Report-Responses to New Transportation Alternatives
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Workshop Report-Responses to New Transportation Alternatives
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.
This Page Intentionally Left Blank
SECTION 2 DYNAMICS AND ITS RESPONSE
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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,
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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).
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to 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
Dynamics and ITS: Behavioural Responses to Information
(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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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-
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Dynamics and ITS: Behavioural Responses to Information
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 Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Dynamics and ITS: Behavioural Responses to Information
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Research into A TIS Behavioural Response: Areas of Interest and Future Perspectives
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Research into A TIS Behavioural Response: Areas of Interest and Future Perspectives
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Research into A TIS Behavioural Response: Areas of Interest and Future Perspectives
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Research into A TIS Behavioural Response: Areas of Interest and Future Perspectives
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.^
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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-
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
Emerging Travel Patterns: Do Telecommunications Make a Difference?
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.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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<)ur trio
« Next trip or contact
km
Figure 2 Part of the Notebook for the Intrapersonal Context Following the pilot survey, the questionnaire was modified for further surveys. The first dataset was collected in Germany, consisting of about 2,000 households and 4,400 individuals. These individuals reported roughly 16,000 trips and 8,000 contacts. The second data-set resulted from a survey performed in Seoul, Korea - a mega-city with a widely heterogeneous population of 11.5 million. Seoul is also a core development area in terms of the government-led efforts to build an information society. It boasts of high
188
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
penetration of various communication media and holds great potential for the spread of cutting-edge teleservices. The recruiting of survey participants was intended to obtain representative data from a variety of respondents. Proportional to the population of each district, a number of persons was selected randomly from the telephone directory in Seoul. Their willingness to take part in the survey was confirmed by telephone calls. Finally a questionnaire with a self-addressed envelope was sent to the 900 participating households. The return rate was 71.4%, which is equivalent to the travel and communication data of 644 households, 2,476 persons and 10,819 activities. For the analysis of the transition probability to teleservice, an additional subsample of interactive interviews was formed on the basis of the distribution of trip purposes for each person group. These interviews were carried out at places like banks and department stores with customers who had just finished their business being relevant in terms of potential transition to a teleactivity in order to maintain realism. This subsample was selected randomly based on quota planning considering key variables such as age, gender and occupation. A third data-set would result from a survey with a sample size of 2,500 individuals conducted by SIKA in Sweden in 1997 (Widlert, 1997).
FIRST RESULTS
Germany The use of modes and media. The analysis of the intrapersonal context starts with the examination of the present influence of media on transport behaviour. In Table 1, the frequency of use and the spatial field of activities given by the transport system and media are compared. The figures clearly indicate, that: • •
The frequency of using media is still lower than the use of transport modes, but this is due to a higher share of "immobile persons" (non-active relative to the medium) The spatial field of activities covered by media is already higher compared to trips.
Transport and Telecommunication: First Comprehensive Surveys
189
Table 1 The Use of Modes and Media per Person Groups Trips
person group share of immobile persons
1 employed high inc. 1 employed low inc. 1 student 1 housewife/ -man and jobless 1 retired all
Contacts
distance # of trips per person per person in terms of and day in km in * persons terms of * persons *— *— *= *_ all mo- all mobile bile
share of immobile persons
#of contacts per person in terms of * persons *= all
*=z
mobile
distance 1 per person and day in km in terms of * persons *= all
*= 1 mobile
5.0 %
4.7
4.9
46.3 48.7
24.1 %
5.5
7.3
211
278
7.9 %
4.3
4.6
32.5 35.5
30.8 %
3.3
4.8
56.6
82
4.1%
5.0
5.3
23.4 24.4
30.0 %
1.8
2.6
129
184
4.9 %
4.0
4.2
22.5 23.7
25.7 %
1.8
2.4
98.6
133
7.3 %
3.6
3.9
36.1 38.9
33.6%
1.4
2.1
104
157
5.7 %
4.4
4.6
32.0 33.9
29.0%
2.7
3.8
122
171
Thus, a first hint is given that media contribute much more to covering activities with higher distances than transport modes do. It has to be clarified whether this is the aggregate result of people compensating (substituting) between transport and communication or of individuals combining modes and media in order to increase their spatial field of activities. Interactions between transport and communication. When analyzing interactions between new telecommunication facilities and transport we must not forget that telecommunication has already been in existence for quite some time (letters, phones, faxes, television, interactive media etc.) and that it has already been influencing our transport behaviour in an unknown way. It is of great interest now whether new telecommunication devices affect our transport volume to a positive or negative extent. To answer this question, we look again at Table 1,
190
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
which contains information about the use of modes and media for different person groups. The figures distinctly show that people with high physical mobility tend to have a complementary high level of telecommunication use. Consequently, the complementary factor seems to be predominant over the substitutional one (Zumkeller, 1997). This is reflected by both the lower share of immobile persons and the average distance covered per day by the more active person groups. A glance at the combined patterns, Analysts of trip patterns know (Chlond, Lipps, Zumkeller, 1997) that the major part of all trips (80 - 90 %) can be modeled by only few basic patterns (10 - 20 %). For this reason, the frequency of contacts for the trip patterns already known was examined. Table 2 shows this relationship for employed persons categorized by their level of education. The results allow the following observations: • the combination of trips and contacts results in patterns of considerable complexity • there is neither indication that short trip patterns are linked with many contacts, nor of the reverse • a high level of contacts often occurs together with a high level of trips per day. This is valid for both the higher and the lower education levels.
1#
Table 2 Activity Patterns versus Frequency of Contacts employed (high education level) employed (low education level)
1 ^
length of trip pattern (# of trips)
share in%
# of contacts
length of trip pattern (# of trips)
share in%
#of contacts
2 3 4 5 6 7 8 9 11 12 14
13.3 13.3 25.0 13.3 13.3 10.0 3.3 3.3 1.7 1.7 1.7
0.9 0.6 2.3 1.4 3.1 0.3 0.0 3.0 0.0 4.0 1.0
2 3 4 5 6 7 8 9 12 13 14
17.4 13.0 17.4 4.3 17.4 4.3 8.7 8.7 4.3 4.3 0.0
1.0
1
1.0
1
5.0 0.0
1 1
6.0 1.0 6.0
1 1 1
1 ^ 1 ^ 1 ^ 1^ 1 ^ 1 ^ ' 1 ^ 1 ^ 1 ^^ 1 11
0.0 1.0
0.0
:
These observations provide more detailed insight into the manner in which patterns of trips and
191
Transport and Telecommunication: First Comprehensive Surveys
contacts are composed. Table 3 presents typical trip patterns of employed people, for which it is known that the ten most frequent patterns already cover 72% of the total (Zumkeller, 1993). As a result of the survey, roughly half of these patterns have been combined with contacts as shown in Table 3. These typical patterns exhibit the following features: • contacts are tied to very specific locations like home and workplaces • unlike trips, contacts occur not as sequences but as clusters depending on the communication facilities available at certain locations • contacts are apparently allocated more frequently to the later part of the day. They are more flexible in terms of "departure time", frequency, their composition and local destination, but not in terms of the location of their origin. Table 3 The Structure of Combined Trip/Communication Patterns
1 ^ 1Trip patterns
employed persons share of share of patterns all with contacts patterns
1 ^
H-W-H
38%
0.6
2
H-W-H-L-H
12%
0.5
1 ^
1 ^ 1 ^ 1 ^ 1^ '8
H-W-H-S-H H-W-H-W-H H-S-H H-W-S-H H-L-H H-S-H-L-H
6% 4% 3% 3% 3% 2%
9
H-W-H-S-H-L-H
1%
I 11^^1
H-W-W-W-H Remainder
1% 28%
typical patterns H-W-H-C
H-W-C-H-C
Legend:
H-home,
W-work,
0,4 0.5 0.4
j
1
H-W-C-H-L-H H-W-H-L-H-C same as 1
1
1
H-S-H-several c j same as 1 || H-L-H-several c 1 many patterns with series of 1 contacts at home and work place
0.4
S-shopping,
L-leisure,
C-communication
The spatial view. The cumulative frequency of distances per person-day (Figure 3) indicates that the range of activities covered by media is much wider than that of physical trips.
192
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Roughly, it can be stated that the distances covered by media are about three to four times of the distances covered physically.
cumdativefrequencyof distances per person - day 100% 90% 80% -
trips
70% -
^
60% -
" " contacts
a 40% -
_ ^ - ^ "
30% 20% 10% 0% -
/
/
/
/
/ y
— "^ <50
/
J
^ ^ ^
/
^ <100
<150
<200
<250
<300
<350
<«»
<450
^500
distance in km
Figure 3 Distances Covered by Modes and Media Can we conclude that telecommunication contributes to a higher level of spatial dispersion? To examine this suggestion, different spatial prototypes with respect to the level of concentration are defined as follows: • densely populated region • medium densely populated region • suburban region • rural region According to Figure 4, the number of trips is nearly independent of the spatial prototype and there is no discernible regularity concerning the number of contacts per spatial prototype.
Transport and Telecommunication: First Comprehensive Surveys
193
Number of trips and contacts by region 391
4 3,5 3 fe 2,5 E
2
3 = 1.5 1 0,5 0
ill
Li densely populated region
3,98
3,87 n number of trips per person per day
2,02
1ii_
® number of contacts per person per day
1,37
medium densely populated region
suburban region
rural region
type of region
Figure 4 Number of Trips and Contacts by Type of Region
distances of traffic and telecommunication by region type CI person distances travelled
170,4
180 n
^telecommunication distances
160 140 120 100 80 60 40 -
451 29,9
'^'^nM
20-
^Am hm
n d( se^ populated region
n edum densely populated region
u a region
type of region
Figure 5 Distances of Trips and Contacts by Type of Region However, considering the distances covered and the range of telecommunication, a dependency on spatial prototypes becomes apparent (see Figure 5). The distances covered by person and day are increasing with decreasing dwelling density by 50 percent, whereas the telecommunication distances are decreasing continuously from 170 km per day in densely populated regions to 95 km per day in rural regions. This manifests completely different individual combination strategies in the complementary usage of the two different "spaceovercoming instruments." Moreover, this is a hint for complex and heterogeneous influences
194
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
of both travel and telecommunication on spatial concentration and dispersion, so that further analyses are necessary (Mokhtarian, 1993).
Seoul (Korea) and Cross Cultural Comparison The results of the Seoul survey provide several options to deal with the following questions: • Are there any similarities in the transport and communication context between Korea and Germany? • Are there obvious explanations for these similarities and/or differences? • The penetration of society by telecommunication is a progressive process. Do the data indicate different degrees of penetration? Examining Figures 6 and 7 seperately, similar behavioural patterns appear regarding the level of trips and contacts as well as the distances covered by both systems. However, the connection of the two figures reveals a difference in the average distance covered in a single trip (Germany: 7.3 km per trip, Korea: 16.0 km per trip) and in the average distance associated with contact (Germany: 45.0 km per contact, Korea: 30.1 km per contact) suggesting that the use of the transport system in Germany is ecologically more efficient and that the use of the communication system is spatially more extended, which might be attributed to a higher degree of penetration.
trips and cent acts per person per day b -
4,4
^Germany
4
32
aSeou[
1 i 2,7
3 2 1
0
I
...ps
contacts
Figure 6 Tr ip and iContact Fr equency
Transport and Telecommunication: First Comprehensive Surveys
140 -
195
spatial field of activities in km per person per day 121,5
120 -
n Germany
100 -
^ Seoul
80 60 40 20 -
0
32
•1
•
66.3
51,2
[^i
travel-km
contact-km
Figure 7 Spatial Field of Activities in Km per Person and Day
On the other hand, Figure 8 does not support this interpretation since the distribution of different modes of communication is similar except for the rapidly developing market of faxmachines and mobile phones.
100% 80% ^ Email
60%
• Fax
40%
H Letter D Mobile phone
20%
H Pager m Telephone
0% Seoul
Germany
Figure 8 Distribution by Mode of Communication for Contacts To provide more insight into the telecommunication penetration process (Selvanathan and Selvanathan, 1994), a differentiated typology of person groups was developed on the basis of the Seoul data. Classified by position within life-cycle and level of education and employment, Figure 9 suggests nine person groups corresponding to the well-known transportation
196
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
typologies. Only the "male under 30" group constitutes an exception; it was set up separately due to its quite different communication patterns. Population > 13 year M
^ \
r
.
in Education
,„.„i„
Puj>i( ,.
„..
\_
| i.„..
i in Employment ^
College Student 1 „„„,
Male
1 1
L
under30
|
i
i
1
wh^ecoSar
i
I
i not in Employment
i
1
1
Female
: Housekeeper 1
1
i„
blue collar
i without (^lldrenl
i_
J
_
1 retired
..„_
with children |
Figure 9 Classification of Person Groups Table 4 provides more detailed information and shows noticeable differences between person groups in communication and mobility patterns. In terms of frequency, "males with white collar occupation", who use telephone and fax most often, contrast sharply with "the retired", "the unemployed" and the "full-time housekeepers", who are nearly abstinent from telecommunication. Concerning the use of e-mail along with the computer, "college students" are leading by far. Due to their high trip frequency, they decidedly prefer paging service to telephone. The purposes of mobility are also widely different according to the person group. "Male with white collar occupation" scored highest in trip length with a lot of business trips and "the retired" and "the unemployed" scored lowest, as was the case for frequency in using telecommunication modes. Once more, the qualitative indicators for the use of modes and media per person group (Figure 11) suggest the following: • the use of modes and media is complementary in both countries (high physical activity is associated with high virtual activity and vice versa) • the level of use of both systems is slightly higher in Germany • the usage patterns of both systems show distinct similarities. • the use of telecommunication services is clearly favored by younger individuals and thus is a matter of generation and individual experience.
Transport and Telecommunication: First Comprehensive Surveys
197
Table 4 Characteristics of Person Groups Person group
Characteristics
Male under 30
- Employed young male with much computer experience
Male with "White Collar" occupation
- Clerks, Professionals and managers
Male with "Blue Collar" occupation
- Elementary occupation, Service worker. Sales, Craft and Operator Young office lady
Employed female without child
Employed female with child Housekeeper
Retired/ Unemployed
Student
Communication behaviour - High use of telecommunications
Travel behaviour
- Very high use of telephone & fax - Very high mobile phone accessibility - Average use of telephone & fax
- High trip frequency for job related activities - Meetings, business trips
- High use of telephone
- Very high trip frequency - High trip frequency for leisure activities - Department store, theater - High trip frequency for Shopping activities - Visit, department store
- Elder employed woman - Very dependent on child - Young housewife, dependent on child - Older housewife
- Average use of communications
- Pensioner or unemployed
- Rare use of telecommunication
Student
- Very low use of telecommunication
- Highest level of emailuse - Higher level of paging use instead of telephone - Highest computer accessibility
- High trip frequency - High travel length for leisure activities
- Average trip frequency - High trip frequency for leisure activities
- High trip frequency - High trip frequency for leisure activities - Shopping, Visit - Very low trip frequency - High trip frequency for leisure activities - Hobby, Pharmacy - Higher trip frequency - High trip frequency for leisure activities - Library, private institute
198
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Male, w hite collar Brnployed w omen
College Student
ill Contacts per person per day: Germany B Contacts per person per day: Seoul » Trips per person nfir d a y nprmanv
t
Unemployed \ Housekeeper
2 number of trips
^
Figure 10 Trips and Contacts by Person Group
Male, blue collar Male, white collar Employed women,children Employed women
Contact length per person per day: Germany • Contact length per person per day: Seoul • Trip length per person per day: Germany m Trip length per person per day: Seoul
Male < 30 College Student Pupil Unemployed Retired Housekeeper
100
120 km
140
160
Figure 11 Lengths of Person Activities by Person Group
180
200
220
Transport and Telecommunication: First Comprehensive Surveys
199
THE CONCEPT OF A MODEL Modeling the relative amount of substitutional and complementary contributions of new telecommunication facilities to the transport and communication sector (Mokhtarian and Solomon, 1994) requires deeper insight into the characteristics of potential interactions between modes and media. The following definitions of combined transport and communication activity patterns are meant to be an aid in the process: activities
teleactivities
travel
travel modes
in- home activities
telecom
no external contact
media modes
Thus, an activity is described by the purpose and by the associated spatial element. The resulting activity patterns then form a sequence of activities containing the information per activity already mentioned (see Table 3 and the following example). activity pattern purpose
home
work
private contact
home
1 spatial element
residence
work place
person in zone x
residence
potential travel/c Dmmunication choices tele-work home choice A: no travel work home car 1 choice B: car travel work home bus 1 choice C: bus travel
phone car
visit
bus
phone
1
home 1 car
Figure 12 Example for Combined Travel/Telecommunication Activity Pattern
home home
200
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Consequently, the inclusion of a combined trip/communication generation process is a necessary prerequisite for the modeling of interactions between modes and media. The basic characteristic of this module should be that, among other elements: • either trips can be omitted due to contacts • or trips can be generated due to contacts. A suitable method for such a module might be a rule-based approach that recognizes the number of options and constraints associated with combined trip/communication patterns. But the necessary prerequisite should be further analysis of the building process of these patterns. As a first step, a descriptive data set has been formed which helps to realize the model concept shown in Figure 13. A) Combined trip/communication generation
I
B) Existing m icro
com bined trip/communication patterns
..„..„:.....,. i
' '
1
microscopic trip planning and performance module e.g. E U R O S C O P E (Axhausen.Ayerbe e t a l , 1991) (Schwarzmann, 1995) with an extended modes and media choice model
Figure 13 The Concept of a Model
Transport and Telecommunication: First Comprehensive Surveys
201
The result is a data set providing all relevant information at the individual level, which is necessary to determine behavioural changes. A first application was performed with respect to the potential of telebanking and teleshopping to substitute travel in Seoul.
FIRST APPLICATION IN SEOUL Preliminary resuhs from a comprehensive field application in Seoul are reported in this section.
Teleactivities - Ultimate Ratio? The efforts to increase accessibility and to reduce energy consumption and emissions would suggest greater "virtualization" of activities. Table 5 illustrates several possibilities for teleactivities that are currently performed or could be performed in the future. Telebanking and Teleshopping were chosen first for further investigation from among new telecommunication modes.
Transition Probabilities As mentioned before, interactive interviews were conducted in banks and stores. Figure 14 shows one of the results: the current level of perception and experience with telebanking. 91% of the interviewees had heard about telebanking, 51% knew how to use it and only 13% actually had experience with telebanking. After an explanation about telebanking and the Internet, the majority (56%) of the persons asked would have preferred telebanking to the activity just completed in the bank. This theoretical transition probability of 56% is considerably higher than those 13% of respondents having ever used telebanking before. This discrepancy indicates the extent of uncertainty associated with the measured transition probabilities, but it quantifies at the same time the potential for future development. In view of this uncertainty, it may be helpful to differentiate the potential transitions according to the previously defined person groups (Figure 15). With regard to their communication behaviour, "males under 30" have the highest transition probability to telebanking whereas "the unemployed" and "the housekeepers" have the lowest.
202
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Activity
1
Table 5 Teleactivities and Associated Infrastructure potential technical infrastructure teleservice present future
shopping
homeshopping
videotext, internet
internet
business
teleconferencing
videoconference
videoconference
banking
telebanking
phone
internet, security e-mail
personal contacts
videophone
imagephone, e-mail, internet
culture
video on demand
phone, fax, e-mail, internet pilots
medical services
telediagnosis
pilots
videoconferencing, internet
education
teleleaming
pilots
internet
booking services
telebooking
phone, fax, Internet
phone, fax, internet
sports, entertainment
virtual reality
pilots
cyber technique
VOD, HDTV
Similar patterns on a much lower level show the results of the interviews concerning teleshopping (Figures 16, 17). Ongoing work is now concentrating on the application of these transition probabilities in different alternatives to trips with the associated trip purposes at their destinations in order to quantify potentials to substitute travel. The major advantage of these simulation runs is seen in: • the ability to run sensitivity tests related to the potential measurement errors • the expectation to produce spatially differentiated impacts in an infrastructure network • the ability to include demographic attributes in forecasts to reflect the differing use of telecommunication across generations.
Transport and Telecommunication: First Comprehensive Surveys
100%
9170
80% 60% -
t)17o
40% 100/
20% 0% ^ heard
know how to use
experienced
Figure 14 Current Attitude to Telebanking
Figure 15 Transition Probability to Telebanking by Person Group
203
204
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
80% 80%
]
60% -
41% 40% 20% -
8%
1
0% " heard
know how to use
1
experienced
Figure 16 Current Attitude to Teleshopping
25%
-
20%
-
23%
15% -J
10%
12%
12%
c olleg<5 studen t
e rn p l o y e d Vi^omen witl- out child
14%
14%
employe 3d w o m e n v\/ith child
m ale < 30
13%
6%
5% ^
0% ^ housekeeper
m a e , "white collar"
average
Figure 17 Likelihood of Teleshopping by Person Groups CONCLUSION AND OUTLOOK There is a strong interrelationship between telecommunication and travel behaviour going far beyond those interactions that are currently taken into account in planning methods and considerations (Janelle, 1995). We are challenged to answer the question of how new telecommunication media are changing travel behaviour. However, already one finding that already emerges is that further increases can be expected in both telecommunication usage as well as travel mobility, in light of increasing dispersion of labor and progressive scattering and specialization in private life (Figure 18). Evidently, there is little room for substitution theories, or hopes of decreasing traffic volumes. There is instead a strong requirement to solve the
Transport and Telecommunication: First Comprehensive Surveys resulting and difficult task of simulating the telecommunication services.
complementary
impacts
205
of new
Preparation long distance journey
thesis for future development
A telecommunication » A transport
Figure 18 Spatial Action Fields for Transport and Communication
206
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
ACKNOWLEDGEMENT This publication is based on research work supported in major part by the Rochling-Stiftung in Mannheim (German data) and the Krupp-Stifitung in Essen (Korean data).
REFERENCES Axhausen, Ayerbe, Bannelier, Berkum, Billotte, Goodwin, Herry, Katteier, Mede, Meurs, Polak, Schwarzmann, Selva, Yune and Zumkeller (1991). "Towards a Dynamic and Activity-based Modelling Framework", In Advanced Telematics in Road Transport, proceedings of the DRIVE Conference Brussels, Elsevier. Cerwenka, P. (1992). Verkehrsentwicklung im ZivilisationsprozeB, Internationales Verkehrswesen, Deutscher Verkehrsverlag GmbH, Hamburg, November. Chlond, B., O. Lipps and D. Zumkeller (1997). The German Mobility Panel: Options, Limitations and the complementary Use of secondary Data. Paper for the International Conference on Transport Survey Quality and Innovation: Transport Surveys: Raising the Standard, Grainau, Germany. Garrison, W. and E. Deakin (1988). Travel, Walk and Telecommunications: A View of the Electronics Revolution and its Potential Impacts, In Transportation Research, vol. 22a, no. 4, Great Britain. Heinze, W. (1985). Zur Evolution von Verkehrssystemen, In Perspektiven verkehrswissenschaftlicher Forschung, hrsg.: Sigurt Klatt, Dunker und Humboldt, Berlin. Heldman, R. K. (1995). The Telecommunications Information Millenium.-Washington: McGraw-Hill, Inc., 230 S. Janelle, D. G. (1995). Metropolitan expansion, telecommuting, and Transportation. In The geography of urban transportation 17, S. 407-435. Kohler, S. (1993). Interdependenzen zwischen Telekommunikation und Personenverkehr, Diss., Universitat Karlsruhe. Mokhtarian, P. L. (1993). The travel and urban form implications of telecommunications technology. Discussion paper for FHWA/LILP Workshop. Mokhtarian, P. L. and I. Solomon (1994). Modeling the choice of telecommuting: Setting the context. In: Environmental and Planning A, Vol. 26, S.749-766. Salomon, I. (1986). Telecommunications and travel relationships: A Review. In Transportation Research A, Vol. 20A, S.223-238. Schwarzmann, R. (1995). Der EinfluB von Nutzerinformationssystemen auf die Verkehrsnachfrage, Diss., Universitat Karlsruhe.
Transport and Telecommunication: First Comprehensive Surveys
207
Selvanathan, E.A. and S. Selvanathan (1994). The demand for transport and communication in the United Kingdom and Australia. In Transportstion Research B, Vol. 28B, S.1-9. Widlert, S. (1997). Survey of communication habits - a short description of purpose and design. Unpublished paper by the Swedish Institute for Transport and Communications Analysis, Stockholm. Zumkeller D. (1997). Sind Telekommunikation und Verkehr voneinander abhangig? Ein integrierter Raumiiberwindungskontext", In Internationales Verkehrswesen (49) S. 16 21. Zumkeller D. and H. Seitz (1993). Aufbereitung vorhandener Daten fiir Verkehrsplanungszwecke als Ersatz fiir neue Befragungen. Forschung StraBenbau und Strafienverkehrstechnik 642. Bundesminister fiir Verkehr Bonn/Bad Godesberg.
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
TELECOMMUNICATIONS - TRAVEL INTERACTION: WORKSHOP REPORT
David A. Hensher and Jackie Golob
BACKGROUND As we approach the new millennium, the opportunity for virtual accessibility is well advanced. The higher definition of audio and visual interactive communications in time and space is improving at a fast pace with the support of digital technology and spatial determination tools such as global positioning systems (GPS). The opportunities to appraise the potential for telecommunications to facilitate and/or enhance the exchange of information with/without travel are opening up continuously. For example, teleconferencing can now deliver gaming capability, as well as screen, white board and graphics capture from a remote site. The introduction of technology continues at a rapid rate; in contrast behavioral response is somewhat slower. Opportunities presented to individuals to change travel behavior through telecommunications, in the form of travel substitutes and complements, as well as travel modification and value-enhancing travel, are extensive, but are not well understood. The Mokhtarian and Salomon (2001) paper appearing in this book summarizes the state of current knowledge of the impact that telecommunications has had on travel behavior. This chapter consolidates ideas which were discussed over three days in the Austin workshop on telecommunications-travel interaction. The chapter is organized as follows. We begin with a definitional statement, followed by some ideas for an integrating conceptual framework which
210
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
was necessitated by the ambiguity in the relationship between the contents of three critical constructs: communications, activities and travel. This conceptual framework served to bridge the disparate sets of interests and concerns drawn from the broad and often unrelated literatures on telecommunications and travel. This then leads on naturally to consideration of the analytical framework and data needs. The last substantive section summarizes a research agenda. We also emphasize that although the behavioral responses of agents (ie individuals, households, firms) is the primary focus of this report, that the outputs from such an effort are but one source of inputs into a broader benefit-cost framework designed to evaluate the impact of alternative telecomunications scenarios on the environment, the efficiency and effectiveness of private and public sector organisations, household quality of life and the economy as a whole (Shafizadeh et al 1997).
DEFINITIONS Travel is, with rare exception, a derived demand. In this sense it is a derived activity which has historically dominated the ways in which people and commodities are (re)positioned in time and space. We view the underlying attributes driving this repositioning of agents and commodities as being a muhi-dimensional vector of contents, whose dimensions may simultaneously include information content (IC), physical content (PC), social content (SC), material content (MC), aesthetic content (AC) etc. An activity is then defined by its contents. Most importantly activities are allowed to be multi-dimensional. These activities occur within communication space which provides an attractive context within which to study the role of telecommunications in travel behavior research. Communication space is a global construct capturing all forms of contents' exchange. The definition of communication space is sufficiently general to incorporate near or local distance such as an in-home activity, and nonlocal distance (Zumkeller 2001). An activity becomes the overarching unit of analysis, differentiated by the mix of contents. For example, the activity 'shopping' might be described by an internet order placed at home, followed by a car trip to the shop, a social meeting with a friend at the shopping center, the collection of the ordered shopping and a car trip home. The social content is high, the information content may not be as high as ordering on site where the merchandise can be felt (but this may depend on the nature of the goods being purchased and one's overt experience with them). The physical content in terms of physical exertion is high compared to a homedelivery; the aesthetic content may be not applicable, and the material content is high because of the travel component and the greater ability to do comparative shopping.
Telecommunications-Travel Interaction: Workshop Report
211
Alternative ways of shopping are particular specifications of the contents mix. For example, the internet order may generate a home-delivery instead; however the social content of the activity would have been zero, and the physical content high in the form of a commercial urban goods movement rather than a person-shopping trip, but the physical content would not be high from the perspective of the decision maker. Hence, the low physical content (from the individual's perspective) may be a factor favoring the choice of teleshopping or a factor against it, for some. Here we have substitution between a trip by a household agent and a trip by a firm's agent. Telecommuting or teleworking, as an example of an alternative to commuting, fits nicely within this framework. Table 1 illustrates how we might describe four work week activity alternatives - telecommuting 1 and 2 days a week, and traveling to work by car and by bus. Each generic content dimension (eg information) has been decomposed into a set of elemental contents (eg information access, information quality, information type). Traveling to work has higher information levels and physical exertion, external social interaction and travel costs than telecommuting, and telecommuting has higher household interaction and telecommunications costs. The idea of content mix can be applied to a simple activity such as home-walk-home and a complex activity (or chain) such as home-walk-restaurant-walk-home or home-teleshopdelivery to home-eat, all of which have physical, material, social, aesthetic and information content. Content mix also helps with the problem of simultaneous activities e.g. driving to work while using a cell phone to plan social activities accompanied by a colleague with whom one is discussing strategy for an immediate meeting. It is the content mix which defines the sources of relative utility associated with simple and complex activities. This gives the analyst the ability to represent multiple simultaneous activities in utility space.
212
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
Table 1 An Illustration of The Contents-Mix of Alternative Activity Profiles for the Work Week Contents
TC -1 day/week
TC - 2 days/wk
Travel by Car
Travel by Bus
Infonnation Access Information Quality Information Type Physical Exertion Materials - non online Travel Cost - $'s Travel Cost - Stress Social Interaction Household Interaction Aesthetic Content Telecom Costs to Ind.
lower lower low low low access 20% lower ? low high depends high
lowest lowest lowest lowest lowest access 40% lower
high high high high high high high high low depends 0
high high high highest high low high high low depends 0
7
lowest highest depends highest
Notes to Table 1: TC = telecommuting which may be at a remote site office or a home location. The dimensions of content can be quantitative and qualitative. The scales we have used to measure most of the attributes are illustrative only, and are open to debate. The three dimensions of IC capture the ease with which information can be accessed, the quality of the information available via this alternative, and the type of such information (eg working on a project at home may limit information to on-line inputs and data that are available at home, missing the quality of face to face interaction and debate which may also affect information type)
A CONCEPTUAL FRAMEWORK TO CAPTURE THE CONTENT-MIX OF AN ACTIVITY The proposed approach assumes that agents act as if they maximize the utility of an activity (or set of activities), and that (indirect) utility is a function of the content of the activity on the set of dimensions identified above. Conceptually we might define the demand for communications by agent ^ as a function of the contents exchange demanded by the agent located at spatial point /, the generalized goods content located at spatial pointy, and the generalized cost of exchanging contents between / andy. A specific ij pair can be defined very finely as the same location (eg an office and kitchen at home) or a different location (eg office / at home and kitchen y at home, office / and restaurant 7 in the same building, or a home at location / and store at location 7). The activity choice set embodies alternative amounts of content exchange. The choice of an alternative activity is a function of, amongst other influences, the nature of exchange content, the degree to which this content can be obtained remotely ('at a distance' without travel), and the (dis)utility associated with the sources of benefits and costs (eg the trade-off between the utility of information content with the utility of social content). Formally, an indirect utility function associated with an agent q and an activity a is defined as: Vqa =f(ICqa, PCga, MCga, SCqa, ACga, othcr bcncfits, travcl time/stress, telecom cost, telecom time, other costs, socioeconomic characteristicSq, constraints....).
Telecommunications-Travel Interaction: Workshop Report
213
In recognition that the communication choices made by individuals are influenced by other members of the household (such as teleshopping, telebanking, tele-entertainment), and/or an employer (notably for teleworking), the indirect utility function may be generalized further to recognize 'interactive' agency effects: ^qa = ^(Vq'a, ICqa, PCqa, MCqa, SCqa, ACqa, othcr bcnefits, travcl timc/strcss, tclccom cost, telecom time, other costs, socioeconomic characteristicsq, constraints....). Each indirect utility expression can represent a simple or complex activity in which both travel and telecommunications are inputs into the creation of a content-mix. Taste weights can be attached to each of the underlying content attributes to identify their contribution to relative indirect utility. The framework can be generalized to incorporate dynamics in recognition of the importance of changing awareness, ability to adopt and adapt to experience with new telecommunication alternatives and the potential impacts on activity chains. We hypothesize that the taste weights will vary over time quite dramatically as the opportunities for complex activity chain switching takes place in the presence of greater opportunities for telecommunications - travel substitution and complementarity, travel modification and value enhancement of travel. Thus the mixed-content framework is an evaluation of benefits and costs (or disbenefits). For example, traveling to work has higher information levels (a benefit) and physical exertion (a disbenefit), external social interaction (a benefit) and travel costs (a disbenefit) than telecommuting, and telecommuting has higher household interaction (a benefit) and telecommunications costs (a disbenefit). The choice to telecommute will only be made when the benefits outweigh the disbenefits .This mixed-content methodology is inherently a valuelatent approach. The blurring of the temporal and spatial boundaries between activity purposes linked to traditional trip purposes is expected to increase with the advances in telecommunications. For example, cellular (or mobile) phones already are enabling some types of work activity to occur during the commute, blurring the distinction between travel-leisure and travel-work trade-offs and time allocation. Behavioral values of leisure travel time savings may decrease as we recognize the productivity component of commuting activity. The content-mix approach is well positioned to handle this blurring of traditional classifications. The telecommunications-travel interface is essentially a rearrangement of the weights that agents can place on the content of activities.
214
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
Methodological Issues The extant literature has recognized the need for a broadening of the research effort to better understand the dynamics of the interactions between travel and telecommunications. The very nature of behavioral response in the context of a fast changing technological choice set, places limits on the ability of single cross-section data paradigms to reveal sufficient understanding of why, how and when agents rearrange travel-linked activities to take advantage of the contentsmix of the continuously evolving universal communication activity choice set. To date the most comprehensive empirical tests of the degree of substitution and complementarity between travel and telecommunications simply have been broad expenditure studies at a highly aggregate level. While the empirical tests have not been limited to the expenditure studies, those have been the most comprehensive empirical tests. The telecommuting and teleworking pilot studies have been useful but limited in the insight they can offer into this critical behavioral and predictive issue (see Lyons et al. 1997 for a recent example, and Mokhtarian and Salomon 2001 for references to earlier studies). The need for an appropriate analytical framework within which cohort effects, period effects, and adoption cycles can be studied is needed - what we have before us is a technology-led transition process in which the preferences of agents are almost certainly changing through time (although the extent of instability is unknown). The recognition of interactions between agents opens up opportunities to move beyond the simplifying assumption of perfect competition which has dominated travel choice analysis. Although there are some efforts to study the choices of agents such as employees and employers in the context of telecommuting and teleworking (eg Mahmassani et al 1993, Bernardino and Ben-Akiva 1996), the two agents have been studied in a static context where a set of exogenous variables define the influence of one agent set on the other agent set. Brewer and Hensher (2000) propose a more dynamic and interactive game theoretic approach which draws on ideas of negotiation, bargaining and arbitration. Stated choice methods can be extended to incorporate sequential and simultaneous interactive choice experiments in which each agent initially makes a choice either with or without knowledge of the other agent's preferred choice. In subsequent rounds of preference review, each agent considers his/her prior position in the light of the other agent's willingness or otherwise to revise their preferences. A set of cooperative and non-cooperative solutions are available which provide the distribution of probabilities that each agent will cooperate under particular content-mixes. This approach reveals the set of constraints that deny cooperation.
TelecommunicationS'Travel Interaction: Workshop Report
215
and signals the nature of incentives to resolve non-cooperation. This method is potentially very powerful in the study of telecommuting and teleworking for the employee-employer agent pair, and for teleshopping, telebanking, teleservices and tele-entertaining for the interaction between household members. It raises challenging questions about the definition of the agents in an interactive game; for example whether negotiations on telecommuting might be investigated with an employer and a team of employees given that there are limits on the incidence of telecommuting across a team.
Data Issues To enable the empirical determination of the contents-mix approach, some very specific data issues need to be addressed. These include the detail of an activity, especially the amount of information on the telecommunications opportunities, and the measurement of exchange content once a set of definitions of each content dimension is agreed on. There is a need for increased collection of pecuniary data associated with telecommuting. The collection of this data would be implemented in the mixed-content approach by assigning actual values to costs and benefits, eg telecom costs or travel costs. Zumkeller (2001) proposes a distinction between a trip and a contact (i.e. a communication) as a way of designing a survey diary which captures the necessary content of travel and telecommunications behavior. For each telecommunications contact (ie a phone call, a fax, a letter), information is sought on when the activity occurred, the purpose, the aim of the contact (eg exchange of information, social conversation, to organise a visit or meeting), and distance in kilometers to the location of the contact. Such data when combined with trip data can produce trip-telecommunication chains. Data can be divided into a number of broad categories. These include descriptive items such as a record, in some detail, of the activities that an agent undertakes in-homeplace, in-workplace, and out-of-home and work places. Such data can desirably be extracted from a generic activity survey which recognises the detail necessary to capture the broader communication activities that impact in various ways on travel. A longitudinal (panel) approach is essential if we are to understand the role of cohort effects, period effects, early and late adopters etc. In addition there is an ongoing need for specialized data to provide a capability in understanding behavioral response ('what will happen if....') to changing circumstances. This includes the need to obtain a better understanding of the constraints which limit the ability of individuals to use the growing array of telecommunications alternatives in both work and nonwork activities (Handy and Yantis 1997). The development of interactive agency choice
216
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
experiments grounded in a game-theoretic framework as proposed by Brewer and Hensher (2000), offers a way forward to empirically identify the constraints and associated incentives necessary to increase the probability of cooperation between the interacting agents in general and specifically for telecommunication inputs into activities which impact travel. An alternative means of gathering data to illuminate the understanding of dynamic adoption and diffusion of technology experiences is through the use of accelerated learning in controlled experiments. These methods borrowed from market research could help identify defining elements and thresholds. Further experimentation with interactive interviewing is also likely to play an important role in pursuing the understanding of the complexity of effects as new and increased flexibility and costs are added into content mixes of activities. The addition of telecommunication variables to data collection needs adds to the burden of the data collection task. Distinguishing time spent using e-mail communication and its associated purposes and/or quality of content is but one example of the measurement challenges presented to those interested in pursuing the links with travel behavior. It will be particularly important to be farsighted in the design of longitudinal studies to ensure that questions are in place to distinguish the acquisition of forms of communication media and the growth of knowledge and experience of use. Data need to be collected not, for example, as days "telecommuted" or "teleworked," but by objective criteria such as the hours worked, time of day and location (Pratt 1997, 1997a). Particularly at the level of population census, questions with regard to work status and location must be re-formulated if we are to relate disaggregate findings to aggregate population data.
FUTURE DIRECTIONS: A RESEARCH AGENDA This report has evolved out of a workshop in which the participants sought out opportunities to clarify the definitions of key constructs, which seemed to be impeding the development of a conceptual framework. The conceptual framework proposed offers a way forward in integrating the elements of communication, activity, and travel. The idea of contents-mix seems to hold the key to this integration, at least conceptually. Measuring content is more difficult; although we are convinced that this should occur within a panel-based activity framework in which the evolution of behavioral response is captured as we learn more about the role of telecommunications in facilitating the dimensions of travel.
Telecommunications-Travel Interaction: Workshop Report
217
There appears to be consensus that even though technological change and innovation are occurring rapidly, behavioral change and adaptation occur much more slowly. The primary policy question for the transportation planning community is: to what extent will telecommunications have impacts which are complementary to existing activities or will they prove responsible for substitutions of activities which previously involved travel? The motivation for answering this difficult question is the need to plan and budget for investments in transportation infrastructure and services. It is premature to attempt to answer such questions on the basis of pilot projects which have been limited in scope and frequently focused on specialized groups of "early adopters" and in many cases dealing with only one application at a time. However work to date has illustrated the need for more comprehensive study frameworks which capture the full complexity of the impact of telecommunications media on our lives. This awakening recognition of the need for a conceptual framework which can deal with activities in the sense of their multi-dimensional content as well as capture the role of the media in the exchange of information seemed to the group to be a positive breakthrough. Leaving behind such single dimensional concepts as journey purpose or imposing a constraint on definitions of activities limiting them to a single topic allows instead for a multidimensional concept of activity which reflects the various qualities of its content. One hopes this will lead to a greater understanding of how and why individuals respond to changing telecommunication availability. The following summarizes future study topics which will undoubtedly take us into the new millennium in our search for understanding of the impact of telecommunications on travel behavior. • Operationalize the conceptual framework to capture the content mix of activities and the manner in which content is exchanged. Develop further the understanding of communications space. • Investigate the role of telecommunications in non-work activities for both individuals and households. • Pursue the concept of the social content of activities and the relative significance for the understanding of the travel component. • Incorporate an understanding of quality of life issues whose importance may have been underestimated. (Two examples discussed in the workshop illustrated anticipated or unanticipated benefits of teleworking (Gordon 1997). In a demonstration program in the U.K. (Lyons et al 1997) a respondent indicated a belief that the positive benefits of teleworking had been a reduction in stress, which allowed her to become pregnant.
218
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges Other workshop members had anecdotal evidence of the benefits of teleworking contributing to both human and animal wellbeing including reduced costs from puppy damage to household furnishings).
•
Land use issues and population shifts in response to telecommunications opportunities will continue to be a provocative area of study requiring finer levels of geographic detail and longitudinal data.
•
Broadening the understanding of intergenerational constraints and the tracking of transitions to new forms of communication following exposure and experience can be tackled through panels and accelerated learning experiments. Such simple techniques as the application of focus groups to investigate activity complexity and telecommunication choices will also have a role. Interactive agency and game theory linked to stated preference and revealed preference studies can be used to investigate the bounds of both individual and group choice making with regard to changing activity locations. The type of data collected in population surveys should also not be ignored and attempts should be made to address the need to modify classifications over time to recognize such things as more complex work practices. Also the availability of pc's, internet interfaces, interactive t.v., cell phones, etc. should be included. Investigate in greater depth the value of applications from social and psychological studies of communication, work and leisure time activities. Interdisciplinary studies present special challenges because of the extent of the very disparate literatures. Identify concepts that show most promise for supporting the travel behavior study field while avoiding information overload, and language and definitional confiision.
•
•
•
ACKNOWLEDGEMENTS This report is the product of a joint effort of the authors, and workshop members: Glenn Lyons, Pat Mokhtarian, Joanne Pratt, KevanShafizadeh, Ilan Salomon, Jin-Ru Yen, Dirk Zumkeller, Ahmed Abdelghany. This report was prepared remotely as a teleworking activity.
REFERENCES Bernardino, A. and M. E. Ben-Akiva (1996). Demand for telecommuting: modeling the adoption process, in Hensher, D.A. , King, J. and Oum, T. (eds.) World Transport Research Vol. 1, Pergamon Press, Oxford, 241-253.
Telecommunications-Travel Interaction: Workshop Report
219
Brewer, A. and D. A. Hensher (2000). Distributed work and travel behavior: the dynamics of interactive agency choices between employers and employees, Transportation 27 (1), 117-148. Gould, J. and T. Golob (1997). Shopping without travel or travel without shopping? An investigation of electronic home shopping, Transport Reviews, 17 (4). Gordon, G. (1997). Telecommuting Review newsletter, No.9, p. 15. Handy, S. and T. Yantis (1997). The impacts of telecommunications technologies on nonwork travel behavior. Presented at 76^^ Annual Meeting of the Transportation Research Board, Washington D.C. Lyons, G., A. Hickford and J. Smith (1997). The potential impacts of teleworking on travel: results from a longitudinal UK case study, IATBR'97 Conference Pre-Prints, Workshop on Telecommunications-Travel Interactions, Austin, Texas, September 21-25. Mahmassani, H., J-R. Yen, R. Herman and M. Sullivan (1993). Employee attitudes and stated prefrences towards telecommuting: an exploratory analysis, Transportation Research Record 1413,31-42. Mokhtarian, P. L. and I. Salomon (2001). Emerging travel patterns: do telecomunications make a diference? Chapter 7 in this volume. Pratt, J. H. (2000). Asking the right questions about telecommuting: avoiding pitfalls in surveying home-based work. Transportation 27 (1), 99-116. Pratt, J. H. (1997). Counting the New Mobile Workforce. U.S. Department of Transportation, Bureau of Transportation Statistics, April. Shafizadeh, K., D. Niemeier, P. Mokhtarian and I. Salomon (1997). The costs and benefits of telecommuting: an evaluation of macro-scale literature, IATBR'97 Conference PrePrints, Workshop on Telecommunications-Travel Interactions, Austin, Texas, September 21-25. Yen, J-R. (2000). Interpreting employee telecommuting adoption: an economics perspective. Transportation 27 (1), 149-164. Zumkeller, D. (2001). Transport and telecommunications: first comprehensive simulation approaches, Chapter 8 in this volume.
This Page Intentionally Left Blank
SECTION 4 TRAVEL BEHAVIOUR-LAND USE INTERACTIONS
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
10
TRAVEL BEHAVIOUR - LAND USE INTERACTIONS: A N OVERVIEW AND ASSESSMENT OF THE RESEARCH
Susan L. Handy
INTRODUCTION The link between land use and travel behaviour has been of interest to transportation researchers for many decades. Mitchell and Rapkin's oft-cited Urban Traffic: A Function of Land Use, published in 1954, reflected the emerging interest in forecasting travel demand and stated that "more knowledge is needed about the precise nature of the relationship between land use and movement and about the extent of their mutual effects" (pg. 3). That statement is as true today as it was then. In the intervening time, our knowledge of the nature of the relationship between land use and movement has certainly increased, thanks to a long series of studies on the topic (Handy 1992a; T C R P 1995). Early studies on the link between land use and travel focused maximizing the efficiency of the transportation system and determining transportation needs (e.g. Levinson and Wynn 1963, Levinson and Roberts 1965). Studies from the late 1960s and early 1970s often focused on energy consumption, reflecting the growing concern over resource limitations at the time (e.g. Gilbert and Dajani 1974). Studies from the 1980s reflected a shift towards concerns over congestion and the impact on the individual commuter, although energy and air quality concerns also played a role (e.g. Cervero and Griesenbeck 1986). Most of these studies characterized land use - or urban form more generally^ - using relatively "macro" measures
224
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
such as density, activity mix, job decentralization, overall structure (e.g. monocentric vs. polycentric), or city size. The need for more knowledge today is partly driven by the fact that the questions have changed as the policy focus has changed. ISTEA, for one thing, has put renewed emphasis on planning for non-motorized modes such as walking and biking and has generated projects designed to increase use of these modes. In addition, as the new urbanism movement has grown, communities throughout the U.S. have taken a new look at their land development policies and considered or even adopted changes to these policies that enable or even encourage forms of development more supportive of pedestrians and transit users. The hope, supported by the claims of new urbanism proponents but not much else, is that pedestrian- and transit-oriented development will lead to a reduction in automobile dependence and thus automobile travel. In both cases, the question is much different than before: how "micro" elements of the environment influence travel choices. The movement towards micro-simulation as a tool for forecasting travel demand has also created a need for a better understanding of how the micro elements of urban form influence travel behaviour. Some agencies have succeeded in incorporating more and better land use variables in their traditional forecasting models, thereby increasing the sensitivity of these models to land use policies. But the coarseness of these traditional models means that there is limited benefit to worrying about urban form at a very detailed level. With micro-simulation models, however, it now becomes both important and profitable to incorporate specific aspects of urban form that influence individual decisions about travel. The question is how "micro" elements as well as "macro" elements of urban form influence travel choices. Although the questions have changed, the research methodologies have not, or are at least changing slowly. For the most part, research-to-date demonstrates correlations between travel patterns and characteristics of urban form, with urban form accounting for a limited but statistically-significant share of the variation in travel patterns. But it has shed very little light on the behaviour underlying these correlations and thus on causal relationships - the kind of knowlegdge needed to predict the impact of new urbanist policies or to build more accurate micro-simulation models. Although some progress has been made, this body of research is sorely in need of more theoretically-based, exploratory approaches. In this chapter I first review the types of research approaches that have been used, then discuss a long list of issues that have yet to be adequately addressed.
Travel Behavior-Land Use Interactions: An Overview and Assessment RESEARCH
225
APPROACHES
Most of the recent research on the link between urban form and travel behaviour that is used to support policies as a way of shaping travel behaviour falls into three categories (Handy 1996a). In the first type, traditional transportation models are used to predict differences in total travel between typical suburban neighborhoods and hypothetical neighborhoods of higher densities or more traditional design. In the second type, aggregate level data are used to compare average travel characteristics in neighborhoods of different design or cities of different densities. In the third type, disaggregate data are used to test differences in individuals' travel choices between neighborhoods and the relative importance of a variety of urban form factors in those choices. Studies of the first two types generally show significantly lower levels of automobile use, due to fewer automobile trips and/or shorter trip distances, in areas of higher density or traditional design than in typical suburbs, but studies of the third type are less conclusive and often suggest that the results depend on various factors not accounted for in the first two approaches. Studies of both the second and third types suggest that the results depend on what characteristic of urban form and what aspect of travel are analyzed. Simulation studies assume certain relationships between urban form and travel patterns and then use these assumed relationships to predict the implications for travel of alternative forms of development; they do not empirically test the relationship between urban form and travel behaviour. In most cases, hypothetical cities or neighborhoods, or hypothetical changes in real cities or neighborhoods, are tested using a traditional transportation planning model. Two important limitations of this approach should thus be noted: the accuracy of the assumed relationships between urban form and travel behaviour, and the appropriateness or generalizability of the assumed land use development. For the most part, these studies provide some general insights into the potential effect on travel patterns of different types of development but do not contribute to an understanding of the relationship between urban form and travel behaviour. Until recently, simulation studies tended to focus on the overall structure of a city or metropolitan area, in terms of the distribution of employment and residential activities and/or the structure of the transportation network (e.g., Levinson and Roberts 1965; Schneider and Beck 1973; Kulash 1987). The well-known "Costs of Sprawl" study by the Real Estate Research Corporation (1974) used this approach to test the implications for travel times, energy use, and air pollution in six different community development patterns. More recent studies have used this approach to test the impact of neo-traditional development ~ or certain aspects of it ~ on travel. Douglas (1991), for example, tested alternative future development patterns
226
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
for the Middlesex-Somerset-Mercer, NJ, Region. In more limited studies, McNally and Ryan (1994) and Rabiega and Howe (1994) focus on the structure of the street network within a neighborhood. A large segment of the research on the link between urban form and travel patterns, including many of the recent studies, falls into the category of aggregate analysis. This type of research characterizes both urban form and travel, for cities or neighborhoods or zones, using aggregate measures and tests the strength of the relationship using simple comparisons, correlations, or regression procedures. Many of these studies have shown significant relationships between density or other measures of urban form and trip frequency, average trip lengths, mode split, or total automobile travel. This approach has thus provided promising evidence of the potential effectiveness of land use policies in reducing automobile dependence. But because of the focus on the relationship between urban form and aggregate patterns of travel, this approach on its own does not allow for an exploration of underlying factors and the mechanisms by which urban form influences individual decisions. Of the studies that compare one city to another, Newman and Kenworthy's (1989) study may be the most well known. More recently a number of studies have compared travel patterns in preWWII "traditional," gridded suburban communities and post-WWII "sprawling," curvilinear suburban communities. The intent of this work is, for the most part, to test claims that are made about the potential impact of new urbanism strategies on travel behaviour. These studies together provide convincing evidence that automobile travel is lower in traditional-style neighborhoods, but only a limited understanding of the specific elements of characteristics of these neighborhoods that lead to these differences. However, this research has become more sophisticated over time in terms of increasingly comprehensive characterization of urban form. Examples include Holtzclaw (1990, 1994), Friedman et al. (1994), Frank and Pivo (1994), Cervero and Gorham (1995), and the LUTRAQ study for Portland (Parsons, Brinkerhoff, Quade & Douglas 1993). Cambridge Systematics' (1994) study differs from these others in its focus on the effects of land-use-related demand management strategies at employment sites in the Los Angeles area, but is notable for its comprehensive approach to characterizing urban form at the work site. In contrast to aggregate analyses, disaggregate analyses use individual and household socioeconomic and travel characteristics, rather than zonal averages. Analysis-of-variance or regression models are estimated to test the strength of the relationship between socioeconomic, urban form, and travel characteristics, but in this case for the individual or the household rather than for zonal averages, thus accounting for within-zone variations. Some of
Travel Behavior-Land Use Interactions: An Overview and Assessment
227
these studies also incorporate disaggregate measures of urban form into the analysis; although many aspects of urban form are most appropriately measured at an aggregate level, for the neighborhood as a whole, certain characteristics such as distance to nearby activities can vary significantly for households within the neighborhood. These studies show not only that travel differs between different types of communities, but also begin to reveal some of the complexities in behaviour underlying these differences. These disaggregate studies, however, do not directly reflect theories of choice processes. Hanson's study in Uppsala, Sweden, in the early 1980s is often cited in more recent studies of this type (Hanson 1982). This study characterized urban form in terms of accessibility to various activities and found significant links between accessibility and certain aspects of travel behaviour. Work by Frank and Pivo (1994) and for the LUTRAQ study (Parsons 1993), which included disaggregate analysis along with aggregate analysis, show that urban form variables account for a significant portion of the variation in travel behaviour. Ewing, et al. (1993), Handy (1992b), Cervero and Radisch (1995), Rutherford, et al. (1996), and Cervero and Kockelman (1996) show that travel patterns differ more significantly between neighborhoods of different type than among residents within each neighborhood and begin to explore urban form characteristics that may lead to these differences. The study by Kitamura, et al. (1994) is notable for its exploration of the role of attitudes in explaining travel behaviour. This study showed that characteristics of urban form account for a significant portion of the variation in travel behaviour, but that attitudes accounted for a larger portion of the variation. A more recent study by Handy (1996b) explored the role of perceptions of urban form characteristics in explaining patterns of walking.
ISSUES The research-to-date on the link between urban form and travel behaviour has contributed to a substantial base of knowledge, yet leaves a number of important issues unresolved. What we know is that there are significant differences in travel patterns between different types of communities. What we don't ftilly understand is the reasons for those differences and the importance of urban form relative to other factors in explaining those differences. Although some studies claim to show that some measure of urban form (density, for example) "explains" a certain portion of the variation in travel behaviour, the explanation is a statistical one, not a behavioural one. To better understand the behaviour and thus the causality underlying the differences in travel patterns that have been observed, researchers would do well to take
228
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
another look at travel behaviour theory and the kinds of questions it raises about the ways in which urban form may influence travel behaviour. Despite its limitations, discrete choice theory provides a useful framework for thinking about the causal relationship between urban form and travel behaviour (Handy 1996c). In short, this theory says that the probability of an individual making a particular choice, out of some set of possible choices, depends on the utility of that choice relative to the utilities of all possible choices. The utility of any choice depends on the characteristics of that choice as well as the characteristics of the individual, since everyone evaluates choices in different ways. When applied to travel behaviour, this theory suggests that three elements must be considered: the set of possible choices, the characteristics of those choices, and the characteristics of the individual. Within this context, urban form is evaluated in terms of the sets of choices that it provides and the way it influences the characteristics and quality of those choices. Some elements of urban form determine what choices are available to an individual. Most basically, the pattern of land uses - what activities are located where - determines the set of possible destinations, which will be different for each activity, and the distances to them. Other elements of urban form determine the quality or characteristics of the choices that are available for destination and for mode: are there sidewalks, are they shaded, is the scenery interesting? is the store surrounded by parking or situated on the street? To understand the role that urban form plays in travel choices it is necessary to evaluate the ways in which urban form shapes the choice sets available and the characteristics of the choices within those sets. A closer look at these two elements - choice sets and choice characteristics - suggests a number of issues with respect to urban form that need to be addressed, and these issues pont to the need for different research approaches.
Choice Sets The relevant choice set depends most fundamentally on the choice being made. Most studies have focused on trip frequency or mode choice or on total vehicle travel (a combination of frequency, mode choice, and distance as determined by destination choice), but it is widely recognized that different kinds of choices are not always independent. For example, the choice as to mode of travel may by tied to the choice of when to travel (e.g. go during the day and walk vs. go at night and drive), or the choice of how often to go may be tied to the choice of where to go (e.g. a closer store more often versus a farther store less often). Choices may be
Travel Behavior-Land Use Interactions: An Overview and Assessment
229
made simultaneously or sequentially or both. In any event, the structure of choices is more complex and more variable than most models reflect. As Heggie (1978) put it: "To characterize travel choices in terms of trip frequency, time of day, destination, mode and route is thus unduly simplistic. It ignores the wide range of substitutes actually available to the individual and the linkages which often cause an (expected) response to lead to another (unexpected) one" (pg. 113). One issue that is important in understanding whether certain aspects of urban form can reduce automobile dependence is whether or not walking trips substitute for driving trips. The usual assumption is that the issue is one of mode choice: whether to walk or drive. In this case, a decision to walk would mean one less decision to drive and thus a reduction in total driving. It is also possible, however, that the choice is one of whether to take a walk or to stay at home. In this case, urban form can be said to "induce" trips, that is, enable or encourage trips that would not be made if walking were not an option. Traditional travel diary surveys show what choices have been made but do not provide a basis for understanding the set of alternatives considered in making that choice. Similarly, the relationship between long term choices, such as residential and work location choice or automobile ownership, and day-to-day choices about travel must be considered: long-term decisions determine what choices are available on a day-to-day basis. Where one chooses to live, for example, will determine whether or not stores are within walking distance or whether transit service is available. This suggests another, indirect role for urban form in travel choice, in influencing these longer term decisions that determine the choice sets available in the shorter term. This relationship also raises the possibility of self-selection: that those individuals who want the option of walking or taking transit purposefiilly select a residential location where they will have that option. In this case, a correlation between urban form and travel does not reflect a day-to-day influence on travel behaviour but rather an impact on the longer term decisions that determine day-to-day choice sets. It is almost too obvious to state that different types of trips (work versus shopping, grocery shopping versus fiimiture shopping) will have different choice sets. Yet many of the urban form studies analyze total travel or focus on work trips or on nonwork trips. Jones (1978) points out the behavioural differences between work and nonwork travel behaviour: "Nonwork travel decisions are usually much more complex than their work journey counterparts, because of the wider range of real choice potentially available (including mode, timing, frequency, destination and route), the greater variability of travel behaviour (across individuals and over time), and the diverse interrelationships between sub choices: for example, choice of
230
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
mode predetermines the availability of routes, travel times and feasible destinations" (pg. 266). But the difference in terms of flexibility and complexity between different types of nonwork trips may be as great as the difference between work and nonwork trips. It is important to understand differences in the structure of choices and the definition of choice sets for specific types of trips, and the different role that urban form plays for different types of trips. As noted earlier, many recent studies focus on the influence of urban form within the neighborhood. This means that travel choices are evaluated within the context of the choices available within the neighborhood, but not beyond. Presumably the greater the number of destination choices available within the neighborhood, the more likely that a destination within the neighborhood will be selected. But it is also probable that the more choices available outside the neighborhood, the more likely that a destination outside the neighborhood will be selected. Limiting the evaluation of urban form to neighborhood boundaries is an artificial restriction from the stand point of travel behaviour. In addition, many recent studies evaluate the choices available rather coarsely; for example, in terms of the amount of retail employment within the neighborhood. But if the neighborhood retail activity does not include a supermarket, the fact that the neighborhood has lots of retail activity will have no impact on grocery shopping trips. Both of these simplifications mean that the choices residents make are not being studied in the context of the true set of choices available. It is also widely recognized that perceptions, attitudes, and preferences influence each individual's choice set, for both long-term and day-to-day choices about travel. Two neighbors may, objectively, have the exact same choice set in terms of possible destinations and travel modes, yet one may not be aware of one of the local shops while the other may not consider riding the bus an option. In addition, individual and household constraints vary: limited income, extra household responsibilities, or physical limitations may mean one neighbor has fewer options than the other. Thus it is important to understand not just how urban form objectively shapes choice sets (e.g. what destinations are located at what distance away), but also how perceptions about urban form moderate the choice sets that individuals perceive.
Choice Characteristics Urban form also shapes the characteristics of the choices available, most directly for choices about mode and destination. As for choice sets, the role of urban form in influencing the characteristics of choices can most accurately be understood when specific types of trips are examined. Different aspects of urban form may matter to different degrees for different types
Travel Behavior-Land Use Interactions: An Overview and Assessment
231
of trips. For work trips, for example, the primary concern is generally speed, and design elements may have little influence on the choice of mode. For trips to the convenience store, design elements may be the critical difference between the choice to walk or the choice to drive. Another important question is whether specific characteristics of urban form influence choices independently or whether sets of characteristics work together to influence travel choices and if so, what sets. Many of the studies have simply compared travel in one type of community to another. This leaves open the question of what it is about these communities that leads to the differences in travel. Another approach is to use very general measures of urban form - density for example. But two communities with the exact same density may be very different in other ways. The problem is not an easy one to resolve. As Parsons, et al. (1996) conclude after testing the relative influence of density, land use mix, and urban design, "it is difficult to sort out the roles of land use mix and urban design in shaping travel behaviour because of their strong correlation with density... collectively they have a clear impact. The difficulty lies in determining the effects of each separately." It may not be essential to determine the effect of each aspect of urban form separately, but some effort should be made to better understand what the important elements are, help to guide decisions about land use policies and urban design guidelines. A more challenging issue is how characteristics of urban form should be measured. Urban form is most appropriately measured in terms of what matters to the individual. For example, does an individual consider the density of the neighborhood when deciding whether to take the bus or to walk? More likely, she considers how close the destination is, how close the bus stop is, how frequently the bus runs, how congested the roads are - all characteristics which are correlated with density. Does she consider whether the street network is a rectilinear grid? More likely, she considers how direct a route, whether a variety of routes are possible, how much traffic. In other words, elements of urban form should be translated into what they mean for residents and what kinds of choices they make available. Taking the issue one step fiirther, the choices that individuals make depend not on an objective evaluation of urban form but on their perceptions of and responses to urban form: "it is generally accepted that an individual's perception of an event or attribute is the appropriate dimension for explaining behaviour" (Goodwin and Hensher 1978: 17). One neighbor may perceive the neighborhood to be a safe place to walk, the other may not. One neighbor may enjoy looking at houses and gardens on his walk, the other may be oblivious to her surroundings. The urban design literature, including work by Appleyard (1981), Gehl (1971),
232
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
and Schumacher (1986), suggests ways in which urban form shapes the perceptions of the physical environment and how perceptions of the physical environment shape activity within that environment, including walking and "hanging out." Although researchers can objectively observe different design elements, they can't so easily evaluate the different ways in which different individuals interpret and respond to those elements.
Changes Research in this area often makes the leap that differences in travel behaviour in different types of analysis mean that a change in urban form will lead to a change in travel behaviour. The error is one of assuming that cross-sectional studies demonstrate causality. Determining whether a change in urban form leads to a change in travel behaviour requires evaluating how the change in urban form alters the choice equation, by expanding or contracting the choice set or by increasing or decreasing the utility of difference choices. Whether or not an individual actually changes his or her decision depends on how the new or altered choice now stacks up versus other options. If, for example, the city were to provide sidewalks from a residential area to a new commercial development on an infill site, would residents now walk there? If a new local store opens up, will residents now shop there rather than farther away? Of course, even if the change leads to a better alternative, an individual may still not change his behaviour. Travel behaviour is to a large extent habitual, and it may be that the individual does not acquire the information about the change or even stop to reconsider his original decision; "most choices are only infrequently evaluated" (Heggie 1978: 116).
Research Approaches Several promising research approaches have been underutilized in studies of the link between urban form and travel behaviour. These approaches have the potential to provide a significantly better understanding of the ways in which urban form influences travel choices and to provide insights beyond those provided by the approaches most widely used. Travel choice models, based on discrete choice theory, predict the probability of an individual choosing a particular alternative based on the utility of that alternative relative to others. In most cases, urban form is implicitly represented in the models, which include travel costs and destination attractiveness in the measurement of utility for mode or destination. In this way, these models demonstrate the importance of a limited set of urban form factors, given the
Travel Behavior-Land Use Interactions: An Overview and Assessment
233
influence of socio-economic characteristics. It is important to note, however, that the use of multinomial logit techniques does not necessarily equate with the development of a choice model; a true choice model reflects a sound theory about the contribution of the independent variables to the choice process. Thus, to build better travel choice models reflecting a greater range of urban form variables, researchers need to develop a better basic understanding of the ways in which urban form influences choice sets and the characteristics of the choices. Activity-based analysis constitutes "a more exploratory methodology... in which the researchers attempted to start with open minds and to examine travel behaviour in terms familiar to the traveler, in the wider context of his daily patterns of behaviour" (Jones , et al. 1991); this approach "enables the researcher to escape the constraints of a highly structured, but possibly wrong, established theory and way of thinking." It involves an even more complex treatment of both socio-economic characteristics and travel characteristics and a focus on constraints placed on an individual by his or her characteristics, those of his or her household, and the environment. Activity-based analysis has been associated with a more holistic approach to evaluating socio-economic characteristics, by combining them to define "roles" and "life-cycle stages." Similarly, rather than focusing on particular travel characteristics or even a set of simultaneous decisions, activity-based analysis often looks at the total pattern for an entire day. Relationships are not always tested statistically, but may be qualitatively evaluated. Unfortunately, few of the studies in this area have been designed to explore the role of urban form in travel behaviour, although a series of studies by Hillman, et al. (1976) in the U.K. may provide a useful model for ftirther research of this type. Many of the issues outlined above demand research approaches not commonly used in travel behaviour research. Qualitative methods, although they do not provide the sort of statistical testing and generalizability of the studies mentioned here, can be used to explore questions about the structure of choices, appropriate measures of urban form, and the role of perception, among others. Techniques such as focus groups, structured interviews, and open-ended travel diaries may begin to provide explanations for the behaviour underlying the differences in travel patterns that have been observed. Techniques from other disciplines, such as urban design, geography, and environmental psychology, might also contribute to the knowledge base. A strong dose of creativity would go a long way toward curing the shortcomings of research on the link between urban form and travel behaviour.
234
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
CONCLUSIONS What we know so far is that there are significant differences in travel behaviour in different kinds of communities. But we also know that the answers are not that simple, that the behaviour underlying those differences is complex and variable and the role of urban form not always straightforward. What we don't know about the role of urban form in this behaviour is due partly to data limitations (the problem of needing extensive data about urban form in an area for which extensive travel data are also available (Handy 1996a)), and it is also partly due to our limited understanding of travel behaviour in general. Although each study in this area contributes to the body of knowledge, by confirming earlier work or revealing another aspect of the relationship or pointing to a new complexity, the next generation of research must consider more exploratory approaches to developing a deeper understanding of the behaviour that underlies the observed link between urban form and travel patterns.
REFERENCES Appleyard, D. (1981). Livable Streets. Berkeley: University of California Press. Cambridge Systematics, Inc. (1994). The Effects of Land Use and Travel Demand Management Strategies on Commuting Behavior. Final Report. Prepared for the U.S. Department of Transportation and the U.S. Environmental Protection Agency. November. Cervero, R. and R. Gorham (1995). Commuting in Transit Versus Automobile Neighborhoods. Journal of the American Planning Association, Vol. 61, No. 2, pp. 210-225. Cervero, R. and B. Griesenbeck (1986). Commuting Behavior in Suburban Labor Markets: A Case Analysis of Pleasanton, California. Research Report UCB-ITS-RR-87-3, Institute of Transportation Studies, University of California at Berkeley. Cervero, R. and K. Kockelman (1996). Travel Demand and the 3Ds: Density, Diversity, and Design. Paper prepared through research support of the University of California Transportation Center. Cervero, R. and C. Radisch (1995). Travl choices in Pedestrian Versus Automobile Oriented Neighborhoods. Working Paper 644, Institute of Urban and Regional Development, University of California at Berkeley, July. Douglas, G. B. (1991). Planning on the Fringe: The Impact of Land Use Strategies on Congestion. Prepared for the Third National Conference Transportation on Solutions for Small and Medium-Sized Areas, October. Ewing, Reid, P. Haliyur, and G. W. Page (1994). Getting Around a Traditional City, a Suburban Planned Unit Development, and Everything in Between. Transportation Research Record 1466, pp. 53-62.
Travel Behavior-Land Use Interactions: An Overview and Assessment
235
Frank, L. D. and G. Pivo (1994). Impacts of Mixed Use and Density on Utilization of Three Modes of Travel: Single-Occupant Vehicle, Transit, and Walking. Transportation Research Record 1466, pp. 44-52. Friedman, B., S. P. Gordon, and J. B. Peers (1994). Effect of Neotraditional Neighborhood Design on Travel Characteristics. Transportation Research Record 1466, pp. 63-70. Gehl, J. (1971). Life Between Buildings. New York: Van Nostrand Reinhold, Translated by Jo Koch. Gilbert, G. and J. S. Dajani (1974). Energy, Urban Form, and Transportation Policy. Transportation Research, Vol. 8, pp. 267-276. Goodwin, P. B. and D. A. Henser (1978). The Transport Determinants of Travel Choices: An Overview. In David A. Hensher and Quasim Dalvi, eds, Determinants of Travel Choice, New York: Praeger Publishers, pp. 1-65. Handy, S. (1996a). Methodologies for Exploring the Link Between Urban Form and Travel Behavior. Transportation Research Z), Vol 1, No. 2, pp. 151-165. Handy, S. (1996b). Urban Form and Pedestrian Choices: A Study of Austin Neighborhoods. Transportation Research Record, No. 1552, pp. 135-144. Handy, S. (1996c). Understanding the Link Between Urban Form and Nonwork Travel Behavior. Journal of Planning Education and Research,Yo\ 15, No. 3, pp. 183-198. Handy, S. (1992a). How Land Use Patterns Affect Travel Patterns: A Bibliography, CPL Bibliography No. 279, Council of Planning Librarians, Chicago, IL. Handy, S. (1992b). Regional Versus Local Accessibility: Variation in Suburban Form and the Implications for Nonwork Travel. Unpublished PhD dissertation. University of California at Berkeley. Hanson, S. (1982). The Determinants of Daily Travel-Activity Patterns: Relative Location and Sociodemographic Factors. Urban Geography. Vol. 3, No. 3, pp. 179-202. Heggie, I. G. (1978). Behavioural Dimensions of Travel Choice. In David A. Hensher and Quasim Dalvi, eds, Determinants of Travel Choice, New York: Praeger Publishers, pp. 100-125. Hillman, M., I. Henderson and A. Whalley (1976). Transport Realities and Planning Policy: Studies of Frictions and Freedoms in Daily Travel. London: Political and Economic Planning, Vol. XLII, No. 567 (December). Holtzclaw, J. (1990). Explaining Urban Density and Transit Impacts on Auto Use. Presented by the Natural Resources Defense Council and the Sierra Club to the State of California Energy Resources Conservation and Development Commission, April 19. Holtzclaw, J. (1994). Using Residential Patterns and Transit to Decrease Auto Dependence and Costs. Natural Resources Defense Council for California Home Energy Efficiency Rating Systems, June. Jones, P. M., et al. (1991). Understanding Travel Behaviour. Aldershot, England: Gower Publishing Co., Ltd. Second Edition. Jones, P. M. (1978). Destination Choice and Travel Attributes. In David A. Hensher and Quasim Dalvi, eds. Determinants of Travel Choice, New York: Praeger Publishers, pp. 266-311. Kitamura, R., L. Laidet, P. L. Mokhtarian, C. Buckinger, and F. Gianelli (1994). Land Use and Travel Behavior. Report No. UCD-ITS-RR-94-27, Institute of Transportation Studies, University of California at Davis, October. Kulash, W. (1987). "Comparison of Activity Center Development Versus Sprawl," ITE Compendium of Technical Papers, 57th Annual Meeting, August.
236
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
Levinson, H. S. and K R. Roberts (1965). System Configurations in Urban Transportation Planning. Highway Research Record 64. Levinson, H. S. and F. H. Wynn (1963). Effects of Density on Urban Transportation Requirements. Highway Research Record!, pp. 38-64. McNally, M. G. and S. Ryan (1993). Comparative Assessment of Travel Characteristics for Neotraditional Designs. Transportation Research Record 1400, pp. 67-77. Mitchell, R. B. and C. Rapkin (1954). Urban Trajfic: A Function of Land Use. New York: Columbia University Press. Newman, P. W. G. and J. R. Kenworthy (1989). Gasoline Consumption and Cities: A Comparison of U.S. Cities with a Global Survey. Journal of the American Planning Association, Vol. 55, No. 1, pp. 24-37. Parsons, Brinkerhoff, Quade & Douglas. (1993). Volume 4A: The Pedestrian Environment. LUTRAQ Project, 1000 Friends of Oregon, Portland. Parsons, Brinkerhoff, Quade & Douglas, Robert Cervero, Howard/Stein-Hudson Associates, Inc., Jeffrey Zupan. Influence of Land Use Mix and Neighborhood Design on Transit Demand. Prepared for the Transit Cooperative Research Program, Transportation Research Board, National Research Council, TCRP Project H-1, March. Rabiega, W. A. and D. A. Howe (1994). Shopping Travel Efficiency of Traditional, NeoTraditional, and Cul-De-Sac Neighborhoods with Clustered and Strip Commercial. Paper presented at the Annual Meeting of the American Collegiate Schools of Planning, Tempe, AZ, November. Real Estate Research Corporation (1974). The Costs of Sprawl: Environmental and Economic Costs of Alternative Residential Development Patterns at the Urban Fringe. Prepared for CEQ, HUD, EPA. Washington, DC: U.S. GPO. Rutherford, G. Scott, Edward McCormack, and Martina Wilkinson (1996). Travel Impacts of Urban From: Implications From An Analysis of Two Seattle Area Travel Diaries. Paper prepared for the Travel Model Improvement program Conference on Urban Design, Telecommuting and Travel Behavior, Williamsburg, VA, October. Schneider, J. B. and J. R. Beck (1973). Reducing the Travel Requirements of the American City: An Investigation of Alternative Urban Spatial Structure. Research Report no. 731, U.S. Department of Transportation. Schuler, H. J. (1979). A Disaggregate Store-Choice Model of Spatial Decision-Making. Professional Geographer, Vol. 31, pp. 146-156. Schumacher, T. (1986). Buildings and Streets: Notes on Configuration and Use. In Stanford Anderson, ed. On Streets, Cambridge: MIT Press. Transit Cooperative Research Program (1995). An Evaluation of the Relationship Between Transit and Urban Form. Research Results Digest, No. 7, June. "Urban form" is used here as a more inclusive term than "land use": urban form includes both macro and micro characteristics of both land use patterns (what activities are located where) and the transportation system (the links between activities). Design elements are an important part of urban form.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
11
COMPARATIVE NEIGHBORHOOD TRAVEL ANALYSIS: A N APPROACH TO UNDERSTANDING THE RELATIONSHIP BETWEEN PLANNING AND TRAVEL BEHAVIOUR
Roger Gorham
INTRODUCTION The intuitive argument that land-use and transportation are rekted is not new^, but it is the focus of renewed debate. Many environmentalists, urban designers, and planner-activists contend that, for a multitude of reasons, automobile dependency has been built into people's lives in the United States through poor urban planning and land-use decisions. The proof, these New Urbanists say, is in the statistics: America has the highest automobile mode share of any industrial country, and Americans travel more in the course of a day than any other people, a function of the extent to which the automobile has shaped urban settlement patterns. What is needed, they argue, is a re-orientation of urban development toward more compact and walkable cities, in order to encourage more non-motorized and public transit trip-making. These assertions seem to many to be intuitively true, yet researchers continue to find it difficult to prove a causal link between urban form on the one hand, and how people travel around on the other. What is missing is an understanding of the point of influence. Does the form of compact cities itself induce people to travel less by car, or is it that people who generally populate compact cities—those who live, work, or otherwise recreate there—have a preference
238
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
or some other constraint which causes them to travel less? Research on the subject has become increasingly sophisticated. Much of the early focus of this research was related to the effects of urban density on aggregate indicators such as modal shares, aggregate gasoline consumption or total VMT (Pushkarev and Zupan 1977, Newman and Kenworthy 1989). Subsequent studies have attempted to describe the urban environment albeit with quantitative measures - in more meaningful and sophisticated ways than simply density would allow. Recent studies have included a wide range of variables to represent different aspects of urban form, such as jobs-housing balance (Ewing et al. 1994, Frank and Pivo 1994, Messenger and Ewing 1996, Cervero 1995), the nature of the street network (Handy 1992, Cervero and Gorham 1995, Crane and Crepeau 1996), quantitative measures of degrees of mixture of use (Frank and Pivo 1994, Messenger and Ewing 1996, Crane and Crepeau 1996, Kockelman 1996) and general indicators of the quality of the pedestrian environment (PBQD 1993). The motives, methods, and conclusions of all these studies vary, but they share a common trait: a restless drive to displace density as the proxy for all things "land-use" and replace or supplement it by attributes which more adequately describe land-use or urban form. Some of these attributes are actually quite sophisticated. Measures such as jobs-housing balance or degrees of mixture of use are difficult to quantify, and not entirely lucid to the lay person. For example, a simple ratio does not suffice for the former, since there is a qualitative element whicji must be taken into account—of interest is the balance between particular kinds of jobs and housing which will contain people suited for those jobs. In addition, the degree of mixture-of-use is often represented as an entropy index of some kind (Frank and Pivo 1994, Kockelman 1996), not at all an accessible concept for the uninitiated. Data requirements to generate such indicators also require a certain degree of sophistication, utilizing Geographic Information Systems (GIS). As the cost of desk-top computing continues to fall, more sophisticated GIS applications — and more sophisticated quantitative formulations of urban form phenomena — are likely to be used in research into the travel behaviour/urban form nexus. For this reason, it is crucial that researchers keep their goals clear. We are interested in determining the nature of the interaction between the form of human spatial settlements and the travel behaviour of people who live in these various settlements. It is the human element which creates the connection; people react to the built environment, take their cues from it, and engage in a number of behaviours - of which travel is only one - which make them comfortable. Densities, entropy indices, and jobs-housing balance ratio do not induce travel behaviour. These are simply tools and methods our profession has and is currently developing to try to measure those phenomena which we believe people are reacting when they make their
Comparative Neighborhood Travel Analysis: An Approach to Understanding 239 travel and lifestyle choices. The challenge for researchers is to represent and interpret what it is that people do perceive in the urban environment which influences their travel decisions. No doubt, aspects of urban form, such as population or building density, degrees of mixture of use, amount of shading, or "griddedness" of the street network, all contribute to the total experience of travel, and researchers should continue working to understand how they influence travel behaviour. But there also needs to be efforts at understanding how the totality of the experience of a neighborhood - both in and of itself, and with respect to the region in which it lies - influences traveler perceptions and behaviour. Perhaps researchers on urban form and travel behaviour should be turning for insight to the pioneering studies of Kevin Lynch and Donald Appleyard on city imaging and perceptions (for example. Lynch 1970). One fairly accessible method of simulating urban perceptions is to use generalized categories of neighborhoods that serve to encompass a range of attributes which group and define neighborhoods according to different criteria. These neighborhood typologies are, by definition, inaccurate descriptions of neighborhoods, since attributes which identify a neighborhood are never exactly identical in any two - assuming for the moment that even the concept "neighborhood" can be discretely defined. Nevertheless, typologies might actually be a more accurate way of describing human perception of neighborhoods — and associated travel behaviour therein ~ since people may associate different neighborhoods simply because they group them together typologically, whether they actually share similar attributes or not. A neighborhood type, then, is an abbreviation for a set of urban^rw characteristics for a given geographic unit of urban space, that is, formal characteristics at the most local level. They not only identify physical qualities about these units; they may also help to identify them in a hierarchical structure of space perception ~ a hierarchy of use-concentration, for example ~ by residents of a region or travelers within it. The way these types inter-relate spatially ~ that is, the physical relationships between neighborhoods of distinct types ~ begins to describe the structure of the region. I maintain this distinction ~ between neighborhood (micro) form on the one hand and regional (macro) structure on the other ~ throughout this chapter, because I believe it is an important one. Previous research (Cervero and Gorham 1995, Handy 1994) has suggested that regional structure may be a more influential aspect of urban settlement pattern on travel behaviour than neighborhood form, yet it has thus far been the focus of less interest. Previous use of typological approaches related to urban form and travel have focused primarily on household energy consumption and carbon emissions. In a French study of Grenoble, for
240
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
example (Hivert and Le Dily 1994), neighborhoods in the region were grouped into three categories (City, Inner Suburb and Outer Suburb) and household travel-related energy and carbon "budgets" compared among the types. A similar study in Paris (Gallez 1995) used a similar methodology, with an eight category typology. In these studies, however, the typology corresponds more to zones which have been differentiated based on distance to city center than on distinctive types per se.
METHODOLOGY AND BACKGROUND This study applies a typological approach to two regions, the San Francisco Bay Area (SFBA) and the Stockholm Metropolitan Region (SMR), as a way of examining the interaction between urban form and travel behaviour in a comparative context. The initial purpose was simply to examine whether similar neighborhood "types" show common travel behaviour characteristics among residents who live in two distinct countries and cultures. The regions were chosen because they share many qualities which make them well suited to compare. They differ however, in a key respect: in the author's opinion, the San Francisco Bay Area represents a region which has had minimal planning intervention at the regional level, and as such is relatively ahierarchical in structure, while the Stockholm Metropolitan Region has had a tradition of strong urban and regional planning, and structural hierarchies between the neighborhoods have been maintained. (For a detailed discussion about both the history of planning in the two regions and the nature of regional structure, see Gorham 1996). Planned development in the SMR since the Second World has focused on new communities, integrated into the new heavy rail system to allow for "centralized decentralization", while still maintaining the importance and dominance of central Stockholm to the region (see Cervero (1995) for a detailed description of these "Satellite Suburbs"). The present case study compares a snapshot of travel behaviour between the SMR and the SFBA, as recorded by regionally administered household travel surveys in the late 1980s and 1990. SMR data come from the 1986/87 Resvaneundersokningen (RVU). This was a one year household survey, in which each participant household maintained a travel diary for the period of one week. The administration of this survey consisted of a household sampling-base component, an individual sampling-base component, and a single-person household samplingbase component. For comparability, I have used only the household sampling-base component, and only weekday responses. The SFBA data come from the 1990 Bay Area Travel Survey (BATS), a household samplingbase survey conducted in late spring and early autumn of 1990. This survey was a telephone
Comparative Neighborhood Travel Analysis: An Approach to Understanding 241 survey recording travel behaviour from household members for either a single-day or several days within a five-day period. Weekend travel was not sampled. For comparability with the RVU, only the single-day survey results were used. Additionally, both data sets have been filtered to insure maximum comparability between them. The typology was derived from close historical and formal evaluation of the two regions. Neighborhood types were developed to correspond to the period of initial urban development (with the exception of outlying neighborhoods which are predominantly non-residential, which were grouped into a separate category regardless of when they were initially developed.) The types used for this analysis were: Center City neighborhood Urban Cottage/Apartment District^ Tower-in-the-Park (post War, mostly high-rise, housing district) Suburban Residential (single family house, open lot) Outlying (suburban or rural) Town or Village Center Rural Non-Residential (outside of CBD) ~ i.e. industrial or office use Every census tract in both regions ~ in some cases, an even finer grain of geography ~ has been assigned to one of these neighborhood types. (Detailed definitions of these types, as well as some comments on limitations of the methodology, can be found in the appendix.) Survey respondents were then linked to a neighborhood type based on the census tract of their primary residence, which was included in the survey data sets.
RESULTS
Preliminary Tests Limitations in both the data and the census geography used to derive the neighborhood types (see appendix) prevented us from being able to actively control for the effects of confounding influences, such as income or lifecycle stage. Nevertheless, one of the first steps in the analysis was to examine these factors and test for correlation between them and neighborhood type. The lifecycle categories used were: • One adult, under 35, no children
242
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Two or more adults, referent under 35, no children One adult, with children (persons under 18) Two or more adults, with children One adult, 35 or over, but under 60, no children Two or more adults, referent between 35 and 60, no children Third agers (referent between 60 and 75) Elderly
Chi square tests showed that, in both cases, there was statistically significant but weak correlation between neighborhood category and lifecycle stage at the 99 percent confidence level? Next, income distributions were examined and compared for each of the principal neighborhood types, using income data from the household travel surveys. It was found that, perhaps not unexpectedly, incomes were slightly more stratified in City Center neighborhoods in the SFBA than in the SMR. Incomes in the Suburban Residential neighborhoods in the SMR were slightly skewed toward the upper two quintiles, while those in the SFBA were more evenly distributed. Of particular interest were the distributions of incomes in Town Center neighborhoods, shown in Figure 1. • Stockholm U San Francisco
i
^ •
1 1 1 1i 1 •_ \
•
"""in
1 1
Figure 1 Distribution of Incomes in Town Center Neighborhoods Source: Bats, 1990 Stockholm RVU, 1987
Comparative Neighborhood Travel Analysis: An Approach to Understanding 243 If the income distribution in Town Centers were exactly the same as income distributions for the region as a whole, the bars in this figure would be exactly even. The results show that Town Centers in the SMR have a mix of incomes that look more like that of the region as a whole than Town Centers in the SFBA. The former have a slight predominance in the middle incomes (the lowest and highest percentiles are each below 20%), while the latter, of lower income people. It is difficult to know whether and how these patterns affect the observed travel behaviour for each neighborhood type, but their impacts are probably weak enough that the examination by neighborhood type still reveals real relationships.
Trip Generation On average, residents in the SMR make 3.0 trips per day, while those of the SFBA average around 2.4. There are small variations among neighborhood types of each region, but not in any easily comprehensible pattern. Consequently, it is not possible to draw any inferences about neighborhood form and trip generation per se from the available data. It is with other indicators, therefore, that any influence of neighborhood type on travel behaviour will be detected.
Average Trip Distance Average home-based trip distances by neighborhood category of residence type are presented in Figure 2. The results for the most part conform to intuitive expectations. Residents of the SMR make significantly shorter trips than residents of the SFBA, as do Center City and Cottage District residents of both regions relative to everyone else. The average trip distance of a Center City resident in SMR and the SFBA is 66% and 63%, respectively, of the region-wide average. Trip distances in outlying Town Centers in Stockholm are, interestingly, about 15% higher than the region-wide average, although they are slightly lower than the region-wide average in the San Francisco Bay Area. These differences in the average trip distances are statistically significant at the .05 alpha level. (Stockholm Anova F Statistic = 38.85, DF = 9, Bay Area Anova F Statistic = 119.70, DF=6). In both regions, almost all paired differences in average trip distance are significant.^
244
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Figure 2 Average Distance per Trip by Neighborhood Type Source: BATS, 1990 Stockholm RVU, 1987
One would expect to glean more information about the relationship between neighborhood type and distance per trip by examining particular trip purposes — that is, those which are predominantly local in character. Table 1 summarizes the ratio of distance per home-based trip relative to the region-wide mean for four trip purposes considered to be relatively local in nature: convenience services, convenience shopping, eating out and groceries. (The category "Convenience Shopping" is not available for the BATS data, so "Nonfood Shopping" has been used instead."^) 1 1 1 1
Stockholm Convenience Services Convenience Shopping Eat Out Groceries
Center City 0.89 0.63 0.55 0.50
Cottage District 0.75 0.57 0.88 0.62
Tower/Park 0.90 1.08 1.23 0.98
Suburban 0.92 1.32 1.17 1.16
Rural 1.73 1.83 2.06 1.74
Bay Area Convenience Services Nonfood Shopping Eat Out Groceries
0.66 0.54 0.59 0.55
0.77 0.84 0.75 0.76
#N/A #N/A #N/A #N/A
1.04 1.00 1.07 1.02
1.14 1.45 1.07 1.54
Town Center | 1.28 0.72 0.89 1.26 1.09 1.15 1.08 1.17
Table 1 Ratio of Average distance per Trip by Neighborhood Type Relative to Region-Wide Mean Source: BATS, 1990 Stockholm RVU, 1987
\ 1 1
Comparative Neighborhood Travel Analysis: An Approach to Understanding 245 These data show that trip distances for residents of Center City and Cottage District type neighborhoods are significantly lower than the regionwide averages. For example, a resident of central San Francisco, Oakland, San Jose, or Stockholm travels half the distance as the regionwide average to get groceries. However, the ratios of trip distances for these trip purposes among residents of Town Centers suggest that the proximity advantages postulated by the New Urbanists either do not exist or are of little value to the people who live there. Only in the SMR do residents of outlying Town Centers make shorter home-based local trips than the region-wide average, and even these only for convenience shopping and eating out. Otherwise, average trip distances in both regions for residents of Town Centers are either as long or longer than the regionwide average, hi fact, average trip distances for convenience services (banking, dry cleaning, etc.) are shorter in Suburban Residential neighborhoods than they are in Town Center neighborhoods, a somewhat surprising finding in the case of Stockholm. In the SFBA, Town Centers seem to provide no distance advantage whatsover to local activities over their Suburban Residential counterparts. This cursory examination of the two regions suggests the dominant influence of regional structure on trip-making patterns, even for "local" trips such as convenience shopping or services. Trip distances in outlying Town Centers seem more influenced by the "outlying" than by the "Town Center" aspect of the name. It is possible that splitting local from more regional trip purposes is not entirely capturing true variation in trip-making patterns attributable to neighborhood type, since it does not take into account the range of choices available to residents in the different types. In the SFBA, for example, the sheer dearth of people actually living in outlying Town Centers probably impacts the range of supply choices available to those who do. If so, the results indicate an example of the dominance of regional structure over local neighborhood form.
Modal Shares A second aspect of travel behaviour as affected by urban form or neighborhood type is the structure of modal activity. In this section, I will be referring to trip as opposed to travel mode shares^, since the former is more of a reflection of behavioural choices. Figure 3 shows mode shares for non-motorized trips, and Figure 4, those for public transit trips. These two figures reveal the strong difference in proclivity to use alternatives to the car in the two regions. The difference in reality is probably even more pronounced than the data suggest; the Stockholm RVU was a year-long, rolling survey, whereas the BATS survey was administered in the late spring and early autumn — the period of the Bay Area's best weather! Taken together, these figures suggest that there are some modest benefits on modal structure of outlying Town
246
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Centers, particularly in the SMR. Residents of outlying Town Centers there make about 43% of their trips by car, compared with 55% of all trips by car for residents of Suburban Residential neighborhoods in the SMR. Most of this benefit is attributable to non-motorized vehicle trips; there is only a modest gain of about 1 percentage point in public transit mode share when one moves from Suburban Residential to outlying Town Center type in the SMR.
I esFBA I
HH
Center City
Cottage District
1
Suburban Residential
Figure 3 Mode Shares of Non-Motorized Trips Source: BATS, 1990 Stockholm RVU, 1987
Suburban Residential
Town Center
Figure 4 Mode Shares of Public Transit Trips Source: BATS, 1990 Stockholm RVU, 1987
Comparative Neighborhood Travel Analysis: An Approach to Understanding 247 This finding, plus the relatively modest, almost minimal, modal advantages of outlying Town Centers in the SFBA serve once again to highlight the importance of regional structure. In the SMR, neighborhood form can and does influence the proclivity to walk, but regional structure is a much more important aspect of modal choice. Consequently, neighborhood type of residence has relatively little influence on transit trips in the SMR. In a region such as Stockholm, where a great deal of planning emphasis was placed on transit access into the center city, this probably means that whether a resident is facing a particularly pleasant walk to the transit station has little bearing on whether he or she decides to use transit. In the SFBA, non-motorized trip shares are only slightly higher in outlying Town Centers than in Suburban Residential neighborhoods. Again, this suggests that structural relationships among activity centers region-wide is having a stronger effect than localized elements of the uban landscape. We get an even stronger sense of the importance of regional structure by comparing modal shares between the two regions relating to trips to and within the Central Business Districts (CBD), as shown in Tables 2a and 2b.^ Modal Shares of AH Trips (%) Mode SFBA SMR 28.4 Walk 1 9.5 Bike 1.3 5.2 Transit 4.8 18.7 35.2 Car Driver 68.8 Car Passenger 14.4 11.5 Other 1.1 1.1
| 1
Modal Shares of Trips to or within the CBD Mode SFBA SMR 31.1 32.1 Walk 0.5 2.3 Bike 29.9 33.3 Transit 29.2 25.6 1 Car Driver 8.4 5.8 Car Passenger 0.8 0.9 Other % of Trips to / 27.9 3.8 1 in CBD
Tables 2a and 2b Modal shares of trips, comparing all trips with those to the CBD Source: BATS, 1990 Stockholm RVU, 1987 When only trips to and from the CBD are taken into account, modal shares for the two regions are actually quite similar. However, as the table shows, only about 4% of all trips are CBDrelated in the SFBA, compared with nearly 30% in the SMR. The structure of the region itself ~ the pattern of trip origins and destinations — seems to be a determining factor in mode choice.
248
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Trip Duration Another important aspect of travel behaviour — perhaps the most important, from the point of view of the transport consumer - is trip duration. One might speculate that, by and large, the trip durations one observes from a household travel survey are those which the traveler has assessed to be the shortest for the ensemble of all the trips he or she makes during the day or part of the day. If so, then differences in trip durations can be interpreted as access time differences to various activities, goods, and services among the different neighborhood types. Figure 5 shows, that, on average, residents of the more urban neighborhoods of the SFBA, and residents of the rural parts of the SMR have longer access times than residents of other parts of their respective regions. In fact. Cottage District residents in the SFBA take on average about 5 minutes more per trip than residents in Suburban Residential neighborhoods. All in all, though, trip durations are remarkably similar both between the regions and across neighborhood types. Trip makers probably have strong expectations of how long and short a given trip should be; these expectations might be the controlling factor in the decision of when, where, and how to make a trip. Trip duration data, however, are notoriously inaccurate. Respondents often round when asked either how long a trip took, or when they left and arrived.
Figure 5 Average Duration of Trip by Neighborhood Type. Source: BATS, 1990 Stockholm RVU, 1987
Carbon Budgets The product of the number of trips an individual makes per day, the distance per trip, the proportion of trips made by different modes, and a carbon emissions factor for each of those
Comparative Neighborhood Travel Analysis: An Approach to Understanding 249 modes, makes up what can be called a "carbon budget" for that individual. It represents the amount of carbon released into the atmosphere as the sum total of transportation decisions that an individual has taken. A comparison of the carbon budget for an average resident of each of the neighborhood types is shown in Figures 6 and 7, the former representing the SMR, and the latter the SFBA. The figure shows the average amount of carbon emitted in grams per day. The vertical lines represent the 95% confidence interval around the reported mean. These are shown to highlight the difference in precision of the final result that can be calculated fi*om the number of observations in each neighborhood type. Details on the derivation of the carbon emissions factors are shown in the appendix.
Q te 1600
I 1400 a>
OL
S; 1200
a
940
^ 1000 6 ui c 800 o 5
614
600 400
^t5L
ZZ2
Center City
Cottage District
-484
428
549
^(M-
200
Tower / Park
Suburban
Rural
Town Center
Nonresidential
Figure 6 Average Daily Travel-Related Carbon Emissions per Person and Associated Confidence Interval, by Neighborhood Type, SMR Source: Stockholm RVU, 1987 and Storstockholm Lokaltrafik
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
250
o
© 3500
^8Ha_
I I
^256^ 2000
% 1500l-^5a
Center City
^mr
2tfl5
^mr
1501
Cottage District
Suburban
Town Center
Nonresidential
Figure 7 Average Daily Travel-Related Carbon Emissions per Person and Associated Confidence Interval, by Neighborhood Type, SFBA Sources: BATS, 1990, FTA Section 15 Data, Bell, Gupta, et al. 1995 Anova tests indicate that the differences among the neighborhood types are significant at the .05 alpha level (F statistic for the SMR with 9 df: 20.59. F statistic for the SFBA with 6 df: 58.42). Residents of Town Centers seem to have lower daily travel-related carbon emissions than residents of Suburban Residential neighborhoods in both regions. However the spread of the confidence interval for Town Centers is significantly wide to introduce some doubt as to this difference; at any rate, it is statistically significant for the SMR but not for the SFBA at the .05 alpha level. What is clear from the data is that the average resident of the SFBA emits over four times as much carbon into the atmosphere in the course of his or her travel day than an average resident of the SMR.
Comparative Neighborhood Travel Analysis: An Approach to Understanding 251 Neighborhood Type Center City Gbttage District Tower / Park Suburban Rural Town Center Nonresidential
SFBA 0.59 0.71 #N/A 1.06 1.33 1.01 0.95
SMR 0.64 0.66 0.86 1.23 1.88 0.961 1.10
Table 3 Ratio of Average per Capita Carbon Output per by Neighborhood of Residence to Regional Average Table 3 summarizes the relative average daily carbon outputs by residents of the different neighborhoods for the two regions. It shows the ratio of carbon output for each neighborhood relative to the regional mean, for both the SFBA and the SMR. The table suggests that while average carbon output is significantly higher in the SFBA than in the SMR, the relative environmental benefits of Center City and Cottage District neighborhoods taken together are roughly the same. Carbon emissions in Suburban Residential neighborhoods are relatively worse in relation to the regional average in the SMR than in the SFBA, but this is primarily because of the larger proportion of people living in Suburban Residential neighborhoods in the SFBA. These considerations suggest that efforts to influence the structure of the region may be an effective means to effect change in household travel energy consumption, but that changes in localized, neighborhood form are likely to only have a marginal effect. Nevertheless, the two may work together in a complementary way.
CONCLUSION The indicators examined in this study all suggest that there are some similarities in travel behaviour between "equivalent" neighborhood types in the two regions examined, but that these similarities are strongly circumscribed and dampened by aspects of the regional structure. Many of these indicators show that travel behaviour in the various neighborhood types conforms to intuitive expectations, and to the benefits which advocates of compact urban form attribute to them. However, a frequent exception seems to be outlying Town Centers, which generally show little consistent benefit over Suburban Residential neighborhoods in the indicators at which we have looked. Average trip distances in Town Centers in the SMR were actually higher than the region-wide average. For local, home-based trips, there was no proximity advantage whatsoever for Town Centers in the SFBA, and the results were mixed in the SMR.
252
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Outlying Town Centers have more of an advantage in terms of modal shares. Both regions show residents with greater proclivity to walk, bike, or use public transit in outlying Town Centers than in Suburban Residential neighborhoods, although significantly less proclivity than in Center City or Cottage District neighborhoods. Still, in a region such as San Francisco, the greater proclivity to walk, bike, or use transit, is swamped by the relative unimportance of nonmotorized modes. 87% of trips region-wide are by car, while 84% of trips in Town Centers are by car. The travel activity of residents of Center City and Cottage District neighborhoods results in significantly lower average weekdaily carbon emissions than the regionwide averages. This lower carbon-emission benefit appears to extend to residents of outlying Town Centers in the SMR as well, but it is a weak benefit. Outlying Town Centers in the SFBA have no carbonemissions advantage, however. All of these results point to the importance of regional structure as a determinant of travel behaviour patterns. In many cases, the particular form of a neighborhood is almost insignificant. In the Stockholm Metropolitan Region, where concerted planning effort was taken to insure that the Central Business District retained its importance in the region (Sidenbladh 1964), the choice to use public transit varies relatively little according to neighborhood type, compared with the SFBA. In both regions, the relationship among neighborhoods is more important than the form of any individual neighborhood. For this reason, modal behaviour between the two looks startlingly similar when we isolate trips which reflect structurally equivalent relationships — that is, trips to and within the CBD. There is minimal variation among the neighborhood types in terms of trip duration. Most trips are around 21 minutes in length, except for trips by urban residents in the SFBA and those by rural residents in the SMR, which are about 5 minutes longer. Regional structure may also be having an important impact here, too. Travelers may be adjusting their travel patterns — when, where, and how to travel — according to whichever choices will allow them to access goods, services and activities in an average of 21 minutes. This, in turn, will depend on what is feasible given overall regional structure and the level of accessibility of different origins and destinations.
Comparative Neighborhood Travel Analysis: An Approach to Understanding 253 FURTHER WORK The results presented raise a number of interesting questions. First, the behaviour of travelers who reside in outlying Town Centers is not at all clear cut. I have speculated that this is, at least in part, because of the overwhelming importance of the structure of the region — the degree to which it can be said to be hierarchically organized. Yet it may also have to do with the inappropriateness of census geography to designate neighborhood types. A more apt method of using neighborhood typologies is to designate them prior to conducting the household surveys, and then using these types to segment households for the study. This method would allow the researcher to do focused studies specifically about urban form per se. Geography could then be delineated specifically for the purpose of this study. A second question for further study is whether the delineation of neighborhood typologies is, in fact, a potential means for understanding better regional structure and its impact with local neighborhood form. One potentially promising method is to construct trip origin-destination matrices using neighborhood types of the end points of the trip. (In the case of the neighborhood categories used in this study, this would be a 7 X 7 matrix.) Indicators similar to those used in the present study would then be examined for each of the cells. The goal would be to see if identifiable patterns emerge. Finally, it would be interesting and important to expand on the typologies presented here both methodologically and definitionally. Methodologically, the generation of the typology might focus on empirical research on imaging studies and mental mapping of city residents. Definitionally, these types could be expanded or refined to include far less crude definitions of neighborhoods, and to capture finer nuances between different types of neighborhoods. In short, a more methodological approach from start to finish might reveal even more interesting relationships than the present study has uncovered.
APPENDIX The Center City type refers to the very urbanized areas which form the core of the central cities, and which form the earHest parts of the regions to urbanize. It is defined as Stockholm Inre stade, central San Francisco (northeast quadrant, including the Mission, and going out to about Araguello Street to the west), and the Central Business Districts (CBDs) of Oakland and San Jose, as well as those areas immediately adjacent to the CBDs. These are areas that can be said to have been urbanized before or during the first phase of rail transit deployment (roughly prior to 1900). About 16 % of residents of the Stockholm region and 7% of residents of the San Francisco region live in this neighborhood type.
254
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
The Cottage or Apartment District type —characterized by urban neighborhoods in more outlying parts of the central cities ~ largely refers to those areas developed during the main period of expansion of the streetcar and inter-urban network beyond the original city boundaries. These neighborhoods were often developed as "Additions", or appendages to the existing plat, with the pressure for the development of these previously rural areas coming from the expansion of streetcar and inter-urban lines themselves. In both regions, these neighborhoods were often developed by the transit companies themselves. (For an overview of this period in the United States, see Warner, 1971). These neighborhoods were identified defmitionally by examining contemporary and historical maps of the two regions. The street patterns of Center City and Cottage Districts tend to be rectilinear; however, the built densities of the latter tend to be lower, and the degree of mixture of land-uses more sporadic. Often, land-uses were mixed (residential and commercial) only at transit hubs, such as major stations or where two streetcar lines crossed. Cottage Districts in both regions tend to have a significant number of what were originally owner-built cottages — California bungalow-style houses — as well as apartment houses with significant frontage along the street and ample pedestrian permeability. Residents in this category constitute about 11% and 5% of the San Francisco and Stockholm regions, respectively. The Tower-in-the-Park neighborhood type refers primarily to the planned, postwar suburban high density housing district found in the Stockholm region. These districts generally have groups of mid- or high-rise buildings clustered around either greens, common areas, or parking lots. These common areas were usually planned and developed at the same time as the buildings themselves. Unlike apartment buildings of the Cottage and Apartment District type, these buildings generally do not form a continuous frontage along the street, and have a low degree of pedestrian permeability. In fact, pedestrian entrances are usually set back from the main street. Most of the districts of the "Satellite Suburbs" (e.g. Vallingby, Farsta) tend to be of this type, but there are many examples from all over the region which are not satellite suburbs. For the purpose of this research, this neighborhood type is not present in the San Francisco Bay Area. To be sure, there are examples of Tower-in-thePark type neighborhoods in the region (Saint Mary's Houses in San Francisco, for example). The census geography, unfortunately, does not allow these areas to be isolated statistically. This neighborhood type houses about 20% of the Stockholm region's population. The Suburban Residential category refers to neighborhoods consisting mostly of single-family dwellings, oriented primarily, if not exclusively, to private motorized transportation. Streets tend to be non-continuous and curvilinear, and the street networks in these neighborhoods tend to follow fairly rigid hierarchies of street importance. Unlike those in the Urban Cottage Districts, houses in Suburban Residential areas tend to be less orientated toward the street. This type represents the represent the predominant neighborhood type in San Francisco (about 3/4 of the region's population), but there are quite a number of examples in Stockholm as well (about 1/3 of the region's population). Rural neighborhoods are those that remain primarily agricultural or undeveloped, containing 5% and 3.5% of the San Francisco and Stockholm regional population respectively. The outlying Town Center type generally refers to one of two types of areas. For the most part, it refers to areas that were outlying business, cultural, retail, and administrative centers prior to the motorized expansion of the
Comparative Neighborhood Travel Analysis: An Approach to Understanding 255 regional agglomeration. We define these neighborhoods historically for consistency; a "Town Center" neighborhood may not have retained its status as a business, cultural, retail, and/or administrative center, but it would have filled these functions at the time of development, prior to and through the Second World War. The second type of area to which the "Town Center" type can refer is an area which has been spQcificaWy planned to be business, cultural, retail and administrative centers. These are referred to in Swedish documents as "centrums" of the satellite suburbs. For the purposes of this study, these types of town centers are defined as the centrums, plus those residential census tracts immediately adjacent to them. There are currently no equivalents of this second type of "Town Center" in the San Francisco Bay Area. Several caveats to the methodology presented above are in order. First, the use of census tracts in both regions is a less-than-ideal, but necessary, method of partitioning neighborhoods for the purpose of the typology. Census tracts are the primary means of identifying where in the region the household resides for both data sets. In the SMR, they tend to be drawn small by discrete land-use type. However, in the SFBA, they are often either geographically large, or encompass too many different land-uses and urban forms to be of use. Where this is the case, we tried to use block groups, the next smaller unit of census geography. Unfortunately, the BATS data is inconsistent in reporting block groups for all respondents, so, with about 202 data points, it was not possible to identify precisely the neighborhood type of residence. In some cases in the SFBA, even the block group proved too large an area to adequately isolate a distinctive neighborhood type. In this case, the typology was assigned probabilistically where this was feasible. In a few cases, census tracts were left unassigned. Households in the BATS data were then assigned to a neighborhood-type based on the census tract/block group of residence. A second caveat related to the methodology discussed in this chapter relates to data weighting. In the ideal situation, the neighborhood categories would be used to segment a program of primary research. The best way to simulate this ideal, given that the surveys had already been conducted, would be to re-weight the data based on either the neighborhood category definitions or the census tract itself. The existing weights correct for sampling bias to the "superdistricf' level in the SFBA, and the level of municipalities in the SMR. Unfortunately, time and resources did not allow for such a reweighting of the data. Short of actually reweighting the data, is not possible to estimate either the direction or the magnitude of any bias resulting from the analysis. A third caveat is that the data come from two distinct administrations of two distinct surveys, for different reasons, using different surveying methods, and engendering different underlying definitions and assumptions. Where possible, these differences have been corrected for, but many of these differences — such as sampling technique and sampling frame ~ simply mean that at some level, the data are not entirely compatible. The carbon emissions factors developed for the carbon budgets express grams of carbon emitted per passenger kilometer traveled for each mode. In the case of the San Francisco Bay Area, this was actually specifiable for each mode and transit operator. These carbon factors were then multiplied by the trip distance, for the appropriate mode, giving an overall carbon emission figure for a given trip. These, in turn, were summed for each individual, and the sums averaged over all individuals, both collectively and by neighborhood type, giving the comparative average emissions figures discussed in chapter 5. The carbon factors were calculated for the collective (public) transit modes as follows:
256
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Cp = CvO V
^PKM
o=-^ Y^VKM V
where Cp = carbon (passenger) factor Cv = carbon(vehicle) factor for each mode and operator O = vehicle occupancy e = IPCC carbon emission factor, by fUel type, per terajoule of energy consumed E = energy consumed, in terajoules PKM = reported passenger kilometers VKM = reported revenue vehicle kilometers V = vehicle of a particular fuel type, by operator The IPCC carbon emission factor refers to published emissions factors from the Intergovernmental Panel on Climate Change. (IPCC, 1995). For the United States, the remainder of the input data comes from published figures of fuel consumption (to calculate energy consumed), revenue vehicle kilometers, and passenger kilometers, as reported by the United States Department of Transportation, under Section 15 reporting requirements. These summary statistics are reported for each individual (major) operator, and for each mode (i.e. diesel bus separate from trolley bus.) For modes powered by electricity (BART, Muni, etc.), the carbon emission factor e is a weighted average of IPCC emissions factors, based on the primary fuel mix of electricity production in California in the year 1990. (lES, 1996) For Sweden, the inputs come from data provided by the SL on vehicle kilometers, passenger kilometers, and fuel consumption for the years 1986 and 1987, the years of the RVU survey. For modes powered by electricity, the carbon emission factor e is a weighted average of IPCC emissions factors, based on the primary fuel mix of electricity production in Sweden in the years 1986 and 1987. (IBS, 1996) Carbon factors for automobiles were calculated as follows. For the Bay Area Travel Survey, in which model year of the vehicle used is available, vehicle emissions factors were developed by car model year, from the MOBILES emissions model, as reported in Bell, Gupta, and Greening (1995). Carbon content was extracted from the published figures in the source, which list emissions by major greenhouse gases and local pollutants. Carbon content, therefore, is probably slightly underestimated for automobiles, since it was not possible to include carbon from non-methane hydrocarbons in the total. Passenger emissions factors were calculated based on the vehicle occupancy of the individual trip, which is reported in the BATS data. In the Stockholm RVU data, model years of vehicles used on each trip is not reported. Consequently, I was obliged to use a Swedish fleetwide vehicle emissions factor, as given in lES (1996). These vehicle emissions factors were then divided by the overall Stockholm travel load factors of 1.27, to come up with an automobile carbon (passenger) factor. The final calculated carbon factors for both regions are presented in Table 4. They are presented in order, from the
Comparative Neighborhood Travel Analysis: An Approach to Understanding 257 least through the most emitting mode. Carbon factors for private automobiles in the Bay Area are absent because they depend on the age of the vehicle and vehicle occupancy of the trip.
Transit System
Mode
Storstokholm lokal-trafikforbund Rapid Rail Rapid Rail Bay Area Rapid Transit San Francisco-MUNI Trolleybus San Francisco-MUNI Cable Car San Francisco-MUNI Streetcar Storstokholm lokal-trafikforbund Motor Bus Golden Gate Transit Motor Bus Commuter Rail CalTrain Motor Bus SamTrans Santa Clara County TD Streetcar Private Motor Car ~ Stockholm Motor Bus AC Transit San Francisco-MUNI Motor Bus Santa Clara County TD Motor Bus Motor Bus Santa Rosa Transit Golden Gate Transit Ferryboat County Connection (CCCTA) Motor Bus Demand SamTrans Private Motor Car - San Francisco Bay Area
Travel Emissions Factor* 0.0 11.7 15.3 18.0 20.7 25.7 28.7 37.3 37.9 45.8 48.4 48.7 50.5 58.4 83.6 90.0 122.7 166.4 N/A
•Grams of Carbon per Passenger Kilometer Table 4 Carbon Factors for Various Transit Agencies and Modes.
REFERENCES Cervero, R. and R. Gorham (1995). "Commuting in Transit versus Automobile Suburbs" in Journal of the American Planning Association, Vol 61 No. 9, Spring. Cervero, R. (1995). "Sustainable New Towns: Stockholm's Rail-Served Satellites" in Cities Vol. 12, No. 1. Crane, R. and C. Richard (1996J. Does Neighborhood Design Influence Travel?: A Behavioral Analysis of Travel Diary and GIS Data. Toronto: Conference paper at Association of Collegiate Schools of Planning National Conference. Ewing, R., P. Haliyur and G. W. Page (1994). "Getting Around a Traditional City, a Suburban Planned Unit Development, and Everything in Between", in Transportation Research Record. Volume 1466. Frank, L. and G. Pivo (1994). Relationships Between Land Use and Travel Behavior in the Puget Sound. Olympia: Washington State Department of Transportation [Springfield, VA: Available through the National Technical Information Service].
258
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Gallez, C. (1995). Budgets Energie Environnement des Deplacements (BEED) en Ile-deFrance: Analyse de la Depense Energetique et des Emissions Polluantes Liees a la Mobilite des Franciliens. Arcueil: Institut National de Recherche sur les Transports et leur Securite. Gorham, R. (1996). Regional Planning and Travel Behavior: A Comparative Study of the San Francisco and Stockholm Metropolitan Regions. Master's Thesis in the Department of City and Regional Planning, University of California at Berkeley. Groth, P. (1995). "Old Additions Versus New Packaged Districts," lecture at Department of Architecture, University of California, Berkeley, February 16, 1995. Handy, S. (1993). "Regional versus Local Accessibility: Implications for Nonwork Travel" in Transportation Research Record. No. 1400. Hivert, L. and P. Le Dily (1994). Budgets Energie — Pollution: Bilan de la Mobilite des Menages de VAgglomeration Grenobloise. Arcueil: Institut National de Recherche sur les Transports et leur Securite. Intergovernmental Panel on Climate Change (IPCC) (1996). Policies and Measures for Mitigating Climate Change. Paris: Organization for Economic Cooperation and Development. International Energy Studies Group (lES) of the Energy Analysis Program of the Ernest O Lawrence Berkeley National Laboratory (1996). Database of International Energy Statistics for the OECD. Kockelman, K. (1997). "Travel behavior as a function of accessibility, land use mixing and land use balance: evidence from the San Francisco Bay Area". Washington, D.C.: Transportation Research Board. Lynch, K. (1970). The Image of the City, in M.I.T. Press Paperback Series 11. Cambridge: M.I.T. Press. Messenger, T and R. Ewing (1996). Transit-Oriented Development in the Sunbelt: Get Real (and Empirical)! Washington: Transportation Research Board. Newman, P. and J. Kenworthy (1989). Cities and Automobile Dependence. Brookfleld (VT): Gower Publishing. Parsons Brinkerhoff Quade and Douglas (PBQD) (1993). The Pedestrian Environment. Volume 4A in series Making the Land Use Transportation Air Quality Connection. Portland: 1000 Friends of Oregon. Pushkarev, B. and J. Zupan (1977). Public Transportation and Land Use Policy. Bloomington: University of Indiana Press. Sidenbladh, G. (1964). "Planning Problems in Stockholm" in Stockholm Regional and City Planning: Seven Articles on Planning Problems in Greater Stockholm. Stockholm: Planning Commission of the City of Stockholm. Warner, S. B. (1978). Streetcar Suburbs: the Process of Growth in Boston, 1870-1900. Cambridge: Harvard University Press. ' For brevity, I will refer to these types as "Cottage Districts", although I do not mean to minimize the prevalence of Apartments in this category. The term Cottage District I have borrowed from Groth (1995). ^ The strength of the association ((|) coefficient) was . 19 for the SFBA, .40 for the SMR. ^ Exceptions are, for SMR: Tower-in-the-Park and Nonresidential, Suburban Residential and Nonresidential, Town Centers and Nonresidential; Center City and Cottage Districts. For SFBA: Suburban Residential and
Comparative Neighborhood Travel Analysis: An Approach to Understanding 259 Nonresidential; Town Centers and Nonresidential. These tests are based on a simple difference of means t-tests. However, sequential difference of means t-tests increases the probability of an experiment-wise Type I error. The Tukey decomposition method of pairwise comparisons can be used to control the overall, experiment-wise alpha level (again, .05). Using this method, however, revealed no significant differences; neither the BATS nor the Stockholm RVU data were sufficiently robust for this technique. ^ This category will also contain some Comparison Shopping trips, which we speculate are less local in nature. ^ Trip mode share means the proportion of all trips which was made by a certain mode. Travel mode share means the proportion of all travel ~ the sum distance of all trips - which was made by a particular mode. ^ CBD for the SFBA is considered to be the Northeast quadrant of the City of San Francisco
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
12
TOWARDS A MICROECONOMIC FRAMEWORK FOR TRAVEL BEHAVIOUR AND LAND USE INTERACTIONS
Francisco J. Martinez
INTRODUCTION The understanding of the urban system has received the contribution of several disciplines: geographers, transport engineers, economists and urban planners, among others. They have emphasised the use of certain techniques focused either on the transport subsystem (T), regarding land use as exogenous, or on the land use (L) development where transport is always considered but highly simplified. Their integration in a common comprehensive framework has proven to be elusive due to several complexities and differences: Time: activities do take time, both to perform and to travel to them. The scale of time in transport, in the range of hours, days or weeks, is different from land use, where location decisions are made in terms of years. The main theoretical implication of this first complexity is that the concept of general urban equilibrium would combine different time scales, which calls for the study of the dynamics of the system. Secondly, in transport, time is recognised as a constrained resource of consumers which plays an important role in users behaviour (transport demand), in project evaluation (where ways have been found to estimate "value of time") and in operational optimisation (road pricing).
262
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Conversely, time is not seen as a relevant constraint in land use frameworks, apart from its role in T, as time expenditure in activities is usually taken as exogenous. A third more technical difficulty has been the treatment of time as a continuous variable, which in transport is overcome by using a discrete time variable called periods, which brings up the costs of aggregation, averaging and the usual assumptions of independence between them. However, it has not been straightforward to extend it to the notion of access, including accessibility and attractiveness, frequently used in location analysis. Space: non homogeneous urban space introduces a sort of mathematical complexity which was long ignored in urban economics under Alonso's (1964) assumption of a "featureless city." The latter eliminates urban spatial diversity, other than distance, generated either from the natural or the built environment, inexorably yielding a monocentric city. It also eliminates all potential external effects associated with the fact that location choices are strongly dependent on the built environment, which technically implies the reduction to a linear system. Again the answer has been a discrete space, called zones, which is used both in L and T. Spatial interaction is a category of studies dealing with space and location in a rather physical approach, as opposed to a behavioural approach. Consumers: Called users in transport, are taken as individuals that share socioeconomic attritbutes with a family group; it is only in commercial transport where the consumer unit is taken as the firm. In contrast, in land use consumers are called locators or agents, which represent family or firm units. Therefore, consumers in L are an aggregation of those in T which introduces theoretical and technical difficulties ; for instance, whose accessibility is more relevant in the choice of a household location ? Summarising, time scale, spatial context and the aggregation of time, space and consumers, are all factors that have prevented (or delayed) the development of a comprehensive and consistent unified behavioural framework for L&T studies. The aim of this chapter is to underline a behavioural framework whose consistency is drawn from a constant valid to L and T, that is locators and transport users are all the same. This fact provides natural consistency between L and T under the assumption of rational behaviour: consumers should be consistent in all their choices, regardless of time scale or space description; moreover, consistent rational behaviour should prevail regardless of whether the consumers play different roles: as an individual or as member of a family or a firm. This provides an approach to develop an integrated framework, based on microeconomics and behavioural studies, looking at the source of any action or activity (taste, constraints, etc.) instead of its effects (trips, consumption and location).
Towards a Microeconomic Framework for Travel Behaviour
263
The main concern of this chapter is to gather and integrate otherwise scattered contributions that provide elements for the development of a microeconomic framework of land usetransport interactions.
MICROECONOMICS OF LAND USE This L&T microeconomic approach has been emerging in the last decade or so. The transport behavioural approach, mostly developed in the 80's (see Ben-Akiva and Lerman, 1985), is based on disaggregate choice modelling and highly focussed on destination and mode choices, with less development on trip generation and attraction models. Although the terminology used has intentionally abandoned the name of 4-stage transport models, studies continue to recognise mode and destination as stages in demand modelling, while trip generation and attraction are largely seen as part of the interaction with land use. In this section we concentrate on the development of a land use behavioural approach which can be combined with transport demand modelling to produce a consistent framework.
Consumers' Behaviour Land is demanded by agents to perform activities, either residential or commercial. Location choices are made by family units (residential) or firms (commercial), which may be classified into homogeneous categories of agents {h); this classification may be as fine as required. Behavioural studies on residential location have been developed using discrete choice models, for example by McFadden (1978) and Anas (1982) among others. The indirect utility function of households (F) is assumed to be a function of zone (z) and dwelling characteristics {d), additionally, some sort of transport cost attribute (t) and socio-economic characteristics are included. Morisugi and Yoshida (1986) formally derive the indirect utility function including time constraints. However, as in many applied studies the arguments of function V are only intuitively introduced (see for example Miyamoto, 1993 ; Martinez, 1992). In a recent study, Jara-Diaz and Martinez (1999) proposed that utility is extracted from activities performed, with activities described by quality and time spent. Since quality is assumed dependent on environmental conditions, the indirect utility function explicitly depends on the natural and built environment where the activity is performed (z). For activities performed away from home, the combination of the benefit obtained at the destination which depends on
264
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
environmental (zone) characteristics (Ufz)), and transport costs, justifies the use of accessibility measures (ace). Accessibility is carefiilly defined as a measure of the expected total locator's net benefit (benefit minus cost) obtained from activities performed away from home (Martinez, 1995). Therefore, the indirect utility function for residential choice conditional on location s is: V(I-rs,T,P,Zs,d,acCs,p)
(1)
with / the family income (income constraint); r^ the rent at location s for a dwelling described hy d'yT available time after discounting time spent on compulsory activities^ ; P is the price of a composite good; p is the set of taste parameters. Following Jara-Diaz and Martinez's framework, fimction V provides a usefiil formal interpretation for the intuitive use of location attributes (z, d, ace), but it also introduces the role of available time constraint (7) and disposable income (/-r^). It also defines the role of accessibility in location choices, including aggregation of trip benefits across family members and different activities, plus alternative destinations available for performing each activity. Most relevant is the link provided by accessibility between location choice and individuals' transport choices, because it is embedded within a unified microeconomic framework of individual rational behaviour. From equation (1) and following Rosen (1974) and Martinez (1992), the consumer maximum willingness to pay for location s, (WPs), subject to achieving a utility level (w), given the family income (7), available time constraint (7) and composite prices (P), is obtained by inverting V in the location rent. Then : WPs(zs, ds, accs, I, T, P, u, P)
(2)
which represents the expenditure the consumer is willing to pay at each location, described by Zs as to achieve same level of satisfaction (u). WPftinctionsdescribe a family of indifference curves relating attributes (z) to money, usually labelled as "hedonic ftinctions". As income is always linear in equation (2) and the inverse of the utility in income is the expenditure ftinction (e), then: WPs-1 - e(zs, ds, accs, T, P, u*, P)
(3)
Towards a Microeconomic Framework for Travel Behaviour
265
with w* the maximum utility achievable subject to exogenous prices P. Based on the fact that willingness to pay is the expenditure function and measuring consumers' surplus as the compensating variation-defined as a change in expenditure-it can be proven that a strict measure of benefits, in the economic sense, is given by: CSs-WPs-rs
(4)
In what follows, it is useful to present consumers' maximum utility behaviour in its equivalent terms, the maximum consumers' surplus approach (Martinez, 1992). This states that the best locator choice is one which yields the agent the maximum surplus, so the consumer's choice problem is: CS'^^MaxiWP^-r^)
(5)
sen
with Q the choice set of alternative location options. The advantage of equation (5) compared to the classical utility approach, is that it allows a direct comparison with the alternative bid-auction approach.
Land Market and Bid Auctions The consumer's behaviour follows the standard utility maximising approach, which assumes that goods are homogeneous. However, urban locations are quasi-unique, meaning that location amenities cannot be produced as in normal products to satisfy demand, therefore location rents are not associated to a production cost but to the scarcity of the land. This characteristic is shared by other unique goods, which are sold in auctions and prices are called endogenous; thus, prices reveal the consumers' value or maximum willingness to pay (see McAffe and McMilland, 1987). Urban economists have discussed whether the urban land market should be modelled as normal competitive market or as bid-auction market, as proposed by Alonso (1964) and applied in a discrete logit model by Ellickson (1981); a notable discussion in favour of the competitive approach is developed by Anas (1982). Martinez (1992) sheds some light by demonstrating that, subject to some assumptions, both approaches are equivalent under the necessary -but not sufficient- condition that the market be at a walrasian equilibrium (no excess of demand or supply). Nevertheless, the general conclusion is that the bid-auction approach applies to a wider range of urban conditions; moreover, it allows for an explicit treatment of
266
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
the effects of market thickness (number of bidders against number of supply options), and of speculative behaviour. The bid-auction approach states that the location is sold to the best bidder and the rent is defined by the highest bid. Bids represent consumers' valuation reduced by a speculative factor {w) which depends on the auction conditions (selling mechanism, information, entry costs, etc.) and market thickness (McAfee and McMillan, 1987; Levin and Smith, 1994). Then consumer h bid for location s is given by : Bhs=WPhs-Whs
(6)
and the location or land use problem is the result of the supplier's search for the best bidder at each location among a set of A^ bidders. Note that A^ is a subset of total potential bidders A^, which defines demand thickness and may also depend on the auction's conditions. Replacing equation (2) in (6) and calling^ the set of auction conditions, we get: Bhs = WPs(Zs, ds, acchs, Ih, Z P, Uh\ Ph) ' ^hs(A, N^)
(7)
which defines consumer /z's behaviour under the bid-auction approach with speculation, including the consumer's perception and valuation of: i) Exogenous conditions, such as the building quality (d) and composite good price (P) ; ii) Endogenous attributes (z), generated by the built environment, hence produced by the location process, iii) Accessibility {ace), which depends both on transport supply (represented by index s) and on the consumer's trips demand (index h); iv) Socio-economic characteristics: consumer's income, time budget (/, T, w*) and taste parameters {p); v) Land market conditions {A, A^); vi) Equilibrium conditions, which define the utility level achievable after the market clears.
Rents Within the bid auction approach rents are directly defined by the best bid rule, then r^^=Max {WP,^-wJ heN^
(8)
Towards a Microeconomic Framework for Travel Behaviour
267
which lead us to the following conclusions : i) Rents are functions of bids, hence they depend directly on consumers' behaviour. Additionally, there is an indirect effect of supply, through the consumers' speculative behaviour via the market thickness. A third component in land formation is the utility level, which explicitly links rents with the equilibrium outcome. This model of rents differs from those where prices are obtained out of the assumption of a competitive equilibrium, because they ignore the quasi-uniqueness of urban locations. ii)
iii)
The functional form for rents is well defined by the original utility function, which defines the WP function form, and the form of the speculation function w(A,N^). This second result contradicts the usual argument of urban economists, which produce the so-called hedonic rent models for urban land. They argue that rents have no underlying theoretical functional form, hence the rent model should be specified based solely on empirical evidence (see Halvorsen and Pollakowsky, 1981). What equation (8) says is that a freely specified rent function inevitably defines an underlying utility function which should comply with the normal set of conditions. Urban land economists should constrain their models so that estimated parameters simultaneously replicate location and rents. An example of this type of consistent models are shown by Martinez and Donoso (1995) for the city of Santiago.
Equation (7) and the maximisation of equation (8) yields the following rent function : r / = r(zs, ds, acchs, h, T P, UH^ A . ms(A, N^X N^)
(9)
where the market thickness A^ plays a dual role : the direct effect of the number of bidders at the auction and the indirect effect on consumers' speculative behaviour.
Location Externalities One type of location externality is called agglomeration economies, usually referring to the benefits obtained by non-residential activities from the presence of other commercial activities in the neighbourhood as a result of a higher attraction to consumers. Another known type is neighbourhood quality, which depends on the land use in the area and associated location quality variables such as the average income level, density, etc. More generally, the land use pattern determines several variables that are relevant for locators, represented in our notation as elements of the zone characteristics vector z.
268
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
The effect of location externalities is that location depends itself on the location of other agents, i.e. on the land use. This produces an endogenous location equilibrium effect which complicates the performance of the market and the mathematical solution of the location models. Indeed, the effect of location externalities is mathematically described as a non-linear fixed-point problem defined by : NHs-L(Nh'i)
(10)
where Nhs is the number of agent type h located at zone ^ by a location function Z, and N^'f is the distribution of all other agents and zones. A known example of this function is the logit location function and its fixed-point problem described by Martinez and Donoso (1995). Location externalities are highly relevant to understanding consumers' behaviour. They not only introduce cultural differences between cities, which are crucial in this kind of studies, but they make individual choices mutually dependent, which defines one of the main dynamic processes of city development. For example, studies in Santiago City show that the main location attribute is the average residents' income in the zone, which depends on the location distribution of residents.
Equilibrium and Spatial Constraints There are two equilibrium conditions: i) Every agent locates somewhere, with the total number of agents by socio-economic cluster given exogenously. ii) The Walras equilibrium, which states that supply meets demand, with supply given by a function of rents and exogenous construction costs, iii) Allocated land does not exceed total land available in each zone (Qj), iv) Land use is subject to planning regulations and economic incentives for each land use category (Rhi). Therefore, equilibrium conditions introduce the following variables into the location outcome: supply Sdi(P, rdi, y), land availability Q=(Qi,, Qi,-) and regulations R =(Ru ,..., Rhu-); with / a set of parameters of the profit maximising suppliers' (developers) flinction (see an example in Martinez and Donoso, 1995). It is possible to formulate an optimisation problem where the objective function is defined by the maximum bid paradigm and the solution is constrained to a solution space defined by the above conditions. This optimisation process adjusts the
Towards a Microeconomic Framework for Travel Behaviour
269
maximum utility achievable by agents, which in the model means to adjust bids; this will in turn affect rents and supply to comply with equilibrium constraints.
Summary It is now possible to summarise the set of urban variables that govern the microeconomic location framework. Indeed, the number of agents from category h located in zone / and dwelling type d depends on consumers' bidding functions, supply function and land constraints, hence: Nhs (z, d, ace, I T, P, w* fi w(A, N^), y, Q(R))
(11)
Thus, the final location of any agent depends on: i) all agents' perceptions, their time distribution among activities, their utility levels and ii) iii)
their speculative behaviour (J3, T,u*, w)\ transport costs (time and fares) {ace); all suppliers' parameters (y);
iv) all zones' environment quality (z), land available ( 0 and land use regulations R\ v) land market selling mechanisms {A) and market thickness (A^^ ; vi) regional macroeconomic conditions {P, I) ; Additionally, rents depend on the same set of urban variables.
LAND USE-TRANSPORT INTERACTION The link between land use and transport models is two-fold. The simple one is the definition of the land use scenario in which the transport market operates. The more complex one is how location choices are affected by transport facilities. The difficulty is to define how consumers perceive transport facilities at each location, or how to measure the attribute loosely called accessibility (ace). Rigorously, however, what should matter to a rational locator is the differential in transport benefits associated with alternative locations. Transport-related benefits may be divided in two types, according Martinez (1995): accessibility (ace), or the benefits associated with the trip generator activity, and attractiveness (att), or the benefits associated with the visited activity. It
270
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
depends on the type of activity to be located whether accessibility, attractiveness or both are most relevant. A number of access measures have been proposed based either on some definition of transport generalised costs or its combination with land use; see for example Jones (1981) and Morris et. al. (1979). Those based only on transport costs ignore the variability of land use, therefore are of limited use. Combined measures of transport and land use have been developed based on the notion of economic net benefit, which attempts to measure the difference between the benefit obtained at the trip destination {bj) and the disutility caused by the transport generalised cost required to obtain them (cy). In the case of attractiveness, it measures the benefit obtained by the visited activity at the trip destination, depends on the number of visitors which in turn depends on bj and Cy. The utility obtained at the destination depends on the quality and quantity of the required activities, described by land use at the destination (o/j. Therefore, the relative access between two zones is: acchij =f(aj, Cij, 4)
atthij = g(aj, cy, p^)
(12)
with h the trip maker agent for ace, or the visited agent for att; funtions / and g and parameters S and p are defined by transport demand models, with an important role played by the trip destination choice (distribution) model. Integral measures of accessibility are obtained as the expected benefit derived from a given trip taking into account the probability of visiting alternative destinations. Integral economic measures of accessibility can be obtained from transport demand models, by calculating the transport users' benefit. Neuburger (1971) and Williams (1976) developed measures of accessibility for entropy maximising transport demand models which happened to be expressed by the known balancing factors of the trip distribution model. Domencich and McFadden (1975), Williams (1977) and Ben-Akiva and Lerman (1979) describe measures of users' benefit based on the utility maximising logit model; in this case these measures are directly provided by the known logsum function (the expected minimum transport generalised cost). Then integral measures are : acchi =f(a, cu A^
attt = g(a, Ct, y^)
(13)
with the/' and g' functions derived from the trip demand models and the definition of expected user benefit.
Towards a Microeconomic Framework for Travel Behaviour
271
These accessibility measures have been applied as transport-related attributes in some land use microeconomic models. For example, RURBAN (Miyamoto, 1993) uses the logsum measure for residents' ace, while MUSSA-ESTRAUS (Martinez and Donoso, 1995) uses balancing factors for ace and att. In any case, the land use model has to identify the appropriate aggregation from trip related benefits to a family -or firm- daily trip pattern. Such measures can be easily calculated and allow planners to asses the impact of transport projects in location choices, including route or link capacity changes, new modes or operational methods, pricing policies, etc. Measures of users' benefits are readily available from transport models. However, despite the attempt of Williams and Senior (1978) to interpret their results in terms of users' benefits and land rents, it is not straightforward to identify separate measures for ace and att from the transport users' benefits measures. These require careful understanding of the demand model and its assumptions (Martinez, 1995). Accessibility is the only willingness to pay attribute that is both location and agent specific. Indeed, while transport costs differ by location, benefits differ by agent as they are affected by consumers' valuation of visiting activities. This double dimension in the variability of accessibility induces some complexity. For example, very different values may be obtained for families with similar socio-economic characteristics, located in the same zone, but with different daily trip patterns.
EXTENSION TO THE ANALYSIS OF LOCAL AREAS The framework presented above provides consistency in the analysis of consumers' behaviour across their L&T choices. It has been partially applied for strategic studies in large metropolitan areas. However, there is no equivalent framework to study L&T interactions in a local context. This section discusses the potential extension of the framework to a local area, where cycling, walking, pedestrianisation, location of bus and Metro stations and concentration of retail and services, are some of the issues that affect short distance travel behaviour. Here, the type of questions are not which areas of the city will develop more rapidly in the next ten years, but where the location of a Metro station in a given area will have the greatest impact. The approach followed is to assume that the general framework presented is valid in this context although the interpretation and focus are different. In other words, it is assumed that the general urban economic mechanism operates at any spatial level. The modelling approach is
272
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
spatially hierarchical. The upper level solves major location tendencies in a city-wide context, providing total residents and firms as well as the expected rents in each zone. This level deals with the dominant location effects of the future increases in population and the development of major strategic transport projects, achieving demand-supply equilibrium conditions subject to space constraints and land use regulations. At the lower level, a complementary local model, called the local L&T model, defines the final distribution of activities and rents within the zone and for a more detailed spatial description (blocks), subject to holding the total number locators and associated average rents constant.
Attributes for the Local Model The scale of the local model requires an appropriate adjustment of the characteristics that are relevant for local distribution of activities and rents. In terms of the economic framework, it means to identify those attributes that may describe consumers' variability in willingness to pay for locations. These attributes should belong to the WP functions, therefore they can be found in the location expression (equation 11): Nhs (z, d, ace, 1 T, P, w* p, w(A, N^), y, Q(R)) Experience in defining such local attributes can be drawn from real state agencies, urban planners and architects, in addition to surveys. Variables may be described as : i) Land lot attributes: front width, total size, view, etc.; ii) Surrounding: quality of buildings, density of commercial activity, presence of nuisance industry, etc., is usual in this context; iii) Transport: traffic and pedestrian flow levels, the width of the front street, access to public transport, distance to mass transit stations, etc. The distribution of these variables within the local area induces differential willingness to pay across the space and agents, since residents and commercial activities have different valuation of these variables. Indeed, it is important for the modeller to be able to introduce variables that provide variability in agents utility, therefore to explain variability in local location and rents. This reduction in the scale of attributes provides a first consistent link between detailed characteristics associated with local location choice and the L&T framework.
Towards a Microeconomic Framework for Travel Behaviour
273
Land Rents A known technique to study land use at the local level is the estimation of hedonic rent functions. These functions are consistent with our L&T economic framework but their interpretation is slightly different. The hedonic function parameters are assumed to represent directly the agents' valuation of attributes, and the set of variables included in the hedonic function are specified primarily according to the modeller's intuition. In the L&T framework, rents represent a complex but well-defined function of agents' willingness to pay, consistent with a bid-auction selling mechanism. The definition of attributes is supported by theoretical considerations. Nevertheless, once estimated, both may look very similar. The estimation of rent functions in the local model should comply with theoretical considerations: parameters are required to reproduce location and rent distributions simultaneously and total locators and average rents should reproduce values provided by the upper level of the L&T model. The importance of land rents in local models lies in the fact that land price data are usually available in good quality with spatial variability. Secondly, they provide evidence of the capitalisation of urban externalities and public investment into land values. A third and more technical point is that location data only allow us to calibrate relative willingness to pay fiinctions, while rent data let us adjust them to absolute figures.
FUTURE R E S E A R C H
In this section some research issues are identified in connection with both metropolitan and local L&T models.
Spatial Context The role of environmental attributes in location and rents raised the question about the extent to which the zoning system biases the parameters or reduces the prediction accuracy. Some have intended to find the best zoning system that maximises prediction ability. Another approach, called spatial econometrics (Anselin, 1988), introduces tests of spatial autocorrelation and correction procedures by analogy to times series studies.
274
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
An emerging technique is to replace econometric models by neural network models, though these lack a theoretical basis for model specification. Research is needed to apply and evaluate these techniques in the context land use planning.
Land Use, Transport and the Environment The L&T model needs to be expanded to cope with environmental impacts in the urban context. Some research has been directed towards that objective (see contributions in Hayashi and Roy, 1996) and some L&T models have developed the ability to assess environmental impacts generated by land use and transport policies (Wegener, 1994); these are called LT&E models. A step forward has been the integration of the environmental model in a LT&E interaction model, which can assess the feedback impact of environmental policies, such as costs and constraints, back to the land use and transport sub-models (O'Ryan, et. al. 1996).
Dynamics Perhaps the main criticism of L&T models is the application of the concept of static equilibrium in a process where time is relevant, while those using a dynamic approach are criticised because of the absence of any equilibrium point to which urban economic forces tend to move. There is not much progress in this long lasting and relevant discussion. A possible move forward is to develop a mixed concept of dynamic equilibrium, which allows variables and parameters to be time specific and equations to develop over time. Here the concept of dynamic utility functions is the main driving element. A complementary element is to integrate microeconomic studies with research on the evolution of macroeconomic variables. As noted above, some WP attributes, like P and /, are defined in the macroeconomic context and can be directly connected with such studies. Extensions can also be made on the supply side by making the land use and transport investment obtained from the corresponding models consistent with the regional forecast of investment in these sectors. As macroeconomic variables are normally defined by medium term economic policies, they are not only predicted but also common for all activities within a state.
Towards a Microeconomic Framework for Travel Behaviour
275
ACKNOWLEDGEMENTS The author thanks the research funding provided by FONDECYT N' 1950280.
REFERENCES Alonso, W. (1964). Location and Land Use, Cambridge, Harvard University Press. Anselin, L. (1988). Spatial econometrics : Methods and Models, Kluwer Academic Publishers, London. Anas, A. (1982). Residential Location Markets and Urban Transportation, Academic Press, London. Ben-Akiva, M. and S. R. Lerman (1979). Disaggregate travel and mobility-choice models and measures of accessibility. Behavioral Travel Modelling, (ed.) D. A. Hensher and P. R. Stopher, Croom Helm, London. Ben-Akiva, M. and S. R. Lerman (1985). Discrete choice analysis: theory and application to travel demand modelling. The MIT Press, Cambridge. Domencich, T. A. and D. McFadden (1975). Urban travel demand: a behavioural analysis. North Holland, Amsterdam EUickson, B. (1981). An alternative test of the hedonic theory of housing markets. J. of Urban Economics 9, 56-79. Halvorsen, R. and H. O. Pollakowsky (1981). Choice for functional form for hedonic price equations. J. of Urban Economics 40, 37-49. Hayashi, Y. and J. Roy (1996). Transport, Land-Use and the Environment. Kluwer Academic Publishers, London. Jara-Diaz, S. R. and F. Martinez (1999). On the specification of indirect utilit>^ and willingness to pay for discrete residential location models. Journal of Regional Science 39 (forthcoming). Jones, S. R. (1981). Accessibility : a literature review. TRRL Laboratory Report 967. Levin, D. and J. L. Smith (1994). Equilibrium in auction with entry. American Economic Review, 84 (3), 585-99. Martinez, F. J. (1992). The Bid-Choice land use model: an integrated economic framework. Environment and Planning v4 75,871-885. Martinez, F. J. (1995). Access, the economic link in Land Use-Transport interaction. Transportation Research 29B, 457-471. Martinez, F. J. and Donoso, P. P. (1995). MUSSA Model : The theoretical framework. ModelHng Transport Systems. Proceedings 7th World Confence on Transportation Research WCTR), Vol. 2, (eds.) D. Hensher, J. King and T. Oum, Pergamon, 333-343. McAfee, P. and J. McMillan (1987). Auctions and Bidding. J. of Economic Literature, XXV, 699-738. McFadden, D. L. (1978). Modelling the choice of residential location. In Karlqvist et. al. (eds). Spatial Interaction Theory and Planning Models, North-Holland, Amsterdam, 75-96. Miyamoto, K. (1993). Development and applications of a land-use model based on random utility/rent-bidding analysis (RURBAN). Thirteenth Meeting of the Pacific Regional Science Conference Organisation, Whistler, British Columbia, Canada.
276
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Morisugi, H. and T. Yoshida (1986). Forms of utility function for residential behavior analysis and neighbourhood benefits estimation. Environment and Planning, 18, 53-62. Morris, J. L., P. L. Dumble and M. R. Wigan (1979). Accessibility indicators for transport planning. Transportation Research A, 13A, 91-109. Neuburger, H. (1971). User benefit in the evaluation of transport and land use plans. J. of Transport Economic and Policy, 5, 52-75. O'Ryan, R., F. J. Martinez and L. Larraguibel (1996). A Neural Network Approach to Evaluating Environmental Urban Policies : The Case of Santiago, Chile. Urban Transport and the Environment. Rosen, S. (1974). Hedonic prices and implicit markets: product differentiation in pure competition. J. of Political Economy 82, 1, 34-55. Wegener (1994). Operational models: State of the art. J. of the American Planning Association, Winter, 17-28. Williams, H. C. W. L. (1976). Travel demand models, duality relations and user benefit analysis. J. ofRegional Science, 16, 2, 147-166. Williams, H. C. W. L. (1977). On the formation of travel demand models and economic evaluation measures of user benefit. Environment and Planning A, 9, 285-344. Williams, H. C. W. L. and M. L. Senior (1978). Accessibility, spatial interaction and the evaluation of land use transportation plans. Spatial Interaction Theory and Planning Models, A. Karlovist, L. Lundovist, F. Snickars, J. W. WeibuU (Eds). North Holland, 253-287. Compulsory activities are those with fixed time allocation, whose location are assumed exogenous to the residential location choice.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
13
LAND USE - TRANSPORTATION INTERACTIONS: WORKSHOP REPORT
Ed Weiner and Roger Gorham
LAND-USE IMPACTS ON TRAVEL BEHAVIOUR That land-use has an impact on travel behaviour is a basic and long-standing tenet of transport planning. Land-use is a, if not the, principal exogenous component in classical metropolitan transport planning models, providing the key input to the heuristic variable of trip-generation rate, on which the remainder of the transport forecasting system is based. Land use — particularly net residential density ~ has also been a key component in public transport planning and ridership assessment (Pushkarev and Zupan 1977). Empirical studies show clearly the importance of urban and regional form in influencing travel choices. In the San Francisco Bay Area, the mode shares of trips to and within the Central Business District (CBD) of San Francisco proper have been shown to resemble those of European cities — about 30% by car (Gorham 2001), which is in the same range as figures reported by Monheim (1996) for German and Italian cities, and Stockholm (Gorham 2001). However, only about 4% of all trips in the region actually go to or are within the CBD, compared with 28% for the Stockholm region. In other words, people in the San Francisco region use their cars substantially less when they go downtown — but they rarely go downtown. Post-war planners in the Stockholm region, and other parts of Europe, such as Paris and Lille, were aware that the relationship of the outlying parts of the region to the CBD was crucial to maintain a mode share which is not overly skewed toward car use. They purposely sought to maintain the hegemony of the CBD - central Stockholm - as the economic and cultural centre of the region (Gorham 1996). They sought to influence land-use as a matter of policy to
278
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
influence travel behaviour. The idea of manipulating land-use for transport policy has been regaining currency amongst planners and lay people in the last ten years, spearheaded by American planners associated with the New Urbanist movement (Peter Katz, Andres Duany, Elizabeth Plater-Zyberk, Peter Calthorpe, and others). The idea, however, is controversial, especially in the United States and Australia. A debate touched off ten years ago by Australian researchers Peter Newman and Jeffrey Kenworthy led to a flurry of research on the subject throughout the 1990's. Newman and Kenworthy (1989) argued that, in cities, petrol consumption per capita was highly correlated with the overall density of the city. Consequently, creating cities with compact urban form (i.e., higher densities) would resuh in reduced petrol consumption, which they took to be a proxy for car use (or more specifically, according to their formula, car dependence). Their methods as well as conclusions have been highly criticized (Gordon and Richardson 1989, Schipper 1993). Results given in Fouchier (1997) suggest that for the inner ring of suburbs of Paris, doubling of net densities (both population and employment) would result in a savings of less than 24% of total travel. For all but the most outer parts of the region, densification, in addition to being highly controversial, may be less effective than Newman and Kenworthy suggest. Recent research on the subject, however, has de-emphasized the role of density per se, because, first, as a concept and measurement tool, it is often vague and improperly used (Fouchier 1997), second, it may not be a particularly accurate measurement tool for describing those features of land-use which most influence travel behaviour (Handy 1996, Gorham 1996), and third, there may be other land-use policy tools which are more effective at inducing sustainable travel behaviour than simply manipulating density. Other aspects of urban form which have gained prominence as determining variables include: degree of land-use mix, orientation of buildings toward the street, street pattern and layout, street widths, and various other urban design micro-characteristics. Overall, the results of this research show that different aspects of urban form are associated with more or less sustainable travel choices, suggesting (but by no means proving) that manipulation of elements of urban settlements may be at least somewhat effective in influencing travel choices. Gorham (1996) argues that a comparison of the San Francisco and Stockholm regions shows that concerted, long-term, region-wide planning and land-use/transport co-ordination can lead to sustainable travel choices. This research also suggests that the structure of the metropolitan region (land-use on a macro level) is more important in influencing travel behaviour than the localized or micro elements enumerated above (Gorham 1996, Cervero and Gorham 1995).
Land Use-Transportation Interactions : Workshop Report
279
RESEARCH ON THE INFLUENCE OF LAND-USE ON TRAVEL BEHAVIOUR Much of the workshop discussion focused on the methodologies and needs of researchers looking into the question of whether a causal link between human settlement patterns in urbanized regions and travel behaviour exists. There was widespread agreement in the group both that the interactions between these phenomena was quite complicated, and that (consequently), land-use policy alone will be insufficient at reducing car dependency. Such an acknowledgement, however, poses a significant dilemma for researchers in the field. If it is clear that land-use alone is not sufficient to influence travel behaviour - i.e. it must be done in concert with other policies - how, then, can researchers effectively isolate the effects of landuse? In one form or another, this question underlay much of the group's discussions and attention.
Important Distinctions and Definitions Early on in the discussion, the group made two important distinctions that helped organize much of what followed.
Two FUNDAMENTAL APPROACHES TO THE QUESTION
Heuristic Approaches This approach involves using land-use variables in concert with other variables, usually sociodemographic ones, to predict or otherwise explain travel behaviour. Much of the "newness" of this approach comes from the development of newer and more refined indicators of land-use. It is the more prevalent approach in the literature (Frank and Pivo 1996, Cervero and Gorham 1995, Kockelman 1996, PBQD 1994). While it is a reasonably accessible method of evaluating the impacts of land-use and other variables on travel behaviour, it is often inadequate in its ability to elucidate or explain the relationship between land-use and socio-demographic or other variables.
Behavioural Approaches These methods investigate the link between urban settlement patterns and travel behaviour in
280
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
more detail than the heuristic approaches. While breadth of potential research methods is potentially very large, the small body of extant research in this approach has tended to focus on how people are observed to behave in different urban environments (Handy 1994) and how people's attitudes change according to different residential urban environments (Kitamura 1997). There was significant consensus in the group that one area to which this research approach needs to pay more attention in the future is residential and business location choice and the effects of urban-environment self-selection as a factor in travel behaviour. Three Types of Research. The group distinguished between three broad categories or types of research. They are presented sequentially, as each builds on knowledge gained in an earlier phase. Exploratory. Here, the researcher is looking at relationships between and among various phenomena and factors. There is a minimum of theory at this point; researchers are in the process of developing hypotheses for later testing. The absence of theory at this stage is a strength, in that it encourages creative experimentation both in the design and analysis of studies. Explanatory. In this phase, researchers endeavor to prove or supply evidence to their hypotheses as they develop new or expound on expound on existing theory. Theory is an important organizing element to the research, but the researcher's attitude towards it is either critical, or, at least, questioning. Predictive/simulative. In this type of research, the student is using accepted theory and a body of knowledge to simulate and/or predict behaviour, both as a result of a natural evolution of transport and land-use systems, or resulting from the introduction of new physical or policy interventions into those systems. Again, theory organizes the nature of the simulation, the researcher's aim, however, is not to validate the theory, but rather to examine outcomes under different scenarios under the assumptions that the theory used is the best explanation of observed behaviour. Each of these broad categories of research was represented in papers presented at either the workshop or plenary sessions of the conference. However, there seemed to be a consensus among participants that much more emphasis needed to be placed on exploratory research, especially in light of the previous discussion of the need for more behavioural approaches to research in transport/land-use interactions, where theory is not well developed. The external demands on the research community, however, creates potential conflicts with this need for more exploratory research. In the United States, transportation legislation (ISTEA) requires the simulation of air quality at 20 year horizons, based on expected changes in and interventions to the transportation and land-use systems. This requirement puts pressure
Land Use-Transportation Interactions : Workshop Report
281
on researchers to concentrate on predictive or simulative projects, even though the working group had a strong feeHng that existing theory or understandings of the relationships was inadequate to the task.
Needs and Directions for the Field Two sets of organizing concepts were developed to help organize our ideas and recommendations. The first was the enumeration of the different aspects of travel behaviour research connected specifically with land-use research. The second was the specification of different types of needs for the research agenda. The first group identifies three distinct aspects of travel behaviour research: • Understanding and representing the physical environment, including: • Land use system • Transportation System • Understanding and representing observed behaviour • Understanding and representing user perceptions, attitudes, and intentions The first of these categories can be thought to correspond roughly to the supply elements of the overall system, the second to demand, and the third, to the behavioural mechanisms which link the first two. We do not mean to suggest that these aspects are discrete elements; rather, they are presented in such a manner to help organize the discussion that follows. The workshop also identified three different levels of needs for researchers in transportland/use interactions: • • •
Conceptual needs - gaps in our understanding of how to conceptualize problems, phenomena, or systems (understanding what needs to be on the research agenda) Methodological needs - gaps in the tools available to address the research agenda Data needs - gaps in the availability of the raw material of research: information of various kinds
These two sets of categories create the outline of a matrix in which we organize our particular suggestions. This matrix is presented as Table 1.
282
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
Physical Environment 1 Conceptual Needs
Methodological 1 Needs
1 Data Needs
Harmonize measurement concepts (density, mixture of use, etc.) harmonize geographic measurement unit understand better systemic relationships among different parts of a region, esp. for polycentric or acentric structures workable definitions of neighborhoods Extending and organizing the use of existing resources, including historical data Better integration of micro-economic theory into land-supply models and analysis Resolution of econometric issues related to spatial analysis Expand use of neural networks as an analytical tool More disaggregate land use data tied to household or address better definition of "mixed-use"
Understanding/Representation of: Observed Behaviour
mobility territory (as distinct from accessibility territory)e.g. use of space in daily and weekly activities, including dynamic and stochastic aspects of this mobility territory is there a need to rethink traditional activity definitions to account for alternative uses of space?
Understand and elucidate the linkages with environmental factors (pollutant emissions, C02, energy consumption) Better and more vigorous longitudinal (before and after) studies; need to seek out opportunities Openness to other approaches beside traditional activity/travel survey
Consistency of (geo-coded) location info in travel/activity surveys Geographically localized data sets of sufficient quantity to permit statistically valid and robust comparisons of local neighborhood form characteristics --e.g. localized "over-sampling" in a regional household activity/travel survey
| Perceptions, Attitudes/Intentions 1 Understanding space perceptions and needs Understanding choice set generation • lifestyle • residential location • destination • mode Which occurs when? Defining neighborhoods and neighborhood typologies Defining and measuring accessibility Use of techniques from other disciplines • behavioural and social psychology • urban design research More extensive use of qualitative methods • in-depth interviews • focus groups Combine empirical with attitudinal data for more accurate forecasts
1 Very little data available at present
Table 1 Matrix of Travel Behaviour/Land Use Interactions Conceptual Needs for the Physical Environment. As research shifts from focusing primarily on density to more comprehensive understandings and classifications of the physical environment, researchers will need to grapple with definitional and conceptual issues which, until now, have been somewhat elusive. A fair amount of exploratory research is called for in this area, since so many aspects of the descriptive system of the physical environment remain
Land Use-Transportation Interactions : Workshop Report
283
to be operationalized. This operationalization must include developing better measurement concepts for physical places, such as density, mixture-of-use, permeability and interface between public and private places, as well as harmonized understanding of the geographic units, such as neighborhoods or districts. In the longer run, there needs to be a better understanding of the systemic relationships of different parts of an urban region - that is, a better conceptualization of how local spaces form together to give structure to the region as a whole. Methodological Needs for Representing the Physical Environment. Methodological needs for understanding and representing the physical environment involve better hamassing of existing resources, which may not have been previously used as extensively as possible in the research of land-use/travel behaviour research. Theoretical and practical techniques from other disciplines need to be tapped and brought to bear on the subject, in better application of microeconomic theory into land-supply models and analysis. A number of econometric issues related to spatial analysis, including mutli-collinearity between spatial and non-spatial variables, need to be resolved. Methodologies also need to evolve to make better use of existing resources, since existing analytical techniques can be hungry for data and information which has not necessarily been gathered. Finally, advances in computing techniques, such as neural networks, can be increasingly integrated as an analytical tool. Data Needs for Representing the Physical Environment, Data requirements for understanding the physical environment as it affects travel behaviour must become as extensive and detailed as the that for travel behaviour research as a whole. The state-of-the-art for the latter is increasingly disaggregated, both in terms social organization (down to the household or individual level) and in terms of the unit being measured (individual activities). A similar degree of disaggregation is needed in describing the physical environment. Land-use data needs to be tied to the household or address, not simply the traffic analysis zone. Such disaggregate physical information will also help allow for a better workig definition of the concept of "mixed-use", where existing levels of physical disaggregation in data collection often do not allow for quantitative definitions of mixed-use to correspond to the concepts researchers are pursuing. Conceptual Needs for Representing Observed Behaviour. The workshop identified the need to better understand behavioural distinctions between accessibility and mobility behaviour. The relationship between mobility and access, and the extent to which mobility or the desire for movement is an inherent human need requires further elucidation. In particular, the concept of
284
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
"mobility territory" - how space is used in regular activities - proposed by Boulahbal, should be developed to include dynamic and stochastic understandings of mobility territory. Methodological Needs for Representing Observed Behaviour. The workshop identified three important methodological development needs to help researchers characterize observed behaviour. First, they would need to become more pro-active about identifying and pursuing opportunities for longitudinal studies with respect to changes in the transport / land-use system. This is particularly crucial to explanatory-types of research, where empirical tests will need a more scientifically rigorous basis than has been pursued in the past. Second, methodologies for associating observed travel behaviour resulting from land-use patterns with social and environmental factors - such as pollutant and C02 emissions, energy consumption, distribution of poverty, and public health - need to be refined and developed. Finally, researchers need to be open to approaches other than quantitative activity or trip-based surveying techniques. Data Needs for Representing Observed Behaviour. The primary development needs of data collection techniques in order to help better represent observed behaviour as identified by the workshop were: 1) enhancing the quality and consistency of geo-coded location information in the travel or activity survey instruments themselves; and 2) enhancing sampling in small, geographically localized areas so as to allow for statistical significance in the results by microneighborhood. Taken together, these recommendations imply a need for survey designers to take spatial aspects of the process into account beyond simply associating respondents with points on the transport network. Conceptual Needs for Representing Travelers' Perceptions, Attitudes, and Intentions. Some of the most important gaps in research knowledge on the influence of land-use on travel behaviour relate to conceptual shortcomings in understanding and characterizing perceptions. These perceptions related to four categories in particular: how people perceive space and their own need for it, how they perceive neighborhoods and differentiate between them, how they understand and behave in different accessibility sytems, and how they derive their choice sets. Understanding how people understand and perceive space, neighborhood, and accessibility, is the attitudinal counterpart to observed "mobility space" identified above under observed behaviour. It is also crucial in the development of a concept of a cohesive neighborhood as discussed above under representing the physical environment. Consequently, it is an important and thus far missing aspect of understanding the influence of urban form on travel behaviour. Both the conceptual and methodological bases for efforts to understand perceptions can be drawn from the planning literature - the pioneering work of Kevin Lynch or Donald Appleyard are particularly promising - but the need to associate stated perceptions with observed
Land Use-Transportation Interactions : Workshop Report
285
behaviour remains paramount. Understanding choice set generation - for mode, destination, residential location, and even lifestyle choice - is one of the key links in this association, but researchers' understanding of how travelers put together or conceive these choice sets has historically been poor; researchers' often substitute their own perceptions for those of the subjects. The influence of spatial form - and particularly the perception of spatial form - on the perceived viable choice set is of crucial importance, and may be a crucial key to understanding and measuring accessibility. Methodological Needs in Representing Traveler Perceptions, Attitudes, and Intentions. Developing methodologies to better gauge traveler perceptions and attitudes is an important part of the needs identified by the workshop. These methodologies will need to draw on techniques which have been developed and refined in other disciplines, including behavioural and social psychology, and urban design research. A key innovation, as suggested above, will be the association of perceptions, attitudes, and intentions with observed behaviour. This implies a need to combine empirical with attitudinal data to produce more accurate forecasts. The nature of the subject matter also suggests that more extensive use of qualitative research techniques - in depth interviews and focus groups, for example - will need to be made than has been used in mainstream travel behaviour research in the past. Data Considerations in Representing Traveler Perceptions, Attitudes, and Intentions. The relative absence of previous research into travelers' perceptions of the spatial environment means that existing data are almost non-existent. The workshop therefore was unable to discuss refinements to existing data or data-gathering techniques. The nature of the subject, however, implies the need for researchers to be able to synthesize information and data from a variety of qualitative sources, information which may not lend itself naturally to such comparisons. An interesting example of such a synthesis approach can be found in the Royal Automobile Club's 1995 report Car Dependence (Goodwin, et. al. 1995).
ACKNOWLEDGEMENTS Participants included: M. Boulahbal, K. Clifton, E. Deakin, S. Forward, L. Frank, R. Gorham, S. Handy, S. Kaul, J. Ma, F. Martinez, L.-G. Mattsson, L. Schipper, K. Shriver, A. I. J. M. van der Hoom, E. Weiner, and E. Wong.
286
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
REFERENCES Cervero, R. and R. Gorham (1995). "Commuting in Transit Versus Automobile Neighborhoods" in Journal of the American Planning Association. Vol. 61, no. 2 (Spring 1995). Fouchier, V. (1997). Les Densites Urbaines et le Developpement Durable: Le Cas de Vilede-France et des Villes Nouvelles. Paris: Secretariat General du Groupe Central des Villes Nouvelles. Frank, L. and G. Pivo (1994). Relationships Between Land Use and Travel Behavior in the Puget Sound. Olympia: Washington State Department of Transportation [Springfield, VA: Available through the National Technical Information Service]. Gorham, R. (2001). "Comparative Neighborhood Travel Analysis: An Approach to Understanding the Relationship Between Planning and Travel Behavior". Chapter 11 in this volume. Gorham, R. (1996). Regional Planning and Travel Behavior: A Comparative Study of the San Francisco and Stockholm Metropolitan Regions. Master's Thesis, Department of City and Regional Planning. Berkeley: University of California at Berkeley. Handy, S. (1996). "Methodologies for Exploring the Link between Urban Form and Travel Behaviour." in Transportation Research Part D: Transport and Environment. Volume ID, No. 2, December. Handy, S. (1993). "Regional versus Local Accessibility: Implications for Nonwork Travel" in Transportation Research Record. No. 1400. Kitamura, R. (1997). "A micro-analysis of land use and travel in five neighborhoods in the San Francisco Bay Area" in Transportation Vol. 24 No. 2, May. Kockelman, K. (1997). "Travel behavior as a function of accessibility, land use mixing and land use balance: evidence from the San Francisco Bay Area". Washington, D.C.: Transportation Research Board. Monheim, R. (1996). "Parking Management and Pedestrianisation as Strategies for Successful City Centres" in Sustainable Transport in Central and Eastern European Cities: Proceedings of the Workshop on Transport and Environment in Central and Eastern European Cities, 28th - 30th June 1995, Bucharest. Paris: OECD Publications. Newman, P. and J. Kenworthy (1989). Cities and Automobile Dependence: An International Sourcebook. Aldershot (UK): Gower Publishing Company, Ltd. Parsons, Brinkerhoff, Quade, and Douglas (PBQD). (1994). The Pedestrian Environment. Volume 4A in series Making the Land Use Transportation Air Quality Connection. Portland: 1000 Friends of Oregon. Pushkarev, B. and J. Zupan (1977). Public Transportation and Land Use Policy. Bloomington: University of Indiana Press. Royal Automobile Club (RAC). (1995). Car Dependence: a Report for the RAC Foundation for Motoring and the Environment, P. Goodwin, ed. London: RAC Foundation for Motoring and the Environment.
SECTION 5 TIME USE
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
14
EMERGING DEVELOPMENTS IN TIME USE AND MOBILITY
Nelly Kalfs and Andrew S. Harvey
INTRODUCTION Over the past three decades there has been growing attention paid to the collection, analysis and use of time use data in social, economic and transportation measurement and reporting. Time use data show for an individual what activities were done during a defined period, how many times, in what order, at what time, for how long, where, with whom and other objective and subjective information connected with the activities (such as the degree of pre-planning, feelings about what was done, etc.) given the temporal and spatial context of society and personal and household characteristics. Collection of time use data in transport research was driven by the recognition that travel demand is difficult to understand by examining the characteristics of the transportation system and its travellers only. Travel is derived from the activities people wish to perform, implying that travel can only be understood from the total activity pattern of individuals and households. The variety in household types and activity patterns has increased due to societal trends such as individualisation, changing labour market participation of women, flexible working hours and such technological developments as telematics and information. As a consequence, the understanding of the transport problem has become rather complex and hard to solve. Especially since this problem is no longer a matter of just adding infrastructure. The solution may well lie outside the transport system, i.e. in reducing the need for travel altogether (e.g. by
290
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
reducing the distance between living and working places, flexibilisation, telecommuting, etc.). Therefore, a broader perspective is necessary (Van der Hoom, 1997). This broader perspective is better known as the activity based approach. This approach is increasingly evolving as a potential application to evaluate policy directions (Stopher and LeeGosselin, 1997; Ettema and Timmermans, 1997). From these readings it is clear that a diversity of concepts and methods is embraced by the activity based approach. Recurrent themes are time allocation and duration, scheduling of activities in time and space, constraints on movements and activity choice, interactions between decisions (over the day and week) and different individuals, household roles and structures, and the processes of adaptation and change. It deals with activity and trip-chaining and in-home/out-of-home substitution of activities. To date, a considerable body of knowledge exists regarding several aspects of activity and travel patterns. At the same time research has been very fragmented and an overall framework that links research in different areas is still missing (Ettema and Timmermans, 1997). The same holds for the data collection efforts. In the time use field much experience has been gained with time use data collection. It is from this perspective that this chapter is written. The chapter explores the roots and the potential of time use data collected by time use researchers in terms of the contribution it can make to understanding mobility. First a brief overview of the activity based approach and identification of the key elements in an activity based framework are noted. Then data collection issues are discussed in terms of the standards that have been developed in the time use field and their appropriateness within the framework. The efficacy of using time use data to explore travel behaviour is illustrated by showing some (cross-) national trends in mobility that are based on data from the Multinational Longitudinal Time-Budget Archive and the Dutch Time Use Survey. These trends are discussed within the context of changing labour market participation of women and changes in leisure time and travel.
TIME USE AND TRAVEL According to Pas and Harvey (1997) the first, and only, revolutionary shift in the framework under which travel was analysed came about in the 1970's, the origins of what has become
Emerging Developments in Time Use and Mobility
291
known as the activity-based approach to travel demand analysis. This activity-based approach spans a variety of theoretical and methodological approaches. Themes recurring in this work include: (1) analysis of the demand for activity produced participation-and the analysis of travel as a derived demand, (2) the scheduling of activities in time and space, (3) the constraints—spatial-temporal and interpersonal—on activity and travel choice, (4) the interactions between activity and travel decisions over the course of a day—or longer time period, as well as the interactions between different persons, and (5) the structure of the household. The major difference between the activity-based approach and the trip-based approach is in the treatment of time (Pas and Harvey, 1997). The activity-based approach incorporates the fullness of the time dimension as addressed below. The many elements of the activity based approach are incorporated in activity settings which emerge from time diary data (Harvey, 1982; Harvey, 1997). The major elements in the activity settings framework are actors, activities, space (location) and time. Figure 1 shows how these interact to provide a framework for analysing travel behaviour and its context (Harvey, 1997).
Actors
1 Social Contacts
Activities
2 Activity Patterns
3 Activity Transition
4 Individual Spatial Behaviour
5 Activity Location
6 Travel
7 Individual Temporal Behaviour
8 Activity Timing
9 Use Density
Actors
Activities
Space
Space Time
Figure 1 Key Elements of the Activity Based Approach The framework is usefiil both in highlighting the elements and relationships necessary to understand activity and travel behaviour and to understand the data collected via time diaries. The elements of the framework are highlighted below, following which data collection issues are addressed.
292
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Actors The significance of sex and employment status in determining the activity patterns of individuals pervades the time use and travel literature (Clark et al., 1980; Aas, 1982; Robinson et al., 1992). The presence of young children has been shown to be an important influence on time allocation to family, recreations and related diversions (Chapin, 1974). Given the constraints imposed by an individual's place in the family and the society work structure, and given the fact that approximately three quarters of all activity episodes are routine (Cullen and Godson, 1975), it is most fruitful to begin an analysis of behaviour patterns by focusing on groups in terms of life cycle and employment. The identification of such groups is generally rather straightforward, so they provide an operational starting point for activity analysis. Kutters (1973) offers a categorization which identifies separately housewives, pupils, students, and retired persons, thus extending the breakdown of those traditionally treated as unemployed. While not explicitly incorporating age groups, Kutter (1973) and Janelle et al. (1988) have an implicit life cycle built in, ranging from youth to retirement. The groupings by Janelle et al. (1988) were empirically derived thru the use of hypercodes (Clark, Elliot and Harvey, 1980). This approach yields operationally useful subpopulations consistent with behavioural findings.
Activity Activity definition and identification are difficult in that activities are multidimensional (Scheuch, 1972). The typical approach to data has been adjusting the criteria or boundaries (and, thus, definitions) to fit the problem being examined. Serious activity analysis will continue to suffer until appropriate taxonomic methods or schemata are developed, possibly along lines similar to those developed for industries, occupations and commodities. To date, the coding approach which was used for the multinational comparative time budget survey in the 1960s has provided some order and has become a de facto standard in the time use field (Szalai, 1972). It contains about 100 activities and travel is distighuised into 10 categories. That system, however, must be often collapsed into a more workable structure for analysis and presentation purposes. Previous work has suggested that a compromise between meaningful detail and workability can be achieved with eight categories of activities, excluding travel (Procos and Harvey, 1977). Their classification reflected economic concerns where regular and other market work covered market production, where household duties and family care covered non-market production.
Emerging Developments in Time Use and Mobility
293
and where sociological and planning concerns were reflected in home centered and non home centered discretionary activities. In terms of daily time allocation, life is dominated by sleep, paid work, housework and active leisure. The notable point of most coding schemes is that coding is based only on activity content or spatial (home/non-home) orientation. When and/or with whom the activities occur is essentially ignored.
Time Time has three primary dimensions that are relevant: 1. position, the point at which actions occur (e.g. weekday or weekend, morning or evening); 2. duration, the period during which actions occur (minutes are used as the duration metric) 3. sequence (before or after, past, present, ftiture). Each of these dimensions is relevant to transportation since each can affect or be affected by planning and design considerations.
Space Space has three distinct aspects of relevance: 1. geographical or banal space, represents arrangement and expanse, and is commonly thought of as form 2. adapted space, represents locations that have through design, development or mere habitual use, become the sites of continual, regular or recurring activities 3. channel space, serves to link adapted spaces, permitting linkage of activities within the city, road, rail, bike and walking routes. A space is not only a geographic location, it is a temporal location as well. This is clearly reflected in such terms as 'daytime population' or 'bedroom community'. Temporal location of a site may be established by any or all of the three temporal dimensions discussed above. Behavioural modeling must account for behaviour in terms of each of the elements discussed above. At one level, the interaction of all of these elements is manifested in individual behaviour, while at another level it is manifested in the patterns of the whole. Figure 1 shows the relevance of the linkages among the elements. Considerable work already exists in some of the areas identified, such as cell 4, individual spatial behaviour; and cell 6, travel behaviour. However, there has been little work in other
294
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
areas, such as cell 1, social contacts; cell 3, activity transition; cell 7, individual temporal behaviour; and cell 8, activity timing. These understudied cells promise to hold many of the insights needed to understand travel behaviour in a changing society. For instance, in the Netherlands research was commissioned to one of the authors of this chapter to address both individual temporal behaviour, activity timing and social contact to investigate the trend towards a 24 hour society. Questions to be answered were: •
Should working from home be promoted?
•
Should government transport policy be designed to encourage non-standard work patterns?
•
Should evening, Saturday and Sunday shopping be promoted for behavioural and travel reasons?
The 24-hour society project addressed, mainly, two key issues. First, what are the implications of emerging information technology and other forces of social change for patterns of work and shopping behaviour. Secondly, how might passenger travel be affected by the emergence of new locational and temporal patterns of work. In addressing the second question, the study explored the relationship between social contact and travel, a relationship of considerable concern because of anticipated negative social affects of emerging work and shopping patterns. Many of the measures relevant to address the above questions could have been derived from trip diary data. However, some could not. In particular, the location and timing of work at home would not typically be provided by trip diaries.
TIME DIARY DATA Time diaries can provide the information required to fill virtually all the cells of Figure 1, though for varying degrees of completeness depending on the specific survey content. The strength of the time diary is that it takes the respondent through the day in a natural progression. Nothing is to be omitted from the day. This helps the respondent to be more accurate in their reporting and it helps the coder or editor to be more accurate in their review and processing. Gaps and inconsistencies can be more readily observed. The best design is that used three decades ago in the multinational time use project (Szalai, 1972). The project aims were to systematically study and compare daily activities in different countries, to develop data collection methods and to promote cooperation, standardisation of
Emerging Developments in Time Use and Mobility
295
research techniques and the exchange of quantitative data at an international comparable level. It has subsequently formed a model for at least 50 national studies in various parts of the world (Gershuny, 1995). In the multinational time use diary respondents are asked to step through the day reporting what they were doing, what else they were doing, where they were, who they were with. These are the main transitions and they can be captured most completely and most accurately with time diaries since respondents are informed that a change in any of them should trigger a new diary entry. Two examples can clarify this. Often respondents are doing two things at the same time. For example, they might report "reading the newspaper" and secondarily "going to work on the bus." This report enables the researcher to record a trip, since all travel can be assigned primary status. Respondents may be travelling home from work and pick up their spouse or a child on the way. This would be recorded as two separate events since there would be a change in the "with whom." In the multinational time use diary, a comment column was provided for each entry to permit reporting other descriptive information. The diaries were open interval diaries, which means that the start and end time of each entry was supplied by the respondent. This approach is preferable to one using fixed intervals where entries are registered in terms of fixed, say 10 or 15 minute, time slots. Respondents were asked to fill in all activities taking about 10 minutes or more. However, they were told to record any trips or changes of location, regardless of the length of time used. Properly designed and executed time diaries can provide accurate travel data from the viewpoint of collection, data capture and analysis. From time diaries one can measure the number of trips, trip purpose, origin and destination information, trip mode and trip duration. Additionally one can derive chains in terms of each of these dimensions. Additionally, time diaries are better able to capture all activities undertaken at each location with all the attendant detail including who one is with. This locational and social contact information is essential to properly track the impact of many changes taking place in the spatial-temporal regimen. The travel impact of tele-activities such as telework, teleleaming, teleconsulting can be examined and activities performed at the same location can be measured. In brief, diaries capture all of the elements, in terms of the activity setting (Harvey, 1982; Harvey, 1997).
296
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Time-space diaries have been undertaken within the framework of the multinational time diary proscription (Elliott, Harvey and Procos, 1976). Such studies greatly enrich the opportunities for spatial and travel analysis (Goodchild and Janelle,1982; Janelle and Goodchild, 1988). They permit study of travel directions and speed, integrated with all other aspects of daily life. The multinational time diary design was weak in some respects. Typically, waiting for transport or a walk between modes might well be missed. Respondents might or might not report mode changes. Additionally, the multinational study did not distinguish between driver and rider in car travel. These shortcomings, however, are due as much to implementation as to design. Some studies have captured such information with little difficulty. Further, it is now known that it is desirable to have data for more than one day (Pas, 1986) and for all household members (Jones, et. al, 1983). However, the multinational study collected only one diary per household and one day per respondent. Subsequent studies have captured multi-person household data and multi-day respondent data. One additional shortcoming of timediaries is that their demanding nature typically precludes data collection for the sample sizes typically used in travel studies. However, the quality and extent of the detail they provide somewhat compensates for the smaller sample size. Often time diary studies fail to collect background data, such as number of cars and number of drivers in the household, which are important in studying travel. However, this can be overcome by sensitizing designers of time-budget studies to the type of data needed. Data from the multinational time use study has been brought together with data from a broad range of subsequent studies, approximately 40 country/year combinations, in the Multinational Longitudinal Time-Budget Archive (MLTBA), now housed at the Data Archive at the University of Essex (Gershuny, 1995). That archive, however, does not contain episode or trip level data. It contains a limited number of background variables and aggregated durations of time allocated to 40 different activities including work travel, shopping travel and other travel, as three of the 40 categories. Trip level data for the multinational study plus many of the country/year combinations in the MLTBA is contained in an archive being developed by one of the authors of this chapter as part of the time use research program to Saint Mary's University. This data permits a ftill exploration of all aspects of daily behaviour in real time.
Emerging Developments in Time Use and Mobility
297
Although many different time use surveys have been conducted in many different countries over time, and many time use datasets are available for analysis, it must be noted that the nature of (travel and spatial) information in the data varies considerably from study to study. In summary, time diary studies can be an invaluable source of data for travel analysis, especially if designers of time budget studies are sensitized to the type of data needed. A time diary provides extensive and accurate data in full context of daily life, thus providing a clear alternative to the traditional, or even enhanced, travel diary. In the next section, some general remarks concerning the efficacy of using time use data to explore travel behaviour are provided. Following this, data from the MLTBA are used to show trends in travel cross-nationally and to illustrate the strengths of the time diary approach: the ability to measure disparate behaviour not easily captured even by enhanced travel diaries. At the same time, attention is turned to trends in activity patterns and mobility between 1975 and 1995 in the Netherlands to get a more detailed insight into the trends. In the Netherlands, most time use and a lot of mobility studies are based on five waves of timebudget data that were collected between 1975 and 1995. These involved respondents from the age of 12 and above. The reporting period consists of one week. Because the survey method has remained more or less the same since 1975, a unique series of repeated cross-sectional data is available. The trends are discussed within the context of changing labour market participation of women and their effect on mobility, and changes in leisure time use and travel.
Exploring Travel Behaviour with Time Use Data The efficacy of using time use data to explore travel behaviour has been shown by many researchers. The earliest application was contained in the report of the multinational time use study conducted in the mid 1960's (Javeau, 1972). Javeau examined problems arising from commuting between home and the workplace. In particular, he explored the time tabling of the journey to work of the Belgian population. Developing analysis for two separate groups, one consisting of the five largest Belgian agglomerations, and the other of small or medium size agglomerations, Javeau found that while the separate areas followed very much the national pattern, there were notable iocal' differences. He found that the evening rush hour was 30 minutes later for the smaller places and the noon patterns were different. Javeau fiirther found that the shortest mean of journeys
298
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
occurs at the modal points of each rush hour. He argued that while time budgets appear to be an important tool for measuring constraints on people, they have some shortcomings. One key shortcoming was the lack of spatial data, while another was the need for better traffic data for various parts of the day. During the late 1970's, 1980's and 1990's time use data were used to explore a number of dimensions of travel behaviour. As noted by Janelle, Goodchild, and Klinkenberg (1988) research addressed a number of travel related issues including the complexity of travel linkages and trip chains (Burnett and Hanson, 1979; Pas 1983; Dangschat et al., 1982); and individual, social group and international variation in travel patterns and trip indicators (Hanson, 1982; Hanson and Hanson, 1980 and 1981; Janelle and Goodchild, 1983; Pas, 1984, and Robinson et al., 1992). Time use data proved to be worthwhile due to the fact that activity diaries give comparable or even better results than travel diaries for the amount of trips made (Van der Hoorn, 1983; Kalfs, 1993; Kalfs and van der Waard, 1994) and the number of different activities in activity diary research is usually much larger than the number of trip purposes in trip diaries. The usefulness of (comparable) time diary data is illustrated by showing trends in travel (cross-) nationally. The trends are discussed within the context of changing labour market participation of women and changes in leisure time use and travel.
Changing Labour Market Participation of Women Average daily travel time per person, as an indicator of differing travel behaviour crossnationally and changing travel behaviour over time, is available for a relatively large number of survey locations and time periods in data from the MLTBA. To enhance comparability, data in the archive has been standardised for the population group aged 20 to 59 years and appropriately weighted to reflect population behaviour. This data shows that, at least for the particular population grouped aged 20-59, variability of travel time is the rule, not the exception. What is particularly interesting is the cross-temporal changes in travel in the several countries for which there is more than one year of travel data from time diary studies. In general, it can be observed that data were available in the respective countries.
Emerging Developments in Time Use and Mobility
299
The archive data also provides an opportunity to explore cross-nationally and cross-temporally the differences in travel time between men and women. From 1961 through to the mid 1970's women's total travel time was about two thirds that of men, Figure 2. However from 1975 through to 1981 women's travel time was about three quarters that of men. Since 1981 women's travel has further increased relative to men's travel and is now in the order of 80 to 90% that of men, Figure 2. Comparable results were found in the Netherlands. From table 1, it is clear that between 1975 and 1995 the travel behaviour patterns for men and women became more similar, although some notable differences in tripmaking between men and women still exist.
Finland 1989 Netherlands 1985 USA 1985 UK 1985 Norway 1981 Norway 1981 Canada 1981 Italy 1980 Finland 1979 Netherlands 1975 USA 1975 UK 1975 France 1974 Australia 1974 East Germany 1965 Czechoslovakia 1965 France 1965 Yugoslavia 1965 West Germany 1965 USA 1965 Hungary 1965 Belgium 1965 Poland 1965 UK 1961
0.0%
20.0%
40.0%
60.0%
80.0%
Figure 2 Female as % of Male Travel Time, Population 20-59 Selected Countries and Years 1961-1989
100.0%
300
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Table 1 Trends in Mobility in the Netherlands 1975,1985,1995 weekly basis
1975
1985
1995
men
women
men
women
men
women
number of trips
31.1
29.5
30.5
30.1
31.5
31.8
time use on trips (hrs)
10.0
7.5
10.0
8.3
11.6
9.7
number of car trips
8.8
4.8
9.5
6.2
11.1
8.4
time use on car trips (hrs)
4.7
2.1
5.0
3.0
6.0
3.8
number of trip chains
3.7
3.9
3.7
3.9
4.0
4.2
Source: MuConsuh (1997)
The weekly number of trips increased for women and remained stable for men between 1975 and 1995. As a consequence, the time spent on travel increased more for women than for men. The same holds for the number of car trips: during the research period for women the number of ear trips increased by 75%, whereas for men it increased by 26%. Gender differences have, therefore, become smaller between 1975 and 1995. The same conclusion holds for the time spent on car trips. With respect to the number of trip chains, no differences were found, neither if 1975 is compared to 1995, nor if men are compared to women. One of the reasons why gender differences have become smaller is that over the past decades, there have been considerable demographic changes in the population of the Netherlands. Changes in the Netherlands only mirror what is happening elsewhere. Among the most significant social themes that have emerged in the latter half of this century are the increasing proportion of women in the workplace, and the changing structure of the family. In the Netherlands, for the population of 15-64 year olds, labour force participation increased between 1975 and 1995 from 54% to 63%. The rate for males fell from 81% in 1975 to 76% in 1996, while for women it increased from 30% to 49% (Sociaal en Cultured Planbureau, 1996). The increased participation rate of women and the decreased rate for men have led to major changes in the type of households. Females in an earning couple have increased from 4.2% of the population to 14.8%. Over the same period 'traditional' male breadwinners fell from 25.4% to 10.2% of the population. However, males in an earning couple are up from 9.1% to 17.4% due to the entry of wives into employment. Changes such as those reflected above, can be expected to have a major effect on time use and travel. Consequently, it will be important to take account of such change in order to understand
Emerging Developments in Time Use and Mobility
301
activity patterns and travel behaviour. It was expected that these changes lead to a growling number of people who combine paid work with taking care of domestic duties and to a growing similarity in the pattern of travel for men and women.
Changes in Leisure Time A particular strength of time diary studies is their ability to capture activities that are far less defined and predictable than are work, education and shopping travel. While it is difficult for respondents to recall or distinguish and report all leisure activities, they can report all activities as a stream of consciousness. That stream can be interpreted and the activities and related travel can be integrated into the analysis of travel behaviour. The data from the MLTBA can be called on again to provide insights into the cross-national and cross-temporal distribution of leisure and of leisure travel. Leisure time is not equally distributed over time and space, at least for respondents 20-59 years of age, as shown in Figure 3. For the sites examined in Figure 3, the Netherlands registered the highest level of leisure time, just over 390 minutes (6.5 hours per day), and France and Norway registered the lowest level of leisure time. In general, men register higher levels of leisure than women, although average leisure time was about equal between men and women in the Netherlands in both 1975 and 1985 and in the US in 1985. Interestingly, there is no clear pattern of change in leisure across the sites portrayed in Figure 3. In some cases leisure increased over time, in others it decreased, changed little, and in still others it went up and down. A different pattern exists for leisure travel. While total leisure remained essentially constant in the Netherlands and Finland between the study periods, leisure travel time increased. Figure 4. Leisure travel time also increased in France and the UK over the study periods. Again, however, there were other patterns. In the US, leisure travel time increased from 1965 to 1975 and fell in 1985 for men. For women it decreased continually over that period. For Norway, leisure travel time fell slightly between 1970 and 1980. From the above, it is clear that leisure behaviour differs both cross-nationally and crosstemporally. In general, leisure travel time increased, although other patterns were found.
302
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Finland 1987 Finland 1979 USA 1985 USA 1975 USA 1965 UK 1985 UK 1975 UK 1961 Norway 1980 Norway 1970 Netherlands 1985 Netherlands 1975 France 1974 France 1965
60
120
180
Male
240
•
300
360
420
Female
Figure 3 Average Leisure Time, Persons 20-59 by Gender, Selected Countries and Years
In the Netherlands, because official working hours declined and the number of individuals not engaged in paid work increased due to disability and early retirement, an increase in leisure time was expected over the years. Figure 3 shows that when a normal week is considered, the expected increase in leisure time failed to materialise. The 1995 time use data show an increase in the time required to meet personal obligations, specifically time spent on paid work. In addition to the stability in leisure time, it was also found that the number of leisure trips remained constant between 1975 and 1995. However, differences were found in the purpose for which trips were made and the time spent on trips. In Table 2 it is shown that trips with respect to passive leisure tend to decrease in number, whereas active leisure trips increased in number as well as in time use. In particular, the leisure time spent outside the municipality of residence increased. The time spent both per leisure activity and leisure trip increased over these years, although the time spent on leisure trips increased more.
Emerging Developments in Time Use and Mobility
Finland 1987 Finland 1979 USA 1985 USA 1975 USA 1965 UK 1985 UK 1975 UK 1961 Norway 1980 Norway 1970 Netherlands 1985 Netherlands 1975
L^%.'g.,^l^l.'»L<s8.'^
|, \ I '\ 11W111f \ '\\ 1 ' 1 1 W 'I'SBBfa pgjiHBBBHBiMiiMi ^gggggggggsBOaSSLm lUhUMMWttWWUUb jBHiBBiiiiBBBBiaBB&ib
r°
Female
^dMMAUUIdhUMUUyiUUiMi yyu^ttilUKttlAUyyyyyiiBli
Male
(sagsBiBBSiBBiiBBb
France 1974 France 1965 10
20 30 Minutes per Day
40
50
Figure 4 Average Leisure Travel Time, Persons 20-59 By Gender, Selected Countries and Years
Table 2 Trends in leisure mobility in the Netherlands 1975,1985,1995 weekly basis
1975
1985
1995
number of trips - passive leisure (e.g. social contact) - active leisure (e.g. sport)
11.5 7.5 2.4
11.3 6.7 2.8
11.4 6.7 3.1
3.3 3.4 2.3 2.2 0.6 0.8 Source: MuConsult (1997)
3.7 2.4 0.9
time use on trips - passive leisure (e.g. social contact) - active leisure (e.g. sport)
303
304
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
From the above it is clear that by applying (comparable) time use data, general trends can be examined (cross-)nationally. Travel time increased during, if not over the period for which data were available. Women's travel time increased relative to men's travel. And leisure travel time increased, although other patterns were found. In time use studies, these trends can be examined more closely in terms of number and time use per trip, per trip purpose or per trip mode, and can be related to societal trends like the changing labour market participation of women, as shown for the Dutch situation.
CONCLUSION This chapter has discussed the origins and the potential of cross-national and cross-temporal time use data to help understand travel behaviour and its context. Properly designed and executed time diaries can provide accurate travel data from the viewpoint of collection, data capture and analysis. Over the past decades many different time use surveys have been conducted in many different countries. This chapter illustrated the exploration of travel behaviour with time use data. To provide more concrete insight into the value of time use data, trends in activity patterns and mobility between 1975 and 1995 for the Netherlands were discussed within the context of changing labour market participation of women and their effect on mobility, and changes in leisure time use and travel. From the results it is clear that labour participation of women has partly led to a more similar pattern of travel behaviour for men and women, although some notable differences between men and women still exist. The expected increase in leisure time failed to materialise. Differences were found in the purpose for which leisure trips were made and the time spent on trips. Although many different time use datasets are available for analysis, it must be noted that the nature of travel and spatial information in the data varies considerably from study to study. The shortcomings are due as much to implementation as to design. Sensitizing designers of timebudget studies to the type of data needed, is therefore an important challenge. Various new technologies also offer considerable opportunities to facilitate the collection, recording, and processing of time use and activity data (Doherty and Miller, 2000).
Emerging Developments in Time Use and Mobility
305
REFERENCES Aas, D. (1982). Design for large scale, time use studies of the 24-hour day. In: Staikov, Z. (Ed.). It's about time. Sofia, Bulgaria. Burnett, P. and S. Hanson (1979). Rationale for an alternative mathematical approach to movement as complex human behaviour. Transportation Research Record 723, 11-24. Chapin, F. S. (1974). Human activity patterns in the city: What do people do in time and space. John Wiley, Toronto. Clark, S. M., D. H. Elliott and A. S. Harvey (1980). Hypercodes and composite variables: Simple technique for the reduction and analysis of time budget data. Paper prepared for meetings of the International Research Group on time Budgets and Social Activities. Sofia, Bulgaria. Cullen, I. and V. Godson (1975). Urban networks: the structure of activity patterns. Progress in Planning A (1). Dangschat, J., W. Droth, J. Friedrichs and K. Kiehl (1982). Action spaces or urban residents: An empirical study in the region of Hamburg. Environment and Planning A 14, 11551174. Doherty, S. T. and E. J. Miller (2000). A computerized household activity scheduling survey. Transportation 27(1), 75-97. Elliott, D. H., A. S. Harvey and D. Procos (1976). An overview of the Halifax time-budget study. Society and Leisure 3, 145-159. Ettema, D. F. and H. J. P. Timmermans (1997) (Eds.). Activity-based approaches. Elsevier Science Limited, to appear. Gershuny, J. (1995). Time budget research in Europe. In: Statistics in Transition 2(4), 23. Goodchild M. and D. Janelle (1982). Diurnal patterns of social group distributions in a Canadian City. Economic Geography 59(4), 403-425. Hanson, S. (1982). The determinant of daily travel-activity patterns: Relative location and socio demographic factors. Urban Geography 3, 179-202. Hanson, S. and P. Hanson (1980). Gender and urban patterns in Uppsala, Sweden. Geographical Review 70(3), 291-299. Hanson, S. and P. Hanson (1981). The travel activity patterns of urban residents: Dimensions and relationships to socio-demographic characteristics. Economic Geography, 57(4), 332-347. Harvey, A. S. (1982). Role and context: Shapers of behaviour. Studies of Broadcasting, 18, 6992. Harvey, A. S. (1997). 24-Hour Society and Passenger Travel. Report Time Use Research Program. Halifax, Canada. Janelle D. and M. Goodchild (1983). Diurnal patterns of social group distributions in a Canadian city. Economic Geography, 59(4). Janelle, D., M. F. Goodchild and B. Klinkenberg (1988). Space-time diaries and travel characteristics for different levels of respondent aggregation. Environment and Planning A, 20, 891-906. Javeau, C. (1972). The trip to work: An essay on the application of the time-budget method to problems arising from commuting between residents and workplace. In A. Szalai, (Ed.) The use of time: Daily activities in urban and suburban populations in twelve countries. Mouton, The Hague.
306
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Jones, P., et al. (1983). Understanding Travel Behaviour. Great Britain:Antony Rowe Limited. Kalfs, N. (1993). Hour by hour: effects of the data collection mode in time use research, PhD dissertation. NIMMO, Amsterdam. Kalfs, N. and J. van der Waard (1994). Kwaliteit van gegevens van tijdbestedings- en verplaatsingsdagboeken. (Data quality in time use and trip diary surveys). In: J. M. Jager, (ed.). Colloquium Vervoersplanologisch Speurwerk - 1994 - Implementatie van beleid. De moeizame weg van voornemen naar actie. C.V.S., Delft. Kutter, E. (1973). A model for individual travel behaviour. Urban Studies 10(2), 235-258. MuConsult (1997). Tijdsbestedingsonderzoek 1995. Ontwikkelingen 1975-1995 mobiliteit en tijdsbestedingen. (Time use survey 1995. Trends and developments 1975-1995 in time use and mobility.) Amersfoort. Pas, E. I. (1983). A flexible and integrated methodolody for analytical classification of daily travel-activity behaviour. Transportation Science 17(4), 405-429. Pas, E. I. (1984). The effect of selected socio-demographic characteristics in daily travelactivity behaviour. Environment and Planning 16, 571-581. Pas, E. (1986). Multi day samples parameter estimation precision, and data collection costs for least-squares regression trip generation. Environment and Planning A. 18, 73-87. Pas, E. I. and A. S. Harvey (1997). Time use research and travel demand analysis and modeling. In: P.R. Stopher and M. Lee-Gosselin (Eds.). Understanding travel behaviour in an era of change. Pergamon Press, Oxford. Procos, D. and A. S. Harvey (1977). Modelling for local planning decisions. Ekistics, 44 (264), 257-266. Robinson, J. P., R. Kitamura and T. F. Golob (1992). Daily travel in the Netherlands and California: A time diary perspective. Hague Consulting Group, The Hague. Scheuch, E. K. (1972). The time-budget interview. In: Szalai (Ed.) (1972). The use of time. Mouton, The Hague. Sociaal en Cultured Planbureau (1996). Sociaal en Cultured Rapport. Rijswijk. Stopher, P. R. and M. Lee-Gosselin (1997) (Eds.). Understanding travel behaviour in an era of change. Pergamon Press, Oxford. Szalai, A. (1972) (Ed.). The Use of Time, Proceedings of the World Time-Budget Study. Mouton Publishers, The Hague. Van der Hoom, A. I. J. M (1983). An empirical model of travel and activity choice: a case study for the Netherlands. Kris repro, Meppel. Van der Hoom, A. I. J. M (1997). Practitioner's future needs. Resource paper prepared for the International Conference on Transport Survey Quality and Innovation, May 24-30, 1997, Grainau, Germany.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
15
TIME USE AND TRAVEL DEMAND MODELING: RECENT DEVELOPMENTS AND CURRENT CHALLENGES
Eric I. Pas
EDITOR'S NOTE. This chapter was still a work in progress at the time of Eric's untimely death. It has all the makings of a Pas classic, seminal work that will be quoted for years to come. I have chosen to leave it unperturbed, in raw form. This chapter is so thoughtfully and carefully constructed, that only Eric could have completed it. It will remain a testimonial to his deep contribution to the field, and an immortal reminder of the spirit he so generously shared with his colleagues. With affection and respect to his memory. HSM
AUTHOR'S NOTE TO THE READER Please note that this paper is incomplete, as indicated at specific places in the text. I apologize for any inconvenience this may cause you, the reader. My personal time allocation model appears to have a substantial, systematic bias - it consistently overestimates the number and duration of activities that I should be able to undertake in each time period (day, week, month and year).
308
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
ABSTRACT This paper examines two recent and continuing trends in research and application at the interface between time use studies and travel demand modeling; namely, the utilization of time use diaries in collecting information about activity participation and the related travel behavior, and the development of models of time allocation to activities, activity duration, and travel behavior. The paper first places these recent and on-going developments in travel demand modeling in their historical context, and then explores the utilization of time use diaries in travel demand analysis and modeling, including an examination of the experiences in recent travel behavior surveys in a number of metropolitan areas in the United States. A number of recent and current efforts to develop time allocation and duration models, sometimes in conjunction with travel behavior models, are discussed. The modeling efforts examined here utilize a variety of modeling methodologies, including structural equation models, hazard-based duration models, and utility maximization models. The paper concludes by identifying some areas of time use studies which present challenges for travel demand researchers and which are suggested as issues for discussion at the.workshop.
INTRODUCTION Travel behavior research, and the development of travel forecasting models, entered a new era in the mid-1970's with the birth of the activity-based approach in the ground-breaking research undertaken at the Transport Studies Unit at Oxford University. The development of the activity-based approach represented a revolutionary change, a paradigm shift, from the tripbased framework that was developed in the earliest urban land-use/transportation studies beginning in the U.S.A. in the mid-1950's.* Much research has been done in the past 20 years and the activity-based approach has led to a much greater understanding of urban travel behavior (for reviews of the activity-based approach to travel analysis and modeling, see Kitamura, 1988; Jones, Koppelman and Orfeuil, 1990; Axhausen and Garling, 1992; Jones, 1995; Kurani and Lee-Gosselin, 1996; Pas, 1996). Until recently, the emphasis in the activity-based approach was on analysis and understanding, and there was little interest among practitioners in implementing the concepts and methods of the activity-based approach. However, the activity-based approach has rapidly become recognized as an important and viable approach to travel demand forecasting in the past few years. The renewed interest in this approach results primarily from the fact that contemporary policy and planning questions require precise and accurate models of travel behavior and thus need to be based on a thorough understanding and representation of this behavior. There has been considerable recent attention to the development of travel forecasting models based on the activity-based travel approach to travel demand, and the recent work employs a rich variety of
Time Use and Travel Demand Modeling
309
methodologies (see Pas (1996) for a discussion of this work, and Bhat (1997) for a review of the recent methodological advances). The fundamental, and yet obvious, premise of the activity-based approach is that travel results from the needs and desires of persons to engage in activities at various locations. Thus, travel is considered a 'derived demand', where the demand for a product or service arises because of the demand for some other, related, product or service. Given the importance of activity participation in shaping travel behavior, is it not surprising that in recent years there has been an increased interest in time use studies among travel behavior researchers and travel demand modelers. At the 6^^ International Conference on Travel Behavior Research held in 1991 in Quebec City, Pas and Harvey (1991) presented a paper in which they reviewed the field of time use research and examined the implications of this work for travel demand analysis and modeling. That paper concluded that mutual benefits would accrue from greater interaction between these related fields of research. The past 6 years have seen increased levels of interaction between the two communities of researchers, culminating in a workshop devoted to time use at the present conference, including the participation of time use researchers^. In fact, during the past few years, the concept of time has moved from relative obscurity to center stage in travel demand analysis and modeling (see Pas, 1997, for a comprehensive discussion of the changing role of time in travel demand analysis and modeling). This paper is focused on two recent and on-going developments in the area at the interface of time use studies and travel demand modeling; namely, the utilization of time use diaries in travel behavior surveys, and the modeling of activity time allocation and duration. The remainder of this paper is organized as follows. First, we discuss the utilization of time use diaries for conducting travel behavior surveys. Then, we describe recent and on-going efforts to model time allocation to activities, and activity duration. The paper concludes with the identification of some current issues in time use studies and travel behavior that should be discussed at the workshop on time use.
TIME USE DIARIES FOR TRAVEL DEMAND STUDIES This section of the paper first provides a very brief overview of the field of time use (time budget) research with a focus on the collection of time use data. Then we describe briefly the development of household travel surveys, specifically in the USA, followed by a discussion of
310
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
the recent utilization of time use diaries in household travel surveys in three metropolitan areas in the USA.
Time Use Surveys and Studies Studies of what people do with their time date back a long time, at least to the early part of this century, and the first time-use study having a specific focus on travel was undertaken more than 50 years ago (Liepmann, 1944). In the late 1960's, a large, multinational comparative time budget project was undertaken in which nearly 28,000 standardized time use diaries were collected in 12 countries (Szalai, 1972). Consistent coding schemes have been developed to classify activities into detailed categories, facilitating comparative studies of these time use data sets. Over the past 20 years numerous time use studies have been conducted throughout the world, including a number of longitudinal studies. Pas and Harvey (1991) present an extensive discussion of time-use studies and their relationship to travel demand modeling. Time use data generally provide information on what individuals do over the course of a day or, in some cases, over several consecutive days. Information about activities in the home, as well as outside the home, is usually included in a time use data set. Some time use data sets contain information on activity location and travel, but most focus on the amount of time devoted to activities. A number of methods have been used for sampling the activities undertaken by the selected respondents. These are described here as follows : • time point, • interval, and • daily (or multiday) record."^ In the time point method, activities are sampled at time points that are either randomly generated or recur at fixed intervals (e.g., every 15 minutes), and the activity being performed by the respondent at each time point is recorded. This sampling method leads to unbiased estimates of the time allocated to each type of activity, but it will lead to an underestimation of the frequency of activity episodes, especially for short duration activities. In the interval-based sampling approach, the representative activity, or activities, are recorded for each time interval. The representative activity is typically defined as the activity to which most time is allocated in the selected interval. When a single representative activity is recorded, this sampling method
Time Use and Travel Demand Modeling
311
leads to the systematic bias that activities of shorter durations are under-represented, both in their frequencies and the amounts of time allocated to those activities. The approach in which sampled respondents are asked to record information on all activities in which they participated over the course of a day, or multiple days, is probably most suited for travel behaviour analysis. In this approach, information is recorded for each activity episode over a day, or multiple days, and the data obtained will reproduce both episode frequency and time allocation properly, provided that the respondents record all activity episodes. Data are collected by this method most typically by asking the respondent to recall and describe (often in an open-ended manner) the activities performed on the day before the survey, activity by activity sequentially as they were performed across the day. The focus of the analysis of time-use data has been on time allocation to selected activities, and the time-use patterns of specific sub-groups of the population. The analyses have mainly been descriptive in nature and little attention has been devoted to the modeling of time-use, activity sequencing, and activity location choice. Of course, time use data facilitate thorough analysis of activity engagement as well as trip making. In particular, time use data allow for analysis of the substitution between in-home and out-of-home activities (see Section 3 for a discussion of the results of such analyses).
The Development of Household Travel Surveys in the USA^ The first household travel surveys in the USA were carried out approximately 50 years ago. In the earliest type of these surveys, known as a home interview survey, an interviewer would visit the home of each selected household, typically on the day following the day for which the household members were asked to report their travel. This reliance on the respondent's ability to recall their travel on the previous day was probably mitigated, at least to some extent, by the advantages of a face-to-face interview and the interviewer's resultant ability to probe for "missing" trips (Lawton and Pas, 1996). The conduct of household travel surveys has changed considerably over the years since the earliest ones were undertaken. First, beginning in the mid to late 1970's, those urban areas which conducted household travel surveys started making do with a much smaller sample size. This trend was made possible by the introduction of disaggregate models, which make far more efficient use of the data as compared with the aggregate models that were in use prior to that
312
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
time. The last home interview surveys in major US metropolitan areas appear to be those undertaken in 1977 in Portland and Baltimore, and the survey in Dallas in 1984. In the early 1980's, other methods for contacting sample households and for retrieving the travel and related data started to be used in household travel surveys. Specifically, the use of the telephone for household travel surveys became popular in the 1980's, based in early experiences in the San Francisco Bay Area (Reinke, 1993). While telephone surveys have a number of potential drawbacks, the disadvantages can be mitigated by careful survey design. In order to help the respondents remember their travel on the survey day, it also became quite common in the 1980's to send respondents a diary, ahead of the survey, and respondents were asked to use the diary to record their trips (see, for example, Stopher and Sheskin, 1982). The idea being that when the interviewer called to retrieve the travel information, the respondent would be able to use the diary to help her recall her trips on the survey day. Both the earlier home interview surveys and the more recent telephone-based household travel surveys, were "trip" surveys in that the key question in these surveys was "Where did you go next". That is, the interviewer prompted the respondents for the information of interest by asking about trips undertaken during the course of the survey period. The earliest travel surveys, in fact, limited the definition of "trip" to include only movements made by a motorized means of travel, and thus trips made by bicycle and on foot were excluded from those surveys. Thus, respondents were asked to remember each trip they made during the survey period, typically the 24-hours preceding the survey, and when travel diaries were introduced they were structured around the trips made by the respondents, with each record in the diary pertaining to one trip. In 1991, a survey was conducted in the Boston area using a fundamentally different approach to asking about the trips undertaken during the survey period. In this survey, the focus shifted from travel to activities, and instead of asking "Where did you go next ?", respondents were asked "What did you do next ?". Because of the focus on activities, Stopher (1992) refers to this survey as an activity-based diary. Based on the results of the pilot for this survey, Stopher (1992) reported that the activity diary appeared able to capture non-home-based trip-making better than travel diaries. He also reported that the overall trip rates (both per person and per household) from the activity diary survey are significantly higher than those measured by most travel diaries. Thus, this study seems to provide evidence that an activity diary is better able than a travel diary to help respondents recall what they did on the survey day, particularly the short, non-motorized trips that would otherwise be forgotten and not reported.
Time Use and Travel Demand Modeling
313
Some recent household travel surveys in the USA have seen yet another important change in the conduct of such surveys. Beginning with a survey conducted in Oregon and Southwest Washington in 1994/95, some household travel surveys in the USA have moved beyond the type of survey conducted in Boston in 1991, by including both in-home and out-of-home activities in the survey. In this sense, these surveys are very similar to "time-use" surveys, and yield essentially the same type of data. The survey in Oregon and Southwest Washington was followed by similar surveys in the Research Triangle Area (Raleigh-Durham-Chapel Hill) of North Carolina in 1994/95 and in the San Francisco Bay Area in 1996. These three surveys are described briefly below.
Application of Time Use Diaries in Three Recent Household Travel Surveys in the USA The household travel surveys conducted recently in Oregon-Southwest Washington, RaleighDurham, and San Francisco, represent the state-of-the-art in the application of time use concepts in household travel surveys in the USA. Before discussing the differences among these surveys, we first note their common features. First, all three of these surveys asked respondents to report their activities and travel for a 48-hour period, rather than the 24-hour period that has commonly been used in household travel surveys. This so-called multiday design was used in these surveys to take advantage of the economies of scale to be gained from a multiday survey and/or to gain information on the day-to-day variability in activity and travel behaviour (see Lawton and Pas, 1996; and Pendyala and Pas, 1997, for discussions of the advantages and disadvantages of multiday surveys). Second, the surveys conducted in Oregon-Southwest Washington, Raleigh-Durham, and San Francisco all collected information on both in-home and out-of-home activities. As noted above, these three surveys are therefore similar to those surveys conducted by time-use researchers in which daily or multiday records of activity participation, episode by episode, are obtained. However, there are some differences, as discussed below, between these household travel surveys and traditional time use surveys. From the point of view of the present discussion, the major difference amongst the OregonSouthwest Washington, Raleigh-Durham, and San Francisco surveys has to do with the way in which in-home activities are treated. In the Oregon-Southwest Washington survey, respondents were asked to report only those in-home activities having a duration of 30-minutes or longer, although the original intention was to have respondents report all activities, both in and out-ofhome. There were two reasons for limiting the reporting of in-home activities to those taking
314
In Perpetual Motion: Travel Behavior Research Opportunities and Challenges
more than a specified period of time. The first reason for the change in plans had to do with the large number of activities that were reported when respondents were asked to record and report all in-home activities. The second reason being that the interviewers found that respondents sometimes has difficulty reporting an activity for periods when they were doing multiple activities at home - for example, during the period immediately after returning from work, when respondents were often engaged in reading the newspaper, interacting with family members, reading the mail, helping with meal preparation, etc. In the Raleigh-Durham survey, rather than limiting the recording and reporting of in-home activities according to some pre-specified minimum duration, it was decided to ask respondents to report all in-home activities but to specifically identify only those in-home activities that could also be conducted out-of-home. Thus, in-home activities such as sleeping, housework, home repair, etc. were to be reported only as "home" activities, while activities such as eating or preparing meals, exercise, etc. were to be specifically identified. This approach seems to have worked well in this survey. The 1996 Bay Area Travel Survey used the same approach as the Oregon-Southwest Washington Activity and Travel Survey, in that respondents were asked to record all activities that lasted at least 30-minutes or that involved travel. Interestingly, the Bay Area Survey included travel as one of the activities, whereas the other two surveys discussed here treated travel separately, as a means of getting to/from activities such as work, school, shopping, etc. Another interesting point to note is that the Raleigh-Durham and San Francisco surveys used activity diaries that were much less intimidating than that used in the Oregon-Southwest Washington survey. NOTE: This section of the paper is incomplete. The final version of the paper will include a comparative assessment of the three household travel surveys discussed briefly above, including a comparison of response rates, activity participation rates, activity durations, trip rates, etc.
MODELS OF TIME USE (TIME ALLOCATION AND ACTIVITY DURATION) IN TRANSPORTATION In the past 3 years, a large number of research efforts have been undertaken by travel behaviour researchers to develop models of how people use their time. In many ways, this research can be seen as taking the activity-based approach to travel analysis and modeling to a new level, since these models aim to predict what people do with their time, that is activity participation, which
Time Use and Travel Demand Modeling
315
is the underlying rationale for travel and the basis for the activity-based approach. This section of the paper discusses models of time allocation to activities in transportation research. Table 1 provides a summary of time use (time allocation and activity duration) models in transportation. The information in this table reveals that while there were some attempts to model time use in the late 1970s and early 1980s, the vast majority of the research in this area has been conducted in recent years, and there has been a considerable amount of such work. As can also be seen in the table, some of the models discussed here deal with the time devoted to a set of activities over a period of a day (or week), while others focus on activity duration (that is, the duration of specific activity episodes). Further, the models discussed here include those that deal only with time allocation as well as those that deal with time allocation and travel behaviour in a joint modeling framework. A variety of modeling approaches have been employed in the recent efforts to model time allocation, including structural equation models, utility maximizing models, and hazard-based duration models. Each of these modeling approaches is described briefly below, and examples of each approach are provided and discussed. The results obtained from such models are described.
Structural Equation Models of Activity Time Allocation and Activity-Travel Interrelationships Golob has pioneered the use of structural equation models in travel demand modeling, and he has applied the technique to various aspects of travel behaviour modeling. In recent years, Golob and his collaborators have begun applying the structural equation modeling methodology to modeling activity time allocation and activity-travel interactions (Golob, 1996; Golob, 1997; Golob et al., 1996; Golob and McNally, 1995). Golob and McNally (1995). Golob and McNally (1995) developed a joint model of out-of-home activity participation and the resultant travel of male and female couples (married or unmarried) who are heads of households. The research aimed at identifying the interactions between activity particpation and travel and between the two individuals being modeled. The research demonstrates the existence of, and provides quantitative estimates of the effects of out-of-home activity participation on travel behaviour and the interdependencies between the male and female household heads in their activity participation and travel.
M
4>
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 - <^ c<3 '-^2
=1 5^
•5-9 Q 13
M O 1 o (U
u c
00
.£
-O 00
D-
o --^ "So 2^ o £;
^
^ 00 ^ c S-55
|f^ >^-§
•^ c3 ^
{•til a,^
llir
a, t
VO
c: ^-^
o 2i
O '-'
CQ O
3
CO
eg 00
o ^ CQ C
S ^
i2 S
^ ^^ 1 c2 A 3
3
2 ^
S
>^ (u > ea 2 ' > - 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:
<
s s r
o CQ
o
o a
o
S
J
JS o (l>
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
U JS
ON
> 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 <2 '^
O HD
c«
D
O
ca
U g
T is § Xi
B 'B g .S ^ "3 .2 ^
O _0 c3 T3 •^ = : j d c x^ c3 O ea
i
en
o ^^ A
C o o
•^=
c
S
13 o
.2 cr
ON
o
-^
<
S
13 -o o c ,o
"S (U
2
"o
0^
on ON
i3 C/5
OS
X) O
o
'o
.^ 1 1 £5 2
317
o
o
2 H
o o u
'o
o —>
^4-1
V2
.2 o O
O 3 = O cd ^
1 D ^
^
*^ -o
\B
\'> -^
Q S
Q 13
a c
.2 =3
cr
3
II >-i
ON
ON
1—' ^
13
wo
§ ^
is
>.^ ^ •§
5 i"
.S"
°^| (L) >
111 3
^
jc c . 2 C/5
Q "5
o
E c .2 "ca a* a>
"3
2i "
o 3 is C/3
[ON ON
c
2 -Q
O m cu
o
-^ - ^
w
^
1^ i-J
c3
3
3
-
•> ^ -S -2 O
(u
(SJ ^ ^
tH
3
O
"
M1^1 O i-j^
^ '15 3
N
a
§
O
C O X) o
ON
<1^
•E -o |Df2
"1
'><
E L'
'N
DO C
E
13
-^ i •> "i -^ ^ C 3
c ^ — "O
1l o
11 s (/5
5 J 1 f*^ ON ON
00
<^ .E £ S
I—>
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 ^O '••3
e«
s
s T3
-2 Z
p S
.2
H
•S
y
=^
=2 • S
1.1 - 1 1 CO fa ^
«^ -3 .sa 6 H,
5 (L>
r-.S
> E2 i 1 •21 -8 i a ^
c O
C
a < u
c
13
^ .2 2 -^ ^ - 0 TJ
•^ i :> "^ ^
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
T3
0
(L>
•^ .5 3
^ i i ^ "i^
E B B T
3gs g
jO 03 1 3 ^- 03 C t P "z; . 5
'
1
-a " ^ 1 -
a- 0
i2 ^
N
0
1 3 ON ON
c
i
1
Time Use and Travel Demand Modeling
S C
Q. C «
c
^
S 2i
c« 0
^ ?5 2 ^ 0 -o c3 5: 0
c: g
3 0 2
g
1 2g ««< "^^ 53 "2 ? •••^ ^0 "^
•5 0 •g 0
§ g « >. ^ .
t> S H c ^ •? -^ b ^ .^ -B 0 ^ ^
Sr
03
0\
8 «
0
i
13
3
alii
5 « -f ^
3 0 1 i .§ I' 1 J9 i | . i 0 g .2 S t5 .>: 0 tS
•^ i i 0 S -2 a, Q 13 ^ c - 2
0
. ^ .ti
.£ ^ 0
^
3
0
Sa TD -§ i ^ ON ON
5 ^a i
2 0 -o x:
1 ^ >- tJH C/3
319
320
.2 o
2 H
O
O
^
5 2:
+-» ON
o D,
Ic o o is
i "§ 'O T3 C
o c GO
§•2 O
11 on
1^ w
f" ON CU
o x:
1 .2 .2. J3
S § E fe
3
3
•
^ O O 5? CJ or) O ^ C C c« T3
.2 'S ^-B "S '"' "^ '^ «. £ .=
13
tr iS 9 GO ">^ .5 ^ .3 o Q £ S i3 ^
IT:
.3 .2
.2 -^ "S '>
.£ .2 Q 13
^ ^ c3 cd
t, 0)
.2 <^ ^ ^ .£ o
"O
g .g- 00
t2 t
« ^ ^ -^
12^ -2^^ ^1 u x: -r « ^ o "o ^
<=^
Vk
1 c « •2 ^
3 T3
£ -c
ON ON
PC
N
•
'O (L> 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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
^
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
<
-a fi e« fi ^o '•*3
s
a H CU
c 3
^
sS3 o wj
.2 ^ ti ® ^ ^ 2
1-
t2 .2 "^ "S ^ 9O CO
o ^
J
C
u
CO
73 C CQ
« «n
H Z C B
§ i
•^^
-o
i8 9 00
t i cd
§3
CO
. 2 t>o -^3 C ^O .i2^ o fc
o c ^ CO
O
^
1 "^ P
O
M 3 ti T3 55 O C3 J3 D,
CO
£ ^
O
"« o
O 3
C O
£
C3 O
lis
H <2 '^ H o IS
s 11 ^? •3 ^ Q 13 S "S
13 O •=
O TD TD
^^ 2 H
CO X
CiD
i
£ -zs
^ 2 o "^ '"5
•E
^3 -a
^! fl e«
fi eQ
8 ^ § i S "^
£ '£
fi£
s O
1.1
1£ 3-
c
<^
'"^
0\
o>
t^
Q ^ Q
3
•^
V9
H
P
U) ^ \Os"^ s \Os""^ H
k^
[i
II
i
1
1
1
Time Use and Travel Demand Modeling
321
322
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Golob (1996) uses the structural equation modeling approach to model demand for activity participation and mobility, and he includes one category of in-home activity (namely, work) in the model. The model is formulated to allow for a number of hypothesized behavioural phenomena including: travel demand derived from activity participation, time budget effects, mobility demand (activity participation affects vehicle ownership), and accessibility (vehicle ownership affects activity participation). Other researchers have also begun applying structural equation models to activity time allocation and activity-travel interrelationships. Fujii et al (1997) developed a structural equation model system that explains commuters' time-use and travel after work hours. Their model shows that if the commute trip were reduced by 10-minutes, slightly more than 7 minutes will be used for in-home activities, thus bringing into question the idea of a constant travel time budget. Lu and Pas (1997a) use a structural equation model to study the relationships among sociodemographics, activity participation and travel behaviour, while Lu and Pas (1997b) use this methodology to study the interrelationships in time allocation to different activities on two consecutive days. The structural equation modeling approach has proven to be a useftil and flexible tool for studying time use and time use-travel interrelationships. A variety of hypotheses and relationships have been investigated using the structural equation modeling technique.
Hazard-Based Duration (Survival) Models of Activity Duration Hazard-based duration (or survival) models were originally developed for, and applied to, problems in the fields of medical science and industrial engineering, but they have also seen extensive application in economics (primarily labor economics) and marketing. Since the late 1980's, hazard-based duration models have also been applied to a number of transportationrelated phenomena, including travel demand. Hensher and Mannering (1994) provide a thorough review of the important concepts in hazard-based duration modeling and examples of the application of these models to transportation phenomena. They argue that hazard-based duration models provide the transport modeler with a powerftil tool and they note that there have been surprisingly few applications of these models in transportation modeling, especially since transportation modelers routinely deal with duration-related phenomena. Hazard-based duration models have been used to model the duration of activity episodes, including home-stay duration (i.e., the time between returning home and leaving on another trip). In this connection, the most pertinent work is that of Neimeier and Morita (1996), Mannering and his associates (Mannering et. al., 1994; Hamed and Mannering, 1993), Ettema
Time Use and Travel Demand Modeling
323
et. al. (1995) and Bhat (1996a, 1996b). (Another possible use of hazard-based duration models would be in modeling the time until the next activity of a particular type occurs. Thus, with the appropriate data, one could model the time between, say, shopping activities.) Mannering et. al. (1994) and Hamed and Mannering (1993) have applied hazard-based duration models to model the length of time a traveler spends at home before making another trip. Specifically, this work deals with the amount of time a commuter spends at home after arriving home from work before leaving home to take part in another out-of-home activity. Neimeier and Morita (1996) developed a model for the duration of particular trip-making activities based on gender. The activities they studied include : household and family support shopping, personal business, and free time. Neimeier and Morita found no significant differences between the durations of men and women for the free-time and personal business activities, but gender was a very significant explanatory variable in the case of the household and family support shopping activities, with women being more likely than men to have longer durations for household and family support shopping activities A recently developed duration model, developed by Ettema et. al. (1995), deals with both activity duration and activity choice by using what is known as a "competing risk" hazard model. The authors estimated the model using data collected from a small sample of students, through an interactive computerized data collection procedure called MAGIC, which they have developed to investigate activity scheduling behaviour (Ettema et. al., 1993). The estimated model parameters show that spatio-temporal constraints such as time of day, opening hours and travel time, play an important role in activity scheduling; Activity duration and type were also found to be dependent on the history of the activity-travel pattern and the traveler's priorities. The authors conclude that the estimated model performs satisfactorily, and holds promise for describing activity scheduling as a continuous decision-making process, although further development is needed to deal with some important technical issues. Bhat (1996a) has recently developed a hazard-based duration model of shopping activity duration on the trip home from work, while at the same time significantly extending the methodology of hazard-based duration models. Bhat (1996b) has also recently developed a multiple durations (i.e. competing risks) model that extends the existing state-of-the art considerably. Thus, there are a number of recent examples of the application of hazard-based duration models to activity duration modeling and examples of methodological developments as well (for a discussion of recent methodological advances in hazard-based duration models, see Bhat, 1997).
324
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Utility Maximizing Models of Time Allocation The concept of utility maximization has been one of the most influential in the development of travel demand models. This concept, of course, is the underlying assumption in the disaggregate choice models, such as the multinomial logit and nested logit models, that became popular in travel demand modeling starting in the 1970's., and which are the most commonly used model types for mode choice modeling, and which are also used in other discrete choice situations. The concept of utility maximization has been applied to the formulation of time use models by a number of researchers in the travel behaviour field, using a variety of modeling assumptions.
NOTE: This sub-section is incomplete. The following models will be reviewed and discussed in this sub-section: Bain (1979), Bhat (1996c), Jara-Diaz (1996), Kitamura (1984), Kitamura, van der Hoom, and van Wijk (1995), Kitamura et. al. (1996), Kraan (1995), Lawson (1996), Yamamoto and Kitamura (1997).
Discussion This section of the paper presents an examination of the large body of recent research works that attempts to model time allocation, either daily (or weekly) time allocation to a set of activity types, or the length of specific activity episodes.
NOTE: This sub-section of the paper is incomplete. The final version of the paper will include a comparative discussion of the large number of recent time use modeling efforts in the travel demand community, with a focus on the strengths and weaknesses of the various methodological approaches.
DISCUSSION AND WORKSHOP CHALLENGES This conference is entitled "Challenges and Opportunities in Travel Behaviour Research and Applications". This theme is particularly appropriate to the area of time use studies, as there are both challenges and opportunities, in both research and applications, at the interface between time use studies and travel behaviour analysis and modeling. This paper has shown that in recent years much research and development has taken place in the application of time use
Time Use and Travel Demand Modeling
325
diaries for household travel surveys and in the modeling of activity time allocation and activity duration. The purpose of this final section of this paper is to suggest some important issues for the workshop on time use to address during the 8^*^ International Conference on Travel Behaviour. The topics listed below should be seen merely as a starting point for discussions at the workshop. As one might expect in a rapidly developing sub-field, such as time use data collection and modeling for travel behaviour studies, some of the issues to be addressed are concerned with terminology and definition. The workshop participants should discuss the fact that there is no standard terminology for the different time use-type surveys that have been undertaken in the travel behaviour field. Stopher (1992) referred to the survey undertaken in Boston in 1991 as an activity diary survey since respondents were asked to record activities, not trips. The recent surveys in Oregon-Southwest Washington, Raleigh-Durham and San Francisco also asked respondents about activities, but asked about not only out-of-home activities (as in the Boston survey) but also in-home activities. Should such surveys be distinguished from the type of survey conducted in Boston ? If so, what term should be used ? (In this paper I have referred to these as time use surveys). Kurani and Lee-Gosselin (1996) and Kurani and Kitamura (1996) have recently raised awareness of the lack of a standard activity typology for use in travel behavior surveys and studies. In order to advance activity-based travel modeling, it is important that a standard activity typology be developed and adopted. The workshop participants should examine the issue of activity typology. Even though time use surveys have been conducted for many years, a debate is raging currently in the time use community regarding the interpretation of the data from such surveys, conducted in different ways, in connection with whether Americans are working more or less than 30 years ago. Thus, there may be opportunities for improving traditional time use surveys, and it is certainly possible that traditional time use surveys should be modified for application in travel behavior studies. The workshop should address the issue of whether and how traditional time use surveys should be modified to suit the needs of travel behavior studies. In the context of time use surveys, the workshop participants should examine whether there is a need to extend the usual time use survey to include information concerning expenditures made during activity participation, the planning of activities undertaken, and the satisfaction respondents feel with particular activity episodes and the day as a whole. Of course, including
326
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
any or all of this additional information would result in an additional burden being placed on respondents in already lengthy surveys. The workshop should therefore also consider how travel surveys could be made less burdensome on the respondents, especially if one or more of the above items of information is to gathered in addition to the time use data. The workshop participants should examine the issue that there is no comprehensive theory of time allocation, and its relationship to travel behavior, and discuss the development of such a theory. Specifically, the advantages and disadvantages of the various alternative approaches to the development of such a theory should be examined. The construction of a theory of activity time allocation can build on the work of Tonn (1984a,b), Garling (1992), Bhat and Koppelman (1993), and Jara-Diaz (1996), in the transportation field, and other work in other fields.
ACKNOWLEDGEMENTS The author gratefully acknowledges partial support for the preparation of this paper through grant number DMS 9313013 from the National Science Foundation to the National Institute of Statistical Sciences for the project "Measurement, modeling and prediction for infrastructural systems". The author also wishes to acknowledge that the contents of this paper have been influenced through discussions and collaborations with a number of colleagues over the years. I would especially like to recognize Chandra Bhat, Andrew Harvey, Nelly Kalfs, Ryuichi Kitamura, and Keith Lawton. Johanna Zmud kindly provided invaluable assistance through her willingness to review with me aspects of the recent surveys undertaken in Oregon-Southwest Washington, Raleigh-Durham and San Francisco. I hope I will not offend other colleagues whose names are omitted here. The author, of course, is solely responsible for any opinions expressed herein as well as for any errors.
REFERENCES (incomplete) Axhausen,K. W. (1995). The data needs of activity scheduling models. Paper presented at the EIRASS Conference on Activity-Based Approaches: Activity Scheduling and the Analysis ofActivity Patterns, Eindhoven, The Netherlands. Axhausen, K., Garling,T. (1992). Activity-based approaches to travel analysis: Conceptual frameworks, models and research problems. Transport Reviews 12, 323-341. Barnard, P. (1986). Use of an activity diary survey to examine travel and activity reporting in a home interview survey: An example using data from Adelaide, Australia. Transportation 13, 329-357. Becker, G. S. (1965). A Theory of the Allocation of Time. The Economic Journal 75, 493-517.
Time Use and Travel Demand Modeling
327
Bhat, C. R. (1997). Recent Methodological Advances Relevant to Activity and Travel Behavior Analysis. Prepared for presentation at the 8^^ International Conference on Travel Behavior, Austin, TX. Bhat, C. R. (1996a). A generalized multiple durations proportional hazard model with an application to activity behavior during the evening work-to-home commute. Transportation Research B 30(6), 465-480. Bhat, C. R. (1996b). A hazard-based duration model of shopping activity with nonparametric baseline specification and non-parametric control for unobserved heterogeneity. Transportation Research B 30(3), 189-208. Bhat, C. R. (1996c). A model of post home-arrival activity participation behavior. Manuscript in review for publication. Bhat, C. R, Koppelman, F. S. (1993). A conceptual framework of individual activity program generation. Transportation Research A 27, 433-446. Bhat, C. R. and Misra, R. (1997). Discretionary activity time allocation of individuals between in-home and out-of-home and between weekdays and weekends. Prepared for presentation at the 8^^ International Conference on Travel Behavior, Austin, TX. Ettema, D. A., Borgers, A.; Timmermans, H. (1995). A competing risk hazard model of activity choice, timing, sequencing, and duration. Paper presented at the 74th Annual Meeting of the Transportation Research Board, Washington, DC. Fujii,S., Kitamura,R., Monma,T. (1997). A study of commuters' activity patterns for the estimation of induced trips. Manuscript, Kyoto University. Garling, T. (1992). Determinants of everyday time allocation. Scandinavian Journal of Psychology 33, 160-169. Golob, T. F. (1997). A simultaneous model of activity participation and trip chain generation by households. Prepared for presentation at the 8^^ International Conference on Travel Behavior, Austin, TX. Golob, T. F. (1996). A model of household demand for activity participation and m^obility. Presented at the Conference on Theoretical Foundations of Travel Choice Modeling, Stockholm, August 7-11. Golob, T. F. (1990). The dynamics of household travel time expenditures and car ownership decisions. Transportation Research A 24, 443-. Golob, T. F., Bradley, M. A., Polak, J. W. (1996). Travel and Activity Participation as Influenced by Car Availability and Use. Presented at the 75th Annual Transportation Research Board Meeting, Washington, DC. Golob, T. F., McNally, M. G. (1995). A model of household interactions in activity participaion and the derived demand for travel. Presented at the EIRASS Conference on Activity-Based Approaches: Activity Scheduling and the Analysis of Activity Patterns. (NOW PUBLISHED IN TR) Gronau, R. (1977). Leisure, Home Production and Work - The Theory of the Allocation of Time Revisited. JPE 85(6), 1099-1123. Hamed, M. M., Mannering, F. L. (1993). Modeling travelers' postwork activity involvement: toward a new methodology. Transportation Science 27(4), 381-394. Harvey, A.S. Guidelines for time-use data collection. Social Indicators Research 30, 197-228. Hirsch, M., Prashker, J. N., Ben-Akiva, M. E. (1986). Dynamic Models of Weekly Activity Pattern. Transportation Science 20(1), 17-28.
328
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Hirschman,E-C. (1987). Theoretical Perspectives of Time Use : Implications for Consumer Behavior Research. Research in Consumer Behavior, Volume 2, 55-81. van der Hoom, T. (1983). Development of an activity model using a one-week activity-diary data base. In: Recent Advances in Travel Demand Analysis. S. Carpenter, P. Jones (Eds.) Gower, Aldershot, 335-349. Jara-Diaz, S. R. (1996). Time and Income in Travel Demand : Towards a Microeconomic Activity Framework. Presented at the Conference "Theoretical Foundations of Travel Choice Behavior", Stockholm, August 1996. Jones, P. M. (1979). New approaches to understanding travel behavior : The human activity approach. In: Behavioral Travel Modelling. D. A. Hensher, P. R. Stopher (Eds.) Croon Helm, London, 55-80. Jones, P. (Ed.) (1990). Developments in Dynamic and Activity-Based Approaches to Travel Analysis. Avebury, Aldershot. Jones, P. (1995). Contribution of activity-based approaches to transport policy analysis. Paper presented at the Workshop on Activity Analysis, Eindhoven, The Netherlands, May. Jones, P., F. Koppelman and J-P. Orfeuil (1990). Activity Analysis: State-of-the-Art and Future Directions. In: Developments in Dynamic and Activity-Based Approaches to Travel. P. Jones (Ed.) Avebury, Aldershot, 34-55. Juster, F. T. (1990). Rethinking utility theory. Journal of Behavioral Economics 19(2), 155179. Juster, F. T., F. P. Stafford (1991). The Allocation of Time : Empirical Findings, Behavioral Models, and Problems of Measurement. Journal of Economic Literature 29(2), 471522. Kalfs, N. (1993). Hour by hour: effects of the data collection mode in time use research. Ph.D. dissertation, NIMMO, Amsterdam Kitamura, R. (1996). Activity-Based Travel Demand Forecasting and Policy Analysis. Presented at the TMIP Conference on Activity-Based Travel Forecasting, New Orleans, LA, June 2-5. Kitamura, R. (1984). A model of daily time allocation to discretionary out-of-home activities and trips. Transportation Research B 18, 255-266. Kitamura, R., E. I. Pas (1995). Time Use Data and Analysis : New Approach to Transportation Planning. Presented at the 74th Annual Transportation Research Board Annual Meeting, Washington, DC. Kitamura, R., E. I. Pas and S. Fujii (1997). Time-use data, analysis and modeling : Toward the next generation of transportation planning methodologies. Forthcoming, Transport Policy. Kitamura, R., J. Robinson, T. F. Golob, M. Bradley, J. Leonard and T. Hoom (1992). A comparative analysis of time use data in the Netherlands and California. Proceedings of Seminar E, 20th PTRC Summer Annual Metting, PTRC Education and Research Services Ltd., London, pp. 127-138. Kitamura, R., T. van der Hoom and F. van Wijk (1995). A comparative analysis of daily time use and the development of an activity-based traveler benefit measure. Paper presented at the EIRASS Conference on Activity-Based Approaches : Activity Scheduling and the Analysis ofActivity Patterns, Eindhoven, The Netherlands.
Time Use and Travel Demand Modeling
329
Kitamura, R., T. Yamamoto, S. Fujii and S. Sampath (1996). A Discrete-Continuous Analysis of Time Allocation to Two Types of Discretionary Activities Which Accounts for Unobserved Heterogeneity. In: Transportation and Traffic Theory, (Ed: Lesort,J.-B.) Elsevier, Oxford, 431-453. Kraan, M. (1995). In search for limts to mobility growth with a model for the allocation of time and money. Paper presented at the EIRASS Conference on Activity-Based Approaches : Activity Scheduling and the Analysis of Activity Patterns, Eindhoven, The Netherlands. Kraan, M. (1996). Time to Travel? A Model for the Allocation of Time and Money. Ph.D. Dissertation, Department of Traffic and Transport Management, University of Twente, Enschede, the Netherlands. Kurani, K. S. and R. Kitamura (1996). Recent developments and the prospects for modeling household activity schedules. A report prepared for the Los Alamos National Laboratory, Institute of Transportation Studies, University of California, Davis, CA. Kurani, K. S. and M. E. H. Lee-Gosselin (1996). Synthesis of Past Activity Analysis Applications. Presented at the TMIP Conference on Activity-Based Travel Forecasting, New Orleans, LA, June 2-5. Lawson, C. (1996). Household Travel-Activity Decisions. Draft Dissertation Proposal, Urban Studies/Regional Science, Portland State University. Lawton, T. K. and E. I. Pas (1996). Resource Paper, Survey Methodologies Workshop. In "Conference on Household Travel Surveys : New Concepts and Research Needs", Conference Proceedings No. 10, Transportation Research Board, Washington, DC. Leach, E. (1966). Two essays concerning the symbolic representation of time in anthropology. Athlone Press, London. Lu, X. (1996). A study of the interrelationships among socio-demographics, time use and travel behaviorr. M.S. Thesis, Duke University, Durham, NC. Lu, X. and E. I. Pas (1997a). A structural equation model of the relationships among sociodemographics, activity participation and travel behavior. Presented at the 76th Annual Meeting of the Transportation Research Board, Washington, DC, January (under revision for forthcoming publication in Transportation Research). Lu, X. and E. I. Pas (1997b). An Examination of Activity Time Allocation on Two Consecutive Days. Prepared for Presentation at the 8^^ International Conference on Travel Behavior Research, September 21-25, Austin, TX. Mannering, F., E. Murakami and S-G. Kim (1994). Temporal stability of travelers' activity choice and home-stay duration : some empirical evidence. Transportation 21(4), 371392. Neimeier, D. A. and J. Morita (1996). Duration of trip-making activities by men and women: A survival analysis. Transportation 23(4), 353-371. Nickols,S. Y. and K. D. Fox (1983). Buying Time and Saving Time : Strategies for Managing Household Production. Journal of Consumer Research 10, 197-208. Pas, E. I. (1997). Time in Travel Demand Analysis and Modeling: From Relative Obscurity to Center Stage. In Theoretical Foundations of Travel Choice Modeling. T. Garling, T. Latilla, K. Westin (Eds.) Pergamon, Oxford, forthcoming.
330
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Pas, E. I. (1996). Advances in Activity-Based Travel Modeling. Presented at the Conference on Activity-Based Travel Forecasting, Travel Model Improvement Program^ U.S.Department of Transportation, New Orleans, (forthcoming in the conference proceedings). Pas, E. I. (1994). Emerging directions in travel demand modeling: a new role for time use studies? In: Fifteenth Reunion of the International Association for Time Use Research. N. Kalfs and A. S. Harvey (Eds.), Amsterdam. Pas, E. I. (1990). Is Travel Demand Analysis and Modeling in the Doldrums? In: New Developments in Dynamic and Activity-Based Approaches to Travel Analysis. P. Jones (Eds.) Avebury, Aldershot, 3-27. Pas, E. I. and A. S. Harvey (1991). Time Use Research: Implications for Travel Demand Analysis and Modelling. Presented at the 6^^ International Conference on Travel Behavior, Quebec City, and published in: Understanding Travel Behaviour in an Era of Change. (Eds: Stopher,P.R.; Lee-Gosselin,M.) Pergamon Press. Pas, E. I. and R. Kitamura (1995). Time use analysis for travel behavior research: An overview. Presented at the 74th Annual Transportation Research Board Annual Meeting, Washington, DC. Pendyala, R and E. I. Pas (1997). Multiday and multiperiod surveys for travel demand analysis and modeling. Resource paper prepared for Workshop 5 of Transport Surveys: Raising the Standard, International Conference on Transport Survey Quality and Innovation May 24-30, Grainau, Germany. Prince, H. (1978). Time and historical geography. In: Timing Space and Spacing Time : Making Sense of Time. Vol. I. T. Carlstein, D. Parkes and N. Thrift (Eds.) Edward Arnold, London. RDC, Inc. (1993). Further Comparative Analysis of Daily Activity and Travel Patterns and Development of a Time-Activity-Based Traveler Benefit Measure. Prepared for the Dutch Ministry of Transport and Public Works, the Netherlands. RDC, Inc. (1995). Activity-based modeling system for travel demand forecasting. Prepared for the Metropolitan Washington Council of Governments, September. Robinson, J. and G. Godbey (1997). Time for Life: The Surprising Ways Americans Use Their Time. The Pennsylvania State University Press, University Park, PA. Stopher, P. R. (1992). Use of an activity-based diary to collect household travel data. Transportation 19, 159-176. Stopher, P. R. and I. M. Sheskin (1982). Toward improved 24-hour travel records. Transportation Research Record 891, 10-17. Suda, K. (1994). Methods and Problems in Time Allocation Studies. Anthropol Sci. 102(1), 13-22. Supemak, J. C. (1990). A dynamic interplay of activities and travel:analysis of time of day utility profiles. In: Developments in Dynamic and Activity-Based Approaches to Travel Analysis. P. M. Jones (Ed.) Gower Aldershot. Szalai, A. (Ed.) (1972). The Use of Time: Proceedings of the World Time-Budget Study. Mouton, The Hague. Tonn, B. E. (1984a). The cyclic process decision-heuristic: an application in time-allocation modeling. Environment & Planning A 16, 1197-1220. Tonn, B. E. (1984b). A sociopsychological contribution to the theory of individual timeallocation. Environment & Planning A 16, 201-223.
Time Use and Travel Demand Modeling
331
Yamamoto, T and R. Kitamura (1997). An analysis of time allocation to in-home and out-ofhome activities across working days and non-working days. Manuscript, Kyoto University. ' Elsewhere, I have argued that development of the activity-based approach is the only paradigm shift in the history of travel demand modeling, and that the introduction of disaggregate choice models in the 1970s did not represent a fundamental change in the framework of travel demand modeling and analysis (Pas, 1990). ^ Travel behavior researchers have also participated in meetings of the International Association for Time Use Research (see, for example, Pas, 1994). ^ This section draws on Pas and Harvey (1991), Pas (1994) and Kitamura, Pas and Fujii (1997). ^ In addition to these sampling methods, time use surveys sometimes ask respondents to report directly the amount of time spent on each of a series of activities, on for example, the previous day.
^ This section draws on Lawton and Pas (1996) ^ This section draws on Pas (1996)
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
16
TIME USE: WORKSHOP REPORT
Ryuichi Kitamura
INTRODUCTION Time use is synonymous with activity engagement; activities an individual engages in over a period of time, say, a day, can be precisely described by knowing how the individual uses time, and his time use can be reconstructed by compiling the set of activities he pursued chronologically. In this sense, time use analysis is nothing new to the field of travel behaviour analysis. In fact, activity-based analyses of travel behaviour should be examining the individual's use of time, both in home and out of home. Many so called "activity-based" analyses of travel behaviour, however, are based on travel survey data that contain information on trips and their purposes, but not on activities engaged in at home, or on multiple activities pursued at a destination. For this reason, these studies have been criticized as only paying lip service to activity-based analysis. The reason for this is obvious; data are rarely available that can be applied to a full-fledged analysis of activity and travel behaviour. Indeed it is difficult to locate a data set that contains information on activities, both in home and out of home, their locations, and attributes of the trips made to the activity locations at levels of detail needed for travel behaviour modeling. Time use surveys and data are rather new to the field of travel behaviour analysis. While there have been a number of comparable studies across many countries, and standard categories have been proposed for activity type classification, it is only recently that travel behaviour researchers have directed their attention to time use data. In fact this conference may be the
334
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
first occasion where a number of studies of travel behaviour that are based on time use data are presented. Using travel survey data in an activity-based analysis of travel behaviour involves the critical limitation that there is no information on in-home activities. Consequently no analysis can be performed on the substitution of activities between in home and out of home. This is critical because it pertains to the generation of trips; to address the issue of why people travel the way they do and gain a thorough understanding of travel behaviour, it is imperative that all activities pursued by an individual, both in home and out of home, be observed and investigated. While recent travel surveys have attempted to capture in-home activities, they are not as thorough as time use surveys in capturing activities, for both in-home and out-of-home activities. The most complete account of the individual's activity engagement can be found in time use data.
WORKSHOP
PAPERS
A substantial collection of papers on time use was presented at the Workshop on Time Use. Of the total of 12 papers presented in the Workshop, six (Bhat & Misra; Chlond & Zumkeller; Fujii & Kitamura; Golob; Kraan & Maarseveen; and Lu & Pas) were concerned with the allocation of time to various activities, including travel. Adding to these, there are two papers on trip timing (Mehndiratta, et al.; and Picado & Deakin), a paper each on measures of the similarity among activity patterns (Joh, et al.), international comparison of indicators of travel patterns (Timmermans & van der Waerden), planning for long-distance travel (Lindh & Garling), and social contact and travel (Harvey & Taylor). Mehndiratta et al. focus on the effects of time of day on inter-city travel. Of particular interest is the disruption of sleep due to traveling. They attempt to quantify the disruption effects and determine how they influence inter-city travel decisions. Picado and Deakin explore the relationship between work schedule flexibility and work trip timing. The findings of the study include that, contrary to what one might expect, workers with flexible work schedules still tend to travel during peak periods. They are also shown to exhibit more day-to-day variability in travel patterns and more trip chaining. Joh et al. address the issue of activity pattern classification by proposing a measure of similarity of activity patterns that accounts for the sequence of activities. The study extends unidimensional "sequence alignment methods," which were originally applied to measure the sequential dissimilarity between two strings of information. Timmermans and van der Waerden are concerned with the association between activity-travel patterns and the properties of urban
Time Use: Workshop Report
335
structure and transportation networks. Through regression analyses of indicators of activitytravel patterns, they show that demographic and socio-economic attributes of the individual and auto availability are important determinants of activity and travel, but "the impact of urban setting" is rather limited. Lindh and Garling investigate how far in advance people plan for "long-range travel" and what aspects they plan. The analysis indicates that plans are made quite far in advance, but there are large variations, and that the party traveling together is connected to the trip purpose and the travel mode is often connected to the destination. In their paper entitled "Activity settings and travel behavior: a social contact perspective," Harvey and Taylor examine the impact of social contact on travel behaviour. Social contacts are important determinants of the context in which an activity is pursued, hence the nature of the activity. Furthermore, many activities take on meaning only when jointly pursued with others. Social contacts thus become important elements in the analysis of activity-travel behaviour. The study examines the social interaction at work as a determinant of travel. With the six papers presented on it, the analysis of time allocation was clearly the principal focus of the workshop presentations. Several approaches are taken in these six papers. Chlond and Zumkeller address the question of how increased income and leisure time will be used and how travel time will be affected. They look at both travel time budgets (as expenditures) and monetary expenditures. Their simulation analysis indicates that the reduction in work duration by one hour would lead to an extra four minutes of travel. Bhat and Misra, and Kraan and Maarseveen, both adopt micro-economic approaches. Time allocation is viewed as a rational resource allocation behaviour in these studies. Bhat and Misra examine the allocation of discretionary time between in-home and out-of-home activities and between weekdays and weekend days. Kraan and Maarseveen formulate a utilitarian model of time and monetary expenditures, then apply it to time-use data, examining the allocation of time to in-home activities, out-of-home activities, and travel. Fujii and Kitamura, Golob, and Lu and Pas, adopt structural equations models. Fujii and Kitamura are concerned with workers' activities and travel after work. Their structural equations model is applied to evaluate the impacts of new freeway lines. Golob's structural equation model system is concerned with activity engagement, travel time expenditure, and trip generation. Trip generation is thus modeled while incorporating time use, with household attributes and accessibility indices used as explanatory variables. Lu and Pas examine the interrelationships among the times allocated to subsistence, maintenance and leisure activities
336
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
on two consecutive days. The results indicate the times spent on in-home activities and out-ofhome activities are interrelated, and so are times allocated to activities on two consecutive days. These studies collectively demonstrate the advances being made in the area of time use and travel behaviour. In particular, the multitude of approaches adopted for the analysis of time allocation attest to the synergetic developments involving a number of researchers. The issue of time allocation appears to have been well explored, and the range of analytical tools and models that have been developed are in the stage for practical application.
FUTURE RESEARCH ISSUES A number of issues that are to be addressed in the future were identified through the discussions at the Workshop. The most salient ones are summarized here. First, there are several issues concerning the definition of activities. One is how to measure and record "multitasking," or "polychronic time use," i.e., engaging in more than one activity at a time. Representing concurrently pursued multiple activities by a single "representative" activity, is problematic not only because information on multi-tasking is lost, but also because it is not always evident which of the mukiple activities is the primary activity and which is secondary. Another issue is that the definition of an activity may depend on the context in which it is pursued. For example, activities pursued during a visit to a shopping mall may be characterized as social, recreational, or shopping, depending on when, for what, and with whom the activities were performed. Defining an activity is not merely devising standardized activity categories, but is concerned with the definition of the context in which the activity is pursued. The next issue is concerned with the determination of the "activity episode," i.e., the continuous engagement in an activity, or a set of activities, at a location or in transit. The models of time allocation can predict the amount of time that will be allocated to each type of activity, but this alone falls short of linking activity and travel. What is needed is to break down the total amount of time allocated to a type of activity to individual activity episodes, and determine where each episode takes place (including in transit in cases like "reading newspaper in a commuter train"). Once this is done, it is then possible to identify trips that are made between activity episodes. This subject area is almost entirely unexplored and remains as a major area for future research in the field of time use and travel analysis. Another important issue that is not well explored is the tradeoffs between activity duration and
Time Use: Workshop Report
337
travel time. This is related to the above issue of defining activity episodes in the sense that a given sum of time may be divided into more, shorter activity episodes at more locations, and therefore more travel, or fewer, longer activity episodes w^ith less travel. A traveler is in general faced with a situation where a better opportunity for an activity can be reached by traveling farther, and is making a tradeoff between the quality of the opportunity and the time spent for activity. This may be characterized as a rational behaviour in which an optimum balance is sought between activity duration and travel time. In any event, the series of activity episodes pursued by an individual needs to be examined in the new light of tradeoffs between activity duration and travel time. This workshop has made evident that healthy progress is being made in the area of time use and travel behaviour. It is hoped the effort will extend to include the area of survey design for effective time use surveys for travel behaviour analysis. It is also hoped that models of trip making be developed in the near future on the basis of the accumulation of empirical findings on time use behaviour and models of time allocation.
ACKNOWLEDGEMENTS Participants included: S. Fujii, T. Golob, A. Harvey, C.-H. Joh, N. Kalfs, R. Kitamura, M. Kraan, C. Lindh, X. Lu, S. Mehhndiratta, R. Misra, E. Pas, R. Picado, A. Suddharth, and P. van der Waerd.
LIST OF PAPERS PRESENTED AT THE WORKSHOP Bhat, C. and R. Misra, "Discretionary activity time allocation of individuals between in-home and out-of-home and between weekdays and weekends" Chlond, B. and D. Zumkeller, "Future time use and travel budget changes—estimating transportation volumes in the case of increasing leisure time" Fujii, S. and R. Kitamura, "A structural equations model system of commuters' time use and travel"
338
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Golob, T., "A simultaneous model of activity participation and trip chain generation by households" Harvey, A. and M. Taylor, "Activity settings and travel behavior: a social contact perspective" Joh, C.-H., T. Arentze, F. Hoffman and H. Timmermans, "Activity pattern similarity: towards a multidimensional sequence alignment" Kraan, M. and M. van Maarseveen, "An individual based model for the allocation of time and money" Lindh, C. and T. Garling, "The time course of household choices of long-range travel" Lu, X. and E. Pas, "An examination of activity time allocation on two consecutive days" Mehndiratta, S., M. Hansen and E. Deakin, "Aggregate influence of time-of-day effects in intercity business travel" Picado, P. and E. Deakin, "Scheduling flexibility and the timing of work trip" Timmermans, H. and P. van der Waerden, "The structure of travel patterns: An international comparison"
SECTION 6 TRAVEL BEHAVIOURMEASUREMENT
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
17
CURRENT ISSUES IN TRAVEL AND ACTIVITY SURVEYS
Tony Richardson
INTRODUCTION The field of travel behaviour measurement has been active in one form or another for about 50 years. However, it has only been in the last 20 years that it has been taken seriously as an area of professional activity. In that time, and parallel to the series of conferences held under the L \ T B R banner, a number of international conferences have been held which have attempted to raise the quality of surveys of travel and activity behaviour. These conferences started in Germany in 1979, and continued in Australia in 1983, in Washington D.C. in 1990, in the United Kingdom on 1996, and most recently returning to Germany in 1997 where a major international conference on quality in travel surveys was held in Grainau in May 1997. As a reflection of current interests in travel survey methodology, the Grainau Conference was structured around twelve workshop themes, complemented by a range of plenary sessions on issues of broader interest. The workshop themes were: 1. Multi-instrument and multi-method surveys 2. Respondent sampling, weighting and non-response 3. Item non-response 4. Quality indicators 5. Multiday and multiperiod data 6. Respondent burden 7. Hypothetical situations 8. Practitioners' future needs
342
In Perpetual Motion: Travel Behaviour Research and Opportunities
9. 10. 11. 12.
Modellers' needs Qualitative survey methods Data presentation Questionnaire design
As a means of focusing attention on some of the key current issues in travel behaviour measurement, this chapter summarizes the major issues arising out of the Grainau Conference. While the Grainau Conference was structured around twelve distinct workshop themes, the following discussion will be structured around the twelve components of the survey process, as identified by Richardson, Ampt and Meyburg (1995) and as shown in Figure 1.
PRELIMINARY PLANNING The idea behind the preliminary planning stage of the survey process is to determine whether a survey needs to be performed and, if so, what the overall guidelines are for the survey design. An objective of the Grainau Conference was to develop a set of "industry guidelines" on the state of practice in travel survey methodology to assist those starting out on the task of designing a travel survey. These "guidelines" would serve three major roles: • to document minimum standards that are considered necessary in any travel survey • to inform clients of what should be expected from "quality" surveys • to provide advice to survey designers on various technical aspects of survey design There was seen to be a major need for "educating" clients as to what should be expected from travel surveys. Too often, clients try to cut comers in an attempt to minimize the short-run costs of the survey. However, this often leads to higher long-run costs and, in many cases, unusable data. By having an industry-approved set of guidelines that would serve as a benchmark against which survey proposals could be judged, it was felt that the interests of clients and survey designers could be best protected. In order to have the guidelines accepted as widely as possible on an international scale, it was felt desirable to have the backing of a reputable "cognizant agency" who would be responsible for the production and distribution of the guidelines. Such an agency could be an organization of the standing of the OECD, or similar body. While the development of these Guidelines was ultimately unsuccessful, one of the specific issues that needed immediate attention was the issue of definitional consistency. Because travel and activity surveys are being conducted globally, it is necessary to ensure that similar
Current Issues in Travel and Activity Surveys Preliminary Planning
Selection of Survey Method
Sample Design
Survey Instrument Design
Pilot Survey
i
Survey Administration
I \
Data Coding
Data Correction and Expansion
Data Editing
Data Analysis
Tidying-Up Presentation of Results Figure 1 The Travel Survey Process
343
344
In Perpetual Motion: Travel Behaviour Research and Opportunities
definitions are being used to enable comparison of results. Benchmarking is becoming an increasingly popular practice, but this is of little use if different definitions are used for key terms in data collection programs. The classic definitional problem revolves around the definition of a "trip". Survey designers employ different definitions of a "trip", compared to other units of travel such as "trip stages", "trip legs", "stops", "journeys", "chains", and "sojourns" (to name just a few of the more commonly used terms). Despite this identified need, confusion still exists in the terminology applied to travel. For example, in the MEST project of long-distance travel currently being conducted in Europe, half the participating countries are assuming that a "journey" is part of a "trip" (in line with tourism survey practice), while the remainder are assuming that a "trip" is part of a "journey" (in line with urban travel survey practice). While each country may know what they mean by their own terminology, outside readers of the documents are completely confused by the inconsistent terminology used within the same project. This problem of consistent definitions applies to many other areas as well, and was seen to be one of the immediate issues to be covered by any set of "guidelines".
SELECTION OF SURVEY METHOD In selecting the type of survey method to be used in any particular instance, four specific issues were highlighted in Grainau: •
the use of muhiple survey methods
• the timing and duration of surveys • the role of qualitative surveys • activity-based or trip-based recording The use of multiple survey methods, sometimes called the Total Design approach (Dillman, 1978), was seen to be a useful method of combining the best features of various designs for a specific situation (Goulias, 1997; Freeth, 1997). For example, respondents with low literacy levels (van der Reis, 1997) in a self-completion survey may be offered the chance to complete the survey in a face-to-face interview situation. Alternatively, telephone interviews may be used as a quality control procedure as a follow-up to a self-completion questionnaire to clarify doubtful information or to supply missing information. The timing and duration of surveys considered a number of related issues; how frequently should surveys be performed, should surveys be continuous, how many days should be covered by each respondent, should surveys be panel or repeated cross-sections, what is the role for before-and-after surveys (Pendyala and Pas, 1997)? In addition, a plenary session paper
Current Issues in Travel and Activity Surveys
345
described the benefits of continuous travel surveys and showed how continuous travel surveys have been successfully implemented in Australia (Richardson and Battellino, 1997). The complementary roles played by quantitative and qualitative surveys were highlighted in the workshop on qualitative surveys (Grosvenor, 1997). It was shown that qualitative surveys are particularly useful in the early stages of a project when ideas are still being formulated, or when in-depth exploration of policy issues was required. In addition, qualitative questions could be used within otherwise quantitative surveys as a means of exploring topics in more depth, or just as a way of allowing respondents to have their own say in their own words. Continuing discussion occurred on the relative merits of using activity-based or trip-based recording in travel surveys (Arentze et al., 1997). Three major options were identified: tripbased recording of travel, activity-based recording of travel (i.e. the focus is still on recording travel, but an activity framework is used to assist respondents in recording the travel), and a fiill activity recording framework (including in-home and in-building activities). The general consensus appeared to be that trip-based recording was far less useful and efficient. However, the choice between the latter two methods depends on the intended use of the data. Full activity diary recording is more onerous for the respondent and more demanding of analytical skills; on the other hand, activity-based trip recording gives better estimates of travel behaviour in a more natural format for the respondent. For most travel behaviour studies, activity-based travel surveys should suffice, providing they can account for two specific types of "in-house" activities; "at-home" activities, such as working at home, which are a clear substitute for corresponding out-of-home activities, and "in-building" activities, such as visiting many shops within a covered regional shopping center, which would be recorded as separate activities if they occurred in an outdoor setting. However, the continuing trend towards activity-modeling may move the balance between activity surveys and travel surveys.
SAMPLE DESIGN The Grainau Conference did not concentrate specifically on sample design as a workshop theme. Rather, issues of sample design entered into each of the workshops as and when required. The important issue to emerge from this was that, when sample design was mentioned, most of the attention focussed on issues of minimizing sampling bias rather than on reducing sampling error. That is, more attention was focussed on getting the correct sample
346
In Perpetual Motion: Travel Behaviour Research and Opportunities
rather than on getting a sample of the "correct" size. Specific issues that arose with respect to sample design included: • the need to adopt a random sampling process, and to stick with it •
the desirability of stratifying the population, where possible, in order to obtain a more representative sample
•
the sampling requirements when the data is going to be used for the development of specific types of models the use of replication techniques for the estimation of sample variance
• •
the issues involved in selection of a sampling frame, and the need to document the quality of the adopted sampling frame.
SURVEY INSTRUMENT DESIGN The issues involved in the design of survey instruments received considerable attention in several of the workshops, perhaps because of the unofficial motto of the conference: "Respondents are Customers". Issues of respondent burden were covered extensively in the workshop on Respondent Burden (Ampt, 1997). In many cases, it was felt that introspection could provide a usefiil check on the burden being placed on respondents (i.e. "do unto others as you would have them do unto you"). However, survey designers also need to realize that respondents d© not always have the same levels of literacy, or interest in the surveys, as the survey designer, and should design and test their surveys accordingly. The length of surveys was considered in terms of its effect on respondent burden, and hence the effects of unit nonresponse and item non-response. Although obvious in hindsight, the conclusion was that survey designers should remove all unnecessary questions from the survey. This sounds so simple as to not be worth saying, but sometimes it is not easy to do when clients want information collected on a wide range of issues. In this content, the "guidelines" were seen to be especially useful in educating clients about the interactions between respondent burden and quality of data obtained. With respect to the design of Stated Preference survey instruments, it was recommended that greater emphasis be given to the needs of the respondent in the design process (Lee-Gosselin, 1997). This means tailoring the instrument to be as realistic as possible from the respondent's point of view, and perhaps restricting the extent of the survey design such that it concentrates on obtaining information on the "main effects" in the choice process.
Current Issues in Travel and Activity Surveys PILOT
347
SURVEYS
The main issue with respect to pilot surveys was simply the need to do them. Almost always, failing to do pilot surveys because of time and money constraints turns out to be a false economy. Once again, the "guidelines" may be particularly useful in convincing clients of the need for performing full pilot surveys.
SURVEY ADMINISTRATION With respect to the administration of surveys of all types, one of the main issues concerned the definition, calculation and use of response rates. In the definition of response rates, it is important to ensure that response rates are defined in a consistent fashion for different types of survey methods. This enables comparisons of response rates between survey methods. In making these comparisons. Figures 2, 3 and 4 (Richardson, Ampt and Meyburg, 1996) provide useful insights into the variety of response (and non-response) mechanisms that apply to different types of survey. Figure 2 shows the response mechanism for a mailout/mailback selfcompletion survey, which involves the use of reminders to non-respondents.
Figure 2 The Response Process for a Mailout/Mailback Questionnaire Survey
348
In Perpetual Motion: Travel Behaviour Research and Opportunities
Figure 3 shows the response process for a personal interview survey, with cluster sampling and a blocklisted sampling frame, and the use of callback procedures for households which are not contacted immediately. It can be seen that the process is very different to that of Figure 2. However, in both cases the response rate would be calculated as the fraction of the original population of households that give complete household responses.
Figure 3 The Response Process for a Personal Interview Survey The response process for a telephone survey of household travel patterns in shown in Figure 4. This survey uses listed telephone numbers for a population of households (using reverse listing to obtain the numbers), a system of call-backs for non-contacted households, the sending of travel diaries to those households which agree to participate in the survey, and the retrieval of information from the households over the phone. To be consistent with the response processes
Current Issues in Travel and Activity Surveys
349
shown in Figure 2 and 3, the response rate for a telephone survey should be based on the number of households from which all diary information was retrieved as a fraction of the total number of households in the original population. However, one will often find response rates for telephone surveys quoted as the fraction of people (not complete households) from whom diary information is retrieved as a fraction of the number of people who agreed to participate in the survey by accepting a diary. Clearly such a definition is inconsistent which the previous definitions, and biased towards the phone survey. Population of households Without phone
With phone
Unlisted
Call-backs
^f^^ No answer
Multiple listed
Listed M/ ^ Initial sample phoned Answering machine
Answer
Agree to participate All diaries completed
ISome diariesi completed
Don't agreel |to participatq No diaries completed
J^
Call-backs •
No answer
^
Final sample] phoned Answering machine
Answer
All diaries retrieved
ISome diarieq retrieved
No diaries retrieved
Figure 4 The Response Process for a Telephone Interview Survey
350
In Perpetual Motion: Travel Behaviour Research and Opportunities
No matter which way response rates are calculated, significant debate occurred at Grainau about the use of response rates as a measure of the quality of a survey. One argument was that high response rates indicate that significant effort has gone into the design of the survey to minimize non-response, and that a high response rate offers less opportunity for nonrespondent bias. The other argument was that the nature of the non-respondents is more important than the number of non-respondents. Clearly, if the non-respondents have the same demographic and travel characteristics as the respondents, then there is no need for concern about the number of non-respondents. However the problem usually is that we do not have enough information about the non-respondents to draw such a conclusion. Granted, many surveys implicitly make the assumption that non-respondents are the same as respondents by not getting any information about non-respondents and simply ignoring them. However, a much better approach is to get information about the non-respondents and then explicitly decide whether the non-respondents really are the same as the respondents. While the need to collect information about non-respondents was the major feature of survey administration discussed at Grainau, a number of other topics were raised. The perennial question of the use of incentives as part of the survey process, as a means of increasing response rates, was discussed with the usual lack of agreement on the merits and demerits of the process. While incentives can raise response rates, this was reported not to be always the case. Examples were given of surveys in which the use of incentives appeared to reduce response rates. More importantly, in light of the above discussion about response rates, serious concerns were raised about the nature of biases introduced by the use of incentives. While incentives might increase response rates, the question was raised as to what type of respondent was attracted, and how this varied with the type of incentive offered. For example, offering free transit tickets as an incentive would most likely attract existing transit riders, offering cash might attract lower income people and so on. While these two examples of bias are obvious, the use of other types of incentive carry more subtle biases and need to be thoroughly investigated. One school of thought is that incentives of this type should be avoided altogether, and that greater emphasis should be placed on good survey design to make completion of the survey as simple as possible, thereby providing a different type of incentive to participate.
DATA CODING One of the main issues discussed under the topic of Data Coding was the increasing use of geocoding as a means of defining the spatial location of various entities in the travel survey. In this context, geocoding means more than the allocation of locations to geographic zones.
Current Issues in Travel and Activity Surveys
351
Rather, it entails the description of every location in the travel survey (e.g. respondents' households and all destinations visited on the travel day) by means of latitudes and longitudes (or other x-y coordinate systems). The use of geocoding has been greatly facilitated by the increased use of Geographic Information Systems in transport planning activities. The use of geocoding allows a far higher degree of spatial checking of the data to be undertaken, and allows calculation of trip distances from a knowledge of the start and end coordinates of the trip. This is particularly enhanced if the GIS package contains a description of the road and public transport networks, allowing "on-road" or "on-board" measurements of distance to be calculated. One issue arising from the use of geocoded locations concerns the ethical question of what level of geographical data should be released to clients. It was recommended that the x-y coordinates of all locations should remain in the database except for those belonging to the respondent's home, which should be aggregated to the finest level of zoning available in the study. In this way, the confidentiality of the respondent is upheld, while the detail in the trip chains is preserved.
DATA EDITING A considerable amount of time was spent at the Grainau conference on the topic of data editing, especially in dealing with missing data. Missing data was defined as including situations in which an answer was not provided to a specific question (such as the income question) and where a respondent had failed to tell us about entire trips (i.e. non-reported trips). Several reasons were suggested for missing data (Zmud and Arce, 1997), including: • lack ofknowledge and recall problems • comprehension of questions • perceived or real respondent burden • desire for privacy • concerns about personal information • deliberate mis-reporting Two major means of dealing with missing data were discussed: reweighting of the data and imputation of missing data (Armoogum and Madre, 1997). Reweighting of the data involves the calculation of specific expansion weights for each analysis to be performed to take account of the particular combinations of missing data present in each sub-set of the data. This process,
352
In Perpetual Motion: Travel Behaviour Research and Opportunities
however, becomes extremely tedious given the multitude of different analyses that are typically performed on a data set. Imputation of missing data involves the estimation of values for the missing data based on some other source of information. This information may come from other respondents in the survey or from other surveys. A wide range of different imputation techniques may be used, including: deductive imputation overall mean imputation class mean imputation hot-deck imputation cold-deck imputation regression imputation multiple imputation stochastic imputation The general conclusion from Grainau was that imputation was better than reweighting, and that imputation methods that preserve the variance in the data set (such as stochastic imputation) were better than methods which substitute a limited range of imputed values (such as class mean imputation). There was also discussion on the order in which imputation was performed for a range of variables, and the effect of using imputed variables in later model-building exercises. Two other major issues concerning data editing were also discussed at Grainau; the use of validation models, and the cross-validation of survey data with external data sources. The use of validation models was seen to be extremely useful in detecting outliers in the data. For example, if a model is constructed in which speed of travel is calculated for different length trips by each mode of travel, then this enables the detection of possible errors in each of these variables. For example, having established the likely values of speed for each distance, a very high speed for a bicycle trip (say 70 kph) could occur because of an error in geocoding (giving rise to a shorter than expected trip), an error in recording of trip start or end times (giving rise to a quicker than expected trip) or an error in recording of mode (perhaps it was really a cardriver trip). It should be realised, however, that outliers are not necessarily wrong; there always have to be outliers on continuous distributions. The above method therefore should only be used to detect outliers, and not to automatically "correct" them.
Current Issues in Travel and Activity Surveys
353
The second issue concerned the cross-validation of survey results with external data sources, such as traffic counts and public transport patronage counts. This should occur as a matter of course to ensure that the survey results are at least in the right ballpark. However, great care needs to be taken to ensure that such comparisons are reasonably valid. For example, the two data sets should be recorded for similar time periods, for similar geographic regions and populations, and they should use similar definitions of major terms. For example, public transport agencies in different cities may have different definitions of "trips" in their counting systems to those used in a household-based travel survey. It is important to realize that, in almost all cases, both the household-based survey and the external data source are based on sample surveys, and therefore neither is necessarily better than the other.
CORRECTION AND EXPANSION Given the earlier discussion of the calculation of response rates and on their use as indicators of survey quality, it was not surprising that the issue of correcting for non-response was keenly discussed at Grainau. Importantly, however, it was felt that more was needed to be known about the actual characteristics of non-respondents before corrections could be made to the data provided by respondents. In addition, it was felt that there were considerable interactions between non-response and the non-reporting of trips. In particular, the previously-held assumption that non-respondents were more like late respondents in a mailback survey with reminders was called into serious question. Rather than late respondents having low mobility, as reflected by their low trip rates, it was argued that they may simply be reporting low mobility (by omitting trips) or perhaps selecting a travel day on which they actually made fewer trips. Under these conditions, assuming that non-respondents have low trips rates (like the late respondents) may be considerably in error and lead to non-response corrections in the wrong direction.
DATA ANALYSIS Apart from the workshop on Modelers' Needs, relatively little time was spent discussing the specific types of analyses that might be performed on travel data (more time is likely to be spent on these topics within lATBR Conferences). However, two issues did emerge. Firstly, it was considered that the quality of demand data, and associated analysis, emerging from current travel and activity surveys should be balanced by an improvement in the quality of supply data, and associated analyses, describing the physical transport and land-use systems. The emergence
354
In Perpetual Motion: Travel Behaviour Research and Opportunities
of GIS descriptions of transport networks was seen as a major opportunity in this area. Secondly, it was considered that the considerable attention being paid to measuring travel behaviour of people should be complemented by the measurement, and analysis, of the travel behaviour of freight and commercial vehicles. Only in this way could a complete description of travel patterns be obtained.
PRESENTATION OF RESULTS Having collected travel behaviour data, it was considered essential that it be presented to the various clients in an appropriate fashion. As a general conclusion in this area, it was felt that one could do little better than to buy Tufte's classic books on the presentation of quantitative data (Tufte 1983, 1990, 1997), and then try to live up to the principles enunciated therein. In addition, considerable discussion took place on the merits of using the Internet as a medium for presenting the results of travel behaviour surveys to a wide range of interested parties.
TIDYING-UP Finally, the topic of Tidying-Up at the completion of the survey received more than its normal share of attention. Five major topics emerged in this area. Firstly, proposals for the archival of data sets were discussed including a central clearinghouse for data sets available for secondary analysis. Secondly, issues of data distribution methods and related topics of privacy were discussed in the light of the greater degree of privatization of the data collection function. Thirdly, the idea of providing feedback to respondents, in the form of summaries of results, was raised as a means of sustaining a high degree of cooperation between respondent base and the survey organization. Fourthly, the concept of conducting meta-analyses of travel surveys was suggested as a means of determining what strategies appear to work in increasing response rates and data quality (Kalfs et al., 1997). Finally, the most over-riding conclusion of the entire conference dealt with the need to document all stages of the survey process. In one way or another, the chairs of all twelve workshops reported on the absolute need to improve documentation standards within their area of interest. Such a recommendation accorded nicely with the theme of one of the Keynote Addresses on Survey Quality, wherein the application of ISO9001 quality standards to travel surveys was suggested (Richardson, 1997), since one definition of the processes involved in ISO9001 (Taormina, 1996) is:
Current Issues in Travel and Activity Surveys
355
• Write down what you do • Do what you write down • Verify the results If an adherence to these three rules were all that came out of the Grainau Conference, then the conference would gone a long way to achieving its objective of Raising the Quality of Transport Surveys.
CONCLUSIONS While the Grainau conference had an objective of developing a short (30-page) set of industry "guidelines", and had two keynote papers (Richardson 1997; Pisarski, 1997) and one resource paper (Kalfs et al., 1997) devoted to the topic of quality and guidelines, the final conclusion was that it was not possible to develop such guidelines. This conclusion was based partly on the continuing differences in practice in different countries, but mainly on the fact that it would be impossible to summarize state-of-the art practice in such a short document. In the past five years, there have been several major texts on travel survey methodology (Richardson, Ampt and Meyburg, 1995; Stopher and Metcalf, 1996; Cambridge Systematics and Barton Aschman (1996)) and to attempt to summarize these in 30 pages would give a misleading impression of simplicity to the task of survey design. It would be like attempting to summarize choice modeling in 30 pages. Nonetheless, arising out of the Grainau Conference, the following issues were seen as being the major current topics of debate in the measurement of travel behaviour: The need for consistent terminology Increased use of Total Design approaches Activity-based versus Trip-based recording Sampling Bias is more important than Sampling Error The need to reduce respondent burden The need to calculate response rates in a consistent fashion The need to find out more about non-respondents The increasing use of Geocoding The use of imputation methods for missing data The need for comparable data on freight and commercial vehicle travel Internet applications for travel survey data The absolute need for better documentation of survey methods Paying increased attention to the above issues over the coming years will help lead towards the improved practice that was originally sought by the conference in the development of industry guidelines.
356
In Perpetual Motion: Travel Behaviour Research and Opportunities
REFERENCES Ampt, E. S. (1997). Respondent Burden: Understanding the People we Survey!. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Arentze T., H. Timmermans, F. Hofman and N. Kalfs (1997). Data Needs, Data Collection and Data Quality Requirements of Activity-Based Transport Demand Models. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Armoogum, J. and J-L. Madre (1997). Item Sampling, Weighting and Nonresponse. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Cambridge Systematics and Barton Aschman (1996). Travel Survey Manual. Prepared for the US Department of Transportation, Washington DC. Dillman, D. A. (1978). Mail and Telephone Surveys, The Total Design Method. John Wiley & Sons: New York. Freeth, S. (1997). Using a Range of Methods to Collect Travel Data: the Experience of the British National Travel Survey. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Goulias, K. G. (1997). Multi-Method and Muhi-Instrument Surveys. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Grosvenor, T. (1997). Qualitative Research in the Transport Sector. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Kalfs, N., H. Meurs and W. Saris (1997). Quality Indicators. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Lee-Gosselin, M. (1997). Hypothetical Situations. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Pendyala, R. M. and E. I. Pas (1997). Muhiday and Multiperiod Data for Travel Demand Analysis and Modeling. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Pisarski, A. E. (1997). Recognizing, Creating and Marketing Survey Quality. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Richardson, A. J. (1997). Guidelines for Quality Assurance in Travel and Activity Surveys. Keynote Paper at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Richardson, A. J. and H. Battellino (1997). Similar, yet Different: some Emerging Trends in Travel Surveys in Australia. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Richardson, A. J., E. S. Ampt and A. H. Meyburg (1995). Survey Methods for Transport Planning. Eucalyptus Press: Melbourne. Richardson, A. J., E. S. Ampt and A. H. Meyburg (1996). Nonresponse Issues in Household Travel Surveys. TRB Conference on Household Travel Surveys: New Concepts and Research Needs, 79-114.
Current Issues in Travel and Activity Surveys
357
Stopher, P. R. and H. M. A. Metcalf (1996). Synthesis on Household Travel Survey Methods. NCHRP Synthesis Report 236, Transportation Research Board, Washington DC. Taormina, T. (1996). Virtual Leadership and the ISO9000 Imperative. Prentice-Hall: Upper Saddle River, NJ. Tufte, E. R. (1983). The Visual Display of Quantitative Information. Graphics Press: Cheshire, Connecticut. Tufte, E. R. (1990). Envisioning Information. Graphics Press: Cheshire, Connecticut. Tufte, E. R. (1997). Visual Explanations. Graphics Press: Cheshire, Connecticut, van der Reis, P. (1997). Transportation Surveys among Illiterate and Semiliterate Households in South Africa. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany. Zmud, J. P. and C. H. Arce (1997). Item Nonresponse in Travel Surveys: Causes and Solutions. Presented at an International Conference on Transport Survey Quality and Innovation, Grainau, Germany.
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
18
MOTIVATING THE RESPONDENT: How FAR SHOULD YOU G O ?
Peter Bonsall
ABSTRACT Two separate but related issues are addressed: the use of incentives to improve response rates and the use of rewards and incentives to encourage 'realistic' responses in experiments. Important questions are raised which have implications for the reliability of data and yet which have been largely ignored in the design of surveys and experiments. The chapter draws on evidence from recent surveys and experiments and on relevant behavioural theory and suggests that there is a fine line to be drawn between sufficient motivation and excessive incentivisation.
INTRODUCTION This chapter deals with two separate but related issues; the use of incentives to improve response rates and the use of rewards and incentives to encourage 'realistic' responses in experiments. Important questions are raised which have implications for the reliability of data and yet which have been largely ignored in the design of surveys and experiments. The chapter will draw most particularly on examples from surveys and experiments concerned with traveller behaviour, since this is the field with which I am most familiar, but will also draw on psychological literature and theory where appropriate.
360
In Perpetual Motion: Travel Behaviour Research and Opportunities
ENCOURAGING A HIGH RESPONSE RATE
Avoiding Non-Response Bias It is widely accepted (see for example Richardson et al, 1995) that, although self-completion surveys have advantages over interview surveys in respect of economy, speed and avoidance of potential interviewer bias, their main drawback is that they can suffer from low response and hence carry a risk of non-response biases, (see for example Brog and Meyburg, 1981; or more generally Horowitz, 1969, who demonstrated that conflicting results of a classic psychological experiment on respondents' attitudes to scary messages were explained by the fact that some studies had been conducted using volunteers while others had used 'captive' subjects). These biases can result if there is any tendency in the target population for a particular characteristic, attitude or behaviour to be correlated with the likelihood of responding. The bias may subsequently be corrected if the characteristic is observable but cannot be corrected if it is unobservable or is related in a complex way to the phenomena being measured. Age and gender are examples of a correctable, 'benign', bias, while any tendency of non-locals to be less likely to respond to a roadside mailback survey of journey origin and destination (as reported, for example, by Sammer and Fallast, 1985, and by Bonsall and McKimm, 1993) would be an uncorrectable ,'malign', bias. Good survey practice requires great care to be taken in the design and implementation of the survey instrument so as to maximise response rates and so minimise the risk of non-response bias.
The Use of Incentives It is widely believed that response rates can be increased by offering some form of incentive (see for example Filion, 1975-76). Although it is clear that appropriate prizes, gifts and payments can increase response and ease recruitment, there is evidence to suggest that the effect may be relatively small and that, in some cases, adverse effects can occur whereby the 'incentive' actually decreases the response. Chipman et al (1992) found that, in the context of a self-completion travel survey, the provision of a free gift (a map) for respondents led to an increase in the female response rate but to a reduction in the male response rate. Similarly, Bonsall and McKimm (1993), in the
Motivating the Respondent: How Far Should You Go?
361
context of a roadside mailback survey, found that a £25 prize draw increased the female response rate but reduced the male response rate. They speculated that the provision of a prize draw might have been perceived by the males as 'cheapening' or 'commercialising' the survey and making it somehow less of a 'worthy cause'. Some members of the public may be more likely to help with a survey which they perceive as being in the public interest than with one which has a commercial aspect.
The Size of the Incentive Although it might be assumed that the bigger the incentive, the higher will be the response rate, evidence from the psychological literature suggests that this may not be the case even after allowing for a decreasing marginal rate of return. Festinger and Carlsmith (1959) were responsible for a classic study of the effect that payment can have on participants' attitudes. In this study each participant was asked to perform a boring task and then tell another person that it had in fact been interesting and enjoyable. They were paid either $1 or $20 for giving this testimonial. All participants were then asked for their own, honest, assessment of the task and whether they would be willing to participate in a similar exercise in the future. The fmal assessments by the $1 participants were significantly more positive than those by the $20 participants. The authors concluded that this was evidence of a cognitive dissonance effect whereby the $1 payment was insufficient for the participants to maintain their (real) dislike of the original task and thus an attitude change was brought about. An alternative explanation of this and similar phenomena is provided by inferred value theory (Freedman et al 1992), which holds that the inherent value of a good is deduced from its price (or, in this case the inherent unpleasantness of a task is deduced from the reward being offered for doing it). Freedman et al demonstrated that participants who were told that the reward for a task was $12 rated it as less enjoyable than those who were told that the reward for the task was $5. Whether we accept the cognitive dissonance explanation or the inferred value explanation, the implication is that the higher the reward the less pleasant the task will be assumed to be. This should lead us to conclude that people who are initially disinclined to respond to a selfcompletion survey might be more successfiilly persuaded to do so by the offer of a small reward than by a big one. A similar outcome might be expected if people who were initially inclined to respond to the survey out of a sense of public duty were to regard a large incentive
362
In Perpetual Motion: Travel Behaviour Research and Opportunities
as a bribe and waste of public money and so become less willing to respond. Some people will be predisposed to respond and others will be predisposed not to respond, but it seems that members of both groups may react better to a small reward than to a big one. The art is to judge the nature of the target population and then set the reward at such a level as to be just big enough but not too big. Prize Draws, Gifts and Services It is fairly common practice to offer cash incentives via a prize draw rather than as small payments to all respondents. This approach is normally adopted because it is recognised that, given a finite limit to the sum of the rewards, the prospect of a 1/n chance of winning £x has more effect than the certainty of receiving £x/n. It may also be imagined that a participation in a prize draw for £x would be perceived by some people as more fun, and that £x/n might be an embarrassingly trivial amount - particularly for people on high incomes. Use of prize draws rather than cash may have further theoretical justification from the observation that the cognitive dissonance and inferred value effects seem to relate less strongly to prizes than to payments. It seems that the risk of making people think the task is unpleasant are less if a large prize is available than if a large payment is made - it is, however, unclear whether the likely higher money value of the prize would more than offset the lower influence per unit money. At a more prosaic level, a very practical reason for using a prize rather than payments is that cash payments may be subject to personal taxation while prizes are likely to be exempt. The preceding discussion has referred to cash payments and cash prizes but some surveys use gifts or services instead of cash. Gifts or services may appear more cost-effective than cash rewards, particularly if the survey organisation can obtain them from suppliers at a discount, but it is important to ensure that the gifts or services are appropriate to the target population (for example; the gift of free bus tickets might appeal to sporadic bus users but would not appeal to season ticket holders or to inveterate car drivers). Although a cash reward would undoubtedly have a more general appeal, it too will be less attractive to some sectors of the population than to others (for example, people with high incomes are less likely to be attracted by cash rewards). A particular problem with the use of prizes, gifts and services as a recruitment incentive or as a reward for participation, is that their use has become so widespread in the mail order and
Motivating the Respondent: How Far Should You Go?
363
publishing industry that many people have come to associate such 'gimmicks' with commercial activities rather than activities concerned with the public-good. Their use may therefore reduce the response from people who would otherwise have responded out of a sense of public duty (The findings by Chipman et al, and by Bonsall and McKimm, mentioned above, may be seen in this light and provide some evidence to suggest that males are more likely to be put off by gifts and prizes than are females).
Incentivised Versus Unincentivised Respondents One might suppose that an unincentivised respondent is responding out a sense of duty, a desire to be helpful or a genuine interest in the subject matter while the incentivised respondent may be responding in order to obtain or have a chance of obtaining, the reward. One might then suppose that the incentivised respondent might be more inclined simply to go through the motions of completing the questionnaire rather than to make an effort to complete it correctly. If this involves them rushing through, ticking boxes and providing answers with little thought, then it is possible that their replies may be detected at the logic check stage (see Bonsall and McKimm, 1993). It might be assumed that any tendency of rewards to encourage hastily completed questionnaires could be overcome by making the reward conditional on "correctly completing" the questionnaire. However there is a danger that such a requirement would simply result in it being difficult to detect questionnaires, which have been filled in with more regard for meeting the preconditions for the reward than for providing 'true' answers. It is also possible that the reward-motivated respondents will go to some lengths to manufacture data in order to ensure their eligibility for the reward - they might for example assume that a "complete" travel diary requires a trip to be recorded on every line of the form. All of the above might tend to dissuade us from using participation incentives but it would be incorrect to assume that, where incentivised and unincentivised groups yield different results, it is necessarily the incentivised group's results that are wrong. It is at least conceivable that it is only by offering an incentive that a natural response bias in favour of dutifiil/helpful/interested respondents is overcome. For example, if a survey is seeking individuals' preferences for alternative price/level of service trade-offs, it would obviously be usefiil to have the views of people motivated by self-gratification as well as of those of the dutifial/helpfiil/interested group mentioned above. The area is of course beset by a host of potential response biases and it is clearly possible that the precise form of any incentive might influence the nature of the bias.
364
In Perpetual Motion: Travel Behaviour Research and Opportunities
Consequences of Differential Response A differential response by different groups within the target population to different types and levels of reward raises an important potential problem; if attitudes to incentives reflect, or are associated with, characteristics which might influence travel behaviour or beliefs, then the provision or otherwise of an incentive could affect the results of the survey. If the characteristics were observable, as theoretically is gender, the problem may perhaps be overcome but, if as is likely, it is to some extent unobservable then we have a problem.
INFLUENCING BEHAVIOUR DURING AN EXPERIMENT
Ecological Validity in Experiments There are many situations in which data is obtained more efficiently via experiments rather than by observation of 'natural' behaviour in a real environment. There are also many situations in which it is simply not possible to observe natural behaviour and in which experiments may provide the only feasible source of data. Laboratory and field experiment have a long pedigree in transport - see for example the simulated living room used in environmental^assessment work by Dawson (1974), the rather alarming use of a slide projector in a moving vehicle employed by Mast and Dallas (1976) to study driver response to road signs and, more generally, the review of early driving simulators by Allen et al (1979). Recent years have seen a particular increase in the use of route choice simulators and other computer-based tools - see for example Bonsall and Parry (1991), Adler et al (1993), Koutsopoulos et al (1993,1995), Bonsall et al (1994,1997), and Firmin (1995). Whole branches of psychology and economics are, of course, based on recording the behaviour of participants in experimental conditions. Most experiments seek to represent or reproduce real-world conditions sufficiently well that the behaviour which results can be taken as representing real world behaviour. In some experiments great efforts are taken to ensure a faithful replication of real world conditions whereas, in others, the design may simply be an analogy of the real world situation - see Sommer and Sommer(1991). Examples of these two extremes in transport research might respectively include on-road experiments with new in-car navigation equipment (Parkes and Burnett, 1993) and route choice research based on gambling games (Davis et al, 1995).
Motivating the Respondent: How Far Should You Go?
365
There has been a long-standing debate among psychologists about the value of experimental paradigms (see for example, Chapanis, 1967). Some authorities are fairly scathing about the ability of laboratory- or desk-based experiments based on analogies to reveal anything very useful about real-world behaviour, whereas others (e.g., Mook, 1983) argue that very useful insights can be obtained even when 'mundane realism' is not maintained. However, it is clear that, if experiments are to replicate real-world behaviour, they must include or replicate the factors that influence that behaviour. They must therefore include appropriate risks and rewards, incentives and disincentives; these are the minimum conditions for the achievement of ecological validity. Since it is impossible to include all these factors in any one experiment some simplification and selection is necessary; the aim is to retain the relevant factors while dispensing with the rest. A problem arises, however, in that some of the real world risks and rewards which one wishes to retain cannot be directly replicated. For example, ethical considerations would make it unacceptable to include lasting risks to the participant's mind, body or finances. Indeed some ethics committees will not approve experiments which include any risk to the participants. The question therefore arises as to whether the participants are appropriately motivated to behave as they would in real life. An obvious precaution, widely accepted, is to ask participants to behave as they would in real life. Another equally obvious but surprisingly rarely adopted precaution, is to ask them afterwards whether they did behave as they would have done in real life. This question was routinely included among the debriefing questions in experiments involving the VLADIMIR route-choice simulator (Bonsall et al, 1994, 1997) and typically revealed that about 5% of participants admitted to not having behaved as they would have done in real life - it was then a simple matter to remove their data from the database.
The Use of Monetary Reward as an Incentive in Experiments The prospect of monetary loss or reward has attractions as an incentive in experiments because it is relatively easily understood by the participant and because money is often a contributory factor in real world behaviour. Using Money in an Experiment to Represent Money in the Real World. The use of money incentives is most obviously justified in the context of experiments representing real world situations in which money is an important factor. Recent work at the Universities of Leeds and
366
In Perpetual Motion: Travel Behaviour Research and Opportunities
Newcastle-upon-Tyne (Bonsall et al, 1998) has made extensive use of money incentives in experiments designed to determine drivers' responses to road-user charging. A number of experiments were designed in which the participants were paid a fee to make a series of journeys but deductions were made from the fee to reflect any road-user charges which they incurred while making the journeys. The first experiment was based on a fully interactive driving simulator and was designed to determine the extent to which driving style was influenced by whether, in what way, and at what rate, the driver was being charged. Participants were recruited on the promise of a fee of 'about £10' for a 40 minute session and, on arrival they were told that although the initial fee would be higher than this that they were going to be charged according to their use of the roads (the initial fee was set such that, if they drove normally, they would be left with somewhat over £10). The resuhs of this experiment (see Bonsall and Palmer, 1997) suggested that the participants' driving style had been influenced by the prospect of the charges. However, it can be argued that the participants may have been over-incentivised because, although the charges were realistic enough, the consequences of dangerous driving were not (a smash in a simulator has less serious consequences - the worst that the participant could suffer would be embarrassment at having failed and loss of the expected £10 fee). This possibility led the authors to warn that their results could be taken only as evidence of the relative riskiness of driving styles with and without the imposition of road-user charges. Detailed interpretation must depend on the differential effect of changes in the net fee payable to participants and in their perception of the consequences of dangerous driving. The second experiment was based on a version of the VLADIMIR route choice simulator, which had been modified to calculate and display road-user charges incurred during the simulated journeys. The participants were told to expect a fee of 'about £5' and then learned that the precise fee would depend on how much of their initial credit they expended by incurring road user charges. Our experimental budget prevented us from giving renewed credit for each of the six journeys made by the participant (each journey would require an initial credit higher than was likely to be expended and a canny driver could end up with quite a large amount of credit remaining). We therefore adopted a system whereby, although the charges would be calculated and displayed for each journey, the participants were told that their final fee would depend on the credit-less-charges for only one of their journeys - chosen in advance but not revealed until the end of the session. Thus, each journey would be made with the prospect that any fees charged would reduce the participant's take-home fee.
Motivating the Respondent: How Far Should You Go?
367
This technique, encouraging participants in repeated-task experiments to imagine that each replication of the task might be the one for which the penalties/rewards might be tangible, has been used elsewhere (e.g. Leeds Eurorisk Conference 1996); but we are not aware of tests having been done to establish whether it incentivises behaviour to the same degree as would the certainty of a penalty/reward for each repetition. One might expect the participants to make some allowance for the fact that although they might be charged for the current journey, it is also possible that they might not be. The extensive literature on people's response to risk and uncertainly is clearly relevant here. Meanwhile, initial results from this part of our study suggest that participants' behaviour was affected by the charges to an intuitively reasonable extent. A third phase of our study involved field trials with private cars equipped with a GPS-linked computer which was programmed to calculate and display charges incurred whenever the car was driven along particular stretches of road at particular times of day. The volunteer drivers whose cars were equipped in this way were told that they would be paid a participation fee from which any charges incurred over a period of two weeks would be deducted (the charged roads were on their routes to and from work), hiitial results (Thorpe, 1997) from these trials suggested that, although the charges could have been avoided by using alternative routes or travelling earlier, some participants claimed to be completely ignoring the charges because they did not regard the charge as coming out of their own pocket. This attitude is consistent with the theory that windfall gains are regarded as less valuable than money which has been earned (see Thaler and Johnson, 1990) but was presumably exacerbated by the fact that the participants had not yet received their participation fee (they were to receive it, net of charges, at the end of the two week period). A revised procedure was therefore adopted whereby the participants were given the cash in advance and told that they would have to pay some of it back at the end of the fortnight. This revised procedure does seem to have caused the drivers to take more notice of the charges. Using Money in an Experiment as a Surrogate for Other Gains and Losses in the Real World. The relative simplicity, universal appeal and apparent neutrality of money has led to it being used as an incentive in a wide variety of experiments even when monetary gain or loss is not directly relevant to the behaviour being studied - see, for example. Hey (1991). A classic example is in aircraft evacuation trials where, since ethical committees tend to discourage the use of real fires, some proxy incentive for rapid evacuation is required and it has been thought reasonable to offer cash rewards to participants who can get out of the aircraft most rapidly. One must obviously question whether a modest financial reward can really induce the kind of panic behaviour that would occur in a life threatening situation and one
368
In Perpetual Motion: Travel Behaviour Research and Opportunities
would suggest that the ability of money to cause people to take risks in exiting the aircraft would not be the same for all participants (compare single-young-males and mothers-withchildren). The prospect of financial gains and losses has an obvious relevance to studies of attitude to risk. Classic experiments (for example Tversky and Kahneman, 1974; Slovik et al, 1979) have revealed tendencies such as an 'irrational' amount of emphasis being placed on rare outcomes (as in the experiment which invites people to choose between a 1% chance of £100 loss and a 10% chance of a £10 loss ) and that attitudes to risk are conditioned by one's recent good or bad fortune ( as revealed by the tendency of gamblers who are on a winning roll to accept higher risks while their fellows who are showing a net loss are more likely to be satisfied with breaking even). These and other findings undoubtedly give valuable insight into peoples' attitudes to risk and uncertainty but must obviously be applied to other, non-money, contexts with great care. The work of Thaler and others has shown that, even in a money context, not all money is treated equally and it is clearly important to consider which 'type' of money provides the best analogy for the context in question. A number of authors have noted the extent to which the use of monetary incentives affects the behaviour of participants in experiments. Stein et al (1978) found that drivers in an instrumented car on a test track drove with greater care if they were being fined for contravening traffic regulations. Damkot et al (1983) found that prospects of a 10 cent per second incentive to maintain a specified constant speed had a significant affect on the average speed driven, the number of times the driver sought to look at the road or instruments, and on the usage of various vehicle controls. They concluded that the incentive had motivated drivers to put greater effort and concentration into the task. Finally, Van der Mede and Van Berkum (1991) found that offering a prize to the participant who achieved the fastest journey time in their driving simulator experiment caused participants to try harder to minimise their journey time. Other situations have been reported where a monetary incentive had no discernible effect. For example, Williams (1980) found that monetary incentives for 'skilfiil' performance during a simulation game had no discernible effects on attitudes formed during the game, while Breaugh and Klimoski (1981) found that monetary incentives had no discernible impact on participants' behaviour or reactions as team negotiators. The failure of incentives to influence behaviour in these experiments may reflect the absence of a single, quantified, measure of performance to which they could be tied or, it may reflect the existence of an alternative, more powerftil, motivation.
Motivating the Respondent: How Far Should You Go?
369
Using the Passage of Time as an Incentive during Experiments A more or less implicit (dis)incentive involved in many experiments is the passage of time; the task given to the participants takes time to perform and it is assumed that they will respond to this in much the same way as they would to the passage of time when performing the equivalent task in the real world. Unless the task is inherently enjoyable, the passage of time will normally be a disutility, which the participant in the experiment can be expected to seek to minimise. In some experiments, for example in full scale driving simulators, time is represented at its real-world level (one minute in the simulator represents one minute in the real world) whereas, in others where the task includes only selected aspects of the real world equivalent, time is compressed such that one minute in the simulator might represent several in the real world. This compression avoids the boredom which would set in if participants were required to spend several minutes doing nothing (while in the real world equivalent task they would be performing other functions) and allows the representation of several tasks within the finite duration of the experiment. It is, of course, important in these accelerated time simulations to ensure that appropriate relativities are maintained. Thus, in the VLADIMIR and TRAVSIM route choice simulators (Bonsall et al, 1994,1997) the time taken to pass along each stretch of road is proportional to what it would be on the equivalent stretches in the real road system. In the PARKIT parking choice simulator (Bonsall et al, 1998) the time spent driving, queuing, searching and then finally walking to the destination are again proportional to their real-world equivalents but, in order to represent the widely accepted differences in the perceived irksomeness of time spent in these different activities, the walking, waiting and searching time are less accelerated than is the driving time.
Implicit and Explicit Use of Objective Functions Even where no tangible incentive or reward is involved, we can assume that participants in an experiment will have some implicit motivation and we can characterise this motivation as a desire to 'succeed' by maximising some objective function. The components of this objective function, and their implied weights, will vary from individual to individual but will be strongly influenced by the presentation of the experiment. The context and purpose of the experiment
370
In Perpetual Motion: Travel Behaviour Research and Opportunities
will determine whether it is necessary, or even desirable, to draw the participants' attention to the components of the supposed objective function. Several route-choice simulators (e.g. those developed by Bonsai 1 and Parry, 1991; Bright and Ayland, 1991; Koutsopoulos et al, 1993; and Firmin, 1995) have sought to draw participants' attention to particular aspects of their 'performance' at the end of each journey. This has often been in the form of a table summarising aspects such as total journey time, total journey distance, some measure of driving style and, perhaps, an indication of performance relative to some optimum. However, it can be argued that this approach artificially overemphasises the variables which the experimental designer has deemed to be influential and so disallows or downgrades any other influences. An alternative philosophy, represented in most of the VLADIMIR work, has been to ask participants to behave as they would in real life and then to include naturalistic feedback on a number of factors without drawing particular attention to any of them. Thus the participant experiences the passage of time along links, the frustration of waiting in queues, the nature of the visible environment and other factors along the way without having their attention drawn to any of them in a summary table. The intention has been that the relative importance of these various factors will be perceived as they would be in the real world, but it is difficult to be sure that this ideal has been achieved; the ability of VLADIMIR to replicate real world behaviour (Bonsall et al, 1997) does, however, give cause for confidence. An alternative approach to the question of objective functions is to make them very explicit in the experimental procedure. The FASTCARS route-choice simulator (Adler et al, 1993) provides an interesting example of this approach; participants are first asked to specify their own objective function and their performance against this function is then brought to their attention as they progress through the specified tasks. With the participants thus sensitised, it is hoped that they will be better able to behave as they would in real life and it may be assumed that by drawing their attention to utility or disutility 'scores' they might be motivated to strive to maximise their net utility score and, if the objective function is correctly specified, thus to replicate their real world behaviour. There is ample evidence from psychological literature to suggest that a natural urge to compete causes people to seek to maximise a score even when there is no tangible reward for so doing.
Motivating the Respondent: How Far Should You Go?
371
Competition and Game Playing Participants' urge to 'succeed' in an exercise can be a powerful motivator in an experimental context and may be strong enough to allow us to dispense with any more tangible rewards. One interpretation of the failure of money rewards to affect the behaviour of participants in the Williams, and Breaugh and Klimoski, experiments referred to earlier is that the participants were already fully motivated by their desire to succeed in the tasks in which they were engaged. Participants' urge to succeed may however bring problems if the psychological rewards for success are not matched by equivalent risks involved in failure. A classic example of a situation in which the real world risks are under represented is in games-arcade "driving simulators" which encourage an unrealistically dangerous driving style. Interestingly, in an attempt to achieve greater realism, the latest generation of these games are apparently (Firmin, 1997) including more hazards for reckless drivers. Game playing does, of course, motivate some real world behaviour and a great deal of research is based on drawing analogies between behaviour in gaming situations and supposed behaviour in the real world. An interesting example is provided by a recent experiment by Davis et. al. (1995), which sought to explore drivers' route choice behaviour via a simple computer game which they regarded as an analogy for the route choice process. The 'routes' were represented by six tubes emitting marbles of different values. The rate at which marbles were emitted from a tube was indirectly proportional to the value of each marble it emitted and the players scored points by catching marbles as they came out of the tubes. They could move their 'marble eater' to catch marbles from any tube. At the end of the exercise they could expect a cash reward proportional to the total value of marbles they had caught. The authors noted that participants did not choose the optimal strategy (which would have been to stay put at any one tube) but chose instead to move their marble eater around from tube to tube. They observed that this behaviour was consistent with route switching observed in the real world and concluded that route choice in the real world can be interpreted as a game against nature rather than a rational attempt to minimise travel time. Their conclusion may well be correct but care must obviously be taken before suggesting that evidence of game playing in laboratory experiment is necessarily indicative of similar behaviour in a real world environment.
372
In Perpetual Motion: Travel Behaviour Research and Opportunities
Incentives, Biases and the ^Good Subject Effect' Several authors (eg Rosenthal and Rosnow, 1991; and Richardson et al, 1995) have reported on sources of experimental bias and it is now widely accepted that great care needs to be taken to avoid, or quantify, various forms of response bias including strategic bias, affirmation bias, social-norm bias, starting-point bias, and unconstrained-response bias. Less attention has however been paid to another type of bias which may affect a wide range of experiments. It derives from a co-operative intent on the part of the participant and is thus related in some ways to affirmation bias. It is caused by a tendency of people, knowing that they are taking part in an experiment, to rationalise the purpose of the experiment, to deduce what it is they are supposed to do, and then to act accordingly. A 'good subject' effect was recognised by experimental psychologists in the 1960s (eg Ome, 1962; Rosenberg, 1966; and Fillenbaum, 1966) and methods of reducing it have been proposed (Rosnow and Davis, 1977). However, although making due allowance for it can have a profound effect on the interpretation of results^ its full significance is widely ignored. Experiments conducted in an artificial or laboratory environment, or which invite a considered response from the participants, may be particularly susceptible to this bias. An experiment, such as a stated preference exercise or simple simulator, which draws the participants' attention to specific variables is, I suggest, very likely to encourage the participant to respond to these variables more than they would in a real-life situation where other factors would be equally visible. This concentration on a subset of variables may be acceptable if the aim of the experiment is to determine the relative importance of specified variables, but it is clearly a serious obstacle if one is seeking behavioural insight. The fact that such techniques may yield strong coefficients for their experimental variables does not prove that important determinants of real life behaviour have been found; it may merely result from the experimental context having been so simplified that the subject can easily understand how he is supposed to respond. I suggest that the tendency to act according to the deduced purpose of the exercise may be present whether the participant is responding out of a sense of duty ('good citizen' becomes 'good subject') or if they feel a sense of obligation to the experimenter who has paid them generously for their assistance (the 'obliged respondent'). The thought process may run as follows:- 'I'm being paid to do something ... it must require some effort on my part or they wouldn't be paying me ... I will put some effort into it'. This motivation may be entirely beneficial but would be counterproductive if it leads respondents to put in more 'effort' than they would in the equivalent real life situation.
Motivating the Respondent: How Far Should You Go?
373
Experiments particularly susceptible to this problem will include those which seek to shed light on the amount of information processing in which people engage before making a decision (eg Walker and Ben-Akiva, 1995; Goodwin and Wright, 1996). The latter experiment, which used paid participants, was designed to study the extent to which people were prepared to accept analytic predictions as a basis for forecasting. The authors noted that most participants sought to improve on the analytic predictions provided rather than accept the predictions at face value and concluded that this was a common trait in real life behaviour. However, it could be argued that some participants felt that simply accepting the analytic prediction would have been too easy and cannot therefore have been what they were supposed to do. The possible existence of 'good subject' effects must worry all users of data from questionnaires and experiments. They must be concerned that the results might be reflecting a pattern of behaviour, which, although logical and consistent, may have little to do with real life. The phenomenon might be minimised by using a recruitment incentive that is just sufficient to neutralise the good citizen's sense of duty while not being so generous as to engender a sense of obligation. However, since each potential respondent will have their own idea as to what a fair level of reward might be, it is inevitable that any chosen level (including zero) will leave some of them responding out of their sense of duty and others feeling obliged - and in either case they may try to act as 'good' subjects. The effect of a 'good subject' bias will be reduced but not eliminated, by asking people to behave 'as they would in real life'. A more certain reduction can, however, only come by improving the ecological validity of the experiment, for example by reducing the unnatural concentration on specified experimental variables and by allowing for a wider range of potential incentives and disincentives.
CONCLUSIONS There is clear evidence to suggest that recruitment incentives can influence response rates, that different people are influenced to a different degree by different forms and levels of incentive and that some are put off by them. It follows that, where participation is voluntary, the use of recruitment incentives may influence the results obtained but it is not immediately apparent whether the incentivised or the unincentivised respondents will yield 'truer' data. It is intuitively reasonable to suppose that the use of explicit incentives will influence the behaviour of people in experiments but we have seen that this influence may be far from
374
In Perpetual Motion: Travel Behaviour Research and Opportunities
straight forward. It is clearly difficult to achieve, or perhaps even to recognise, a balance between risks and rewards that maintains the ecological validity of the experiment for all classes of respondent. It is clear that the explicit use of incentives within experiments raises a number of complex issues and that it appears relatively easy to get it wrong! It would, however, be incorrect to assume that the problem can be avoided simply by eschewing the use of explicit incentives. Any would-be experimenter who seeks to steer clear of potential pitfalls by avoiding the use of incentives altogether would be wrong to imagine that their sample of respondents is thereby necessarily more representative of the general population or that their experiment necessarily has greater ecological validity. Although most respondents and participants can be assumed to be essentially co-operative and to be intrinsically motivated to complete the questionnaire correctly, there is no guarantee that self-motivated responses to a questionnaire or experiment will reflect real world behaviour and attitudes. Many of the issues outlined above have been considered in mainstream experimental psychology but, since some conclusions may be context-specific, it would be useful to conduct further experiments in the context of transport behaviour. An initial series of such experiments is planned in conjunction with colleagues in the Psychology Department at the University of Leeds. They will involve a sequence of driving tasks performed by matched groups of participants: i. volunteers, rewarded on the basis of performance ii. volunteers, no reward for good performance iii. paid participants, additionally rewarded on the basis of performance iv. paid participants, no reward for good performance The nature and size of the payments and rewards will be varied. Results will be published as soon as they are available.
ACKNOWLEDGEMENTS I am indebted to the various friends and colleagues for their participation in discussions on this topic over the last few months. Nick Ward and Daniel Read have been particularly generous of their time and knowledge of experimental practice in psychology and economics respectively. Neil Thorpe, Ian Palmer and Tim Brown have made very useful observations about the attitudes of participants in the road user charging experiments to the fees and charges. Simon Shepherd, John Blundell and Margaret Bonsall have each contributed very useful comments on
Motivating the Respondent: How Far Should You Go?
375
motivation and incentives and Paul Firmin made some very useful comments on a previous draft of the chapter. I must however remain responsible for any distortion or misinterpretation of their various contributions.
NOTES An example of how recognition of the good subject effect can result in reinterpretation of results is provided by Samuel and Bryant (1984) in their reworking of Piaget's classic experiment on children's ability to retain cognition of quantities. Piaget's original experiments had asked children to estimate quantities twice - once before and once after rearrangement of a fixed amount of material and had deduced from differences in the answers that rearrangement of the material had caused the children to reassess the quantities. The revised experiment involved simply asking the children whether they thought the quantities had changed when rearranged. The new result suggested a much greater ability to retain cognition of quantities and suggested that Piaget's original result had been due to the children assuming that, since they had been asked the same question twice, the answer must have changed; why else would they have been asked the same question again?
REFERENCES Adler, J. L., W. W. Recker and M. G. McNally (1993). A conflict model and interactive simulator (FASTCARS) for predicting enroute driver behavior in response to real-time traffic condition information. Transportation, Vol.20 (2), pp.83-106. Allen, R. W., R. H. Klein and K. Ziedman (1979). Automobile research simulators; a review and new approaches. Transportation Research Record no 706, pp 9-15. Bonsall, P. W., H-J. Cho, I. A. Palmer and N. Thorpe (1998). Experiments to determine differences in drivers responses to a variety of road user charging regimes. Paper prepared for WCTR Conference, Antwerp July 1998. Bonsall, P. W., R. Clarke, P. E. Firmin and I. A. Palmer (1994). VLADIMIR and TRAVSIM; Powerful aids for route choice research. Proc 22nd Annual meeting of PTRC (European Transport Forum) Seminar H, pp 65-76, PTRC London. Bonsall, P. W., P. E. Firmin, M. E. Anderson, I. A. Palmer and P. A. Balmforth (1997). Validating the results of a route choice simulator. Transportation Research C, vol 5 , no 6. Bonsall, P. W. and J. E. McKimm (1993). Non-response bias in roadside mail-back surveys Traffic Engineering and Control, Vol.34 (12), pp 582-591. Bonsall, P. W. and I. A. Palmer (1997). Do time-based road user charges induce risk-taking? results from a driving simulator. Traffic Engineering and Control, Vol. 38(4), pp 200208. Bonsall, P. W., I. A. Palmer and P. A. Balmforth (1998). PARKIT - a simulated world for parking choice research. Paper prepared for WCTR Conference, Antwerp July 1998.
376
In Perpetual Motion: Travel Behaviour Research and Opportunities
Bonsall, P. W. and T. Parry (1991). Using an interactive route choice simulator to investigate drivers' compliance with route guidance advice. Transportation Research Board, 1306,pp.59-68. Breaugh, J. A. and R. J. Kiimoski (1981). Social forces in negotiation simulations. Personality and Social Psychology Bulletin ,voi 7(2) pp 290-295. Bright, J. and N. Ayland (1991). Evaluating real time responses to in-vehicle driver information systems, in: EU Commission (ed), Advanced Transport Telematics Proceedings of the DRIVE Conference, Brussels Feb 1991, vol 1, pp70-88, Elsevier. Brog, W. and A. H. Meyburg (1981). Consideration of non-response effects in large-scale mobility surveys. Transportation Research Record 775, pp.34-38. Chapanis, A. (1967). The Relevance of laboratory study to practical situations. Ergonomics, voll0,pp557-577. Chipman, M., M. Lee-Gosselin, C. MacGregor, A. Smiley and L. Clifford (1992). The effects of targeted sampling in a driver exposure survey, in Ampt, Richardson and Meyburg (eds) Selected Readings in Transport Survey Methodology, Eucalyptus Press, Melbourne. Damkot, D. K., R. S. Kirk and M.S. Huntley (1983). Influences of alcohol, monetary incentive and visual interruption upon control use during automobile driving. Alcohol and Alcoholism vol 18 no 1 pp 81-88. Davis, A. and M. Powell with J. Bowers and A. Palmer (1995). The importance of small time savings in driver route decisions. Final report on project F733-03-7 to English Nature and the Countryside Council for Wales. Dawson, R. F. F. (1974). Environmental simulator progress report. Transport Research Laboratory Report LR659, TRRL , Crowthome. Festinger, L. and J. M. Carlsmith (1959). Cognitive consequences of forced compliance. Journal ofAbnormal and Social Psychology, vol 58 pp 203-210. Filion, F.L. (1975-76). Estimating bias due to non response in mail surveys. Public Opinion Quarterly, Vol. 39, pp.482-492. Fillenbaum, S. (1966). Prior deception and subsequent experimental performance: the faithful subject Journal of Personality and Social Psychology, vol 4 pp 532 537 Firmin, P.E. (1995). The use of simulated travel environments to investigate drivers' route choice behaviour. PhD Thesis, Institute for Transport Studies, University of Leeds. Firmin, P.E. (1997). Personal communication on latest generation Grand Prix racing simulators. Freedman, J. L., J. A. Cunningham and K. Krismer (1992). Inferred values and the reverseincentive effect in induced compliance. Journal of Personality and Social psychology, vol 62 (3) pp 357-368. Goodwin, P. and Wright (1996). Judgement in Forecasting, paper presented in Risk and Uncertainty Seminar Series, Centre for the Assessment of Risk, Universty of Leeds, Leeds. Hey, J. (1991). Experiments in Economics, Basil Blackwell, Oxford. Horowitz, I. A. (1969). Effects of volunteering, fear arousal and number of communications on attitude change. Journal of Personality and Social Psychology, Vol. 1, pp.34-37. Koutsopoulos, H. N., T. Lotan and Q. Yang (1993). A driving simulator and its application for modelling route choice in the presence of information. Transportation Research, Vol. 2C, pp 91-107.
Motivating the Respondent: How Far Should You Go?
?>11
Koutsopoulos, H. N., A. Polydoropoulou and M. Ben-Akiva (1995). Travel simulators for data collection on driver behavior in the presence of information. Transportation Research, Vol.3C,pp.l43-159. Mast, T. M. and J. A. Ballas (1976). Diversionary signing content and driver behavior. Transportation Research Report 600, Transportation Research Board, Washington. Mook, D. G. (1983). In defence of external invalidity. American Psychologist, Vol. 38, pp 379-387. Ome, M. T. (1962). On the social psychology of the psychological experiment: with particular reference to demand characteristics and their implications. American Psychologist, pp 176-783 Parkes, A. M. and G. E. Burnett (1993). An evaluation of medium range advance information in route guidance displays for use in vehicles. Proc International Conference on Vehicle Navigation and Information Systems, Ottawa October 1993, IEEE. Richardson, A. J., E. S. Ampt and A. H. Meyburg (1995). Survey Methods for Transport Planning, Eucalyptus Press, Melbourne. Rosenburg, M. J. (1966). The conditions and consequences of evaluation apprehension, in Rosenthal J and Rosnow RL (eds). Artifact in Behavioral Research pp (279-349), New York, Academic Press Rosenthal, R. and R. L. Rosnow (1991). Essentials of Behavioral Research, Methods and Data Analysis, 2nd edition. McGraw-Hill series in Psychology. Rosnow, R. L. and F. J. Davis (1977). Demand characteristics and the psychological experiment, ETC: A Review of General Semantics, 34, pp.301-313. Reproduced in R. Rosenthal and R. L. Rosnow (1991). Essentials of Behavioral Research, Methods and Data Analysis, 2nd edition. McGraw-Hill series in Psychology. Sammer, G. and K. Fallast (1985). Effects of various population groups, and of issue and return methods, on the return of questionnaire and the quality of answers in large-scale travel surveys. In: Ampt, Richardson and Brog (eds). New Survey Methods in Transport. VNU Science Press, Utrecht. Samuel, J. and P. Bryant (1984). Asking only one question in the conservation experiment. Journal of Child Pshychology and Psychiatry and Allied Disciplines, vol 25 no 2 pp 315-318. Slovic, P., B. Fischhoff and S. Lichtenstein (1979). Rating the risks. Environment, vol 21 (3) pp 14-20, 36-39. Sommer, B. and R. Sommer (1991). ^ Practical Guide to Behavioural Research; Tools and Techniques, 3rd edition, Oxford University Press. Stein, A. C.j R. W. Allen and S. H. Schwartz (1978). Use or reward-penalty structures in human experimentation, Proceedings of Nth Annual Conference on Manual Control, University of Southern California, April 1978, pp 267-278. Thaler, R. H. and E. J. Johnson (1990). Gambling with house money and trying to beak even: The effects of prior outcomes on risky choice. Management Science , vol 36, pp 643660. Thorpe, N. (1997). Personal communication on behaviour of subjects in Newcastle trials of road-user charging regimes. Tversky, A. and D. Kahneman (1974). Judgement under uncertainty: heuristics and biases. Science, vol 185 pp 1124-131.
378
In Perpetual Motion: Travel Behaviour Research and Opportunities
Van der Mede, P. H. J. and E. C. Van Berkum (1991). Modeling route choice, inertia and responses to variable message signs. Proc 6th International Conference on Travel Behavior Research (lATBR) Quebec City May 1991, lATBR. Walker, J. and M. Ben-Akiva (1995). Modelling traveler information systems: laboratory simulation of information searches using multimedia technology. Paper presented at Transportation Research Board Meeting, Washington DC. Williams, R. H. (1980). Attitude change and simulation games. Simulation and games, vol 11(2) pp 177-196.
SECTION 7 METHODOLOGICAL DEVELOPMENTS
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
19
RECENT METHODOLOGICAL ADVANCES RELEVANT TO ACTIVITY AND TRAVEL BEHAVIOUR ANALYSIS
Chandra R. Bhat
ABSTRACT This chapter presents an overview of the considerable progress in modeling methodology that has been made in recent years and that is directly relevant to improved transportation policy analysis and travel demand forecasting. The overview is organized under three broad classes of models: discrete choice models, hazard-based duration models, and limited-dependent variable models. Because of the objective of the chapter, the focus here is on the methodological aspects of various studies rather than on the empirical findings from the studies. Some important methodological topics have necessarily been omitted from this review because of space considerations. We have tried to specifically identify these topics at the beginning of the chapter and provide references to recent reviews on these topics.
INTRODUCTION This chapter reviews recent methodological advances relevant to modeling activity and travel behaviour. The overview of the state-of-the-art in modeling is motivated by two considerations. First, discrete-choice models that have been well established in the past (such as the multinomial logit and nested logit models) have been generalized in several ways to make them more realistic in their representations of travel choice behaviour. Second, the increasing realization of the need to model travel as part of a larger (and holistic) activitytravel pattern has led to the analysis of activity attributes (such as activity participation, activity
382
In Perpetual Motion: Travel Behaviour Research and Opportunities
duration, home-stay duration, etc.) either in isolation or jointly with one another. This has led to the adoption of relatively non-traditional (in the travel analysis field) methodologies such as duration analysis, limited-dependent variable models, and computational process models. Several points should be noted before proceeding to the remainder of the chapter. First, this review is not intended to provide detailed information regarding the structures or estimation procedures for the models reviewed. Such a task would not be feasible within the space constraints of a single paper. Second, the review does not touch on methodological developments in survey data collection techniques or analytic imputation techniques for missing data. These issues have been addressed in papers presented at a recent conference in Stockholm (see Stopher, 1996 and Brownstone, 1996). Third, the review does not address methodological advances in joint estimation from revealed preference (RP) as well as stated preference (SP) data. Instead, the chapter focuses on estimation from revealed preference data only. The reader is referred to Hensher (1994a) for an overview of the methods of RP-SP estimation. Fourth, this chapter does not review computational process models since comprehensive reviews of these models have been conducted not very long ago by pioneers of the approach who are more knowledgeable regarding the approach than the current author (see Garling et al., 1994; GoUedge et al., 1994; Kitamura and Fujii, 1996; and Kurani and Kitamura, 1996). An additional reason for not focusing on computation process models (CPMs) here is that while there has been considerable advancement in these methods, some basic issues related to statistical estimation and calibration of CPMs are yet to be defined and resolved (Gollfedge et al., 1994). Fifth, we focus on methods for cross-sectional data rather than longitudinal (or panel) data in the current chapter. In concept, the methods discussed here for cross-sectional analysis can be extended to panel analysis after accommodating the additional econometric issues introduced by panel data. Finally, the chapter does not review advances in methods related to demand-supply interaction analysis or demand-supply equilibration.
DISCRETE CHOICE MODELS The multinomial logit (MNL) model has been the most widely used structure for modeling discrete choices in travel behaviour analysis. The random components of the utilities of the different alternatives in the MNL model are assumed to be independent and identically distributed (IID) with a type I extreme-value (or Gumbel) distribution (McFadden, 1973). The MNL model also maintains an assumption of homogeneity in responsiveness to attributes of alternatives across individuals (i.e., an assumption of response homogeneity). For example, in a mode choice model, the MNL maintains the same utility parameters on the level-of-service attributes across individuals. Finally, the MNL model also maintains an assumption that the
Recent Methodological Advances Relevant to Activity and Travel Behaviour
3 83
error variance-covariance structure of the alternatives is identical across individuals {i.e., an assumption of error variance-covariance homogeneity). The three assumptions together lead to the simple and elegant closed-form mathematical structure of the MNL. However, these assumptions also leave the MNL model saddled with the "independence of irrelevant alternatives" (IIA) property at the individual level (Ben-Akiva and Lerman, 1985).^ In the next three sections, we will discuss generalizations of the MNL structure along each of the three dimensions mentioned above: a) Relaxation of the IID (across alternatives) error structure, b) Relaxation of response homogeneity, and c) Relaxation of the error variance-covariance structure homogeneity. While we discuss each of the dimensions separately, one can combine extensions across different dimensions to formulate several more generalized and richer structures.
Relaxation of the IID (Across Alternatives) Error Structure The rigid inter-alternative substitution pattern of the multinomial logit model can be relaxed by removing, fully or partially, the IID assumption on the random components of the utilities of the different alternatives. The IID assumption can be relaxed in one of three ways: a) allowing the random components to be correlated while maintaining the assumption that they are identically distributed (identical, but non-independent random components), b) allowing the random components to be non-identically distributed (different variances), but maintaining the independence assumption (non-identical, but independent random components), and c) allowing the random components to be non-identical and non-independent (non-identical, nonindependent random components). We discuss each of these alternatives below. Identical, Non-independent Random Components The distribution of the random components in models which use identical, non-independent random components can be specified to be either normal or type I extreme value. Discrete choice literature has mostly used the type I extreme value distribution since it nests the multinomial logit and results in closed-form expressions for the choice probabilities. The models with the type I extreme value error distribution belong to the Generalized Extreme Value (GEV) class of random utility-maximizing models. Five model structures have been formulated and applied within the GEV class. These are: the Nested Logit (NL) model, the Paired Combinatorial Logit (PCL) model, the cross-nested logit (CNL) model, the Ordered GEV (OGEV) model, and the Multinomial Logit-Ordered GEV (MNL-OGEV) model.
384
In Perpetual Motion: Travel Behaviour Research and Opportunities
The nested logit (NL) model permits covariance in random components among subsets (or nests) of alternatives (each alternative can be assigned to one and only one nest). Alternatives in a nest exhibit an identical degree of increased sensitivity relative to alternatives not in the nest (Williams, 1977 and Daly and Zachary, 1978). Each nest in the NL structure has associated with it a dissimilarity (or logsum) parameter that determines the correlation in unobserved components among alternatives in that nest (see Daganzo and Kusnic, 1993). The range of this dissimilarity parameter should be between 0 and 1 for all nests if the NL model is to remain globally consistent with the random utility maximizing principle. A problem with the NL model is that it requires a priori specification of the nesting structure. This requirement has at least two drawbacks. First, the number of different structures to estimate in a search for the best structure increases rapidly as the number of alternatives increases. Second, the actual competition structure among alternatives may be a continuum that cannot be accurately represented by partitioning the alternatives into mutually exclusive subsets. The NL model has been applied to multidimensional choice contexts (for example, see Waddell 1993 and Evers, 1990) and unidimensional contexts where subsets of the available alternatives share common unobserved components of utility (for example, see Forinash and Koppelman, 1993 and Brownstone and Small, 1989). The paired combinatorial logit (PCL) model initially proposed by Chu (1989) and recently examined in detail by Koppelman and Wen (1997) generalizes, in concept, the nested logit model by allowing differential correlation between each pair of alternatives (the nested logit model, however, is not nested within the PCL structure). Each pair of alternatives in the PCL model has associated with it a dissimilarity parameter (subject to certain identification considerations that Koppelman and Wen are currently studying) that is inversely related to the correlation between the pair of alternatives. All dissimilarity parameters have to lie in the range of 0 to 1 for consistency with random utility maximization. In the intercity mode choice empirical analysis of Koppelman and Wen, the PCL model which allows correlation between air and car modes as well as between train and car modes performed better than the nested logit models which nests air and car only or train and car only. Koppelman and Wen derive the expressions for the self- and cross-elasticities in the PCL model and show empirically that the policy impacts suggested by the restrictive MNL and nested logit models can be quite different from those suggested by the statistically superior (in their empirical context) PCL model. Another generalization of the nested logit model is the cross-nested logit (CNL) model of Vovsha (1996). In this model, an alternative need not be exclusively assigned to one nest as in the nested logit structure. Instead, an alternative can appear in different nests with different probabilities based on what Vovsha refers to as allocation parameters. A single dissimilarity parameter is estimated across all nests in the CNL structure. Unlike in the PCL model, the
Recent Methodological Advances Relevant to Activity and Travel Behaviour
385
nested logit model can be obtained as a special case of the CNL model when each alternative is unambiguously allocated to one particular nest? Vovsha proposes a heuristic procedure for estimation of the CNL model. This procedure appears to be rather cumbersome and its heuristic nature makes it difficult to establish the statistical properties of the resulting estimates. The ordered GEV model was developed by Small (1987) to accommodate correlation among the unobserved random utility components of alternatives close together along a natural ordering implied by the choice variable (examples of such ordered choice variables might include car ownership, departure time of trips, etc.). The simplest version of the OGEV model (which Small refers to as the standard OGEV model) accommodates correlation in unobserved components between the utilities of each pair of adjacent alternatives on the natural ordering; that is, each alternative is correlated with the alternatives on either side of it along the natural ordering.^ The standard OGEV model has a dissimilarity parameter that is inversely related to the correlation between adjacent alternatives (this relationship does not have a closed form, but the correlation implied by the dissimilarity parameter can be obtained numerically). The dissimilarity parameter has to lie in the range of 0 to 1 for consistency with random utility maximization. Empirical applications of the OGEV model have not been very successful thus far (that is, the OGEV model was not significantly better than the MNL or the dissimilarity parameter exceeded one). However, it is important to note that only two such attempted applications have been documented to date, both by Small (1987). The MNL-OGEV model formulated by Bhat (1997a) generalizes the nested logit model by allowing adjacent alternatives within a nest to be correlated in their unobserved components. This structure is best illustrated with an example. Consider the case of a muhi-dimensional model of travel mode and departure time for nonwork trips. Let the departure time choice alternatives be represented by several temporally contiguous discrete time periods in a day such as AM peak (6AM-9AM), AM mid-day (9AM-12Noon), PM mid-day (12Noon-3PM), PM peak (3PM-6PM), and other (6PM-6AM). An appropriate nested logit structure for the joint mode-departure time choice model may allow the joint choice alternatives to share unobserved attributes in the mode choice dimension, resulting in an increased sensitivity among time-of-day alternatives of the same mode relative to the time-of-day alternatives across modes. However, in addition to the uniform correlation in departure time alternatives sharing the same mode, there is likely to be increased correlation in the unobserved random utility components of each pair of adjacent departure time alternatives due to the natural ordering among the departure time alternatives along the time dimension. Accommodating such a correlation generates an increased degree of sensitivity between adjacent departure time alternatives (over and above the sensitivity among non-adjacent alternatives) sharing the same
386
In Perpetual Motion: Travel Behaviour Research and Opportunities
mode. A structure that accommodates the correlation patterns just discussed can be formulated by using the multinomial logit (MNL) formulation for the higher-level mode choice decision and the standard ordered generalized extreme-value (OGEV) formulation (see Small, 1987) for the lower-level departure time choice decision {i.e., the MNL-OGEV model). The MNLOGEV structure, in the context of the mode-departure time example, has two dissimilarity parameters: one is associated with the correlation among joint alternatives sharing the same mode, and the other is associated with the increased correlation between adjacent departure time alternatives of the same mode. For consistency with random utility maximization, both these parameters should be less than 1 and the latter dissimilarity parameter should be smaller than the former dissimilarity parameter. The advantage of all the GEV models discussed above is that they allow partial relaxations of the independence assumption among alternative error terms while maintaining closed-form expressions for the choice probabilities. The problem with these models is that they are consistent with utility maximization only under rather strict (and often empirically violated) restrictions on the dissimilarity parameters. The origin of these restrictions can be traced back to the requirement that the variance of the joint alternatives be identical in the GEV models. Non-Identical, Independently Distributed Random Components The concept that heteroscedasticity in alternative error terms {i.e., independent, but not identically distributed error terms) relaxes the IIA assumption is not new (see Daganzo, 1979), but has received little (if any) attention in travel demand modeling and other fields. In fact, the IIA property has become virtually synonymous with the assumption of lack of similarity (or independence of random components) among the choice alternatives in travel demand literature. Four models have been proposed which allow non-identical random components. The first is the negative exponential model of Daganzo (1979), the second is the heteroscedastic multinomial logit (HMNL) model of Swait and Stacey (1996), the third is the oddball alternative model of Recker (1996) and the fourth is the heteroscedastic extreme-value (HEV) model of Bhat (1995). Daganzo (1979) used independent negative exponential distributions with different variances for the random error components to develop a closed-form discrete choice model that does not have the IIA property. His model has not seen much application since it requires that the perceived utility of any alternative not exceed an upper bound (this arises because the negative exponential distribution does not have a full range)."^ Daganzo's model does not nest the multinomial logit model.
Recent Methodological Advances Relevant to Activity and Travel Behaviour
387
Swait and Stacey (1996) allowed heteroscedasticity by specifying the variance of the alternative error terms to be functions of observed alternative characteristics. The error terms themselves are assumed to be type I extreme-value. The scale parameter 6. characterizing the variance of each alternative / is written as 9. = exp(P'z,), where z. is a vector of attributes associated with alternative / and p is a corresponding vector of parameters to be estimated. The resulting model has a closed-form structure, though it also places the restriction that the differing variances of the alternatives can be attributed solely to observed alternative characteristics. Swait and Stacey applied the model to brand choice modeling using scanner panel data. Recker (1996) proposed the oddball alternative model that permits the random utility variance of one "oddball" alternative to be larger than the random utility variances of other alternatives. This situation might occur because of attributes that define the utility of the oddball alternative, but are undefined for other alternatives. Then, random variation in the attributes that are defined only for the oddball alternative will generate increased variance in the overall random component of the oddball alternative relative to others. For example, operating schedule and fare structure define the utility of the transit alternative, but are not defined for other modal alternatives in a mode choice model. Consequently, measurement error in schedule and fare structure will contribute to the increased variance of transit relative to other alternatives. The model has a closed-form structure for the choice probabilities based on convenient distributional assumptions on the random components. However, the model is quite restrictive in requiring that all alternatives except one have identical variance. Bhat (1995) formulated the heteroscedastic extreme-value (HEV) model which assumes that the alternative error terms are distributed with a type I extreme value distribution. The variance of the alternative error terms is allowed to be different across all alternatives (with the normalization that the error terms of one of the alternatives has a scale parameter of one for identification). Consequently, the HEV model can be viewed as a generalization of Recker's oddball alternative model. The HEV model is applied to an intercity mode choice context. The motivation is that unequal error variances are likely to occur when the variance of an unobserved variable that affects choice is different for different alternatives. For example, if comfort is an unobserved variable whose values vary considerably for the train mode (based on, say, the degree of crowding on different train routes) but little for the automobile mode, then the random components for the automobile and train modes will have different variances (Horowitz, 1981). The HEV model does not have a closed-form solution for the choice probabilities, but involves only a one-dimensional integration regardless of the number of alternatives in the choice set. Bhat develops an efficient Gauss-Laguerre quadrature technique to approximate the one-dimensional integral. The HEV model can be modified to
388
In Perpetual Motion: Travel Behaviour Research and Opportunities
accommodate variations in the scale parameter because of observed alternative attributes, as done by Sw^ait and Stacey (1996) .^ The advantage of the heteroscedastic class of models discussed above is that they allow a flexible cross-elasticity structure among alternatives than many of the GEV models discussed earlier. Specifically, the models (except the oddball model) permit differential cross-elasticities among all pairs of alternatives. The limitation (relative to the GEV models) is that the choice probabilities do not have a closed-form analytical expression in the HEV model. Non-Identical, Non-independent Random Components Models with non-identical, non-independent random components use one of two general structures: the first is an error-components structure and the second is the general multinomial probit (MNP) structure. The error-components structure partitions the overall error into two components: one component which allows the random components to be non-identical and non-independent, and the other component which is specified to be independent and identically distributed across alternatives. In particular, consider the following utility function for alternative /: (1) = F;+|Ltr.+8,. where V^ and C,^ are the systematic and random components of utility, and C,^ is further partitioned into two components, [x.'z. and s,. z. is a vector of observed data associated with alternative /, JLI is a random vector with zero mean and density g{\x. | Z), L is the variancecovariance matrix of the vector )LI, and s. is independently and identically standard distributed across alternatives with density function f(.). The component \x'z. induces heteroscedasticity and correlation across unobserved utility components of the alternatives (see Train, 1995). While different distributional assumptions might be made regarding^-) and g(.), it is typical to assume a standard type I extreme value fory(.), and a normal distribution for g(.). This results in an error-components model with a logit kernel. On the other hand, if a standard normal distribution is used forX-)» the result is a error-components probit model. Both these structures will involve integrals in the choice probability expressions which do not have a closed-form solution. The estimation of these models is achieved using logit simulators (in the first case) or probit simulators (in the second case). Different and very general patterns of heteroscedasticity and correlation in unobserved components among alternatives can be generated by appropriate specification of the \x and z,. vectors. For example, z. may be specified to be a row vector of dimension M with each row representing a group m of alternatives sharing common
Recent Methodological Advances Relevant to Activity and Travel Behaviour
3 89
unobserved components. The row(s) corresponding to the group(s) of which i is a member take(s) a value of one and other rows take a value of zero. The vector \x (of dimension M) may be specified to have independent elements, each element having a variance component a^. The result of this specification is a covariance of a^ among alternatives in group m and heteroscedasticity among the groups of alternatives. This structure is less restrictive than the nested logit structure in that an alternative can belong to more than one group. Also, by structure, the variance of the alternatives is different (see Bhat, 1997b for application of this structure to a muhi-dimensional choice context).^ More general structures for
|LI'Z.
in equation
(1) are presented by Ben-Akiva and Bolduc (1996) and Brownstone and Train (1996). The general multinomial probit (MNP) structure does not partition the error terms, and estimates (subject to certain identification considerations) the variance-covariance matrix of the overall random components among the different alternatives (see Bunch and Kitamura, 1990; Lam, 1991; Lam and Mahmassani, 1991; and Chintagunta and Honore, 1996). The advantage of the MNP model is that the structure is more general than the error-components models (the error components structure essentially parameterizes the variance-covariance matrix in an MNP model using an a priori structure). However, McFadden and Train (1996) have shown that the error-components formulation can approximate a multinomial probit formulation as closely as one needs it to. Also, the MNP model introduces several additional parameters in the covariance matrix which generates a number of conceptual, statistical and practical problems, including difficulty in interpretation, highly non-intuitive model behaviour, low precision of covariance parameter estimates, and increased difficulty in transferring models from one space-time sampling frame to another (see Horowitz, 1991 and Currim, 1982). Further, the error-components models can be estimated using simulators which are conceptually simple and easy to program. These simulators involve simultaneous draws from the appropriate density function with unrestricted ranges for all alternatives. Consequently, they are inherently faster than simulators for the MNP model where the range for the random draw of one alternative is dependent on the value of the earUer draws for other alternatives (see Brownstone and Train, 1996).
Relaxation of Response Homogeneity The standard multinomial logit, and other models which relax the IID assumption across alternatives, typically assume that the parameters determining the sensitivity to attributes of the alternatives are the same across individuals in the population. Ideally, we should obtain individual-specific parameters for the subjective evaluations of alternative attributes. However, the data used for travel behaviour modeling is usually cross-sectional. This precludes
390
In Perpetual Motion: Travel Behaviour Research and Opportunities
estimation at the individual level and constrains the modeler to pool the data across individuals. In such pooled estimations, the analyst should accommodate differences in responsiveness to alternative attributes (response heterogeneity) across individuals. In particular, if the assumption of homogeneity is imposed when there is heterogeneity, the result is biased and inconsistent parameter and choice probability estimates (see Chamberlain, 1980). Response homogeneity may be relaxed in one of two ways. In the first approach (which we will refer to as the varying coefficients approach), the coefficients on alternative attributes are allowed to vary across individuals while maintaining a single utility function. In the second approach (which we will refer to as the segmentation approach), individuals are assigned to segments based on their personal/trip characteristics and a separate utility function is estimated for each segment. Each of these approaches is discussed in the subsequent two sections. Varying Coefficients Approach Consider the utility U^. that an individual q associates with an alternative / and write it as: ^^,. =a,.+5;z^+8^,.+ilX,
(2)
where a, is an individual-invariant bias constant, z^ is a vector of observed individual characteristics, 5. is a vector of parameters to be estimated, 8^. is a random term representing idiosyncrasies in preferences, and r|^ is a vector representing the responsiveness of individual q to a corresponding vector of alternative-associated variables x^.. The s^, terms may be specified to have any of the structures discussed in in the second section of this chapter. Conditional on r|^ and the assumption regarding the s^. terms, the form of the conditional choice probabilities can be developed. The unconditional choice probabilities corresponding to the conditional choice probabilities will depend on the response heterogeneity specification adopted for the r\^ vector. Three specifications are possible, as discussed next. The first specification allows for systematic response heterogeneity by writing each element r\^^ of the vector r|^ as a function of a vector w^^ of relevant observed individual characteristics: ^^^ = Y^+P*/(%A)(^^it
represents the response coefficient of the qth
individual to the kth alternative attribute. If we pre-specify a functional form for /(w^^^)' ^^^^ the unconditional choice probabilities take the same structure as the conditional (on r|^) choice probabilities. A problem with the specification of the form r)^^ =y^
+PA/(^^A)»
however, is
that it does not guarantee the correct sign of the response coefficient r|^^ for all individuals.
Recent Methodological Advances Relevant to Activity and Travel Behaviour
391
For example, in a mode choice context, we expect the effect of the travel time and cost variables to be negative for all individuals, which is not guaranteed by writing ^gA - Yt+PA/(%A)-
^ ^ alternative method that accommodates systematic response
heterogeneity and, at the same time, ensures the appropriate sign on the response coefficients is to specify r|^^ = ±exp(Y^+p^w^^). The '+' sign is applied for a non-negative response coefficient and the '-' sign is applied for a non-positive response coefficient. The unconditional choice probabilities with this exponential specification are the same as the conditional probabilities after replacing ri^^^ with ±exp(Y^ +PA%A) • The resulting model, however, now has a non-linear-in-parameters utility function. It is important to emphasize that incorporating systematic heterogeneity may contribute to a more realistic representation of response sensitivity, but does not relax the IIA property (if the IID error assumption across alternatives of the MNL is maintained). The second specification for the response coefficients allows random variation is sensitivity, but does not accommodate differences in sensitivity due to observed individual attributes. One form for the random variation may be r|^^ = Y* + ^qk •> where y^^ is the mean response sensitivity across all individuals in the population and v^^ is a term representing random taste variation of individual q from the mean. Alternatively, if the response coefficient needs to be of a particular sign, then one can use an alternative form: ri^^ = ±exp(Y^ + ^^A) • ^qk is typically assumed to be normally distributed, which implies that the response coefficients are normally distributed if one uses the first form and log-normally distributed if one uses the second form. Applications of random response heterogeneity include Fischer and Nagin, 1985, Revelt and Train, 1996, Train, 1996, Ben-Akiva et al., 1993, Goniil and Srinivasan, 1993, and Mehndiratta, 1996. The v^^ terms in the random response heterogeneity specification represent the random tastes of person q and are common to the utility of all alternatives /. Therefore, variance in the v^^ terms across individuals induces a correlation among the utility of different alternatives (see McFadden and Train, 1996). As a result, the random response specification does not exhibit the restrictive independence from irrelevant alternatives (IIA) property even if the IID error assumption across alternatives of the MNL is maintained. The third (and most general) specification for the response coefficients superimposes random response
heterogeneity
over
the
systematic
response
heterogeneity:
r|^^ =±exp(Y^ +Pl^gA +^9*)' where v^^ is a term representing random taste variations across individuals with the same observed characteristics w^^. Bhat (1996a) adopts such a specification in an intercity mode choice context.
392
In Perpetual Motion: Travel Behaviour Research and Opportunities
Segmentation Approaches Two segmentation approaches may be identified depending on whether the assignment of individuals to segments is exogenous (deterministic) or endogenous (probabilistic). The exogenous segmentation approach to capturing heterogeneity assumes the existence of a fixed, finite number of mutually-exclusive market segments (each individual can belong to one and only one segment). The segmentation is based on key socio-demographic variables (sex, income, etc.). Within each segment, all individuals are assumed to have identical preferences and identical sensitivities to level-of-service variables {i.e., the same utility function). In the exogenous segmentation approach, the assignment of individuals to segments is deterministic and is implicit in the definition of the segments. A choice model is estimated subsequently for each segment. The total number of segments is a function of the number of segmentation variables and the number of segments defined for each segmentation variable. Ideally, the analyst would consider aH socio-demographic and trip-related variables available in the data for segmentation (we will refer to such a segmentation scheme as full-dimensional exogenous market segmentation). However, the full-dimensional segmentation approach has a practical limitation; the total number of segments grows very fast with the number of segmentation variables, creating both interpretational and estimation problems due to inadequate observations in each segment (with typical sample sizes used for mode choice analysis). To overcome this limitation, researchers have resorted to a Limited-Dimensional Exogenous Market Segmentation method, which uses only a subset of the demographic and trip variables (typically one or two) for segmentation. It is not uncommon for the subset of variables to be decided a priori based on judgment, though one could estimate models with different subsets and then select the preferred subset for segmentation. The advantage of the limiteddimensional approach is that it is practical (the parameters can be efficiently estimated with data sizes generally available for mode choice analysis) and is easy to implement. The disadvantage is that its practicality comes at the expense of suppressing potentially higherorder interaction effects of the segmentation variables on response to alternative attributes. In addition, an intrinsic problem with the exogenous market segmentation methods is that the threshold values of the continuous segmentation variables (for example, income) which define segments have to be established in a rather ad hoc fashion. Also, the exogenous approach does not relax the IIA property if the IID (across alternatives) assumption on the random components is maintained. The endogenous market segmentation approach attempts to accommodate heterogeneity in a practical manner not by suppressing higher-order interaction effects of segmentation variables (on response to alternative attributes), but by reducing the dimensionality of the segmentspace. Each segment, however, is allowed to be characterized by a large number of
Recent Methodological Advances Relevant to Activity and Travel Behaviour
393
segmentation variables. The appropriate number of segments representing the reduced segment-space is determined statistically by successively adding an additional segment till a point is reached where an additional segment does not result in a significant improvement in fit. Individuals are assigned to segments in a probabilistic fashion based on the segmentation variables. The approach jointly determines the number of segments, the assignment of individuals to segments, and segment-specific choice model parameters. Since this approach identifies segments without requiring a multi-way partition of data as in the full-dimensional exogenous market segmentation method, it allows the use of many segmentation variables in practice and, therefore, facilitates incorporation of the full order of interaction effects of the segmentation variables on preference and sensitivity to alternative attributes. The method also obviates the need to (arbitrarily) establish the threshold values defining segments for continuous segmentation variables. The approach does not exhibit the individual-level independence from irrelevant alternatives (IIA) property of the exogenous segmentation approach even if a multinomial logit structure is maintained within each segment. A potential disadvantage is that the model estimation can be unstable. However, Bhat (1997c) has recently proposed a stable and effective hybrid estimation approach for the endogenous segmentation model that combines an Expectation-Maximization (EM) algorithm with standard likelihood maximization routines. Other applications of the endogenous segmentation approach may be found in Gopinath and Ben-Akiva (1995), Swait (1994), Gupta and Chintagunta (1994), Dayton and Macready (1988), and Swait and Sweeney (1996).^
Relaxation of Error Variance-Covariance Structure Homogeneity The assumption of error variance-covariance structure homogeneity across individuals can be relaxed either by a) allowing the variance components to vary across individuals (variance relaxation), b) allowing the covariance components to vary across individuals (covariance relaxation), and c) allowing both variance and covariance components to vary across individuals (variance-covariance relaxation). Variance Relaxation Swait and Adamowicz (1996) formulate a heteroscedastic multinomial logit (HMNL) model that allows the variance of alternatives to vary across individuals based on attributes characterizing the individual and her/his environment (the variance, however, does not vary across alternatives). The motivation for such a model is that individuals with the same deterministic utility for an alternative may have different abilities to accurately perceive the overall utility offered by the alternative. The HMNL model has exactly the same structure as the heteroscedastic model described in the second section, though the motivations for their
394
In Perpetual Motion: Travel Behaviour Research and Opportunities
development are different. Swait and Adamowicz apply their model to analyze market structure in a consumer behaviour study and find evidence of varying variance components across individuals. McMillen (1995) also proposes a heteroscedastic model in the context of spatial choice. Both the above studies specify the variance of alternatives to be a deterministic function of individual-related characteristics and do not relax the IIA property if the IID (across alternatives) structure on the random components is maintained. Steckel and Vanhonacker (1988), on the other hand, develop a heteroscedastic logit model that treats the heteroscedasticity across individuals in the variance of alternatives as a random variable. This random variable is assumed to take an exponential distribution, and appears as a parameter in a generalized type I extreme value distribution for the random components of utility. The resulting mixing distribution for the random components of utility provides a closed-form expression for choice probabilities. Steckel and Vanhonacker show that their model is not saddled with the IIA property. Covariance Relaxation Bhat (1997d) develops a nested logit model that allows heterogeneity across individuals in the magnitude of covariance among alternatives in a nest. The heterogeneity is incorporated by specifying the logsum (dissimilarity) parameter(s) in the nested logit model to be a deterministic function of individual-related characteristics. The model is applied to intercity mode choice analysis, where such heterogeneity may be likely to occur. For example, consider a nested model with car and train grouped as surface modes and air treated as a non-nested alternative. The degree of (increased) sensitivity (or cross-elasticity) between the two surface transportation modes relative to the air mode may differ based on characteristics of the traveler such as income (lower income may imply greater sensitivity between the surface modes) and attributes of the traveler's trip such as trip distance (shorter trip distances may lead to greater sensitivity between surface modes). Kamakura et al. (1996) adopt a different approach to accommodating covariance heterogeneity across individuals in their joint product form typebrand choice marketing analysis of peanut butter purchase behaviour (there are two major product forms; creamy and crunchy; and several major brands such as Peter Pan and Skippy). They specify two nesting structures based on whether product form type (brand choice) is at the top (bottom) level of bottom (top) level and then assign individuals to each nesting structures probabilistically. The author is not aware of any study that allows both variance and covariance components to vary across individuals (variance-covariance relaxation), though in concept the extension involves just a combination of the variance and covariance relaxations discussed earlier.
Recent Methodological Advances Relevant to Activity and Travel Behaviour
395
HAZARD DURATION MODELS Hazard-based duration models are ideally suited to modeling duration data. Such models focus on an end-of-duration occurrence (such as end of shopping activity participation) given that the duration has lasted to some specified time (Kiefer, 1988; Hensher and Mannering, 1994). This concept of conditional probability of "failure" or termination of activity duration recognizes the dynamics of duration; that is, it recognizes that the likelihood of ending a shopping activity participation depends on the length of elapsed time since start of the activity. Hazard-based duration models, which had their roots in biometrics and industrial engineering, are being increasingly used to model duration time in the fields of economics, transportation, and marketing (see Kiefer, 1988, Hensher and Mannering, 1994, and Jain and Vilcassim, 1991 for a review of the applications of duration models in economics, transportation, and marketing, respectively). To include an examination of covariates, which affect duration time, most studies use a proportional hazard model which operates on the assumption that covariates act multiplicatively on some underlying or baseline hazard. Two important specification issues in the proportional hazard model are a) the distributional assumptions regarding duration (equivalently, the distributional assumptions regarding the baseline hazard) and b) the assumptions about unobserved heterogeneity {i.e., unobserved differences in duration across people). We discuss each of these issues in the third section of the chapter. The extension of the simple univariate duration model to include multiple duration processes, multiple spells from the same individual, and related issues is also considered in the third section. Baseline Hazard Distribution The distribution of the hazard may be assumed to be one of many parametric forms or may be assumed to be nonparametric. Common parametric forms include the exponential, Weibull, log-logistic, gamma, and log-normal distributions. Different parametric forms imply different assumptions regarding duration dependence. For example, the exponential distribution implies no duration dependence; that is, the fime to "failure" is not related to the time elapsed. The Weibull distribution generalizes the exponential distribution and allows for monotonically increasing or decreasing duration dependence. The form of the duration dependence is based on a parameter that indicates whether there is positive duration dependence (implying that the longer the time has elapsed since start of the duration, the more likely it is to exit the duration soon), negative duration dependence (implying that the longer the time has elapsed since start of the duration, the less likely it is to exit the duration soon), or no duration dependence (which
396
In Perpetual Motion: Travel Behaviour Research and Opportunities
is the exponential case). The log-logistic distribution allows a non-monotonic hazard function. The choice of the distributional form for the hazard function may be made on theoretical grounds. However, a serious problem with the parametric approach is that it inconsistently estimates the baseline hazard and the covariate effects when the assumed parametric form is incorrect (Meyer, 1990). Sometimes, there may be little theoretical support for any particular parametric shape. In such cases, one might consider using a nonparametric baseline hazard. The advantage of using a nonparametric form is that even when a particular parametric form is appropriate, the resulting estimates are consistent and the loss of efficiency (resulting from disregarding information about the hazard's distribution) may not be substantial (Meyer, 1987). Within the nonparametric approach, one may use the partial likelihood framework suggested by Cox (1972) which estimates the covariate effects but not the baseline hazard, or the approach suggested by Han and Hausman (1990) which estimates both the covariate effects and the baseline hazard parameters (also sometimes referred to as the incidental or nuisance parameters) simultaneously (the Han and Hausman approach is an alternative formulation of the approach originally proposed by Prentice and Gloeckler, 1978 and extended by Meyer, 1987). Between the Cox and Han and Hausman (HH) approaches, the HH approach has many advantages. First, in many studies, the dynamics of duration is itself of direct interest; the Cox approach, however, conditions out the nuisance parameters. Second, the Cox approach becomes cumbersome in the presence of many tied failure times (Kalbfleisch and Prentice, 198, page 101). Third, unobservable heterogeneity (which we discuss in the next section) cannot be accommodated within the Cox partial likelihood framework without the presence of multiple integrals of the same order as the number of observations in the risk set at each time period. Estimation in the presence of such large orders of integration is impractical even with recent advances in the computation of multidimensional integrals. In addition, the HH approach is the only appropriate method when duration models are to be estimated from interval-level data arising from the grouping of underlying continuous duration times. The parametric and Cox approaches use density function terms in their respective likelihood functions which are appropriate only for estimation from continuous duration data. If they are used to model grouped (or interval-level) duration data, the resulting estimates would generally be inconsistent (Prentice and Gloeckler, 1978). Most studies of duration to date have made an a priori assumption of a parametric hazard. The most relevant duration studies for activity-travel modeling include a) the homestay duration models for commuters {i.e., the time between coming home from work and leaving home for another out-of-home activity participation) of Mannering et al (1992) and Hamed and Mannering (1993), b) the sex-differentiated shopping duration models of Niemeier and Morita
Recent Methodological Advances Relevant to Activity and Travel Behaviour
397
(1996), c) the shopping activity duration during the evening work-to-home commute of Bhat (1996b), and d) the delay duration model for border crossings by Paselk and Mannering (1992). These studies have been reviewed in greater detail by Pas (1996). Of these studies, Bhat (1996b) uses a nonparametric baseline hazard specification, while others use a parametric baseline hazard specification. Some studies in the marketing literature have used general parametric forms which nest the more frequently used Weibull, exponential and Gompertz distributions. Examples include Jain and Vilcassim (1991) and Vilcassim and Jain (1991).
Unobserved Heterogeneity Unobserved heterogeneity arises when unobserved factors {i.e., those not captured by the covariate effects) influence durations. It is well established now that failure to control for unobserved heterogeneity can produce severe bias in the nature of duration dependence and the estimates of the covariate effects (Heckman and Singer, 1984; Lancaster, 1985; Sharma, 1987). The standard procedure used to control for unobserved heterogeneity is the random effects estimator (see Flinn and Heckman, 1982). This involves specification of a distribution for the unobserved heterogeneity (across individuals) in the population. Two general approaches may be used to specify the distribution of unobserved heterogeneity. One approach is to use a parametric distribution such as a gamma distribution or a normal distribution (most earlier research has used a gamma distribution). The problem with the parametric approach is that there is seldom any justification for choosing a particular distribution; further, the consequence of a choice of an incorrect distribution on the consistency of the model estimates can be severe (see Heckman and Singer, 1984). A second approach to specifying the distribution of unobserved heterogeneity is to use a nonparametric representation for the distribution and to estimate the distribution empirically from the data. This is achieved by approximating the underlying unknown heterogeneity distribution by a finite number of support points and estimating the location and associated probability masses of these support points. The nonparametric approach enables consistent estimation since it does not impose a prior probability distribution. Application of duration models in the transportation field have, for the most part, ignored unobserved heterogeneity (but see Bhat, 1996b and Hensher, 1994b). Researchers in the marketing and economics fields have paid more attention to unobserved heterogeneity. However, even in these fields, most applications have employed a parametric heterogeneity specification (see Gupta, 1991, Manston et al, 1986, Meyer, 1990, Han and Hausman, 1990,
398
In Perpetual Motion: Travel Behaviour Research and Opportunities
all of whom use a gamma distribution). Very few studies have adopted a nonparametric heterogeneity distribution (see Jain and Vilcassim, 1991 and Vilcassim and Jain, 1991). Among the duration studies mentioned above, Bhat (1996b) uses a nonparametric baseline hazard (based on the Han and Hausman approach) and a nonparametric unobserved heterogeneity specification (based on the Heckman and Singer approach). By allowing a nonparametric distribution for both the baseline hazard and unobserved heterogeneity, this chapter sheds light on the importance of allowing a nonparametric specification for the baseline hazard, for unobserved heterogeneity, and for both of these. The finding from the study indicates that, at least in the context of the empirical analysis of the paper, the nonparametric baseline-nonparametric unobserved heterogeneity specification is preferable to other parametric specifications for the baseline or for heterogeneity or both. This result is important. It is contrary to the commonly held view that the choice of the mixing distribution may not be important if the baseline hazard is nonparametrically specified (see Meyer, 1990; Han and Hausman, 1990; Manston er al, 1986).
Multiple Duration Processes The discussion thus far has focused on the case where durations end as a result of a single event. For example, the length of unemployment ends when an individual gains employment (see Meyer, 1990) or home stay duration ends when an individual leaves home to participate in an activity (Mannering et al. 1992). A limited number of studies have been directed toward modeling the more interesting and realistic situation of multiple duration-ending outcomes. For example, failure in the context of unemployment duration {i.e., exit from the unemployment spell) can occur either because of a new job, recall to the old job, or withdrawal from the labor force. Similarly, home stay duration may be terminated because of participation in out-of-home shopping activity, social activity, or personal business. Previous research on multiple duration-ending outcomes {i.e., competing risks) has extended the univariate proportional hazard model to the case of two competing risks in one of three ways. The first method assumes independence between the two risks (see Katz, 1986 and Gilbert, 1992). Under such an assumption, estimation proceeds by estimating a separate univariate hazard model for each risk. Unfortunately, the assumption of independence is untenable in most situations and, at the least, should be tested. The second method generates a dependence between the two risks by specifying a bivariate parametric distribution for the underlying durations directly. For example. Diamond and Hausman (1985) specify a log bivariate-normal distribution for the durations. This method has the result of placing very
Recent Methodological Advances Relevant to Activity and Travel Behaviour
399
strong (and non-testable) parametric restrictions on the form of the baseline cause-specific hazard functions. The third method accommodates interdependence between the competing risks by allowing the unobserved components affecting the underlying durations to be correlated. Cox and Oakes (1984, page 159-161) develop a model which generates a positive dependence between the underlying durations based on common dependence on an observed random variable. More recently, Han and Hausman (1991) propose a model which allows unrestricted correlation in random unobserved components affecting the competing risks. This model permits nonparametric baseline hazard estimation, enables estimation from intervallevel data of the type commonly found in econometrics and other fields, and retains an interpretation as an incompletely observed continuous-time hazard model. A shortcoming of all the competing risk methods discussed above is that they tie the exit state of duration very tightly with the length of duration. The exit state of duration is not explicitly modeled in these methods; it is characterized implicitly by the minimum competing duration spell. Such a specification is restrictive, since it assumes that the exit state of duration is unaffected by variables other than those influencing the duration spells and implicitly determines the effects of exogenous variables on exit state status from the coefficients in the duration hazard models (this situation is analogous to the difference between a general endogenous switching regression equation system and the more restrictive disequilibrium market model of demand and supply; see Maddala, 1983, page 308). Bhat (1996c) considers a generalization of the Han and Hausman competing risk specification where the exit state is modeled explicitly and jointly with duration models for each potential exit state. The resulting formulation follows strictly from the proportional hazard specification for the duration spells. This is in contrast to the Han and Hausman specification which uses an approximation to the proportional hazard specification. The model also extends the Han and Hausman framework to multivariate competing risks.^ Bhat's formulation does not require placing parametric restrictions on the shapes of hazards within discrete time intervals, as required in the specifications of Han and Hausman, 1991 and Sueyoshi, 1992 (Han and Hausman and Sueyoshi maintain an assumption of a constant hazard within each discrete timeinterval in deriving the competing-risk model specification). Another desirable characteristic of the model is that it is a generalized multiple durations model where the durations can be characterized either by multiple entrance states or by multiple exit states or by a combination of entrance and exit states. The focus of econometric literature has been on multiple durations due to multiple exit states {i.e., the competing risk model). However, in many applications, multiple durations may arise because of multiple entrance states. Examples of multiple entrance states include layoffs, being fired, or first-time labor force entry for unemployment duration, activity-type participation choice (shopping, recreation, visiting, etc.) for activity
400
In Perpetual Motion: Travel Behaviour Research and Opportunities
duration, and type of initial acquaintance (in college, though personal advertisement, etc.) for marriage durations. Ignoring the entrance state when there are common unobserved factors affecting entrance status and spell duration will lead to biased and inconsistent hazard model parameters due to classic sample selection problems. In this context, information on the absence of a duration spell itself may be valuable; that is, it may be important to consider the "no-entry" state (for example, the "employed" state in unemployment duration modeling, the "home" state in activity duration modeling, or the "unmarried" state in marriage duration modeling) as an explicit entrance state in modeling durations for other entrance states. Most multiple-duration hazard formulations do not accommodate unobserved heterogeneity because it makes the estimation difficult. However, with the computing capabilities available today, this should not be an excuse for ignoring unobserved heterogeneity. There has also been only limited work in accommodating dependence in the effect of unobserved variables across multiple spells from the same individual. Hensher (1994b) and Mealli and Pudney (1996) have formulated and estimated a competing risks model that captures both unobserved heterogeneity specific to each spell as well as unobserved "fixed" dependence across multiple spells from the same individual. These papers also serve as exhaustive reviews of recent competing risk formulations.^ Other issues in the context of hazard models not discussed here include incorporating the time-invariant effect of time varying covariates or allowing for time-varying effects of time-invariant covariates. For recent work in these areas, the reader is referred to Hensher (1994b), McCall (1994) and Wedel et al. (1995).
LIMITED-DEPENDENT VARIABLE MODELS Limited-dependent variable models encompass a wide variety of structures which may be classified in one of two broad categories. The first category recognizes the discontinuous nature of a variable (such as the ordinal nature of number of activity stops or several zero values for out-of-home activity duration because of non-participation in out-of-home activity). The second category accommodates the interdependence between a discrete choice variable and another related continuous or grouped or ordinal variable (such as the interdependence between mode choice to work and the number of activity stops during the evening commute). In general, the structures for the second category subsume those of the first category as special cases. Thus, modeling out-of-home activity participation (a discrete choice) and out-of-home activity duration (a continuous choice) using a discrete/continuous framework is more general than considering duration as being a discontinuous variable with bunching of values at zero for individuals who do not participate in out-of-home activity. In this chapter, we will focus on the more general second category of inter-related discrete and non-discrete variable systems. The
Recent Methodological Advances Relevant to Activity and Travel Behaviour
401
non-discrete variable can take several forms. However, the three most interesting cases in the context of travel and activity modeling are the continuous, ordinal, and grouped forms. Further, the structure for the discrete/ordinal and discrete/grouped variable systems are very similar; so we will examine limited-dependent variable systems under two headings: discrete/continuous and discrete/ordinal models.
Discrete/Continuous Models The methods developed for, and applications of, discrete/continuous choices can be broadly classified under two categories based on the number of alternatives involved in the discrete choice decision. The first category is the dichotomous alternative case and the second is the polychotomous alternative case. By far, most of the attention to date has focused on the dichotomous case (see Heckman, 1976 and Lee, 1981 for estimation methods and Willis and Rosen, 1980 and Sakamoto and Chen, 1991 for applications; Amemiya, 1985 and Maddala, 1983 provide a review of economic applications, while Winship and Mare, 1992 provide a review of sociological applications). In contrast to the dichotomous case, the polychotomous case has received much lesser attention (see Hay 1980, Dubin and McFadden 1983, Hanemann 1984, and Lee 1983 for estimation methods and Hensher et al 1992, Bhat 1996, Barnard and Hensher 1992, Hamed and Mannering 1993, and Mannering and Winston 1985 for applications; Mannering and Hensher, 1987 provide a review of transportation-related applications). Almost all applications of the dichotomous case have used either a logit or probit approach to model the discrete choice in the discrete/continuous model. Applications of the polychotomous case have generally used a multinomial logit-based approach to model the discrete choice due to the resulting simplicity in structure. Hamed and Mannering (1993) use the discrete/continuous model framework to model activity type choice, travel time duration to the activity, and activity duration. They use a limitedinformation (two-stage) maximum likelihood method in their estimation where all variables specific to (or determined by) the activity type in which the traveler participated and appearing in the continuous travel time/activity duration equations are replaced by their expected values as obtained from the discrete activity type choice model. A similar limited-information procedure is used by Damm (1981) in his study of activity participation choice and activity duration, as well as by Hensher et al (1992), Train (1986), and Mannering and Winston (1985) in their automobile brand type/automobile use models. Barnard and Hensher (1992) estimate a discrete/continuous model of shopping destination choice and retail expenditure. They use Lee's (1983) transformation method for polychotomous
402
In Perpetual Motion: Travel Behaviour Research and Opportunities
choice situations with non-normal error distributions in the choice model. Bhat (1996d) has also used Lee's method for discrete/continuous models, but extends the method to jointly estimate a polychotomous discrete choice and two continuous choices (rather than a single continuous choice). Lee's method has two advantages over the other polychotomous (twostage) methods discussed earlier. First, Lee's method enables full-information maximumlikelihood estimation, while the other methods are two-stage methods in which the discrete choice model is estimated first and then the continuous choice model is estimated using one of several methods to account for selectivity bias (see Dubin 1985, page 158). Thus Lee's method facilitates asymptotically more efficient estimates in the discrete/continuous choice model. Second, the expressions for the asymptotic covariance matrices of the two-stage estimates are very complicated, while the asymptotic covariance matrix in Lee's method can be obtained directly from the maximum likelihood estimation. Lee's method is also very flexible and can accommodate any model formulation for the discrete choice decision with little change in the methodology.
Discrete/Ordinal Models There has been relatively little empirical work in the area of joint discrete/ordinal variable systems compared to the joint discrete/continuous systems reviewed in the previous section. As in the case of discrete/continuous systems, discrete/ordinal models can also be classified under the dichotomous and polychotomous categories based on the number of alternatives involved in the discrete choice decision. The next two paragraphs review studies in each of the two categories. Bhat and Koppelman (1993) estimate a joint model of employment status (represented by a binary flag indicating whether or not an individual is employed) and annual income earnings. Observed income earnings in their data is in grouped form {i.e., observed only in grouped categories such as < 20,000, 20,000-39,999, 40,000-59,999, etc.). Since it is likely that people who are employed are also likely to be the people who can earn higher incomes, the two variables are modeled jointly. Bhat (1996) subsequently has used a similar structure to impute a continuous income value from missing and grouped income observations in a data set. The motivation for this work is two fold. First, while income is measured in grouped form, it is the continuous measure of income that frequently appears as an explanatory variable in labor supply models, market research models and travel demand models (Killingsworth 1983, Koppelman et al 1993, Golob 1989). Second, there might be systematic differences in unobserved characteristics affecting income between respondent and non-respondent households (or individuals). For example, it seems at least possible that households with
Recent Methodological Advances Relevant to Activity and Travel Behaviour
403
above-average income, other things being equal, will be more reluctant than other households to provide information on income (see Lilliard et al., 1986). Bhat (1997e) has recently developed a joint model of polychotomous work mode choice and number of non-work activity stops during the work commute (i.e., the total number of nonwork stops made during the morning home-to-work commute and evening work-to-home commute). The joint model provides an improved basis to evaluate the effect on peak-period traffic congestion of conventional policy measures such as ridesharing improvements and soloauto use dis-incentives. Traditional mode choice models address the question "What is the effect of a change in, say, solo-auto in-vehicle travel time (for example, due to conversion of an existing general-purpose lane to a high-occupancy lane) on work mode choice?". If commute trips were the sole contributors to peak period congestion, then the shifts in work mode choice provide a direct indication of the potential impact on congestion. A more pertinent question to address today, however, is "What is the effect of a change in, say again, solo-auto in-vehicle time on work mode choice and number of non-work stops?" This question is prompted by the recognition that vehicle trips due to non-work stops also add to peak period congestion. Thus, understanding the effect of a policy action on work mode choice and number of non-work stops allows us to evaluate the effect on peak-period congestion through the impact on both direct commute vehicle-trips and additional vehicle-trips due to non-work stops.
SUMMARY AND CONCLUSIONS This chapter has reviewed methodological developments in the econometric field of direct relevance to activity and travel behaviour modeling. Clearly, there has been substantial progress in the development and practical applicability of the methodologies in the recent past. This progress can be traced to at least four factors: a) The need for realistic representations of the behavioural decision processes underlying activity-travel decisions, b) The ability to provide micro-level demographic inputs required by activity-travel models, c) Better tools for data storage and processing, and d) The advent of simulation techniques to approximate multidimensional integrals.
Need for Realistic Representation of Behavioural Decision Processes The travel demand models used widely today were developed in the late sixties and have seen little change over the years. These models were developed primarily to evaluate alternative
404
In Perpetual Motion: Travel Behaviour Research and Opportunities
major capital improvements. While this continues to remain an important objective of travel demand models, there is a shift in emphasis from evaluating the long-term investment-based strategies to understanding travel behaviour responses to shorter term congestion management policies such as alternate work schedules, telecommuting, and congestion-pricing. The traditional travel demand models are not suited to such a task because, due to their many simplifying assumptions and narrow "individual-trip" perspective, they are unable to examine the potentially complex behavioural responses to demand management actions (Spear, 1994). For example, a change in work schedule to an early arrival home may lead to increased tripmaking at the evening because of the additional time available to participate in out-of-home activities. If some of this travel is undertaken during the same time as the PM peak-period travel, the extent of congestion alleviation projected by traditional models will not be realized (see Jones et al, 1990). Similarly displacements of travel (and its associated consequences) to other times of day due to a change in activity patterns caused by adoption of work telecommuting strategies cannot be examined by traditional models (see Mokhtarian, 1993). Also, traditional models do not incorporate adequate richness in the substitution pattern among alternatives or the different sensitivities of individuals to changes in the transportation system. This can lead to inappropriate evaluations of travel demand management policies (Stopher, 1993). Finally, from a transportation and regional planning perspective, reasonably accurate forecasts of travel demand are needed to be better prepared for the future and to endeavor to avoid serious conflicts between transportation supply and demand. Inasmuch as the travel needs of the population is changing rapidly (due to changes in lifestyle, changes in activity needs of particular subgroups such as the elderly, changes in household structure and social environment, changes in urban structure, changes in technology, etc.), it is obvious that models with a sound behavioural casual linkage between travel patterns and the travel environment will be critical to good design and planning of future transportation infrastructure.
Ability to Provide Micro-Level Inputs for Activity-Travel Models The need for realistic representations of activity and travel decisions requires modeling of these choices at the individual (or household) level. Once the individual-level models are estimated from a sample, they can be used to examine the impact of various policies (in the short-term) or forecast activity-travel patterns (in the long term). In either case, detailed disaggregate-level inputs of the characteristics of the decision-making entities and other attributes (such as options available and constraints encountered) of the choice context are required. Oftentimes, such information is not readily available. For example, consider a destination choice model which has been estimated on a sample and is to be applied to study the policy impact of imposing congestion-pricing on selected spatial corridors. The destination
Recent Methodological Advances Relevant to Activity and Travel Behaviour
405
choice model might include household and individual level characteristics as exogenous determinants (for example, older individuals might prefer destinations which are close by or higher income earning individuals may be willing to travel greater distances, etc.). It is quite possible, however, that we will not have observations of individuals in the sample making trips between certain zonal pairs. In such a case, we cannot study the impact of the congestionpricing policy on trip-making between those zonal pairs. Similarly, in a forecasting context, there will be changes in the characteristics of the population and the currently available sample may become unrepresentative of the future population. In both the cases mentioned above, there is a need for a mechanism to generate the appropriate disaggregate-level inputs. This issue has been at the core of the debate on the practical usefulness of disaggregate-level models. Miller (1996) summarizes it well as follows "I believe a strong case can be made that a primary reason for the relatively slow diffusion of disaggregate modeling methods into travel demand forecasting practice is due to the difficulty practitioners have in generating the disaggregate forecast inputs required by these methods". However, with the development of micro-simulation techniques to generate the required disaggregate-level inputs either through updating of the current sample over time or by "synthesizing" a representative sample from other supplementary aggregate-level information such as census data, it is now possible to apply models which are more realistic in their representations of behaviour to policy analysis and forecasting. The reader is referred to the comprehensive review by Miller (1996), and the chapter in this volume (Miller and Salvini, 2001) on techniques and research issues associated with micro-simulation.
Better Tools for Data Storage and Processing The tools available for data storage and processing have seen dramatic improvement over the past few years. Desktop and even notebook computers are able to store data of large sizes and are remarkably fast in the retrieval and processing of such data. This has made possible the estimation of models deemed earlier to be impractical. The improved computer processing capabilities has also spurred the development of new and behaviourally rich model formulations. Another area that has developed quite considerably is Geographic Information Systems (GIS) technology. Fotheringham and Rogerson (1993) discuss the potential of integrating travel analysis methods with GIS technology. A specific application of GIS technology to activity-travel analysis is the development of a measure of spatial accessibility for use in the modeling of multistop and multi-purpose travel (see Arentze et aL, 1994a,b,c; Lee, 1996). Golledge et al. (1994) and Kwan (1994) have used GIS to calibrate a production system model of activity scheduling behaviour. Caliper Corporation's TransCAD GIS software represents an important bridge in linking GIS developments with travel demand modeling
406
In Perpetual Motion: Travel Behaviour Research and Opportunities
practice. Specifically, TransCAD attempts to package advanced econometric modeling techniques within an interface that is user-friendly, enables spatial representation of the transportation network and geographic database management, and allows an intuitive spatial display of the results from the travel demand models.
Advent of Simulation Techniques to Approximate Multi-Dimensional Integrals Recent advances in the field of Monte Carlo simulation methods to evaluate multi-dimensional integrals have contributed considerably to the feasibility in estimating complex discrete-choice and other limited-dependent variable models. Two types of simulators that are of particular interest in the activity-travel area are the probit-based and the logit-based simulators. The former is suitable for discrete-choice structures that use a normal distribution for the random components and the latter is appropriate for various extensions of the multinomial logit structure (see Chib and Greenberg, 1996, Hajivassiliou et al., 1996, and Brownstone and Train, 1996 for reviews of such simulation techniques). The underlying concept in such methods is to approximate the integration by computing the integrand at various values drawn from the appropriate multi-variate distribution of the variable vector over which the integration is being carried out and taking the mean across the computed integrand values. Several issues arise during the actual implementation of the approach, which we do not discuss here. The application of probit-simulators in the travel behaviour field can be found in the work of Mahmassani and his colleagues who have used the multinomial probit structure to examine the day-to-day dynamics in departure time and route choice of commuters (see Lam and Mahmassani, 1991; Mahmassani and Jou, 1996; Mahmassani, 1997). Logit-based simulators have been used in the travel demand field by Brownstone and Train (1996), Bhat (1996a, 1997b), and Ben-Akiva and Bolduc (1996). The formulation and estimation of behaviourally rich models has been greatly facilitated by the developments discussed above. However, the fields of micro-simulation. Geographic Information Systems, and simulation of integrals are continually evolving and by all accounts there is still considerable progress to be made. As these fields develop, and as practitioners and researchers in the activity-travel behaviour field become familiar with them, there is bound to be more empirical applications using these tools. This, along with the need for improved policy analysis and accurate demand forecasting, should contribute further toward the implementation of improved methodologies in the area of activity and travel behaviour research.
Recent Methodological Advances Relevant to Activity and Travel Behaviour
407
ACKNOWLEDGEMENTS This research was supported by National Science Foundation grants DMS 9208758 and DMS 9313013 to the National Institute of Statistical Sciences (NISS). Some of the material in this chapter was developed as part of a presentation at a travel demand analysis workshop sponsored by NISS in April 1997. The discussions with Prof. Eric Pas, Prof. Frank Koppelman, and Prof. Ryuichi Kitamura at this workshop are greatly appreciated. The author would also like to thank Prof. Hani Mahmassani for the invitation to prepare this resource chapter.
REFERENCES Allenby, G. M. and J. L. Ginter (1995). The effects of in-store displays and feature advertising on consideration sets, InternationalJournal of Research in Marketing, 12, 67-80. Amemiya, T. (1985). Advanced Econometrics, Harvard University Press, Cambridge. Arentze, T. A., A. W .J. Borgers and H. J. P. Timmermans (1994a). Geographical information systems and the measurement of accessibility in the context of multipurpose travel: a new approach. Geographical Systems, 1, 87-102. Arentze, T. A., A. W. J. Borgers and H. J. P. Timmermans (1994b). Geographic information systems, spatial opportunities, and multi-stop travel: a new measurement approach, International Journal of Geographical Information Systems, 8, 2. Arentze, T. A., A. W. J. Borgers and H. J. P. Timmermans (1994c). Multistop-based measurements of accessibility in a GIS environment, International Journal of Geographic Information Systems, 8, 343-356. Ben-Akiva, M. and S. R. Lerman (1985). Discrete Choice Analysis: Theory and Application to Travel Demand, The MIT Press, Cambridge. 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 Economics, Massachusetts Institute of Technology, Cambridge. Ben-Akiva, M., D. Bolduc and M. Bradley (1993). Estimation of travel choice models with randomly distributed values of time, Transportation Research Record, 1413, 88-97. Barnard and Hensher (1992). The spatial distribution of retail expenditures. Journal of Transport Economics and Policy, September, 299-311. Bhat, C. R. and F. S. Koppelman (1993b). An endogenous switching simultaneous equation system of employment, income, and car Ownership, Transportation Research, 21A, 6, 447-459. Bhat, C. R. (1994). Imputing a continuous income variable from grouped and missing income observations. Economics Letters, 46, 311-319. Bhat, C. R. (1995). A heteroscedastic extreme-value model of intercity mode choice. Transportation Research, 29B, 6, 471-483. Bhat, C. R. (1996a). Accommodating variations in responsiveness to level-of-service measures in travel mode choice modeling, working paper. Department of Civil and Environmental Engineering, University of Massachusetts, Amherst.
408
In Perpetual Motion: Travel Behaviour Research and Opportunities
Bhat, C. R. (1996b). A hazard-based duration model of shopping activity with nonparametric baseline specification and nonparametric control for unobserved heterogeneity, Transportation Research, SOB, 3, 189-207. Bhat, C. R. (1996c). A generalized multiple durations proportional hazard model with an application to activity participation behavior during the work commute, Transportation Research, 30B, 432-452. Bhat, C. R. (1996d). A model of post-home arrival activity participation behavior, working paper. Department of Civil and Environmental Engineering, University of Massachusetts, Amherst. Bhat, C. R. (1997a). Analysis of travel mode and departure time choice for urban shopping trips, Technical Paper, Department of Civil and Environmental Engineering, University of Massachusetts, Amherst. Bhat, C. R. (1997b). Accommodating flexible substitution patterns in multi-dimensional choice modeling, Technical Paper, Department of Civil and Environmental Engineering, University of Massachusetts, Amherst. Bhat, C. R. (1997c). An endogenous segmentation mode choice model with an application to intercity travel. Transportation Science, 31, 34-48. Bhat, C. R. (1997d). Covariance heterogeneity in nested logit models: econometric structure and application to intercity travel. Transportation Research, 3IB, 11-21. Bhat, C. R. (1997e). Work travel mode choice and number of nonwork commute stops. Transportation Research, 31B, 41 -54 Brownstone, D, (1996). Multiple imputation methodology for missing data, non-random response, and panel attrition, paper presented at the Theoretical Foundations of Travel Choice Modeling Conference held in Stockholm, August 7-11. Brownstone, D. and K. Train (1996). Forecasting new product penetration with flexible substitution patterns, working paper, Department of Economics, University of California, Irvine. Brownstone, D. and K.A. Small (1989). Efficient estimation of nested logit models. Journal of Economics and Business Statistics, 7, 67-74. Bunch, D. S. and R. Kitamura (1990). Multinomial probit model estimation revisited: testing estimable model specifications, maximum likelihood algorithms, and probit integral approximations for trinomial models of household car ownership, working paper. Graduate School of Business, University of California, Davis. Chamberlain, G. (1980). Analysis of covariance with qualitative data, Review of Economic Studies, 47, 225-238. Chib, S. and E. Greenberg (1996). Markov chain monte carlo simulation methods in econometrics. Econometric Theory, 12, 409-431. Chintagunta, P. K. and B. E. Honore (1996). Investigating the effects of marketing variables and unobserved heterogeneity in a multinomial probit model. International Journal of Research in Marketing, 13, 1-15. Chu, C. (1989). A paired combinatorial logit model for travel demand analysis. Proceedings of the Fifth World Conference on Transportation Research, 295-309, Ventura, CA. Cox, D. R. and D. Oakes (1984). Analysis of Survival Data, Chapman and Hall, London. Currim, I. S. (1982). Predictive testing of consumer choice models not subject to independence of irrelevant alternatives. Journal of Marketing Research, 19, 208-222. Damm, D. (1980). Interdependencies in activity behavior. Transportation Research Record, 750, 33-40.
Recent Methodological Advances Relevant to Activity and Travel Behaviour
409
Daganzo, C. (1979). Multinomial Probit: The Theory and its Application to Demand Forecasting, Academic Press, New York. Daganzo, C. F. and M. Kusnic (1993). Two properties of the nested logit model, Transportation Science, 27, 395-400. Daly, A. J. and Zachary, S. (1978). Improved multiple choice models, in D.A. Hensher and M.Q. Dalvi (eds.) Determinants of Travel Choice, Saxon House, Westmead. Dayton, M. C. and G. B. Macready (1988). Concomitant-variable latent class models. Journal of the American Statistical Society, 83, 173-179. Diamond, P. and J. Hausman (1984). The retirement and unemployment behavior of older men, in H. Aaron and G. Burtless (eds.). Retirement and Economic Behavior, Brookings Institute, Washington, D.C. Dubin, J. A. and D. McFadden (1984). An econometric analysis of residential electric appliance holdings and consumption, Econometrica, 52,2, 345-362. Ettema, D., A. W. J. Borgers and H. J. P. Timmermans (1995). A competing risk hazard-model of activity choice, timing, sequencing and duration, paper presented at the Transportation Research Board Meeting, Washington, January. Evers, G. H. M. (1990). The residential location and workplace choice: a nested multinomial logit model, in Spatial Choices and Processes, editors: M.M. Fischer, P. Nijkamp and Y.Y. Papageorgiou, Elsevier Science Publishers, B.V. (North-Holland), 313-329. Fischer, G. W. and D. Nagin (1981). Random versus fixed coefficient quantal choice models, in Manski, C.F. and D. McFadden (eds.) Structural Analysis of Discrete Data with Econometric Applications, MIT Press, Cambridge, Massachusetts, 273-304. Flinn, C. and J. Heckman (1982). New methods for analyzing structural models of labor force dynamics. Journal of Econometrics, 18, 115-168. Fotheringham, A. S. and P. A. Rogerson (1993). GIS and spatial analytical problems. International Journal of Geographical Information Systems, 7, 3-19. Garling, T., M. P. Kwan and R. G. GoUedge (1994). Computational process modeling of household travel Activity scheduling. Transportation Research, 25B, 355-364. Gilbert, C. C. M. (1992). A duration model of automobile ownership. Transportation Research, 26B, 2, 97-114. Golledge, R. G., M. P. Kwan and T. Garling (1994). Computational process modeling of travel decisions using geographical information systems. Papers in Regional Science, 73-99. Golob, T. F. (1989). The dynamics of household travel time expenditures and car ownership decisions, presented at the International Conference on Dynamic Travel Behavior Analysis, Kyoto, Japan, July. Goniil, F. and K. Srinivasan (1993). Modeling multiple sources of heterogeneity in multinomial logit models: methodological and managerial issues. Marketing Science, 12,3,213-229. Gopinath, D. and M. Ben-Akiva (1995). Estimation of randomly distributed value of time, working paper. Department of Civil Engineering, Massachusetts Institute of Technology. Gupta, S. and P. K. Chintagunta (1994). On using demographic variables to determine segment membership in logit mixture models. Journal of Marketing Research, 31, 128-136. Hajivassiliou, V. A., D. McFadden and P. Ruud (1996). Simulation of multivariate normal rectangle probabilities and their derivatives: theoretical and computational results. Journal of Econometrics, 72, 85-134. Hamed, M. M and F. L. Mannering (1993). Modeling travelers' postwork activity involvement: toward a new methodology. Transportation Science, 27, 4, 381-394.
410
In Perpetual Motion: Travel Behaviour Research and Opportunities
Hanemann, W. M. (1984). Discrete/continuous models of consumer demand, Econometrica, 52,541-561. Hay, J. (1980). Selectivity bias in a simultaneous logit-OLS model. Manuscript, University of Southern California. Heckman, J. J. (1976). The common structure of statistical models of truncation, sample selection and limited dependent variables and a simple estimator for such models. Annals of Economic and Social Measurement, 5/4, 475-492. Heckman, J. and B. Singer (1984). A method for minimizing the distributional assumptions in econometric models for duration data, Econometrica, 52, 271-320. Hensher, D. A. and F. L. Mannering (1994). Hazard-based duration models and their application to transport analysis. Transport Reviews, 14, 1, 63-82. Hensher, D. A. (1994a). Stated preference analysis of travel choices: the state of the practice. Transportation, 2\, 107-133. Hensher, D. A. (1994b). The timing of change for automobile transactions: a competing risk multispell specification. Presented at the Seventh International Conference on Travel Behavior, Chile, June. Hensher, D. A. (1996a). Establishing a fare elasticity regime for urban passenger transport: non-concession commuters, working paper. Institute of Transport Studies, The University of Sydney. Hensher, D. A. (1996b). Extending valuation to controlled value functions and non-uniform scaling with generalized unobserved variances, presented at the Theoretical Foundations of Travel Choice Modeling Conference held in Stockholm, August 7-11. Hensher, D. A., J. Louviere, and J. Swait (1996). Combining sources of preference data: the case of the lurking scale parameters, working paper, Institute of Transport Studies, The University of Sydney. Hensher, D. A., N. C. Smith, F. M. Milthorpe and P. O. Barnard (1992). Dimensions of Automobile Demand: A Longitudinal Study of Automobile Ownership and Use, Amsterdam, North Holland. Horowitz, J. L. (1981). Identification and diagnosis of specification errors in the multinomial logit model. Transportation Research, 15B, 5, 345-360. Horowitz, J. L. (1991). Reconsidering the multinomial probit model. Transportation Research, 25B, 6, 433-438. Jain, D. C. and N. J. Vilcassim (1991). Investigating household purchase timing decisions: a conditional hazard function approach, MarA:er/>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.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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 ^and /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 ^and 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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
^U
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,
^U 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
^U
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 ^U ^ ^U
(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)
The Goods/Activities Framework for Discrete Travel Choices
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
^u
Iz-t.
du
du
+ a Ci - W
^i
^
X ^
{l + a w)\ = 0
,
6> > 0
423
(16)
.
(17)
Equations (16) and (17) yield generic solutions for ^and 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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
= - 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 ^is 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
The Goods/Activities Framework for Discrete Travel Choices
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
dt.
dt.
^
^^
dL
dt,
^
'
' ^
On the other hand, after a few manipulations,
ac-
oG
oc.
Out of the many possible combinations on ^and /^ 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 ^and 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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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.
The Goods/Activities Framework for Discrete Travel Choices
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.
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Integration of Choice and Latent Variable Models
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)
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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)
Integration of Choice and Latent Variable Models
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
^I^-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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Integration of Choice and Latent Variable Models
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[W^A+(^ri+^..^i)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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
*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.
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Integration of Choice and Latent Variable Models
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^<0
{1 equation}
(1X1)
Note that the covariances of the error terms in the latent variable structural and measurement model are constrained to be equal to zero (denoted by the "IdiagonaF notation). Likelihoodfunction. f{y,I\X;a,p,Y,^)
= j ( . ^{yiXp.+r *
1
p,)}
I.-Z'a,
^Li^lz'=1 <^n,
Integration of Choice and Latent Variable Models Results,
453
The parameters to be estimated include: P (9 parameters estimated), a (8
parameters estimated, 2 parameters constrained to one for identification, 2 parameters constrained to zero based on exploratory factor analysis), y (8 parameters estimated), and the standard deviations cr^(6 parameters) and cr^(2 parameters), where the covariances of the latent variable equations are restricted to zero. Unless otherwise noted, parameters set to zero were done based on statistical tests and a priori hypothesis about the behaviour. All parameters except the variances are reported. The results are shown in Table 2. Estimation was done via sequential numerical integration: first the latent variable model was estimated, then the choice model (including integration over the latent variable) was estimated. The dataset included 219 observations. The top panel displays the estimation results of two different choice models: the second column is the choice model without the latent variables, and the first column is the choice model with the latent variables. The integrated choice and latent variable model consists of the choice model with latent variables (the first column of the upper panel) and the latent variable model (displayed in the lower panel of Table 2). The table for the latent variable model displays the estimation results of both the structural and measurement equations for each of the two latent variables comfort (the first column) and convenience (the second column). The latent variable model is made up of many equations: one structural equation for comfort, one structural equation for convenience, and six measurement equations for comfort and convenience. Both of the latent attributes have significant parameter estimates. Inclusion of the latent attributes identified by the linear structural equation resulted in a large improvement in the goodness-of-fit of the discrete choice model. The rho-bar-squared for the model with latent attributes uses a degree-of-freedom correction involving two variables beyond those used in the model without latent variables, and thus this degree of freedom adjustment only accounts for the estimated parameters of the choice model. While the indicators used for comfort and convenience in this case study are adequate, the structural equations are not particularly strong because the available explanatory variables of comfort and convenience were limited. In general, it can be difficult to find causes for the latent variables. This issue needs to be thoroughly addressed in the data collection phase.
454
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
Case Study 2: Employees' Adoption of Telecommuting The second case study (Bernardino, 1996) assesses the potential for the adoption of telecommuting by employees. Figure 9 presents the modeling framework. The behavioural hypothesis is that an employee faced with a telecommuting arrangement will assess the impact of the arrangement on lifestyle, work-related costs and income, and then decide whether to adopt telecommuting. Telecommuting is expected to influence lifestyle quality by providing the employee with the benefit of increased flexibility to adjust work schedule, work load, personal needs, and commuting patterns. The perceived impact is expected to vary according to the characteristics of the individual and of the program. Telecommuting is also expected to impact household expenditures, such as utilities, equipment, day care, and transportation. Figure 10 provides a full path diagram of the model, noting the relationships between each variable. The employee's decision to adopt a telecommuting program in a simulated choice experiment is modeled as a function of her/his motivations and constraints, as well as the impacts of the available program on lifestyle quality, work-related costs, and income. Changes in income are included in the telecommuting scenarios, while latent constructs of benefit (i.e. enhancement to lifestyle quality) and cost are estimated. To obtain indicators for benefit, respondents are asked to rate the potential benefits of the telecommuting program on a scale from 1 to 9 as shown in Table 3. These responses provide indicators for the latent variable model. The latent cost variable is manifested by the employees' responses to questions about the expected change in home office costs, child and elder care costs, and overall work-related costs as shown in Table 4. The employee is assumed to have a utility maximization behaviour, and thus will choose to adopt a particular telecommuting option if the expected change in utility is positive. This decision is influenced by the characteristics of the arrangement, the individual's characteristics and situational constraints, and the perceived benefits and costs of the arrangement.
Integration of Choice and Latent Variable Models Table 2 Estimation Results of Mode Choice Model with Latent Attributes CHOICE MODEL
XIO X9 X3 X6 X5 X8 X7 Zl* Z2*
ExDlanatorv Variables Rail constant Cost per person Line-haul time Terminal time Number of transfers Business trip dummy Female dummy Ride comfort (latent) Conyenience (latent) Rho-bar-Sauared
WITH Latent Attributes Est. R t-stat 0.32 LOO -4.10 -0.03 0.20 0.08 -2.60 -1.18 -0.32 -1.70 L33 3.60 2.60 0.65 2.70 0.88 L39 4.10 0.352
WITHOUT Latent Attributes Est. ft t-Stat 2.00 0.58 -0,03 -4.20 -0.41 -1.60 -4.20 -L57 -L30 -0.20 0.94 3.60 2.30 0.47
0.242
LATENT VALUABLE MODEL
X2 XI X3 X6 X5 X4 Xll
77 12 15 16 13
M^
Structural Model (2 eauations total. 1 per column) Age >40 First class rail rider Line haul trayel time (rail-auto) Terminal time (rail-auto) Number of transfers by rail Availability of free parking for auto (Age >40) * (Line haul travel time)
Comfort Z7* t-stat Est. vl -1.40 -0.23 1.00 0.29 -1.30 -0.29
Measurement Model (6 eauations total, one per row) Relaxation during trip Reliability of the arrival time Flexibility of choosing departure time Ease of traveling with children/baggage Safety during the trip Overall rating of the mode
Comfort Z7* Est. a I t-stat 1.00 1.80 0.77
Convenience Z2* Est. v9 t-stat 0.41 3.30
-0.52 -0.05 0.16 -0.04
0.69 1.64
-2.10 -0.60 1.60
-0.10
3.10 _2M
Convenience Z2* Est n7 t-stat 0.17 0.80 1.00 4.30 1.49 1.16 1.16 2.00 0.33 5.90 2.43
455
456
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
Charact. of the Employee S and Attributes of the Telecommuting Program Z
-•{
and Benefits Z / of Program >
Indicators of Cost I22 and Benefit /^^
^
stated Participation Decision y (Participate or not)
Figure 9 Modeling Framework for Employee's Adoption of Telecommuting The adoption of telecommuting model is specified by the following equations. Structural Model. Zj=Xri-^rj,
, / = 1,2 , 7~A^(0,Z diagonal)
{2 equations}
(1X1) (1X14X14X1) (1X1)
{1 equation}
U = XJ3^ + Z*y^2 + ^ , ^ ~ standard logistic (1X1) (1X14)(14X1) (1X2) (2X1) (1X1)
Measurement Model. I^=Z*a^+u^,
{14 equations}
r = l,...,14, t; ~ A^(0, S^ diagonal)
(1X1) (1X2)(2X1) (1X1)
1, i f ^ > o y = [-1, i f ^ < 0 (1X1)
{1 equation}
(1X1)
Likelihood Function.
f{y,I\X;a,p,r,Y) = \[.
1 -{XP,+Z p^)y
1 + exp
nd-^ z;-xr, W
^
1
L-Z'a,
Integration of Choice and Latent Variable Models
457
Tables Indicators of Benefit What type of impact would you expect the telecommuting arrangement to have on: extremely negative 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3
Your schedule flexibility Your productivity Your autonomy in your job The productivity of the group you work with Your family life Your social life Your job security Your opportunity for promotion Your sense of well being Your job satisfaction Your life, overall
4 4 4 4 4 4 4 4 4 4 4
5 5 5 5 5 5 5 5 5 5 5
6 6 6 6 6 6 6 6 6 6 6
extremely positive 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9
Table 4 Indicators of Cost How would you expect the telecommuting arrangement to impact your expenditures on: home utilities: child care: elder care: overall working costs:
Results.
() () () ()
decrease decrease decrease decrease
() () () ()
remain the same remain the same remain the same remain the same
() () () ()
increase increase increase increase
The parameters to be estimated include: p (5 parameters estimated), a (13
parameters estimated, 1 parameter constrained to one for identification), Y(\\ parameters estimated), and the standard deviations cr^(14 parameters) and a^{2 parameters), where the covariances of the latent variable equations are restricted to zero. Unless otherwise noted, parameters were set to zero based on statistical tests and a priori hypothesis. The estimation results are shown in Table 5 (estimated variances of the error terms are not reported). The model was estimated using observations from 440 individuals and employed a simultaneous numerical integration estimation procedure. The top panel displays the results of the choice model, which includes the latent explanatory variables benefit and cost. The lower panel displays the results for the latent variable model. The latent variable model consists of many equations: a structural equation for benefit, a structural equation for cost, 11 measurement equations for benefit (one equation per row), and 3 measurement equations for cost (again, one equation per row).
458
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities ^n\
VA
\XH
Figure 10 Full Path Diagram for Model of Employee's Adoption of Telecommuting (See Table 5 and the model equations for notation.) This model of the employee's adoption decision contains more information and allows for a clearer behavioural interpretation than standard choice models. It demonstrates the impact of different telecommuting arrangements on the employee's lifestyle and work-related costs, as a function of the employee's characteristics and situational constraints. The results indicate that females and employees with young children perceive a higher beneficial impact from telecommuting on lifestyle quality than their counterparts. Note that unlike the other two case studies presented in this chapter, a survey was conducted that was designed specifically for this model, and, as a result, the structural models are quite strong with solid causal variables. For more information on these models and other models for telecommuting behaviour, see Bernardino (1996).
Integration of Choice and Latent Variable Models
459
Case Study 3: Usage of Traffic Information Systems The objective of the third case study (Polydoropoulou, 1997) is to estimate the willingness to pay for Advanced Traveler Information Systems. The model uses data collected for the SmarTraveler test market in the Boston area. SmarTraveler is a service that provides real-time, location-specific, multi-modal information to travelers via telephone. Figure 11 shows the framework for the model, which includes a latent variable of satisfaction as an explanatory variable in the usage decision. Travelers' satisfaction ratings of SmarTraveler are used as indicators of the satisfaction latent construct. Table 6 shows the survey questions used to obtain ratings of satisfaction. The model assumes that each traveler has an underlying utility for the SmarTraveler service. The utility is a function of the service attributes such as cost and method of payment, as well as the overall satisfaction with the service. Since utility is not directly observable, it is a latent variable, and the responses to the alternate pricing scenarios serve as indicators of utility. Respondents were presented with several pricing scenarios, and then asked what their usage rate (in terms of number of calls per week) or likelihood of subscribing to the service would be under each scenario. Two types of scenarios were presented: a measured pricing structure in which travelers are charged on a per call basis (corresponds to SPl responses) and a flat rate pricing structure in which travelers pay a monthly subscription fee (corresponds to SP2 responses). Travelers' revealed preference for free service is reflected by the actual usage rate, which serves as an additional indicator of utility. Figure 12 provides a full path diagram of the model, noting the relationships between each variable in the model.
460
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
Table 5 Estimation Results of a Telecommuting Choice Model with Latent Attributes CHOICE MODEL X8 X9 XI0 ZJ* Z2*
Explanatory Variables Telecommuting specific constant Higher salary to telecommuters (relative to 'same') Lower salary to telecommuters (relative to 'same') Benefit (latent variable) Cost (latent variable) Rho-bar-Squared
Est. (3 2.02 0.50 -2.36 0.99 -0.37 0.35
t-stat 8.94 1.12 -5.78 7.01 -3.12
Est.yl -0.15 0.10 -0.04 -1.02 0.69 0.27 0.55 0.28
t-stat -6.65 3.02 -1.99 -14.75 7.47 3.21 7.46
Est.al 0.59 0.80 0.32 0.41 0.76 0.60 0.92 0.61 1.04 1.07 1.00
t-stat 11.61 18.37 6.19 8.15 14.40 12.51 20.87 12.43 24.86 24.84
Est. Y2 0.39 -0.36 0.76 0.65 0.21
t-stat 2.00 -2.70 2.50 2.91
Est.a2 0.37 -0.11 0.50
t-stat 4.78 -3.07 3.63
LATENT VARIABLE MODEL
XJ X2 X3 X4 X5 X6 X7
// 12 13 14 15 16 17 18 19 110 111
XI1 XI2 XI3 XI4
Structural Model for Benefits Zl* (1 equation) Min # of telecommuting days/week Max # of telecommuting days/week * team structure dummy Max # telecommuting days/week * individual structure dummy Telework center telecommuting dummy Travel time * female dummy Travel time * male dummy Child under 6 years old in household dummy Squared multiple correlation for structural equation Measurement Model for Benefits Zl * Social life Family life Opportunity for job promotion Job Schedule Job Your Productivity Group Sense of well being Job Life
(11 equations)
(1 equation) Structural Model for Cost Z2 * Day care costs proxy Home office utilities proxy Equipment costs Weekly transportation costs Squared multiple correlation for structural equation
Measurement Model for Cost Z2* 112 Day care costs 11^ Home office utilities costs 114 Overall working costs
(3 equations)
Integration of Choice and Latent Variable Models Charact. of the Customer S and Attrib. of the Service Z
Indicators of Satisfaction /^
Stated Preferences: Usage Rate of Service /^' Likelihood of Subscription y^^^
Revealed Preference: Usage Rate of Free Service yRP
Figure 11 Modeling Framework for Usage of SmarTraveler
x,\
^10
r"\ rA\^n X
Y#x
A <'
/.^^^
C--
h < h 14
Is
•4
-4
<
~^ < L
4
h
-4
\ h
-4
'^' \ 1
Figure 12 Full Path Diagram for Model of Usage of SmarTraveler (See Table 7 and model equations for notation.)
',0
-4
461
462
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities Table 6 Indicators of Satisfaction with SmarTraveler Service
Please rate your level of satisfaction with the following aspects of the existing SmarTraveler service. extremely dissatisfied Ease of use 1 2 3 1 2 3 Up to the minute information Availability on demand 1 2 3 Accuracy of information 1 2 3 Level of detail of information 1 2 3 Provision of alternate routes 1 2 3 Hours of operation 1 2 3 Coverage of major routes 1 2 3 Cost of service 1 2 3 Overall satisfaction with service 1 2 3
4 4 4 4 4 4 4 4 4 4
5 5 5 5 5 5 5 5 5 5
6 6 6 6 6 6 6 6 6 6
extremely satisfied 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9 7 8 9
All of the choice variables are ordered categorical. The revealed preference choice (y^) and the stated usage rate iy') can take on the following values: 1, if less than 1 call per week 2, if 1 to 4 calls per week y = 3, if 5 to 9 calls per week 4, if more than 9 calls per week The stated likelihood of subscription (y) can take on the following values: 1, if very unlikely to subscribe 2, if somewhat unlikely to subscribe y = 3, if somewhat likely to subscribe 4, if very likely to subscribe The following equations specify the model of SmarTraveler usage. Structural Model equation} (1X1) (1X13X13X1) (1X1)
Utility equations:
{1+M+Q equations} ^^~A^(0,1) N{0,alp,),
JTRP
(1X1)
1/RP
, ~RP
vRP
p
, 7*
p
, ^
,
^RP
(1X13)(13X1) ( I X D d X I ) (1X1) (1X1)
m = \,...,M
Integration of Choice and Latent Variable Models
463
where: m denotes a particular measured rate scenario q denotes a particular flat rate scenario. The disturbance in the utility equations, s, are made up of 2 components: a respondentspecific component and a dataset/scenario specific component. The random disturbance characterizing each respondent, s^, is constant for any respondent across pricing scenarios, and captures the correlation among responses from the same individual (an "agent effect"). The assumed distribution is s^^ ~ N{0, a^). Measurement Model. I^ =Z*a^-\-u^ , r = l,...,10 , L>~iV(0,E^ diagonal) (1X1) ( I X D d X D
{10equations}
(1X1)
f = l,...,4 m = \,...,M
•^'''=r, ifr,f
r = l,...,4 q = \,...,Q
j.r=r,ifrff
T are unknown threshold parameters, with r^ = -oo, r, = 0 (for identification), Additional Notation. [0, otherwise 0, otherwise ^'
10, otherwise
Likelihood Function. O
nij
o
rrr-v^-s,)^
-o
-o
T^=<X
464
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
n TyT Q
10
Results.
1
o
h
^q
^i
'l.-Z-n'
-0)
Z'-X'^/
dZ d£.r
The parameters to be estimated include: p (9 parameters estimated), a (9
parameters estimated, 1 parameter constrained to one for identification), Y{5 parameters estimated), the threshold parameters r , and the standard deviations cr^ (10 parameters), a^ (1 parameter), <Jsp\i} parameter) cispii^ parameter), cr^(l parameter), where CTJ^ was constrained to one for identification and the covariances of the latent variable equations are restricted to zero. Unless otherwise noted, parameters set to zero were done based on statistical tests and a priori hypothesis about the behaviour. Table 7 shows the estimation results for this model (estimated threshold parameters, r , and variances of the error terms are not reported). The model was estimated using observations from 442 individuals, all of whom are SmarTraveler users, and a simultaneous numerical integration estimation procedure. Results of two choice models are presented: one without the satisfaction latent variable (the right column of the top panel) and one that includes the satisfaction latent variable (the left column of the top panel). The integrated choice and latent variable model consists of the choice model with the satisfaction variable and the latent variable model (one structural equation and 10 measurement equations). We found that inclusion of satisfaction in the utility of SmarTraveler model significantly improved the goodness of fit of the choice model. The rho-bar-squared for the model with latent attributes uses a degree-of-freedom correction involving one variable (for the satisfaction latent variable) beyond those used in the model without the latent variable, and thus this degree of freedom adjustment only accounts for the estimated parameters of the choice model. See Polydoropoulou (1997) for additional model estimation results for this model, and for additional models of non-users and of other behavioural responses to SmarTraveler.
Integration of Choice and Latent Variable Models
465
Table 7 Estimation Results of AXIS Usage Model with Latent Satisfaction CHOICE
Utility of SmarTraveler Service Explanatory Variables X6 Constant for actual market behavior X4 Constant for measured service X7 Constant for flat rate service X5 Price per call (cents/10) X8 Subscription fee ($/10) XI Income: $30,000-$50,000 X2 Income: $50,001-$75,000 X3 Income: >$75,000 Z* Satisfaction Latent Variable Rho-bar-Squared
WITH the Satisfaction Latent Variable t-stat Est.B 0.94 5.20 0.56 3.90 0.10 0.70 -0.31 -15.90 -1.29 -15.50 0.02 0.10 0.32 2.10 0.35 2.40 4.50 0.16 0.65
WITHOUT the Satisfaction Latent Variable Est. 6 t-stat 0.97 5.90 0.59 4.30 0.11 0.80 -0.31 -15.80 -1.27 -1^-30 0.15 LOO 0.37 2.60 0.22 L60 0.49
LA TENT VARIABLE MODEL
X9 Jr70 XII XI2 XI3
II 12 13 14 15 16 17 18 19
no
(1 equation)
t-stat -2.40 -10.50 -8.20 -1.60 -1.40
Squared multiple correlation for .structural model
Est. Y -0.19 -0.86 -1.08 -0.26 -0.24 QAM
Measurement (10 equations) Ease of use Up to the minute information Availability on demand Accuracy of information Level of Detail of information Suggestions of ahernative routes Hours of operation Coverage of major routes Cost of service Overall satisfaction with service
Est. a 0.46 1.26 0.47 1.19 1.10 0.75 0.57 0.59 0.19 1.00
t-stat 7.80 21.60 8.2 23.10 22.60 7.80 7.40 12.60 5.30
Structural Model Gender (male dummy) NYNEXuser Cellular One user Age: 25-45 years Age: >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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
•
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.
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Integration of Choice and Latent Variable Models
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
•
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
(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.
This Page Intentionally Left Blank
SECTION 8 FORECASTING
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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 vs^orkplace, 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), w^orkplace 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).
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Uncertainties in Forecasting: The Role of Strategic Modeling
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
In Perpetual Motion: Travel Behaviour Research Challenges and Opportunities
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.
Uncertainties in Forecasting: The Role of Strategic Modeling
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.
This Page Intentionally Left Blank
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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).
Uncertainties in Forecasting : The Role of Strategic Modeling
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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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
Uncertainties in Forecasting : The Role of Strategic Modeling
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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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.
Uncertainties in Forecasting : The Role of Strategic Modeling
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 t^is 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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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-K^otor.
) + c^)
where
MZf is the modal share of light modes for the purpose m in area /, a^,b^, c^are 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:
Uncertainties in Forecasting : The Role of Strategic Modeling
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^, ^^are 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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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.
Uncertainties in Forecasting : The Role of Strategic Modeling
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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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).
Uncertainties in Forecasting : The Role of Strategic Modeling
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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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.
Uncertainties in Forecasting : The Role of Strategic Modeling
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
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
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 ^
®
^o •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^ <
< 0^
0 0
C^
^
_o
0
0
c
_ ,_
'—1
r-
(N
ON
OJI
00
"* ^
X^ 0 y—*
m
vo
00
r»o m
00
CN
\
r-(N
»n (N 0 0
rTf
1
00
•n
— ' ' m
CN 00
^ (N
(N
Os
«n (N
0
r-
r—l
•—I 1
»o
00
VO '—I 1 CM
m *o
r-_W_
VO
M
«n
^
_W_ _wj_wj_W_
(U
^£^ 2
ON
15 1 >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
03
"0 ^ 03
0)
a>
op
0
'£
0
ON
r--
c^ m
0. r> cn Tt
-o
^0 2 U a 0
B
03
a, c
0)
cd 0
«+H'
c
0 0
cd
03
0 on
0
c
1) 1
•0
^ 1
t5 cd
s
00 f^
0 -0
00
C/5
0
a.
'5.
00
a>
^ ,2 "S
S *^ ^00 U (N >-. ^^
11
T3
^
1
0 03
-a
tn
'-^ s -0 «U
— ' *S T3
.•ti
-0
1 (N
s (U
."ti u-^
s
cd
C ^0 0 :3 -0
>^ rn
^ 's c
.2 "uc5 i 00
«
11
03
is
0 03 en
0
C 0
0 0
&0
u^
VO
"n 0
«+-;
O
0
0
00 0 -O
T3 (U "0
»3
t03 i
m
11 v;
0 0 G
c
(U 00
c
(U a> ^ ^ ;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<-j
d 03 03
!/3
0
0
0
"^—' II
^^ •r^
c 0
1 ©
0
0.
Tt;
Uncertainties in Forecasting : The Role of Strategic Modeling
t~-
m 00 (N 00 (N
00 c<^
'vf
^ 5
0
T3 (U
0
TD
cx
s 003^— 0
13
cd
s > s
—
(LT
GO
.2
00
0 r o h^
^ "5- Is(iT =3 0
T3
JZ
(73 (D ftJ)
S
2
a-
LI.
523
524
1
o o o
o ON
p
00
ro
1
m m O
1 "o. C/2
TT
i—1
o
c
2
o o c o
00
c
2
o o c o
roo en »n
^ ro
1—1
Q
en
"•^f
t
00
^ ^ o Tf — ' ' 00
'O »ri
m m o
o
— ' '— ' '
o
CO ON ON
rr-
00 1
00
(N \D
rt '5t
l—<
OO
^
OS VO
c«
CU
a
II
O
X
X
c
u-i
c x:
^
O II CO
o N
C
on
w D.
(U O
c
x: o
B
(1>
O
II
d
_c
d II
c
D, II
o
«+.;
B
cd
1
CO
S
•a o o
o
^ ^ e
-i^
II
d
s
cri
d
c q N
W5
Cu
oo »n
OO
(^
VO
00
— ' '
»o o
'—1
o rr«-» ON
m m CN
00 ON
m
CO
^ (N m Tf ^ :^ ^ ^
C/3
ON
ON
> >
X
ON ON
X
»r>
c3 T3
1c «o
o "3 o
d d
r>
O
T3
o
s o ,fl
o
Q O
o X)
o S 00
Q O
T3
cd
^ X) C/5
•Ti ON ON
C/5
'"' O
d^
C/2
o 00
O
X CO
o o C
o o
>.
en
O
J=
, ' "^D—
»r>
a.
k<
Si
C
w
>^ »n
-C
^D-^,
o
Ui
c«
1
c3
.^
^ u s
H
C8
s
o
ha
s
Q« CI.
>^
e
•a
^
W5
fi ,W)
Ti-
"—I
00
m
"^ 'O
O
— ' ' — ' ' (N
OO (N
m TT
O
o
CO
'^
»—<
VO
•n o ^o •r»
DH
0)
1/5
c^_:
^II
PH
CO
c
00
o o
II
II
Cu
c?i
u o c Si
on
c to c o o
c
B
1/3
o II
to
3 C o
CJ
II
o CJ
Tl-
o •—I
m
<—1
•T)
II
o on C
o o l—1
VO
o
II CO
C
.2 .2 to
^ fO
•Ti
m
v-j
(N
ON
— ' ' ON — ' ' '^ 1
tn
•"^ 1
t^ CN
00 OS
"—I
f^ OO
OO
r~-
VO (N
rsj
r-
^
c
00 *co
c
OO CO CO
c
c
3 O
00 CO
c
c
3 O
CO
c
t
00
3 CJ
c
CO CO
%% c
00
CO
Ui
O
D
C
Oi
CO O
c
U
u O
< <
o o cc oci
1
CO
E E E E E 1/3 CO
< < < d)
u
C T3
CO O
Di
c
1 1 >-• 1 ^
CO O
a> c o o -o OH
CO
00
C
c "5
00
^ fc
c E o
< < < < < < < < <
t-
04
In Perpetual Motion : Travel Behaviour Research Challenges and Opportunities
^ '—1
o
»n o »—1 l—1
c
o o o O a
c
o
.2 .2 "c3 O
(U 00
_3
c« U
c c '5- 'a, a. o o
13
> >
00
«4-;
a, d o B S
U
o
>< 00
c
O
525
Uncertainties in Forecasting : The Role of Strategic Modeling
REFERENCES Ashley, D.J. (1981). Uncertainty in interurban highway-scheme appraisal. In New Horizons in Travel-Behavior Research, Stopher, P.R., Meyburg, A.H., Brog, W. (eds) Lexington Books, pp. 599-615. Bates, J., M. Brewer, P. Hanson, D. McDonald and D. Simmonds (1991). Building a strategic model for Edinburgh, PTRC Summer Annual Meeting. Ben-Akiva, M. E., J. L. Bowman and D. Gopinath (1996). Travel demand model system for the information era. Transportation 23 :241-266. Bonnafous, A. (1985). Simulation du fmancement du transport urbain : le modele QuinQuin, Transports Urbains n°54, Janvier-Mars. Bonnafous, A. and E. Tabourin (1995). Modeles de simulation strategique. Communication a la f"^^ WTCR, Sydney, Juillet 1995, 17 p. Bouf, D. (1989). Un nouvel instrument d'analyse strategique pour la RATP : le modele Gros QuinQuin. These de I'Universite Lumiere-Lyon2, Laboratoire d'Economie des Transports, Lyon. Duffus, L. N., A. S. Alfa and A. H. Soliman (1987). The reliability of using the gravity model for forecasting trip distribution. Transportation 14 :175-192. Faivre D'Arcier, B., O. Andan and C. Raux (1997). Stated Adaptation Surveys and Choice Process: Some Methodological Issues (forthcoming). Goodwin P. B. (1988). Evidence on car and public transport demand elasticities 1980-1988, TSU Ref 427, Oxford, June. Horowitz, J. L. (1981). Sources of Error and uncertainty in behavioural travel-demand models. In New Horizons in Travel-Behavior Research, Stopher, P.R., Meyburg, A.H., Brog, W. (eds) Lexington Books, pp. 543-558. Jones, D., T. May and A. WenBan-Smith (1990). Integrated transport studies: lessons from the Birmingham study. Traffic Engineering-^Control, pp.572-576. Jones, P. M. (1979a). "HATS": a technique for investigating household decisions. Environment and Planning A, vol 11, pp 59-70. Kitamura, R., E. I. Pas, C. V. Lula, T. K. Lawton and P. E. Benson (1996). The sequenced activity mobility simulator (SAMS): an integrated approach to modeling transportation, land use and air quality. Transportation 23 :267-291. Koppelman, F. S. (1981). Uncertainty in methods and measurements for travel behavior models. In New Horizons in Travel-Behavior Research, Stopher, P.R., Meyburg, A.H., Brog, W. (eds) Lexington Books, pp. 577-583. Lee-Gosselin, M. (1990). The Dynamics of Car Use Patterns Under Different Scenarios : A Gaming Approach. In P. Jones (ed.). Developments in Efynamic and Activity-Based Approaches to Travel Analysis, Oxford Studies in Transport, Gower Press, Aldershot, UK. May, A. D., M. Roberts and Holmes, A. (1991). The development of integrated transport strategies for Edinburgh, PTRC Summer Annual Meeting. McKinder, I. H. and S. E. Evans (1981). The predictive accuracy of British transport studies in urban areas, LGORU, Working Note 20. Oldfield, R. H. (1993). A Strategic Transport Model for the London Area, Research Report 376, Transport Research Laboratory, Crowthome, UK. Ortuzar, J. (ed) (1992). Simplified transport demand modelling, PTRC, London, 153 p.
526
In Perpetual Motion: Travel Behaviour Research Challenges andOportunities
Raux C, O. Andan and C. Godinot (1994). The simulation of behaviour in a non-experienced future : the case of road-pricing. Paper to the 7^^ lATBR Conference, 13-16 Juin 1994, Valle Nevado, Chili (to be published). Raux, C. and O. Andan, O. (1988). Les analyses des comportements de mobilite individuelle quotidienne, une synthese bihliographique. Rapport pour ie SERT, LET, Lyon Juillet. Raux, C. and E. Tabourin E. (1992). Congestion et crise du financement des transports a Lyon : vers un peage urbain ? In RAUX C. et LEE-GOSSELIN M. (eds). La mobilite urbaine : de laparalysie au peage ?, editions du PPSH, Lyon. Semaly, L. (1997a). Developpement d'un modele strategique de simulation des deplacements. Guide de Vutilisateur. Fevrier 1997, 33p. Semaly, L. (1997b). Developpement d'un modele strategique de simulation des deplacements. Presentation generale. Fevrier 1997, 28p. + annexes. Skinner, A. and C. Haynes (1992). Birmingham City Centre Transportation Model. The concept, development and use of a multi-mode transportation model. Traffic Engineering + Control, vol.33, n°12, December. Slavin, H. (1996). An integrated, dynamic approach to travel demand forecasting. Transportation 23 :313-350. Spear, B. D. (1996). New approaches to transportation forecasting models. A synthesis of four research proposals. Transportation 23 :215-240. Stopher, P. R., D. Hartgen and Y. Li (1996). SMART : simulation model for activities, ressources and travel. Transportation 23 :293-312. Tabourin, E. (1989). Un modele de simulation du financement des transports collectifs urbains a Van 2000 : le modele QuinQuin, Application a Vagglomeration lyonnaise. These d'Universite, Universite Lumiere Lyon 2, Septembre. Talvitie, A., Y. Dehghani and M. Anderson (1982). An Investigation of prediction errors in work trip mode choice models. Transpn. Res.-A, vol. 16A, No 5-6, pp. 395-402. Talvitie, A. P. (1981). Inaccurate or incomplete data as a source of uncertainty in econometric or attitudinal models of travel behavior. In New Horizons in Travel-Behavior Research, P. R. Stopher, A. H. Meyburg and W. Brog (eds.) Lexington Books, pp. 559-575. ^ The example of particle physics where the concepts of wave and corpuscle, at one time in conflict, gradually seemed like two valid but incompaticble approximations on a macroscopic scale of the underlying nature of the matter components.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
25
FORECASTING: WORKSHOP REPORT
Kostadinos Goulias
SUMMARY OF WORKSHOP ON FORECASTING The scope in this workshop has been to focus on the examination of predictive models from experimental and operational perspectives. Three key issues on complexity, measurability, and transferability over time were given as the guiding directions for lively discussion. The key substantive questions to address with forecasting as defined by the workshop participants were the transportation impacts of: • Sociotechnological changes • Telecommunications and e-commerce • Lifestyle including work hours, daily and weekly rhythms, leisure, and pattern formation including regularities • Environmental policies/concerns such as assessing compact city/neo-traditional neighborhood designs, ability of air quality models to address distributional justice, precursor pollutant predicting and global warming, and assessing the impact of technological advances •
Mobility concerns including negotiation of information acquisition retrieval/exchange and assessment of information quality and compatibility
and
For this workshop, nine papers and a resource paper were prepared and discussed. The participants agreed that car ownership research is indicative and provides a good example of
528
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
research advances in this field. Similar streams of research as in car ownership are expected and needed given the substantive questions above. In addition, many methodological and research questions were raised providing an indication of the substantial amount of research much needed in this arena. A representative sample of the methodological questions include: • Define criteria followed in the distinction of endogeneity and exogeneity and the domain of forecasting models and separate these definitions from the strictly econometric meaning of endogeneity and exogeneity • Develop guidelines for forecasting models in each of the different planning paradigms such as comprehensive, strategic, operations planning/management systems • Create research initiatives on the interface of models that are designed for different purposes to increase reusability (e.g., secondary use of models) • Provide and disseminate model paradigms for behavioural dynamics and policy actions • Incorporate the variety of stages by time scale in models (e.g., very long term, long term, short term, very short term) and the interface among time scales in applications • Initiate research on computation of forecasting errors • Examine the use of forecasting error "bands" in decisions making • Disseminate information about and use more efficiently data and information from other sources that can be compatible with specific forecasting systems in transportation In light of the discussion above, the workshop participants identified the following as the most important advances in the next five years: • Understanding that there are direct/indirect effects, e.g. car ownership and car use, and a modeling framework accounting for that will/should be created • Greater modeling sophistication made possible by increased computing power (model estimation by simulation, easier software development, reusability). In fact some old ideas can be implemented now and they will be implemented • The field rediscovered and began to use micro-simulation in an expanded way simulating the entire spectrum of human activity • The introducfion of Geographic Information Systems into planning will be followed by large scale applications • More data and more accessible data will help the development of many models in a few countries. However, accessibility is still lacking in many countries and for some data sets • We will see more panels, more repeated cross-sections and other longitudinal data collection • We will confront the issue of developing harmonized data-sets (and data collection) to perform comparisons
Forecasting: Workshop Report •
529
A new battery of models will become available (e.g., models of new/emerging behaviours in telecommunication and travel, responses to traveler information systems and so forth)
Another charge of the workshop was also to identify the most important questions and these are: • How many and which input variables should transportation planners be forecasting (most countries have highly aggregate forecasts of population and some demographics). Few have employment income, prices; almost none have car ownership, drivers' licenses. Most of our models require detailed demographic, economic, and social characteristics for each person. Where do we draw the line? • When do we apply microsimulation and do we apply trend analysis or other more aggregate forecasting methods? • How do we locate forecasts of exogenous to our model variables that can be used in our models? • How do we assess and evaluate errors in our forecasts? • How do we convince jurisdictions to use them?
In terms of the findings, methods and recommendations the workshop identified the following: • Detect/identify social change items relevant for current and future policy issues • Formulate models sensitive to change • Develop Interactive Implementable Forecast (IIF) tools • Decide what is endogenous/exogenous? (what we should produce vs. what we get from other sources)
In terms of findings for practice the following are a few suggestions. Use Employment/Income Data from BE A (historic data since 1950, 50 yr. forecast). Look for official data (BTS in the U.S.A., Essex archives, Eurostat). However, for dynamic models practiotioners will need to collect panel data.
530
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
CONCLUSION: A RESEARCH AGENDA One of the important outcomes in this workshop was the following research agenda: • Relationship between complexity and error needs to be studied in more detail. On the surface, more complex models appear to be harder to assess in terms of forecasting error. • Validation is needed for all models when designed for applications. However, experience with validation necessary for panel data-dynamic models is very limited. • One opportunity lost until now has been the lack of using data about transactions from consumers and businesses. Explore the possibility of generating travel data from transaction records while been aware of privacy and property issues. • Goods movement is an area that has been neglected in forecasting. Explore goods movement issues in light of major changes in that sector. • Review and evaluate forecasting from a theoretical and applications viewpoint. • Study, evaluate, and recommend the data needs for advancing the state of the art. • Develop methods and applications with more geographically disaggregate data. • Study the fusion/assembly of forecasting variables/attributes and their monitoring and validation. • Develop and assemble more longitudinal data (e.g., forecasts for more exogenous variables). • Expand the currently available longitudinal household data to include values, roles, education, employment, number of children, car ownership, housing, consumption and so forth. • Develop longitudinal methods for business data (e.g., panels of establishment surveys).
ACKNOWLEDGEMENTS Participants included: G. Baez, G. Bresson, D. Ettema, K. Goulias, J.-L. Madre, C. Moore, A. Nuzzolo, W. Olsen, J. Pommer, A. Pirotte, F. Ramjerdi, C. Raux, K. Srinivasan, P. Stopher, and T. Yamamoto.
SECTION 9 MICROSIMULATION OF TRAVEL ACTIVITIES IN NETWORKS
This Page Intentionally Left Blank
In Peqjetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
26
ACTIVITY-BASED TRAVEL BEHAVIOUR MODELING IN A MICROSIMULATION FRAMEWORK
Eric J. Miller and Paul Salvini
INTRODUCTION Microsimulation as a method for implementing activity-based travel behavior models for forecasting and policy analysis purposes has been receiving ever-increasing attention, as demonstrated by the microsimulation workshop at the Austin lATBR conference, as well as similar workshops or sessions at other recent conferences.^ Many current efforts to implement activity-based forecasting systems are microsimulation-based,^ while most activity-based models are well suited for implementation within some form of microsimulator. Further, the emerging next generation of network assignment models (which compute network link flows, congestion levels, emissions and energy use — still the "bottom line" for most travel demand analyses) are themselves microsimulation-based,^ implying the need for a similar level/method of analysis for the activity models which "feed into" these network models. The case for microsimulation modeling has been well made elsewhere"^ and will not be repeated in detail here. Table 1 does, however, attempt to summarize from the literature the major potential strengths of microsimulation. Put very simply, the argument is that microsimulation represents an effective (and, in at least some cases, perhaps the only) method for generating policy-sensitive forecasts from disaggregate, activity-based models.
534
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Table 1 Potential Strengths of Micsrosimulation Models O
Generates detailed inputs required by disaggregate behavioral models
O
Supports efficient data structures for large, complex, multivariate problems
O
Potential for emergent behavior
O
Policy sensitivity
O
Transparency to modelers and decision-makers
Despite growing interest in and experience with microsimulation methods, many research issues exist with respect to this approach. The purpose of this chapter is to raise what we believe are some of the major research issues associated with the development and implementation of activitybased microsimulation models, thereby providing one starting point for workshop discussions in this area over the next few days. While primarily keeping the presentation at quite a generic level, we will from time to time also draw on our own experience to date in the development of a microsimulation-based Integrated Land-Use Transportation Environment (ILUTE) model [Miller and Salvini, 1997]. The term "microsimulation" tends to mean different things to different people. Perhaps due to the visibility of projects such as TRANSIMS, many people tend to think of microsimulation specifically in terms of network route choice and performance models. On the other hand, given the use of microsimulation methods to generate the disaggregate inputs required by their models, many activity modelers think of microsimulation as the procedures used in this input data synthesis/updating process. In the social sciences, it is common to distinguish between "static microsimulation" in which a fixed, representative sample is used to test various policy alternatives within a microanalytic framework, versus "dynamic microsimulation" wherein the representative sample changes or "evolves" over time as a function of endogenous and/or exogenous processes.^ In this chapter we adopt a comprehensive definition of microsimulation as a method or approach (rather than a modtX per se) for exercising a disaggregate behavioral model over time. Figure 1 very simplistically but still usefully depicts the general microsimulation process as we understand it. Key features of this process include the following: 1. The model must have as its primary input a disaggregate list of actors (or entities or behavioral units) upon which it operates. This list often consists of a representative sample of individuals drawn from the lelevant population. Alternatively, it is becoming more common to use a "100% sample"; i.e., a list of the entire set of actors in the population. In general, two sources exist for this input list or base sample: a sample of actual individuals
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
2.
3.
4.
535
drawn from the population, obtained through conventional survey methods; or a list of synthetic individuals, statistically constructed from more aggregate data concerning the population being modeled (e.g., census data). Over time, the demographic, social and economic characteristics of the population being analyzed will change. Thus, the list of actors must be updated over time within the model so that it remains representative of the population at each point in time. Processes affecting the evolution of the population over time can be both endogenous (aging, births, deaths, etc.) and exogenous (in-migration, etc.). Once the attributes of the list of actors are known for a given point in time, the behavior of these actors can be simulated using a relevant behavioral model. Depending on the application, this "model" may deal with a single process (e.g., activity/travel choices in response to TDM measures) or many "nested" processes (residential location, employment location, activity/travel, etc.), involving a complex set of interconnected sub-models. In general, this behavior will depend upon past and current "system states" (endogenous factors) as well as exogenous factors. Primary outputs from the microsimulation include both the attributes and behaviors of the actors over time. These outputs are generally expressible both in aggregate terms (link volumes, average modal splits, etc.) and in terms of disaggregate "trajectories" of the individuals being simulated (i.e., the "historical" record of the behavior of each individual over time).
Different microsimulation applications involve different implementations of Figure 1. So-called "static" microsimulations, in particular, do not require updating of the base sample/population, nor do they require iterating the model through time. Similarly, microsimulations,which are based on an observed sample of actors, do not require a synthetic sample to be constructed. In a previous discussion of microsimulation and activity-based forecasting. Miller [1996] identified the following research and development issues: 1. development and testing of population synthesis methods; 2. development and testing of population updating methods; 3. determination of appropriate levels of aggregation, particularly with respect to: (a) treatment of space; (b) treatment of time; and (c) selection of population attributes. 4. demonstration of the statistical properties of microsimulation models; and 5. demonstration of the computational feasibility of microsimulation models. Each of these issues is discussed in some detail in the remainder of this chapter.
536
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges t = to Y
Synthesis of Base Sample fort = to 'w
^>f Endogenous Changes to Sample during this At
^ Exogenous Inputs this At
t = t + At v
>k
Disaggregate Behavioral Model
^
>f Behavior/System State at(t -^At)
Figure 1 General Microsimulation Process STUDY AREA SYNTHESIS
Population Synthesis The first step in preparing a long-run microsimulation is to create a data set containing detailed disaggregate information about the population being studied. Ideally, these base data would consist of actual information about each of the participants in the study area. Realistically, however, obtaining such detailed disaggregate information is not possible. Instead, researchers must rely on a variety of imperfect data sources (disaggregate but sampled, aggregate, crosstabulated, survey, etc.) as model inputs. The purpose of population synthesis is to generate a data set of representative persons (also called actors or trip-makers) that is consistent with the available aggregate data. The general synthesis problem is illustrated in Figure 2. As shown in this figure, there are, in general, two types of information which might be available to characterize the base case: 1. Aggregate tabulations which provide information about the distributions of individual
Activity-Based Travel Behavior Modeling in a Micro simulation Framework
2.
537
variables over the population, as well as, typically, limited information about the covariance of selected variables. The national census is usually the primary source of such information. These data are generally highly reliable at the aggregate level, but do not provide sufficient information to determine with certainty the full joint distribution of all the population attributes of interest over the population. In addition, the specific attributes of any one individual within the population are impossible to identify with certainty from these data. Sample data which provide specific attributes for the observed individuals, obtained from sources such as activity/travel surveys, PUMS files,^ etc. These data provide an estimate of the joint population attribute distribution.
Although variations on the theme always exist within specific applications, population synthesis inherently involves two primary tasks: 1. Estimating the full joint distribution of population attributes; and 2. Drawing a "realization" of a set of individuals with specific attributes from this joint distribution. A current and quite promising example of this process is a two-step iterative proportional fitting (IPF) procedure by Beckman, et al [1995]. This procedure simultaneously estimates the multiway attribute distributions for each census tract within a Public Use Micro Area (PUMA), such that each distribution satisfies the marginal distributions for the census tract (as defined by aggregate census tables) and has the same overall correlation structure as the Public Use Microdata Sample-based (PUMS) multi-way distribution. The problem of estimating the joint distribution of population attributes appears to be a classic application of information theory, which provides formal methods for generating "least biased" estimates of a "system state" (e.g., the population joint distribution) given limited known information concerning this system state (e.g., the marginal distributions provided by census data).^ While Beckman, et al note that their procedure is "consistent with" an entropy maximization (information minimizing) approach, it would seem worthwhile to explore the joint distribution estimation problem explicitly within an information theory framework, in the hopes of developing a generalized, "optimal" approach to this problem.
538
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
1 2 3 4 5+
For each zone in the studyarea, assume we have the following aggregate tables: 0 1 2 3+ 0 1 2+ 1 1 2 3 4 5+
No. of Households by Household Size
2 3+
No. of Households by Household Size & No. of Workers
No. of Households by No. of Workers & No. of Autos
In addition, assume we have a small sample of households which provide an estimate of the joint distribution of household size, no. of workers and no. of autos.
The synthesis process con^rts the aggregate matrix-based data, combined with any small sample data available into a list of synthetic (but representative) households with specific attribute values.
HH. No. 1 2 3
Size 4 5+ 2
No. of Workers 2 1 2
No. of Autos 2 1 1
Figure 2 The Population Synthesis Process
Activity-Based Travel Behavior Modeling in a Micro simulation Framework
539
Synthesizing Other System Entities As is typical in the microsimulation literature to date, the discussion to this point in the chapter has focussed on the synthesis of the resident population who will eventually be called upon within the model to make activity/travel decisions. For many (typically short-run) applications, this may be sufficient. In many other (particularly longer-run) applications, however, many other "system entities" may also need to be "known" at a disaggregate level and hence, also may need to be synthesized. These "entities" might includefirms,residential buildings, non-residential buildings, vehicles, jobs, etc. While the synthesis problem for such entities is, in principle, the same as for the population case, additional issues may arise in at least some cases. These may include: 1. In some cases, limitations may exist with respect to the depth, breadth and/or reliability of data which are available to support disaggregate synthesis activities. In particular, data sources other than the census will almost certainly have to be drawn upon in order to synthesis many of these entities. 2. In at least the case of firms, it may well be the case that significantly greater heterogeneity exists than in is the case for people/households. Firms vary tremendously in terms of size, function, locational preferences and needs, etc. Can we credibly synthesis a disaggregate population of firms for an urban area? If so, at what level of detail/specification? Do we need to work at a "completely disaggregated" level of analysis for firms, or is a more aggregate level acceptable for our purposes? 3. Many of the questions raised in point two are indicative of the lack of experience (and hence insight) which we have with modeling aspects of the urban system other than the distribution of the resident population and their subsequent activity/travel patterns. We have a long tradition of assuming that the "supply" or "attraction" side of the process is fixed and given. Dynamic microsimulation (particularly when performed over medium- to long-term forecast periods) severely challenges this assumption, and ultimately requires us to model "supply" as well as "demand". This clearly represents a major challenge, and one which will require us to collaborate with economists, geographers and others who have experience with modeling firm location, housing supply, etc. We believe, however, that this challenge is well worth taking, since, in the long run, it may well be the "attraction" side of the "equation" which is the dominant driver of activity/travel and/or is the side which is most susceptible to policy initiatives.^ The proliferation of commercial GIS-based databases undoubtedly represents a major resource to support the development of supply/attraction databases and models. As discussed further below, however, the completeness, accuracy, etc. of these databases may be a considerable problem for some time to come. Further, even given excellent databases, one should not underestimate both
540
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
the importance and the difficulty of developing credible supply/attraction components for our activity/travel simulation models.
UPDATING THE STUDY AREA In any microsimulation in which system behavior over a significant period of time is to be modeled, one must account for the structural and demographic changes that occur in the study area over this time period. Aging, residential mobility, job changes, divorces, marriages, births, deaths ~ the occurrence of these events over time cannot be ignored if we are to accurately capture changes in activity and travel behavior. The task for accounting for such changes over time can be conveniently referred to as "updating".
Population Updating The demographic and socio-economic characteristics of a study area population must be updated over time. Changes that affect the population include aging, in-migration, out-migration, marriages, divorces, education, employment, births, deaths, household formation and dissolution, residential mobility, auto ownership, etc. In general, each of these events must controlled by a sub-model within the microsimulation model's "evolutionary engine". Demographic and household structure attributes are generally handled using very simple probability models: either fixed transition rates based on empirical data (e.g. fertility rates for women by age), or simple parametric probability functions (e.g. MIDAS uses simple logit models to determine household type transition probabilities). In all such cases, Monte Carlo simulation methods are used to generate household-specific events (e.g. births) on a household-by-household basis. These events are generated at discrete intervals (usually in one year increments) as specified by the modeler. Treatment of employment status, residential location and automobile holdings varies far more widely across models, depending on their application. Each of these can be a significant part (or even the primary focus) of the behavioral modeling component of the microsimulation. Alternatively, if the application permits, one or more of these might be handled in terms of "transition probabilities" in the same way as the demographic variables described above. As with synthesizing procedures, limited experience exists, at least within the travel demand forecasting community, with demographic/socio-economic updating methods. For examples of
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
541
specific methods used to date, see Miller, et al [1987], Kitamura and Goulias [1991], Goulias and Kitamura [1992], and Oskamp [1995]. All of these examples should be treated as being illustrative and experimental in nature rather than in any way definitive. Considerable experience with demographic forecasting obviously exists among demographers. In addition, a long and extensive tradition of microsimulation exists within the social sciences involving a wide range of socio-economic applications. Beginning with the seminal work of Orcutt [Orcutt, 1957,1976;Orcutt, eM/., 1961,1986], economists and other social scientists have developed, particularly over the last ten years, very impressive policy-sensitive microsimulation models designed to inform politicians and other decision-makers on a wide range of economic and social policies.^ Indeed, these models have reached the point of acceptance where they are routinely used within many countries to analyse the benefits and costs (and the "winners and losers") associated with most major policy issues, including such high profile issues as health care reform in the U.S. and taxation policy in Canada.^^ The social science microsimulation literature is one into which we are just now beginning to tap in our own research work, and we suspect that we are fairly typical within the travel behavior research community in this regard. It is clearly a literature with which we must become familiar and which undoubtedly will provide us with considerable insights into many of the problems which we face in our own models. At the same time, one should note the following: 1. It would appear that many of the operational models are "static" rather than "dynamic" in nature (i.e., they do not evolve or update the base population over time). Thus, it is probably the case that we must work with social science microsimulators in order to learn how to best construct dynamic microsimulation models, rather than be able to borrow from them "off the shelf solutions to this problem. 2. It is also almost certainly the case that the social science microsimulations are either completely aspatial or, at best, operate at an extremely gross spatial scale (e.g., perhaps by state within a national model). Similarly, traditional demographic forecasting typically is undertaken at very gross spatial scales (e.g., state/province or perhaps at a regional level). Travel behavior forecasting ultimately requires a much finer spatial scale, typically down to the traffic zone level. The challenge for travel behavior researchers is to adapt existing methods and/or to develop new ones that can operate reliably at the census tract/traffic zone level required for travel demand forecasting.
Updating Other System Entities All other system entitites explicitly represented within the model (such as possible firms.
542
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
buildings, etc.), which can be expected to change over time, must be updated. As with population updating, this may involve the use of simple transition probabilities (x% of all housing stock of a given type undergoes renovation each year), exogenous inputs and/or "behavioral" models of varying complexity. As has already been discussed concerning the synthesis problem associated with these entities, the development of credible, appropriate updating methods represents a significant research and development effort.
SHORT-RUN/LONG-RUN DYNAMICS Dynamic microsimulation models generate discrete events over time. Different events, however, have different temporal scales. Some events, such as residential mobility decisions, have "longrun" dynamics. Events with long-run dynamics occur infrequently ~ perhaps no more than once a month or even year. It would be wasteful to simulate these long-run processes on an hourly basis. Events with short-run dynamics, however, are events that typically occur within a day. Activity generation and route choice, for example, might occur many times during the course of a simulated day. In our own ILUTE modeling work we have addressed the problem of short-run and long-run dynamics by allowing each event to have its own temporal scale. In particular, the ILUTE model provides a short-run microsimulation of activity-based travel demand operating inside a long-run microsimulation of an evolving population. The design of the ILUTE model is flexible enough to provide multiple temporal event scales. In the extreme, each event being simulated could have its own temporal scale. This flexibility is important since the analysis of different policy decisions might require a different temporal scale for the same event. For example, a researcher looking at seasonal trends in real estate would require weekly or monthly updates to residential mobility rather than yearly ones. Another researcher looking at the long-term effects of adding a new multi-lane highway that bypasses the city might be less interested in simulating residential mobility at such a fine level. Research issues with respect to temporal scale include the following: 1. Developing models which are, in fact, "temporally scalable". While in principle, one should be able to adjust the temporal scale in any model, in practice most models are "calibrated" to a given temporal scale and can be subject to the same sort of aggregation biases with which we are more familiar with respect to spatial scale. For example, a model developed for use with an annual time step may not work well when applied to a monthly
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
2.
3.
4.
543
or weekly time step. In general we have little experience with this, due both to the lack of experience with dynamic models in general and to the lack to date of experimentation with respect to this issue. Associated with the issue of temporal scale, is the definition of the temporal unit being simulated. In a simulation model using a one-year time step, a simulated day (e.g., simulation of a day's activities/travel for each household) inevitably is a "typical" day possessing strong "short-run equilibrium" connotations. Alternatively, another simulation model may focus on simulating day-to-day variations in behavior over an explicit sequence of days. These two models will generally be quite different in structure and cannot be reconciled into a single model structure without the use of complex temporal sampling schemes designed to avoid excessive computing times. Another way of discussing these issues of temporal scale and simulation unit is to recognize that what we are talking about is the question of the appropriate level of temporal aggregation to adopt within the microsimulation model. The usual issues of aggregation bias, statistical efficiency and behavioral representativeness, which we are used to discussing within the spatial context, apply equally to the less familiar temporal case. In addition to our relative lack of experience in dealing with temporal aggregation issues, the problem is perhaps more challenging than the spatial aggregation problem, in that no "base" disaggregate temporal unit of analysis exists. That is, in the case of space, fundamental units of analysis exist such as the individual, the household, the dwelling unit, etc. (i.e., the fundamental behavioral units which occupy space). Once we reach one of these fundamental analysis units, no further disaggregation is possible (e.g., we don't need to consider modeling fractions of people). As a continuous variable, however, time does not possess an equivalent "fundamental unit" and is, in principle at least, endlessly divisible intofinerand finer analysis units, limited only by our computational capabilities, data, and perceived need for further disaggregation. Developing appropriate "interfaces" between time frames. It is easy to talk about "integrated" models of transportation and land use; it is more difficult to actually implement the appropriate 'Teedback" from short-run activity/travel models to longer-run location decisions. These difficulties are both theoretical/behavioral (exactly what information/expectations concerning accessibility, trip-making, etc. do people have/use when making residential or employment location decisions) and practical/computational (do I need to re-equilibrate the transportation network every year or will every five years do?) in nature. Much more investigation into the behavioral aspects of this issue is required, as is more experimentation with alternative computational strategies. Microsimulation (or, more generally, dynamic modeling) brings to the fore questions concerning the role of history, memory and information in the determination of behavior. It is without a doubt true that our behavior at any point in time is affected in countless
544
5.
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges ways by our past behavior and experiences and by the information we have acquired along the way. It is also almost certainly true that in most cases this "state dependency" is more than Markovian in nature: it is our historical "trajectory" through time, space and experience that (largely) determines our future path, not just our current location along that trajectory. This observation again raises both behavioral issues (what is the nature of the path dependency) and practical implementation issues (how much information about the past must be saved?; how do we model dynamic processes given that inevitably the model deals with an arbitrary segment of each individual's life-continuum?). The question of the role of memory and learning in dynamic models is perhaps one of the single biggest challenges in microsimulation modeling. Taken seriously, this issue might well imply the need to synthesize prior "memories" (since the simulation will always begin at an arbitrary point in each simulated person's life history), as well as to explicitly model learning processes and adaptive behavior, — while somehow maintaining data storage and computational requirements within some semblance of feasibility. In addition to a past, all people also have a future, and expectations about this future also clearly influence current behavior. Housing choice decisions depend not only on current needs, but also upon expected future needs, expectations concerning market prices, etc. Today's activities and associated travel are determined in part by their implications for future events (if I shop today on the way home from work I can play golf Saturday morning), as well as by constraints imposed by the expected future on current actions (I must shop today on the way home from work because I have to take Junior to the soccer tournament on Saturday). Thus, both "lags" and "leads" generally will exist in dynamic behavioral models: in general we have little experience with either phenomenon, but perhaps especially the latter.
SPATIAL RESOLUTION The ideal travel behavior model would have a spatial scale that is detailed enough to be able to represent without significant error all modes of travel including walking, transit, and auto. Similarly, it would capture all significant heterogeneity in spatially fixed entities (buildings, activities, etc.). Such a model would ideally include an entire three-dimensional representation of an urban area and the buildings and walkways within it. Of course, even if it were possible to construct such a detailed model, it would be prohibitively difficult to evolve it over time. Despite the continuing development of GIS software and databases, most current microsimulation models still operate at effectively a small traffic zone level of spatial aggregation. As computing power increases and as the demonstrated need arises, most state-of-the-art models should be able
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
545
to operate at finer spatial scales. TRANSIMS, for example, has sources and sinks of traffic at "parking accessories" which, once one gets past the jargon, currently look remarkably like census tract centroids. The point of using such an odd term, however, is that, eventually these points could correspond to every driveway, garage, on-street parking spot, etc. at which a car might be parked (and hence define the beginning or end of the vehicle trip). Similarly, in our own work, houses, people and activities are located at x-y geocodes, which, at the moment happen to all correspond to census tract centroids but could "just as easily" correspond to individual lots or buildings. Whether the traffic zone is the appropriate spatial unit for current and future models is a question which is primarily practical and empirical in nature: is it good enough for our purposes? do we have the computational resources to support a finer level of spatial analysis? do we have the data to support a finer level of analysis at any reasonable level of statistical reliability? The question of data support for finer spatial scales of analysis is almost certainly the critical long run question. GIS's are dramatically changing how we think about and analyze space. Possible issues, however, with respect to GIS-based models include: 1. Completeness/accuracy of GIS databases. Many (most? all?) GIS databases vary with respect to accuracy and completeness of their data [Batty, 1995]. For many purposes, this is not a major issue (if I am a telemarketer, information on 90% of the market sounds great to me), but for many modeling purposes it can be problematic. Further, finding and resolving errors, filling gaps, etc. in GIS databases can be a very time consuming, expensive and perhaps never-ending task. The point here is not that GIS databases contain error (all data are subject to error). Rather, the question is whether the errors are sufficiently large that we are paying a high computational price to achieve a "false precision" in our models by working at a spatial scale that, in some cases at least, is not practically sustainable. This issue is relevant regardless of whether the base data we are using is observed or synthesized. 2. Can we update the spatially disaggregated data over time with any reliability (even if we know that there is a Blockbuster Video on this comer in the base year, will it be there five years from now, and if not, what will replace it?). 3. Do we need point-to-point travel estimates or is at least a slightly more aggregate representation adequate, given both our planning needs and other typical sources of error in the model (e.g., given inevitable errors in determining who lives on this block and where they are likely to work, does it really matter whether we put them on the first, second or seventeenth house on the block?)? The spatial component of travel behavior is, in our view, one of the critical reasons why travel is
546
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
such a difficult phenomenon to analyze and model and which sets travel apart from most other "consumer behavior". Part of this is due to the brute size of the problem of representing behavior over two-dimensional space (database size, computational burden, ability to display and assimilate data, etc.). But part is also due to the fact that trying to predict where something will occur is a far more difficuh task than simply predicting whether something will occur. Many current activity-based "travel" models are surprisingly aspatial, typically dealing primarily if not exclusively with what activities will be participated in when. Developing operational models which combine the what, when and where together would appear to be an on-going requirement, whether or not they are to be implemented within a microsimulation framework.
STATISTICAL PROPERTIES AND LINKAGES TO COMPLEX SYSTEMS THEORY Random elements enter virtually all microsimulation models in multiple ways. Synthesized base populations are generally randomly drawn from the known aggregate information. Updating procedures typically really heavily on probabilistic models (simple or complex). And the behavioral "core" of the simulation also is more often than not probabilistic in nature. The result of these observations is that the output from a microsimulation run represents one "realization", randomly drawn from the overall distribution of possible outcomes, where this (unknown) distribution is described by the simulation model as a whole.'' In classical simulation modeling, one would perform multiple runs or "replications" of the model for any given set of inputs in order to trace out the distribution of outcomes and hence to estimate both the expected values of these outcomes and the extent of the variation which is likely to occur around these expected values. While multiple model runs can and should be undertaken as we experiment with and test our models, it is not clear that extensive replications can be performed on a routine, operational basis with larger microsimulation models, given the very high computational burden which this would imply. If replications and associated averaging of results are not performed, however, we must be clear about what exactly it is that our models are generating. At best (and hopefully this is, indeed, the case) the model is generating a "realistic" or "reasonable" or "descriptively representative" feasible outcome, the statistical properties of which, however, are generally unknown. This is a perfectly acceptable situation, as long as we explicitly recognize it as such, and particularly given it is probably better and certainly no worse than the results obtained from conventional models. In addition, in large urban systems within which we are simulating the evolution of and decisions
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
547
by thousands if not millions of people, much of the "averaging out" will occur within the microsimulation itself as the demographic processes, activity decisions, etc. are computed person by person, household by household, thousands of times per time step: if we build our models correctly we should get the "right" number of births per year, the "right" number of shopping trips, etc. Experimentation and testing are, of course, required to establish the validity of this proposition, but it is certainly a reasonable one upon which to proceed. Complex systems theory is a branch of applied mathematics, with applications in an extremely wide variety of physical and social systems. Much of complex systems theory is concerned with the stability of dynamic systems, and, as such, should provide considerable insight into questions concerning the stability/representativeness of microsimulation model results. Among the many aspects of complex systems theory, two noteworthy components include chaos theory and adaptive behavior theory. 1. Chaotic Systems: Many systems are extremely sensitive to their initial conditions, so that very small differences in inputs can lead to very large differences in results. This is particularly worrisome for microsimulation given its heavy dependence on synthesized (or sampled) input data. The field of complex systems theory that deals with this problem is known as chaos theory. The product of chaos is often illustrated by an example known as the "butterfly effect". In this example, it is argued that a butterfly flapping its wings in Brazil can trigger a snowstorm in Alaska. Chaos theory has also been a concern in movies about time travel (such as "Back to the Future") — if you go back in time and change something very small — it is demonstrated that a cascading chain of events will render a significantly different outcome. In some movies, a small change in the past was able to save the planet from destruction! While chaos theory has been an entertaining part of many movies, it is not clear that the real world behaves this way. Furthermore, even if some modeling events exhibit chaotic behavior, there is some optimism that the system itself will remain stable. 2. Adaptive Behavior: Modeling decision-making behavior in a microsimulation is a difficult task. For one thing, different decisions are made at different "levels". Leaving a job, for example, is usually made by an individual; that is, it is made at the person level. The decision to relocate a household, however, is probably made at the household level (i.e. it is made jointly by more than one person within the household). There are other decisionmakers within a microsimulation model: firms, for example, must make decisions about relocating, expanding or reducing the work-force, etc. These decisions are made by a firm. Finally, some decisions might be made jointly by the firm and the person. In finding a job, it is important the that person wants the job and the firm wants the person.
548
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
The decision-makers in a microsimulation should exhibit adaptive behavior. Changes to the decision-maker's environment should resuh in a re-evaluation of relevant decisions. For example, if a firm decides to relocate across town, the employee may choose to relocate as well or find a new job that is closer to home. Microsimulations are ideal for implementing adaptive behavior since actors in the system are modeled individually. By modeling actors individually, it is possible to give each actor its own rules for behavior. Microsimulation involves actors that interact with their environment. These interactions occur in a variety of locations throughout time. Modeling behavior and decision-making requires that sufficient information be available to the decision-making process. The simplest decision-making methods require only information about the current state of the object. For example, the decision to move might be based on "current values" such as present utility, distance to work, etc. Some decisions, however, require that the aspect of history be taken into account. If a couple gets divorced, they go back into the marriage pool; it is important to remember that these two were previously married. While it is not impossible that these two will get back together, their previous relationship shouldn't be unknown to the behavioral process. Finally, the most complex behavioral systems exhibit learning behavior. These system get "smarter" overtime and are unlikely to repeat the same mistake twice. Microsimulation models should be constructed so that they independent of the type of behavioral model in use. In this regard, they should be capable of storing information over time that will help decision-makers with future decisions. In this sense, the decision-maker is provided a "memory" of past events.
COMPUTATIONAL ISSUES Computationally, microsimulation models are quite expensive. Constant leaps in hardware capabilities, however, give promise to the tractability of these problems. Within the next year, hard disk storage capability is expected to increase ten-fold due to new advancements in near-field technology. Similarly, processor speeds are scheduled to double and multi-processor workstations are becoming mainstream cost-effective. There is a reasonable confidence that these systems will make microsimulation an affordable and practical method for performing travel behavior research and modeling. Computationally, it has already been demonstrated that microsimulation models are practical. CORSIM, for example, a microsimulation model developed at Cornell University, showed that it took approximately three
Activity-Based Travel Behavior Modeling in a Micro simulation Framework
549
and a half hours to execute a ten year microsimulation with a population of 180,000 on a 1 OOMhz Pentium computer [Caldwell, 1996]. Ultimately, the computational feasibility of microsimulation depends in no small measure on resolution of some of the statistical/stability issues raised above. In particular, if extensive replications of model runs per test are required to generate credible results, the computational feasibility of many of the more complex microsimulation applications may well be in doubt, despite ongoing improvements in computer performance. Microsimulation models have been programmed in many languages, including Fortran, C and even SAS. As discussed at length in the next section, we believe a strong case exists for the development of microsimulation models within an object-oriented framework using a languages such as C++ and Java. As microsimulation models evolve, the need will also arise to go beyond standard programming languages to the development of higher-level, special purpose languages or "toolkits" for writing microsimulation models that will eliminate the need to develop program code "from scratch". Such toolkits or high-level languages not only greatly increase the efficiency and reliability of model development, but also open up the model development process to a much wider range of researchers possessing a diversity of computer programming skills. This has already occurred in the case of physical simulation modeling, in which a large number of specialpurpose languages are commercially available. It is also starting to occur within the social sciences, where the development of microsimulation "toolkits" is an area of very active research and development. We fundamentally view the microsimulation model and its computational environment (including the model itself, associated databases, ancillary analysis and display software, etc.) as the "laboratory" within which we work. Within this laboratory we conduct many different kinds of experiments. In particular, we experiment with alternative model and sub-model specifications and functional forms, we test alternative hypotheses about human social and economic behavior, and we examine the likely impacts of alternative policies and scenarios of interest to us as both students of and planners for urban areas. In order to turn this ideal of the computer as laboratory into a working reality places several requirements on the modeling software. These include: 1. The model must be highly modular so that sub-models can be altered or even replaced easily, without interfering with the performance of other components of the model. 2. The model must be highly "parameterized" so that the user can readily alter parameters, change defaults, mix and match assumptions, etc. 3. The model should be capable of maintaining at least minimal levels of internal consistency, so that as assumptions, etc. are changed within one part of the model these changes do not invalidate assumptions, etc. in other parts of the model. This ultimately implies the need for knowledge-based artificial intelligence to be built into the model. 4. The model must interface efficiently and effectively with a database management system, a
550
5. 6.
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges GIS, and other software needed for the management, analysis and display of the complex data structures associated with the problem. The model must use modem graphical user interface (GUI) technology to facilitate user interaction. The software architecture must be open-ended to permit the model to grow and evolve over time, as experience, new data, new theories/models and new applications dictate.
OBJECT-ORIENTED APPROACH TO SIMULATIONS^ The purpose of dynamic microsimulation is to model the behavior of actors, or objects, in the real world. In a microsimulation, these objects are persons, firms, households, jobs, autos, dwelling units, etc. The real world consists of these objects evolving and interacting over time. Object-oriented systems employ object-oriented analysis, design, and programming techniques. These techniques were specifically developed to handle complex problems such as large-scale microsimulation. This section will describe the benefits of object-oriented techniques over other methods. In an object-oriented system, the program consists of many objects that have their own state and behavior. There is a one-to-one mapping between objects in the real world and objects in the simulated world. The conceptual benefits of the one-to-one mapping between the real world and objects in an object-oriented system should be clear. Every object in the real world is represented by a similar object in the simulated world - these objects are known as abstractions of their real-world counterparts. The behavior of objects in an object-oriented program is given by a set of methods or member functions. There is a member function corresponding to every behavior that the object exhibits. For example, the decision to change jobs is a behavior that is part of the person object. This decision-making procedure is a member function of the person class. Traditional programming methodologies (such asfiinctionalor procedural programming) are built on data flow diagrams and are not well-suited for microsimulation applications. These methods use an approach known as top-down design with step-wise refinement and require that the program begin with a single problem statement that can be refined over time. Many complex real-world systems, however, do not consist of a single abstract problem - rather, they are comprised of a set of objects that interact in complex ways over time. Object-oriented techniques provide a solution that is conceptually cleaner and easier for non-computer scientists to understand and validate. By matching real-world objects with their
Activity-Based Travel Behavior Modeling in a Micro simulation Framework
551
synthetic counterparts, researchers can focus on understanding the problem rather than programming the solution. When dealing with a system as complex as a microsimulation of events in the real world, any technique that will help to reduce this inherent complexity will surely benefit the project. Object-oriented systems are generally better equipped than procedural systems in solving problems where there is significant complexity in the problem domain. To give an example, we can think of the individual persons in the system. These persons have certain behaviors that can easily be described. In addition, there is information about each person that must be stored ~ this information might include age, gender, education level, marital status, job, etc. Object-oriented systems allow us to encapsulate behavior and data together in the object. The object's behavior is responsible for changing its state. For example, a person will have the behavior that he/she ages over time. This aging process is a behavior that is programmed into the person object that updates the data corresponding to the individual's age. Other behaviors are responsible for moving, finding jobs, getting married, forming households, etc. In addition to the conceptual clarity of object-oriented systems, there are other benefits to object-oriented systems. For example, there has traditionally been a problem in computer science that programmers and users lack each other's expertise. In addition to lacking expertise in their respective areas, programmers and users have traditionally spoken different languages which is sometimes referred to as an "impedance mismatch" between programmers and users. With traditional programming methodologies, the analysis and design of a problem are done in the language of the solution domain rather than the language of the problem domain. In contrast, object-oriented analysis and design can be performed in the language of the problem domain which is easier for users to understand. The result is often a higher quality product since there is less chance of a misunderstanding between the conceptualizers of the system and its implementers. Another advantage of the object-oriented approach to microsimulation is the ability for reusing objects that are created as part of the development of the system. In procedural systems, the architecture or structure of the program is centered around the problem that the particular program is trying to solve. In order to write a program to solve a different problem, the program must be rewritten since the original design has changed. Object-oriented systems, however, are built starting with the objects that are part of the system. These objects are much less likely to change over time than what we are trying to model with them. It is hard to image a travel behavior system that will not involve people, autos, and a transportation network! By designing these objects in an object-oriented manner, these objects can be reused in the construction of new modeling systems. Object-oriented development does not follow the traditional waterfall approach of conventional systems. Instead, the development of an object-oriented system is incremental and iterative. Users
552
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
can start with a very simple model and evolve it slowly over time to become more and more complicated. This evolution of the model is not possible in a traditional system as it requires frequent changes to the design of the system. The iterative approach is particularly attractive to researchers who want a system up and running quickly that can be made more accurate over time as various behavioral components are researched and implemented. Support for class inheritance is another significant advantage of object-oriented systems. With inheritance, it is easy to create a class that defines the contractual obligations of its subclasses. For example, once a dwelling unit class has been defined, we can create many kinds of dwelling units as its subclasses. Each dwelling unit subclass will have some unique properties and behaviors, while sharing others with its overall class. The object-oriented programming paradigm has been in existence for at least thirty years. It is only recently, however, that object-oriented methods and languages have become pervasive in the programming community. The recent popularity of C++ and Java are just two examples of growing support for the object-oriented paradigm. In our own research, the ILUTE model has been designed "from the ground up" using object-oriented technology and respecting the state-of-the-art in object-oriented analysis and design techniques.^^ Many of the objects in the microsimulation represent real-world objects in the study area. This one-to-one mapping between real-world objects and programmed objects reduces some of the inherent complexity in the system's design. C++ was chosen as the language for the project given its familiarity to the researchers, excellent performance, and proven track record in the development of complex systems. C++ is a superset of C that includes object-oriented extensions (including some from Simula 67 - an earlier language used in simulation applications). Other object-oriented languages, such as Smalltalk and Java, were not chosen due to the run-time performance penalty that they incur.
IMPLEMENTATION OF ACTIVITY-BASED TRAVEL MODELS Several other chapters in this volume are concerned with activity-based travel models, and no attempt is made here to summarize the state-of-the-art in this area, or even the state of implementation of these models in microsimulation frameworks. General points to note, however, concerning such implementations include the following: 1. The "inputs" required by activity-based models must be made very clear to ensure that the microsimulator is generating these inputs adequately. These inputs include both the demographic and socio-economic attributes of the decision-makers and the "supply-side"
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
2.
3.
4.
553
information concerning activity locations and attributes (shopping locations and their attributes; job locations and their attributes), etc. As has already discussed, we have some experience with simulating the former, but we have relatively little experience with "supply-side" simulation ~ something which may, in the long run, prove to be the more difficult of the two problems. Turning the problem around, forseeable limitations in what "inputs" can be reliably simulated may impose constraints on activity model development. For example, if it is clear that we cannot simulate activities over space at a level finer that gross SIC code and that we cannot simulate labour force participation at a level finer than broad occupational categories, then developing an activity model which assumes a more detailed representation of occupation will be of little operational use. At the risk of oversimplifying, it can be argued that much of the progress in activity-based modeling has dealt with agenda formation (i.e., what activities do people participate in) and scheduling/sequencing of activities and trips (i.e., what activities actually get selected to be undertaken, when, and in what order), often with surprisingly little attention being paid to the spatial component of the choice process (i.e., where the selected activities will occur). Modeling either the temporal or the spatial component of the process is a sufficiently daunting task. Obviously the two must ultimately be brought together in a unified spatial-temporal model of activity and travel choice. Some discussion has occurred in this chapter concerning the need to better understand the "feedback" (which can consist of both "lags" and "leads") between "short-run" activity/travel behavior and longer-run decisions concerning residential and employment location, auto ownership, etc. A similar argument exists for better integration between activity/travel choice models and network assignment models. With the notable but very limited example of dynamic assignment models which endogenously estimate trip departure times (given spatially fixed vehicle O-D trip matrices as inputs), these two types of models tend to be treated as very separable problems. Given the complexity of both model components, this is an understandable, pragmatic approach. It is, however, ultimately a very limiting one. Activity scheduling obviously depends, among other factors, on travel times within the transportation system; these travel times, of course, depend upon the flow levels determined by these scheduling decisions in a classic supplydemand relationship which we have been wrestling with for forty-plus years in our conventional static equilibrium models and which is no less pertinent within a disaggregate, dynamic microsimulation environment. Another, and perhaps behaviorally more fundamental, way of putting this is that within a dynamic microsimulation model we are ultimately interested in tracing trip-makers' trajectories through time and space. Very simply, this means we want to know where people are at each point in time (and where they will be next in general is conditioned by
554
5.
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges where they are currently, among many other factors). Ultimately, the only way we can do this, and the only way we can model the activity/travel choices in a truly dynamic way is to integrate the activity/travel and route choice modeling components. It is perhaps useful to note with regard to this point that we have very slowly been succeeding in weaning ourselves from the old four-stage way of speaking/thinking with respect to the first three stages (i.e., generation, distribution and mode split) of the "process", but that we still seem to be (too) comfortable with maintaining both an operational and conceptual separation between the "demand" component of the problem (i.e., the activity/travel paradigm) and the "network" component (i.e., the route assignment problem). Although the word "household" has been used a few times within this chapter, the discussion has predominantly been couched in terms of the simulation of individual persons. This directly reflects the state of the art in activity/travel models, which, almost without exception, are formulated at the level of the individual person and the individual trip-maker. We firmly believe that this represents a serious mis-specification of the problem in that activity/travel decisions (as well as most longer-run decisions of interest to us such as residential location and auto ownership) fundamentally involved householdlevel interactions, constraints and decision-making. Adoption of a household-level analysis for decisions such as residential location and auto ownership is clearly an absolute necessity. We would argue strongly that it is just as essential to the proper understanding of activity scheduling, modal choice (i.e., who uses the household cars, when), tripchaining, etc. In our own research we have committed ourselves to a household level of analysis throughout our modeling system, particularly in terms of explicitly dealing with how the household context constrains and conditions individual decision-making. We would strongly urge others to consider, wherever possible, similarly adopting a householdbased approach to the problem.
All modeling exercises involve a complex interplay between available theory, data and "methods", where in the last term we include both the mathematical formulations used to operationalize conceptual theories and the computational procedures and capabilities available for implementing the math in computable form. Clearly one of the overriding questions concerning the microsimulation approach is "does it compute in reasonable time?". Growing experience (and expertise) with microsimulation, in combination with continuing dramatic increases in costeffective computing power is increasingly indicating that the answer to this question is likely to be "yes", at least as long as we are somewhat "sensible" in terms of the level of detail we try to achieve in spatial, temporal and behavioral terms. This, of course, simply reformulates the question in terms of what is a "sensible" level of model design. On the one hand, there is an "internal logic" of microsimulation which, once one adopts
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
555
the approach, tends to drive one towards seemingly ever-increasing levels of disaggregation. This tendency is evident throughout this chapter, in which the "need" has been rather blithely postulated for everything from dynamic models of learning, to integration of virtually every conceivable component of the problem, to synthesis of a potentially huge set of "initial values" (including, it has been suggested, perhaps "memories/histories" of events which occurred prior to the simulation time period). This "infinite regression" into finer and finer levels of disaggregation, however, is tempered at any point in time by practical constraints defined by current theory, data and methods, which define what is operationally feasible. All models are abstractions and simplifications of reality (even, despite seeming occasional pretensions otherwise, microsimulation models), which, among other things, means that they all operate at some level of spatial, temporal and behavioral aggregation: if we adopt a one-year time step, then we will inevitably be aggregating over intra-year "microdynamics"; if we define occupations on a very broad category basis, we may end up treating doctors, lawyers and even travel demand modelers as "being the same" in terms of their employment opportunities, residential preferences, etc. The basic microsimulation modeling research and design issue is to find the "right" balance between disaggregation and aggregation, between precision and accuracy, between "behavioral realism" and computational feasibility.
ACKNOWLEDGEMENTS Research associated with this chapter has been funded by a Natural Sciences and Engineering Research Council (Canada) Collaborative Project Grant. The authors gratefully acknowledge the helpful comments of two anonymous reviewers of this chapter.
REFERENCES Anderson, W. P., P. S. Kanaroglou and E. J. Miller (1996). "Measuring External Effects of Transport with a Land Use Approach", paper presented at the Neuviemes Entretiens du Centre Jacques Cartier, Montreal: October. Ash, R. B. (1965). Information Theory, New York: Dover Publications. Axhausen, K. (1990). "A Simultaneous Simulation of Activity Chains and Traffic Flow", in Jones, P. (ed) Developments in Dynamic and Activity-Based Approaches to Travel Analysis, Aldershot: Avebury, pp. 206-225. Barrett, C, K. Berkbigler, L. Smith, V. Loose, R. Beckman, J. Davis, D. Roberts and M. Williams (1995). An Operational Description ofTRANSIMS, LA-UR-95-2393, Los Alamos, New Mexico: Los Alamos National Laboratory.
556
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Beckman, R. J., K. A. Baggerly and M. D. McKay (1995).] "Creating Synthetic Baseline Populations", paper submitted to Transportation Research, LA-UR-95-1985, Los Alamos, New Mexico: Los Alamos National Laboratory. Booch, G. (1994). Object-Oriented Analysis and Design with Applications, Second Edition, Redwood Ca.: The Benjamin/Cummings Publishing Co. Booch, G. (1996). Object Solutions, Managing the Object-Oriented Project, Reading, Mass.: Addison-Wesley Publishing Co. Caldwell, S. B. (1996). "Dynamic Microsimulation and the CORSIM 3.0 Model", Ithaca, N.Y.: Institute for Public Affairs, Department of Sociology, Cornell University. Citro, C. F. and E. A. Hanushek (eds.) (1991). The Uses ofMicrosimulation Modelling, Volume I, Review and Recommendations, Washington, D.C.: National Academy Press. Chapleau, R. (1986). Transit Network Analysis and Evaluation with a Totally Disaggegate Approach, Publication #462, Montreal: Universite de Montreal, Centre de recerche sur les transports, April. Chung, J. H. and K. G. Goulias (1997). "Travel Demand Forecasting Using Microsimulation: Initial Results from a Case Study in Pennsylvania", Transportation Research Record 1607, pp.24-30. Goulias, K. G. and R. Kitamura (1992). "Travel Demand Forecasting with Dynamic Microsimulation", Transportation Research Record 1357, pp. 8-17. Goulias, K. G. and R. Kitamura (1996). "A Dynamic Model System for Regional Travel Demand Forecasting", chapter 13 in Golob, T., R. Kitamura and L. Long (eds.) Panels for Transportation Planning: Methods and Applications, Kluwer Academic Publishers, pp 321-348. Harding, A. (ed.) (1996). Microsimulation and Public Policy, Amsterdam: Elsevier. Harvey, G. and E. Deakin (1996). Description of the Step Analysis Package, draft manuscript: Hillsborough, N.H.: Deakin/Harvey/Skabardonis. 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, pp. 46-56. Khoshafian, S. and R. Abnous (1995). Object Orientation, Second Edition, New York: John Wiley & Sons. Kitamura, R. and K. G. Goulias (1991). MIDAS: A Travel Demand Forecasting Tool Based on a Dynamic Model System ofHousehold Demographics and Mobility, Projectbureau Integrale Verkeer-en Vervoerstudies, Ministerie van Verkeer en Waterstaat, the Netherlands. Lambert, S., R. Percival, D. Schofield and S. Paul (1994). "An Introduction to STINMOD: A Static Microsimulation Model", STINMOD Technical Paper No. 1, Canberra: National Centre for Social and Economic Modelling, Faculty of Management, University of Canberra. Mackett, R. L. (1990). "Exploratory Analysis of Long-Term Travel Demand and Policy Impacts Using Micro-Analytical Simulation", in Jones, P. (ed.) Developments in Dynamic and Activity-Based Approaches to Travel Analysis, Aldershot: Avebury, pp. 384-405. 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, PTRC, Seminar H: Transportation Planning Methods: Volume II, pp. 53-64. Miller, E. J. (1996). "Microsimulation and Activity-Based Forecasting" paper presented at the TMIP Conference on Activity-Based Travel Forecasting, New Orleans: June. Miller, E. J., P. J. Noehammer and D. R. Ross (1987). "A Micro-Simulation Model of Residential Mobility", Proc. of the International Symposium on Transport, Communications and
Activity-Based Travel Behavior Modeling in a Microsimulation Framework
557
Urban Form, Vol 2: Analytical Techniques and Case Studies, pp. 217-234. Miller, E. J. and P. S. Salvini (1997). "The Design and Evolution of an ILUTE Dynamic Microsimulation Framework", paper presented at the 77^*^ Annual Meeting of the Transportation Research Board, Washington, D.C., January. Orcutt, G. (1957). "A New Type of Socio-Economic System", Review of Economics and Statistics, Vol.. 58, pp. 773-797. Orcutt, G. (1976). Policy Evaluation through Discrete Microsimulation (2nd edition), Washington, D.C.: Brookings Institute. Orcutt, G., M. Greenberg, J. Korbel and A. Rivlin (1961). Microanalysis of Socioeconomic Systems: A Simulation Study, New York: Harper and Row. Orcutt, G., J. Merz, and H. Quinke (1986). Microanalytic Simulation Models to Support Social and Financial Policy, New York: New-Holland. Oskamp, A. (1995). "LocSim: A Microsimulation Approach to Household and Housing Market Modelling", paper presented to the 1995 Annual Meeting of the American Association of Geographers, Chicago, March 15-18, PDOD Paper No. 29, Amsterdam: Department of Planning and Demography, AME - Amsterdam Study Centre for the Metropolitan Environment, University of Amsterdam. RDC, Inc. {\995). Activity'Based Modeling System for Travel Demand Forecasting, DOT-T-9602, Washington, D.C.: U.S. Department of Transportation. Shannon, C. E. (1948). "A Mathematical Theory of Communication", Bell System Technical Journal, Vol. 27, pp. 379-423, 623-656. Reprinted in C. E. Shannon and W. Weaver (eds.). The Mathematical Theory of Communication, Urbana, 111.: University of Illinois Press, 1949. Snickars, F. and J. W. WeibuU (1977). "A Minimum Information Principle: Theory and Practice", Regional Science and Urban Economics, Vol. 7, Nos. 1/2, pp. 137-168. Spear, B. D. (1994). New Approaches to Travel Demand Forecasting Models, A Synthesis of Four Research Reports, DOT-T-94-15, Washington, D.C.: U.S. Department of Transportation. Taylor, D. A. (1990). Object-Oriented Technology: A Manager's Guide, Reading, Mass.: Addision-Wesley Publishing Co. Troitzsch, K. G., U. Mueller, G. N. Gilbert and J, E. Doran (1996). Social Science Microsimulation, Heidelberg: Springer Verlag. Webber, M. (1977). "Pedagogy Again: What is Entropy?", Annals ofthe Association of American GEographers, Vol. 67, No. 2, pp. 254-266 Wegener, M. (1994). "Operational Urban Models: State of the Art", Journal of the American Planning Association, Vol. 60, No. 1, Winter, pp. 17-28. Wilson, A. G. (1967). "A Statistical Theory of Spatial Distribution Models", Transportation Research;Vo\. 1, pp. 253-269. Wilson, A. G. (1970). Entropy in Urban and Regional Modelling, London: Pion. Wilson, A.G. and C. E. Pownall (1976). "A New Representation of the Urban System for Modelling and for the Study of Micro-level Interdependence", Area, Vol. 8, pp. 246-254. Van Aerde, M. and S. Yager (1988a). "Dynamic Integrated Freeway/Traffic Signal Networks: Problems and Proposed Solutions", Transportation Research A, Vol. 22A, No. 6, pp. 435443. Van Aerde, M. and S. Yager (1988b). "Dynamic Integrated Freeway/Traffic Signal Networks: A Routing-Based Modeling Approach", Transportation Research A, Vol. 22A, No. 6, pp. 445-453.
558
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
^ For example, the TMIP Conference on Activity-Based Travel Forecasting in New Orleans, June 1996, as well as recent TRB Annual Meetings. ^ For example, STEP [Harvey and Deakin, 1996], MIDAS [Goulias and Kitamura, 1992, 1996], MIDAS/USA [Chung and Goulias, 1997], and AMOS [RDC, 1995]. Proposed designs for larger-scale microsimulation systems with embedded activity-based travel models include SAMS [RDC, Inc., documented in Spear, 1994], SMART [Lousiana Transportation Research Center, also documented in Spear, 1994] and ILUTE [Anderson, et ai, 1996; Miller and Salvini, 1997]. For a summary of European experience with microsimulation, see Axhausen [1990]. ^ Among the more notable of these are TRANSIMS [Barrett, et al. 1995], DYNASM ART [Mahmassani, et aL, 1994; Hu and Mahmassani, 1995], INTEGRATION [Van Aerde and Yager [1988a, 1988b], and, with respect to transit assignment, MADITUC [Chapleau, 1986]. ^ See, for example, Mackett [1990], Goulias and Kitamura [1992, 1996], and Miller [1996]. ^Lambert, e/a/. [1994]. ^ Public Use Microdata Sample (PUMS) files consist of 5% representative samples of "almost complete" U.S. census records for collections of census tracts. ^ For discussions of information theory see, among many others. Shannon [1948], Ash [1965], Wilson [1967, 1970], Webber [1977], Snickars and Weibull [1977]. ^ For a review of microsimulation applications in land-use modeling, see Wegener [1994]. ^ See, for example, Lambert, et al. [1994], Caldwell [1996], and Troitzsch [1996]. In addition, a useful newsgroup dedicated to microsimulation in the social sciences is found at: [email protected]. 10
Among many others. For a discussion of applications, see, for example, Citro and Hanushek [1991].
^^ Microsimulation is, however, also used to calculate system steady states. A common example of this is a dynamic microsimulation network assignment model which is iterated until convergence to a steady-state solution is obtained. ^^ For an extremely accessible introduction to object-oriented programming concepts, see Taylor [1990]. ^^ See, for example, Booch [1994, 1996] or Khoshafian and Abnous [1995].
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
27
COMPLEXITY AND ACTIVITY-BASED TRAVEL ANALYSIS AND MODELING
Pia M. Koskenoja and Eric L Pas
ABSTRACT Urban travel modeling and forecasting is undergoing a major revolution, primarily in response to new planning and policy analysis needs'. These planning and policy analysis needs are generally motivated by concerns about air quality, sustainability and congestion. The current paradigm shift in urban travel modeling is bringing about a change in the overall approach, as well as in specific modeling techniques. In an effort to support the current efforts to develop a new generation of travel forecasting methods, this chapter presents the results of a selective examination of a collection of concepts, approaches and methods generally known as the "complexity approach". The chapter identifies areas in which these concepts and methods might be useful to travel modelers seeking approaches and methods to implement the new directions in travel forecasting that are currently emerging. The chapter concludes that the complexity approach has useful insights and methods to offer those developing new travel forecasting methods, but that there remain important issues to be addressed.
INTRODUCTION AND BACKGROUND Mathematical models have been used since the mid-1950's to forecast future urban travel patterns in support of urban transportation planning and policy analysis. The planning and policy analysis environment has changed considerably over the years, and so have the models. However, the models used in practice have largely retained the framework that was developed in the earliest land-use/transportation studies conducted in urban areas in the USA about 40 years ago. The transportation planning and policy analysis context has changed considerably in the 1990's, leading to major changes in the demands placed on the urban transportation
560
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
models. Changes in transportation planning and policy objectives have resulted primarily from concerns about air quality and sustainability, and to a lesser extent congestion. The new planning and policy analysis environment requires models that can provide more accurate and precise estimates, than the conventional forecasting models, of the travel and related impacts of changes in the land-use and transportation systems, and socio-demographic characteristics. Most importantly, the models need to be able to deal with plans and policies aimed at reducing travel, especially peak-period travel in single occupant automobiles, and they need to provide estimates of travel that are suitable for input to emerging air quality models. As a result, the models need to provide a far richer description of the behaviour being modeled than is possible with the conventional trip-based framework, and there is a need to be able to represent travel with much greater temporal and spatial precision. Urban travel modeling and forecasting is currently undergoing a major change, a series of paradigm shifts, in response to the needs resulting from the new planning and policy analysis environment. This is not to imply that all the ideas currently being advanced are new. In fact, many of these ideas have been around for quite some time. However, current planning and policy analysis issues have motivated and expedited the implementation of ideas and methods that have been in the research environment for quite some time. For example, the activitybased approach to travel analysis and modeling has long been advocated by researchers who believed that the conventional, trip-based framework was inadequate to properly represent urban travel behaviour, and especially inadequate to predict changes in behaviour resulting from changes in the travel context. The changed planning and policy environment has brought about a renewed interest in recent years in the activity-based approach among both researchers and practitioners, and a considerable quantity of recent work aims at the development of forecasting models based on this approach and employs a variety of methodologies (Pas, 1997). While the activity-based approach was, in some sense, developed before its time, other ideas currently being advanced in urban travel modeling and forecasting are relatively new to the field. For example, the idea of microsimulation modeling is relatively new in urban travel modeling and forecasting, and it represents a major change in the overall modeling methodology (see Miller, 1997 for a recent overview of the use of microsimulation in travel forecasting in USA, and Algers et al. 1997 for a review of European microsimulation models). Similarly, the use of computational process models is a rather recent development in travel forecasting (for two recent examples of the application of computational process models in travel behaviour, see Kitamura, 1997 and Kitamura and Fujii, 1998).
Complexity and Activity-Based Travel Analysis and Modeling
561
Almost 10 years ago, a conference on "Dynamic and activity-based approaches to travel analysis" was held at Oxford University. Many ideas and buzzwords can be found throughout the papers that were presented at that conference, concepts such as: the interaction of microlevel and macro-level processes (Deloucas 1990), micro-simulation (Mackett 1990), possible multiple equilibria, path dependency, speed and of adjustment, boundedly rational behaviour, search strategies, threshold levels, asymmetries, capability and institutional constraints, habits, and inertia (Goodwin et al. 1990). In the summary of the proceedings, the conference organizer, Peter Jones, voices an underlying theme of the meeting: each of the separate presentations was looking for a way to integrate the several aspects of the research topic. In a more prophetic way than he may have intended, he suggests that the researchers should not be apologetic about the complexity of the research area, but instead say: "Yes, it is complex. Isn't it good that we have a way of being able to represent, handle and work with complexity!" (Jones 1990, p. 464). In recent years, there has been considerable development, in a variety of fields, of ideas and methodologies that collectively address many of the issues discussed at the 1988 Oxford conference on dynamic and activity-based approaches in travel analysis. These ideas and methodologies are often referred to as the science of complexity. The purpose of the present chapter is to examine potential applications of the complexity approach in travel modeling, especially extensions to activity-based travel analysis and modeling. Complexity has been described as a science of emergent order, which arises is system states between rigid structure and chaos (Waldrop 1992). Complexity deals with non-linear, nested structures, which lead to unexpected higher level behaviours. These macro behaviours are not predictable with the regular forecasting methods, which tend to assume that the whole is a sum of its parts or a transformed fiinction of the sum, and which are often designed to give a stable equilibrium answer. Themes running through the complexity approach include: dynamic, nonlinear self-organization from bottom up, multiple equilibria, explicit recognition of multilevel phenomena with linkages within and between aggregation levels, and emphasis on local and boundedly rational decision making. Mathematically, complexity refers to non-linear systems that are not reaching a stable equilibrium, but are neither totally chaotic. Complex systems are characterized as systems "at the edge of chaos" where the aperiodic systems show "almost periodic" behaviour, even when the evolution path does not repeat itself exactly in a phase diagram (Kaufman, 1995). This lifeimitating property of complex systems has generated a lot of interest in these systems and a
562
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
body of knowledge is growing about the mathematical and statistical properties of complex systems and their implications to applications in social sciences 1. "Complexity approach" is a colloquial term. It does not refer to a unified theory, but rather a collection of new ideas that support and draw power from each other, and which collectively form a way to see the world. Complexity has been criticized based on the approach's assertion that complex behaviour arises from a small set of simple rules, which cannot be deduced from the observed behaviour. Thus, the suspicion that the various phenomena claimed to be "complex" have in reality very little in common but an empty buzzword (Horgan, 1995). However, the approach has shown its practical applicability in a variety of sciences. Just to name some examples from fields related to activity-based travel analysis and modeling , complexity has been utilized in operational research (Reyck and Herroelen, 1996), decision sciences (Norman and Shimer, 1994), management (Langlois and Everett, 1992), finance (Gilmore, 1993), economics (Brock, 1991; Dopfer, 1991; Norman, 1994), environmental modeling (Young et al., 1996), and regional science (Batten and Karlsson, 1996; Maier, 1995), and urban planning (Wu, 1997)2. The science of complexity started in less applied sciences such as mathematics and philosophy (Prigogine, 1963; 1976), physics (Davies, 1989), and computer science, and had its first applications in genetic algorithms and classifier systems (Holland, 1975; Holland et al., 1986), which were applied to biology (Perelson, 1988; Perelson and Kauffman, 1990), and economics (Anderson et al. 1987; Arthur 1990). To this day these ideas have not been explicitly applied to activity-based approach to travel demand analysis and modeling. What follows is an overview of popular modeling methods used in complexity research, and applications in economics that, by the subjective judgment of the authors, seem as most promising for extending activity-based travel research. In the third section we discuss some of the possibilities complexity approach has opened in economics. We hope that the similarities of problems in these examples, which were overcome with the complexity approach, will spark interest in complexity approach among activity-based travel analysis and modeling researchers. In the fourth section of the chapter we discuss some specific areas in which the concepts and methods of the complexity approach could contribute to the next generation of urban travel forecasting models. In the final section we summarize the chapter and draw some conclusions.
Complexity and Activity-Based Travel A nalysis and Modeling MODELING METHODS: EVOLUTIONARY COMPUTATION AND BASED
563
AGENT-
MODELING
"Evolutionary computation" is a term that refers to all computer algorithms that take their metaphors from biology. The purpose of this section is to present some popular evolutionary modeling methods that seem potentially useful for activity-based travel analysis and modeling. Cellular automata and genetic algorithms are often used as building blocks in agent-based modeling. Neural networks are related to cellular automata and have already been used in several applications in transportation. Evolutionary programming and evolution strategies are often used in dynamic modeling and are briefly discussed here to point out the basic similarities and differences between these methods and genetic algorithms. Classifier systems, or systems of adaptive agents, are the basis of artificial life simulations and agent-based modeling. A good guide to evolutionary computation is "The Hitch-Hiker's Guide to Evolutionary Computation (FAQ for comp.ai.genetic)", started by David Beasley and maintained by Jorg Heitkotter at [1]. The guide has an appendix, "Encore", which is also available from the same site.
Cellular Automata Cellular automata are decentralized, spatially extended systems consisting of large numbers of simple identical components with local connectivity (Mitchell, 1996). In a simple case, cellular automata is a one-dimensional lattice of length L of identical cells, each of which can be in one or two possible states (e.g. black or white). A collection of all the cells and their particular values is called configuration. The lattice is updated in discrete time steps, where all cells update simultaneously. Radius, r, is the number of neighbors left and right that is used in the update rule.3 With a deterministic update rule the system uses the current state of the cell and its neighbors as input. The output is determined from a lookup table which gives the new state for each of the possible neighborhood configurations a cell can be in. Since there are a finite number of entries in the lookup table, the total number of possible update rules is also finite. Starting from a random initial configuration, iterating the lattice using one of the possible update rules, and plotting the configurations over time, gives rise to a space-time diagram which can display complicated patterns. A good source to find more about cellular automata is the "FAQ About Cellular Automata", maintained by Howard Gutowitz at [2]. Nagel et al. (1996) describe a large-scale transportation simulation based on cellular automata developed in Los Alamos.
564
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Other papers describing that research effort are Nagel and Barrett (1996), Rickert and Wagner (1996), and Rickert and Nagel (1997).
Networks Neural networks are based on an allegory of a brain4. A simple neural network consists of cells and connecting links. Each cell computes a weighted sum of its input-signals from other cells, and outputs either a signal or no signal. This new signal becomes the input of other cells. Neural networks are robust computers of incomplete information, because one malfunctioning cell does not usually change the system wide configuration. Neural networks are selforganizing, but for a specific task they often require a hidden unit as a centralized coordinator, which undercuts the wish to model systems that self-organize from bottom-up. Neural networks have been used for modeling transportation related choices instead of the more conventional discrete choice models (McNally and Lo, 1992). Another application of neural networks to transportation is AMOS (Kitamura et. al. 1996), where connectionist networks are used to model the changes of activity options an individual may conceive as a result of a change in travel environment. Transportation applications have been done also with the spinglass metaphor, and they have been used in search algorithms for planning problems like traveling salesman. Weigend et al. (1991) uses networks to produce short term predictions in chaotic computational ecosystems. A good introduction to neural networks is Hertz et al. (1991).
Genetic Algorithms Genetic algorithms are based on an allegory of chromosomes in a cell. The actual form and behaviour of the organism is called phenotype, while the genetic encoding is referred to as genotype. Each phenotype can have fitness assigned to it. Fitness reflects the organism's ability to survive and reproduce. Evolution can be viewed as a process that searches a fitness landscape of possible genotypes, looking for genotypes that encode highly fit phenotypes. The set of possible solutions is called the representation space. A move operator (or combination of move operators) that indicates which points in the representation space are connected to each other defines a neighborhood relation. For instance, each point that can be reached from a particular point by applying the
Complexity and Activity-Based Travel A nalysis and Modeling
565
move operator exactly once is connected. Finally, a fitness function takes the encoding of a possible solution, converts it into the actual solution it represents, and returns a value that denotes how good the solution is for the given problem. Thus a fitness landscape is defined by the set {representation space, the neighborhood relation, fitness fiinction}. The fitness landscape can be characterized by the number of local optima, distribution of the (relative) heights and locations of local optima, the number of steps of the move operator it takes to reach a local optimum starting at random point in the landscape, and the "ruggedness" of the landscape (i.e. the average fitness differences between the neighboring points), which can be measured by correlation between the fitness values of neighboring points. The research in genetic algorithms has moved on from only developing effective search algorithms to study properties of systems that use these algorithms. Mitchell (1996) reviews some strands in cellular automata development and her group's work, which studies mechanisms of global coordination in cellular automata when only local communication is possible. She and her colleagues use a combination of cellular automata and genetic algorithms to find efficient rules for the cellular automata, and they study the higher order semantics that develop in such systems. These kind of systems have an analogy in activitybased approach to travel demand analysis and modeling, where each individual makes individual decisions that are reactions to her "neighbors", for instance family members, and at the same time acts as a part of a system that coordinates activity locations and timing successfully in a larger scale. In a recent paper Nimwegen et al. (1997) analytically model and identify a general mechanism that causes metastability in the evolutionary dynamics of a simple genetic algorithm. In particular, their model predicts periods of stasis in the fitness distributions, identifies their locations, enables the exact predictions of the metastable fitness distributions during the stasis and gives insights into the nature of the periods of stasis and the innovations between them. The analogy to activity-based approach to travel demand analysis and modeling would be predicting the stability of existing patterns and predicting the development of new patterns. Sadek et al. (1997) uses genetic algorithms for dynamic traffic assignment.
Evolutionary Programming and Evolution Strategies Evolutionary programming and evolution strategies are stochastic optimization strategies that place the emphasis on the behavioural linkage between parent and off-spring, rather than
566
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
seeking to emulate specific genetic operators as in genetic algorithmsS. Like genetic algorithms, evolutionary programming assumes that fitness landscape can be characterized in terms of variables, and that one or more optimum solutions exist. The procedure generates a random set of potential answers, the initial "population". The algorithm then computes the "fitness" for each of the potential answers, and retains a proportion of the more fit answers. Determining the fitness of a potential answer and its subsequent survival in the population is typically done by a stochastic tournament, but it can be done deterministically. Afterwards the surviving answers are replicated into a next generation of solutions by mutating them, and the survival among the offsprings is computed through another tournament. Evolutionary programming does typically not use crossover as genetic operator. Evolutionary strategy differs from evolutionary programming in two aspects: it typically uses deterministic selection process instead of randomized tournaments, and a recombination of several individuals as parents. The different usage of recombinations stem from the assumption that evolutionary programming emulates evolution at the level of populations of the species, while evolutionary strategy emulates evolution at the level of individual behaviour. Recombinations between different species are not reproducible by definition, while the individual can be a product of several parents. Unlike genetic algorithms, the evolutionary programming and strategy algorithms do not require binary string coding of the solutions, and can work with variable length, real valued vectors. The basic evolutionary strategy algorithms have a parent solution that generates one offspring per generation. The offspring is a mutation of the parent, generated as a draw from a multidimensional normal distribution centered at the parent solution, making sure that small mutations are more likely than large mutations. This process is repeated until an offspring performs better than the parent and takes the parents place. Later developments of evolutionary strategy algorithms include recombinations of several parent solutions to generate the offsprings, and allowing both parent and offspring populations compete at the fitness evaluation. Usually the algorithms that do not allow parents to compete find the good solutions faster, because they allow a temporary worsening of the solution and thus avoid being stuck in a local optimum.
Complexity and Activity-Based Travel Analysis and Modeling
567
Classifier Systems Classifier systems are models of cognition, used in artificial life simulations. In the simulation the agent classifies the events in its environment and reacts to them accordingly. The requirements for such system are: 1) an environment, 2) receptors to collect information, 3) effectors to manipulate the environment or the agent's own position in the environment, and 4) the agent itself, who has 2) and 3) and lives in 1). A classifier system consists of a set of {if...then} rules — "classifiers"-- which are represented in an easily manipulated form, for instance as binary stings. The set of strings is a population of classifiers. The basic cycle goes as follows: receptors (2) generate binary messages into a message list within the agent. The messages are matched with the classifiers to list the triggered actions, another list of binary messages. The old messages are replaced by the new messages on the message list. Then, the effectors (3) check the message list and carry out the actions, and the cycle repeats. In the activity-based travel microsimulation literature, classifier systems are called "production system models" or "computational process models". Such classifier systems are life-imitating, but they do not learn nor innovate. Furthermore, they have to have "installed" conflict-resolution strategies to choose between rules, if there are several that match one message on the list. To overcome these major limitations a classifier system can be made to innovate by applying a mutating algorithm (e.g. genetic algorithm) to the rules. The algorithm generates new classifiers by removing, adding, or combining condition/action -parts of the existing classifiers. The classifier system can be made to learn by reinforcing the successful rules. Holland suggested a "fire brigade" rule reinforcement system, where each rule that was active to produce a successful move, would get more relative weight. At the conflict situation, where several rules would apply, the rule with a higher weight is applied. Over time, a rule that is often active during successful moves gets more credibility, and the system "learns" to use the more credible rules, and forms an analogy of a mental map or a default hierarchy or rules. Such a system can start as "tabula rasa" with a random set of classifiers and learn with trial and error to generate an effectively functioning set of rules.
568
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Agent-Based Modeling Agent-based modeling uses the principles of artificial life in the sense that it creates both the agents, that the traditional micro-analysis is concerned about, and the macro-level environment, that each of the agents participate to create, but simulatneously take as a given. The system becomes interesting when it contains several nested agents. In activity-based travel analysis and modeling the obvious nests of interest are individuals within a family, and individuals within companies or other organizations, or individuals within cars within traffic systems. Tesfatsion (1997) lists several extensions of traditional modeling that economists have been able to do with the artificial life paradigm: 1) Much greater attention to the endogenous determination of agent interactions. 2) A broader range of interactions can be simultaneously considered. 3) Agent actions and interactions are represented at a greater degree of abstraction, permitting generalizations across specific system applications. 4) The evolutionary process is generally expressed by means of genetic operations acting directly on agent characteristics. These evolutionary selection pressures result in the continual creation of new modes of behaviour and an ever-changing network of agent interactions. Agent-based modeling could be a truly long-term project if the programs would have to be developed from scratch. Fortunately, this does not have to be the case. Santa Fe histitute develops and supports an agent-based simulation platform Swarm, which is freely downloadable from their web site [3]. Swarm is intended to be general enough for practically any agent-based model. Minar et al. (1996) describe the system: "Swarm is a multi-agent software platform for the simulation of complex adaptive systems. In the swarm system the basic unit of simulation is the swarm, a collection of agents executing a schedule of actions. Swarm supports hierarchical modeling approaches whereby agents can be composed of swarms of other agents in nested structures. Swarm provides object-oriented libraries of reusable components for building models and analyzing, displaying, and controlling experiments on those models. Swarm is currently available as beta version in ftill, free source code form. It requires the GNU C Compiler, Unix, and X Windows." The lively support group discussions on the web indicate that swarm can be run on a PC with Linux. European scientists are organizing to develop an agent-based platform with a less programming-intensive interface than Swarm, specially designed for modeling in social sciences. Their work is accessible at [4].
Complexity and Activity-Based Travel A nalysis and Modeling
569
COMPLEXITY IN ECONOMICS This section discusses some of the recent developments of complexity approach in economics and highlights some of the applications. The discussed applications are chosen completely subjectively, by the wish that they may be of interest to activity-based travel analysis and modeling researchers in search of a new approach to their subject. Agent-based modeling allows tying together the behaviour of individuals and system level changes in a more explicit and quantitative manner. The traditionally non-econometric Austrian economics emphasizes the locality of pertinent knowledge and information, which sets limits to the efficiency of outside direction of the economic process and especially innovation. Starting from the notion that a central planner can not have all the pertinent information to induce an optimal allocation of resources, the Austrian tradition has emphasized the virtues of freely developing markets within a clearly defined legal framework and the driving force of the individual creativity (Klein 1997). These ideas in Austrian economics, especially Hayek's idea of "spontaneuos order" (1964) and Schelling's writings (1971, 1978) emphasize the bottom-up, individual-level self-organization, that is central in the complexity approach. An example of modeling Schelling's urban racial segregation is in Robert Axelrod's (forthcoming) book "The Complexity of Cooperation: Agent-Based Models of Competition and Collaboration," where he has collected previously published articles and exercises in agent-based modeling of social phenomena. The more positivist strings of evolutionary economics approaches the problem of relativism by emphasizing the evolution of relations between entities, instead of looking for one dominant reductionist cause for phenomena (England 1994). This Schumpeterian "creative destruction" emphasizes the never-ending process of evolution. Nelson (1995) reviews recent developments in evolutionary economics. McFadzean and Tesfatsion (1996) have developed a specific agent-based platform, which they describe as: "A general C++ platform for the implementation of trade network game (TNG) that combines evolutionary game play with preferential partner selection. In the TNG, successive generations of resource constrained traders choose and refuse trade partners on the basis of continually updated expected payoffs, engage in risky trades modeled as two-person games, and evolve their trade strategies over time. The modular design of the TNG platform facilitates experimentation with alternative specifications for market structure, trade partner matching, trading, expectation formation, and trade strategy evolution. The TNG platform can be used to study the evolutionary implications of these specifications at three different levels:
570
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
individual trader attributes; trade network formation; and social welfare." A comparable model in activity-based travel analysis would be modeling the choice to participate in an activity, the evolution of activity patterns and strategies to choose the activities, and the aggregate level outcomes. Macroeconomics, which as a science has evidenced growing discrepancies between theories and evidence, welcomed complexity as a quantitative computational modeling approach. Leijonhufvud (1993) explains the general reasoning why economics can benefit from computational approach. Both in macroeconomics and finance the non-linear, evolutionary or dynamic systems theory approach has had a lot of applications6. Often these models deal with fraction-data, where the decisive variables in the simulations are the distributions of characteristics over the population on one hand, and the associated probabilities of particular outcomes of these characteristic combinations on the other. But a lot of those models can be modeled on agent-level. For instance Rust (1996) recommends modeling the general equilibrium as an agent-based model, where the equilibrium prices and quantities emerge endogenously from the decentralized interactions of agents. In microeconomics and psychology, the emphasis has traditionally concentrated on the individual decision makers. However, the assumption of an atomistic or representative individual is often needed for traditional modeling techniques. One remedy for these rather restrictive assumptions has been to treat an aggregate unit — for instance a family - as a single decision unit. With this assumption, it is possible to theorize that the family members's individual decisions would be optimizing the family utility, while the family utility is defined as an aggregate of the individual utilities. However, the theorized utility fijnction structure is often neglected in empirical analysis, because the empirical model runs into an identifiability problem when the model-to-be-optimized is intractable. The most natural agent-based models in microeconomics are applications of game theory. The applications range from repeated games of variations of prisoner's dilemma to modeling adaptive and evolutionary strategies and the effects of communication, learning, and memory. Modeling decision makers with learning and memory allows (and requires) that the agents are heterogeneous, transcending the restrictive assumptions of representative consumers. Often the modeled individual interaction mechanisms are simple, but the emerging order is not apparent from analyzing one decision maker or one type of decision at a time. A good source about learning in games is Levine and Fudenberg (1997).
Complexity and Activity-Based Travel A nalysis and Modeling
5 71
Probably the most exiting feature of agent-based modeling approach in economics has been the transcending of boundaries between traditional macro and micro approaches, and a methodology that allows explicit modeling of the interactions of individuals at multiple aggregation levels. This possibility of explicitness opens the door for formulating the models in a positivist way, and testing behavioural assumptions, for instance theories of decision making with incomplete information, and rationality of decision makers. With the accumulated information of how we act and react to circumstances, we have a better chance to know what we can forecast, and how accurate the forecasts can be. Complexity approach has offered more to economics than a better methodology. There has also been totally new metaphors, like Kaufman's metaphor of an evolving economy as a set of autocatalytic polymers. Kaufman (1991) generalizes his models of the origin of life in which catalytic polymers act on polymers and transform them to other polymers. Polymers can be viewed as symbolic strings, and the transformations they mediate in acting upon one another can be characterized as algorithmic or as implementing a grammar. Under certain conditions a phase transition occurs in which collectively self reproducing, or autocatalytic sets of polymers raise. Kaufman suggest that this approach can model cultural transformations, when an isolated culture comes into contact with another culture, or technological "take-off when an exchange economy reaches a critical complexity of goods and services, which allows the set of new economic niches to explode supracritically. Kaufman lists as benefits of his modeling: 1) the growth of economy is modeled through emerging niches in the economy. 2) phase transitions and supracriticality of the system describe and prescribe the sudden growth spurts that have happened in the history. 3) the model exhibits historical contingency (= path-dependency) coupled with law-like behaviour. The statistics of the process, the size distribution of new sectors of the economy, the numbers of new goods entering and old goods leaving the economy, the changes in richness of interconnection within the web, may all be stable given membership in a regime or class of grammars. 4) The decidability problems in the structure of this approach may logically imply that markets must be incomplete. 5) The decidability may also imply that economic agents must, logically, be boundedly rational.
Incomplete Information and Bounded Rationality Empirical modelers, activity-based travelmodelers included, are always at the mercy of methods and incomplete data. The traditional modeler's lack of information has increasingly
572
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
been joined with a new dimension of incompleteness: the objects of interest are not assumed to have complete information and their tendency to use the information rationally is questioned. Complexity approach has embraced the hypotheses of bounded rationality and limited information in modeling decision-making. For instance, De Vany (1993) uses information theory to measure complexity of an organization, and concludes that large optimal organizations are unboundedly complex and that any real economic organization must be boundedly rational or formed by processes that are boundedly rational. Richter and Wong (1996a, 1996b, 1996c), by concentrating on the implications of dealing with computable numbers instead of real numbers, research whether agents with bounded abilities to compute and describe entities can still behave according to the classical economic theory. Turksen and Wilson (1994) approach the topic of limited information by weakening the assumption of the classical set theory. Working with only ordinal data, they model consumer preferences over fuzzy categories to form partial preference orders. They combine the fuzzy measurement with a vector conjoint model, and improve the predicted rank order considerably when the results are compared to the same model with crisp measurement. The rigidity of crisp sets have also been relaxed by studying methods developed for linguistics analysis: Cohen & Stewart (1994) refer to term "fungibility" to indicate how precisely a concept has to be defined in order for it to be useful. Wu (1996) uses a combination of cellular automata and fuzzy linguistic interface for defining transition rules of different land use categories to simulate urban development. Using fuzzy concepts in a system may stabilize the system and lead to more efficient applications, when the jitteriness of a chaotic system melts down to a more robust and predictable system. Another new area of development in dealing with limited information is rough set theory (Pawlak 1991). Chen and Lin (1997) is an example of use of rough sets to identify patterns in large transportation data, where the patterns have high variability. The rough set theory is still so new that many of the applications are published without comparison to other, more traditional methods. Duntsch and Gediga (1997) is a statistical evaluation of the rough set method to specific data sets. Yet another approach to the effects of limited information has been to focus on the ambiguity of what constitutes the set of choice alternatives and the ambiguity about the kind of consequences these alternatives lead to (Manski, 1996). An example of explicitly modeling a market consisting of agents with limited information and knowledge is Huberman and Hogg (1993), which in the words of the abstract "describes a form of distributed computation in which agents have incomplete knowledge and imperfect
Complexity and Activity-Based Travel Analysis and Modeling
573
information on the state of the system, and an instatiation of such systems based on market mechanisms." The authors study the dynamics of the system and describe a mechanism for achieving global stability through local controls. Huberman and Glance (1994) model two types of agent expectations about the influence of the group's behaviour on individual behaviour: bandwagon and opportunistic expectations. They study how group size and delays in information affect cooperation among the agents with these expectations.
Hypothesis Testing Often computer simulations have focused on setting up a system modeled to obey a small set of simple and plausible rules, which together create a system that emulates complex real life behaviour, but does little to convince that the modeled simulation actually replicates the real life phenomenon it seemingly emulates. However, now the programming abilities and platforms have improved to a stage where the emphasis in simulations has shifted from the effort of setting up a plausible system into designing the system in a way that enables testing interesting behavioural hypotheses. Of course, the art lies in the modeler's creativity and ability to formulate the modeled relationships in a way that allows a meaningftil interpretation in terms of behavioural theories. Most of the time in designing this kind of simulation is spent on formulating the modeled relationships as testable hypothesis (Huberman, 1997). An example of such simulations is Arthur et al. (1996), which models agents' asset price expectations that are dependent on the world consisting of such expectations. This selfreferential reflexivity precludes expectations being formed by deductive means, so that perfect rationality ceases to be well defined. Agents act inductively by generating, testing, acting upon, and discarding expectations. The market is driven by expectations that adapt endogenously to the ecology these expectations co-create. Authors show that when heterogeneity is present the inductive nature of expectations leads to emergence of rich psychological behaviour even under neoclassical conditions: the phenomena explained by sharing of information or expectations, herd effects, behaviourism, or other forms of irrationality need not be the cause of such observations in the market. The study uses genetic algorithms to model preference evolution.
574
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
POSSIBLE USES OF COMPLEXITY APPROACH IN A C T I V I T Y - B A S E D TRAVEL ANALYSIS AND MODELING Our activity patterns are complex. Analysts of urban transportation have struggled with perceived complex interdependencies on one hand and limited data quality and quantity on the other. The modelers have had to use ingenuity to simplify the problem into estimable models in order to make policy recommendations. The current methods used in activity-based travel modeling include computational process models, hazard-based duration models, structural equations models, and combinations of discrete and continuous choice models (Pas, 1997). The researchers have understood the complex nature of phenomena, but the modeling approach has been limited to partial analysis, concentrating on one part of the process. The agent-based modeling enables modelers to combine macro level analysis, which earlier has been addressed with simultaneous equations models; micro level phenomena, which have been addressed by choice models; evolution or time series analysis, which has been addressed by hazard models; and the cognitive processes, which have been addressed by the computational process models. Usually modelers who do partial analysis are aware that what has been left out of the analysis is not random noise, but the task of accounting for any such systematic effects has not been feasible. The elusive interdependencies that activity-based travel analysis and modeling researchers have always aimed to reveal can be explicitly modeled by object-oriented, hierarchical modeling and parallel processing of locally independent submodels. Many real life processes are non-ergodic or non-stationary. The usual statistical models applied in travel modeling assume a stable equilibrium or a stationary process. If the true processes are not stationary, we need an alternative to the assumption of representative sample by assuming explicitly that the collected data only partially represents the rules behind the observed behavioural pattern, even when the sampling is random and the sample is large. The rules governing the agents behaviour govern through the evolution and learning of the agent: the rules may not be computable from a sample, especially if the true processes are nonstationary. Even though activity-based travel research has often emphasized the goal of understanding behaviour against the goal of predicting it, for policy reasons methods that utilize and refer to empirical data have been preferred to simulations with simulated data. But now there is a possibility to model artificial activity-based travel to test behavioural hypotheses the same way that artificial life simulations test evolution strategies. With the new simulation tools one can
Complexity and Activity-Based Travel Analysis and Modeling
575
simulate the evolution of agents as they interact, learn, and choose activities that are complementary in the sense that they are sustainable (both at the individual and aggregate levels), and offer a local or bounded utility maximum over the activity space in a coevolving environment. The need to evaluate validity of applied behavioural rules in current microsimulation models is apparent at recent evaluation (Algers at al. 1997). The earlier works in the activity-based approach have suffered from the indeterminacy of dependent and independent variables. There has not been a clear principle as to what should be explained by what. This indeterminacy between individualistic and structuralist explanations is natural in complexity approach, which emphasizes coevolution of entities, and allows its explicit modeling and testing. As the researcher describes the activity patterns in this framework, he has to explicitly choose his choice of variables at each model level and the interaction mechanisms between the levels. While the choice of variables depends on the phenomena under study, the relative easiness of defining competing activity descriptions and dependencies leads to an opportunity to articulate and test one model structure against a competing one. For instance, activity-based travel analysis researchers may want to explicitly include and test agglomeration effects of activities related to urban structure (as suggested by Hagerstrand 1970), and between-persons non-separability of utility function for different homebased activities (Becker 1991), and the non-separability of production function or team work (as suggested by Alcian and Demsetz 1972). Activity-based travel analysis and modeling acknowledges the family as a decision unit, but empirical analysis often concentrates on analyzing the choices of one individual while treating the other individuals as constraints of the choice set, in a way reducing the analysis to single individual. The other extreme is the Beckerian time allocation approach, where the family is treated as a single utility maximizer, with the assumption that the family members allocate their efforts between home and work activities according to their relative market power. The approach hypothesizes that more equal market power between spouses leads to less differentiated time use. Beckerian analysis is equilibrian in the sense that it explains status quo but not how the change in time allocation happens. It would be of interest to further this type of analysis to model and test different decision making strategies concerning choice and timing of the decisions to acquire market power. Activity-based travel analysis researchers have faced the multidimensionality of the data, and often found themselves in the paradoxical situation of simultaneously struggling with the large size of stored data and not enough of information. Fortunately, other scientists have faced similar problems. Computing power and data storing capability have increased to a level
576
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
where large amounts of detailed information can be analyzed. This is required by the agentbased modeling approach. In activity-based travel analysis and modeling, this feature allows modeling explicitly the effects of different time cycles: life-cycle, yearly, seasonal, weekly, and daily cycles and their impacts on choices can be quantified.7 It also allows modeling in the micro level the structural/environmental/supply effects of the choices and their consequences. And most importantly, for activity-based travel analysis and modeling, the agent-based approach allows modeling the interactions between individuals in micro level. Even though this approach is computer intensive, it emphasizes simple behavioural rules, that produce intricate large-scale outcomes. This reduces some of the computing requirements, as it changes the procedure from solving a single large optimization problem (often with large inverted matrices) into series of simple, repeated, parallel-processed tasks. Activity-based travel analysis and modeling has often been limited by data frequency and/or duration: activity data are not collected fi*om a long enough period so that patterns and evolutionary paths could be reliably detected. The available data may often have been too aggregate to capture the interesting phenomena. Now the data collection methods have reached a highly automated level, and data can be collected more reliably and continuously through for instance satellite tracking devices and automatic sensors, as was done in the Lexington, KY study in fall 1996 (Murakami and Wagner 1997, this conference). With the adoption of automated data collection we will see much higher observation frequencies per object, nearlycontinuous-time-data streams. The richer data sets will support model validation, which has been the Achilles' heel of many simulation models. Combined with the new type of data, the traditional point or aggregate data can be used to validate the distributions of simulated outcomes at aggregate level, or used in modeling as indicator variables at different aggregation levels. Activity-based travel analysis and modeling research can also benefit from the new pattern recognition and computer search methods that have been developed for very large data sets, for constrained optimization problems, or for hard problems. Wilson (1996) has made available an application of pattern recognition algorithms based on ftinctional distance measures and sequence alignment procedures. The Artificial Intelligence Applications Institute lists several commercial and non-commercial constraint optimization tools [5]. Examples of work using such tools can be found in the Xerox Park website [6]. Heitkotter and Beasley (1997) list several evolution strategy and genetic algorithm system packages.
Complexity and Activity-Based Travel Analysis and Modeling
577
As in the examples in applications in economics, the object-oriented modeling approach both gives an opportunity and calls for utilizing different techniques at different aggregation levels of the model in an integrated way. With these new technical possibilities and theoretical developments in understanding complex systems, the activity-based approach to travel analysis can proceed beyond descriptive analyses and start dealing directly with quantifying the behavioural rules governing our everyday choices.
SUMMARY The science of complexity is inherently multidisciplinary, and as such has been hard to incorporate into the institutional boundaries of academia. But that is changing rapidly. Of several institutions have formed settings for scientists interested in complexity and there is a growing body of literature of applications of complexity approach in social sciences, notably in economics. The problems that complexity approach addresses are very much the same that urban travel modelers encounter, especially if they use the activity-based approach to travel modeling. The modularity and nestedness of agent-based modeling provides a way to incorporate large, multidimensional, and multilevel data in a natural way. And it allows incorporating different models at different levels of the architecture. For instance, the monetary costs of transportation could be included via separate effects of short-term out-of-pocket costs and longer term fixed costs, in decision models that have different time horizons, but affect decisions simultaneously. Also, now that ITS applications are generating detailed data about car movements and the credit and commercial institutions are generating detailed financial information, the ability to model time-money allocation in meaningfully micro level is possible, at least in theory. Agentbased modeling both incorporates and generates time series data, which allows simulating individual strategies and evaluating them on both individual and aggregate level, as well as testing behavioural hypotheses in a way that includes the context or situation where the behaviour occurs. These possibilities come with their own challenges. The method is data, programming, and computer time intensive. It requires skillful formulation of hypothesized behaviour. And for policy models, it requires building the models in a way which allows validating the results. But we would like to join Peter Jones and say: Isn't it good that we have a way of being able to represent, handle and work with complexity!
578
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
ACKNOWLEDGEMENTS The authors gratefully acknowledge the support obtained through Grant DMS 9313013 from the National Science Foundation to the National Institute of Statistical Sciences. Of course, we are responsible for the arguments presented herein.
LIST OF WEB-PAGES REFERRED TO IN THE TEXT [1] http://research.Germany.EU.net:8080/~joke/iview.html [2] http://alife.santafe.edu/alife/topics/cas/ca-faq/cafaq.html [3] http ://www.santafe.edu/proj ects/swarm [4] http://www.uni-koblenz.de/~moeh/personal/framework.html [5] http://www.aiai.ed.ac.uk/~timd/constraints/cpstools/cpstools.html [6] ftp://parcftp.xerox.com/pub/dynamics/dynamics.html [7] ftp.aic.nrl.navy.mil:/pub/galist/digests/index.html REFERENCES Alcian, A. and H. Demsetz (1972). "Production, Information Costs and Economic Organization," American Economic Review, December: 777-795. Algers, S., E. Bemauer, M. Boero, L. Breheret, C. Di Taranto, M. Dougherty, K. Fox and J-F. Gabard (1997). Review of Micro-Simulation Models. Smartest, Simulation Modelling Applied to Road Transport European Scheme Tests, Project Deliverable D3. Downloadable at http://www.its.leeds.ac.uk/smartest. Anderson, P., K. Arrow, D. Pines (Eds.) (1988). The Economy as an Evolving Complex System. Santa Fe Institute Studies in the Sciences of Complexity, vol. 5. Redwood City, CA: Addison-Wesley. Arthur, W. B. (1990). "Positive Feedbacks in the Economy," Scientific American: February: 92-99. Arthur, W. B., J. H. Holland, B. LeBaron, R. Palmer, and P. Tayler (1996). "Asset Pricing Under Endogenous Expectations in an Artificial Stock Market," Working Paper 96-12093, Santa Fe Institute Economics Research Program. To appear in The Economy as an Evolving, Complex System II, edited by Arthur W.B., Lane D., and Durlauf S.N. Axelrod, R. (forthcoming). The Complexity of Cooperation: Agent-Based Models of Competition and Collaboration. To be published by Princeton University Press, in fall 1997. Batten, D. and C. Karlsson (1996). Infrastructure and the complexity of economic development. Berlin and New York: Springer Verlag. Becker, G. (1991). A treatise on the Family. Enlarged ed. [1981], Cambridge, MA: Harvard University Press.
Complexity and Activity-Based Travel Analysis and Modeling
579
Brock, W. A. (1991). "Understanding Macroeconomic Time series Using Complex Systems Theory," Structural Change and Economic Dynamics, Vol. 2, 1:119-141. Chen, R. and T. Y. Lin (1997). "Finding Reduct in Very Large Databases," Fifth International Workshop on Rough Sets and Soft Computing, March 2-5, 1997, Research Triangle Park, NC. Cohen, J. and L Stewart (1994). The Collapse of Chaos ~ Discovering Simplicity in a Complex World. New York: Viking Penguin. De Vany, A. (1993). "Information, Bounded Rationality, and The Complexity of economic Organization," downloaded from http://cce.sscnet.ucla.edu/. Davies, P. (Ed.) (1989). The New Physics. New York: Cambridge University Press. Dopfer, K. (1991). "The complexity to economic phenomena: Reply to Timbergen and bQyond,'' Journal of Economic Issues, Vol. 25, 1:39-79. Duntsch, I. and G. Gediga (1997). "Statistical evaluation of rough set dependency analysis," InternationalJournal of Human-Computer Studies, Vol. 46, pp. 589-604. England, R. W. (1994). Evolutionary Concepts in Contemporary Economics, Ann Arbor: The University of Michigan Press. Feldman, B. (1992). "A Game Theoretic Interpretation of the Spin Glass," in 1992 Lectures in Complex Systems, Eds. L. Nadel & D. Stein, SFI Studies in the Sciences of Complexity, Lect. Vol V, Addison-Wesley. Gilmore, C. G. (1993). "A New Approach for Testing for Chaos, with Applications in Finance and Economics," International Journal of Bifurcation and Chaos in Applied Sciences and Engineering, Vol. 3, 3:583-588. Hagerstrand, T. (1970). "What about People in Regional science?" Papers of the Regional Science Association, Vol. 24, pp.7-21. Hayek, F. A. (1964). "Kinds of Order in Society," New Individualist Review, 3:3-12. Heitkotter, J. and D. Beasley, (Eds.) (1997). Hitch-Hiker's Guide to Evolutionary Computation (FAQ for comp.ai.genetic). Downloadable as ftp://ftp. Germany.EU.net/pub/reserach/softcomp/EC/FAQ/hhqtec.ps.gz . Hertz J., A. Krogh, R. G. Palmer (1991). Introduction to the theory of Neural Computation. A lecture notes volume I in the Santa Fe Institute studies in the sciences of complexity. Holland, J. H. (1975). Adaptation in Natural and Artificial Systems. Ann Arbor: University of Michigan Press. Holland, J. H., K. J. Holyoak, R. E. Nisbett and P. R. Thagard (1986). Induction: Processes of Inference, Learning, and Discovery. Cambridge, MA: MIT Press. Hordijk, W. (1997). "The Structure of the Synchronizing-CA Landscape," to be published in Proceedings of the International Workshop on Control Mechanisms for Complex Systems, held in November 1996, Las Cruces, New Mexico. Horgan, J. (1995). "From Complexity to Perplexity," Scientific American, Vol. 272, 6:104109. Huberman, B. A. (1997). "Computation as Economics," Presentation at the annual meeting of American Economic Association, January 4-6, New Orleans, LA. Huberman, B. A. and T. Hogg (1993). "The Emergence of Computational ecologies," in 1992 lectures in Complex Systems, Eds. L Nadel & D Stein, SFI Studies in the Sciences of Complexity, Lect Vol V, Reading, MA: Addison-Wesley. Huberman, B. A. and N. S. Glance (1994). "Beliefs and Cooperation," Paper presented at the "Chaos and Society" International Conference, June 1994. Downloadable at the Xerox Park web SAXQ ftp://parcftp.xerox.com/pub/dynamics/dynamics.html.
580
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Jones, P., F. Koppelman, and J-P. Orfeuil (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, England: Gower. Jones, P., (Ed.) (1990). Developments in Dynamic and Activity-Based Approaches to Travel Analysis. Aldershot, England: Gower. Kauffman, S. (1991). "Random Grammars: A New Class of Models for Functional Integration and Transformation in the Biological, Neural, and Social Sciences," 1990 Lectures in Complex Systems, SFI Studies in the Sciences of Complexity, Lect. Vol. Ill, Eds. L Nadel and D. Stein, Reading, MA: Addison-Wesley. Kauffman, S. (1995). At home in the universe: the search for laws of self-organization and complexity. New York, NY: Oxford University Press. Kitamura, R. (1997). Activity-Based Travel Demand Forecasting and Policy Analysis. Proceedings of the TMIP Conference on Activity-Based Travel Forecasting, New Orleans, LA, June 2-5, 1996. Kitamura,R and S. Fujii (1998). Two computational process models of activity-based travel behavior. In T. Garling, T. Laitila and K. Westin (Eds.) Theoretical Foundations of Travel Choice Modeling. Oxford: Pergamon Press, 251-280. Kitamura, R., E. Pas, C. V. Lula, T. K. Law^on and P. E. Benson (1996). The sequenced activity mobility simulator (SAMS): an intergrated approach to modeling transportation, land use and air quality. Transportation 23: 267-291. Klein, D. (1997). "Convention, Social Order, and the Two Coordinations," forthcoming in Constitutional Political Economy. Langlois, R. N. and M. J. Everett (1992). "Complexity, Genuine uncertainty, and the Economics of Organization", Human systems management, 11, 2:67-76. LeijonhufVud, A. (1993). "Towards a Not-Too-Rational Macroeconomics," Southern Economic Journal, Vol 60. Downloaded from http://cce.sscnet.ucla.edu/. Levine, D. K.*and D. Fudenberg (1997). Theory of Learning in Games. Forthcoming, MIT Press. Available at David Levine's web site http:// levine.sscnet.ucla.edu/. Maier, G. (1995). Spatial search: structure, complexity, and implications, Heidelberg: PhysicaVerlag. Mannering, F., E. Murakami and S-G. Kim (1994). "Temporal stability simulations," downloaded from http://www.santafe.edu/projects/swarm/. Manski, C. F. (1996). "Treatment under Ambiguity," A Santa Fe Institute working paper 9610-078. McFadzean D. and L. Tesfatsion (1996). "A C++ Platform for the Evolution of Trade Networks," Iowa State University Economic Report No. 39, December 1996. To appear in Computational Economics, Special issue on Programming Languages. McNally, M. G. and Z-P. Lo (1992). "Prediction of Discrete Choice Via Neural Network," paper presented at the 72"^ Annual Meeting of the Transportation Research Board, Washington, DC. Miller, E. J. (1997). Microsimulation and Activity-Based Forecasting. Proceedings of the Conference on Activity-Based Travel forecasting, held in New Orleans, LA, June 2-5, 1996. Minar N., R. Burkhart, C. Langton and M. Askenazi (1996). The Swarm Simulation System: A Toolkit for Building Multi-Agent Simulations. Downloadable from http://www. santafe. edu/projects/swarm/.
Complexity and Activity-Based Travel Analysis and Modeling
581
Mitchell, M. (1996). "Computation in Cellular Automata: A Selected Review," to appear in Schuster, H. C. and Gramss, T. (Eds.), Nonstandard Computation. Weinheim, Germany: VCH Verlagsgesellschaft. Murakami, E. and D. P. Wagner (1997). Comparison Between Computer-Assisted SelfInterviewing Using GPS with Retrospective Trip Reporting Using Telephone Interviews. Draft, to be presented at this conference. Nagel, K., M. Rickert, and C. L. Barrett (1996). "Large Scale Traffic Simulations," Santa Fe Institute working papers 96-10-080. Nagel K. and C. L. Barrett (1997). Using microsimulation feedback for trip adaptation for realistic traffic in Dallas. A paper posted on Santa Fe Institute web page, dated March 27, 1997. Nelson, R. (1995). "Recent Evolutionary Theorizing About Economic Change," Journal of Economic Literature, Vol. 33, pp. 48-90. Norman, A. L. (1994). "Computability, Complexity and Economics," Computational Economics, Vol. 7, 1:1-21. Norman, A. L. and D. W. Shimer (1994). "Risk, uncertainty, and complexity," Journal of economic dynamics & control, Vol. 18, 1:231-249. Pas, E. I. (1997). "Recent Advances in Activity-Based Travel Demand Modeling," Proceedings of the Conference on Activity-Based Travel forecasting, held in New Orleans, LA, June 2-5, 1996. Pawlak, Z. (1991). Rough sets: Theoretical aspects of reasoning about data, Boston, MA: Kluwer Academic Publishers. Perelson A. S. (Ed.) (1988). Theoretical Immunology, Parts One and Two. Santa Fe Studies in the Sciences of Complexity, Proceedings vols. 2 and 3. Redwood City, CA: AddisonWesley. Perelson, A. S. and S. A. Kauffman (1990/ Molecular Evolution on Rugged Landscapes: Proteins, RNA, and the Immune System. Santa Fe Institute Studies in the Sciences of Complexity, Proceedings vol. 9. Redwood City, CA: Addison-Wesley. Polak, J., P. Jones, P. Vythoulkas, R. Sheldon and D. Wofinden (1993). "Travellers' Choice of Time of Travel Under Road Pricing," Proceedings ofPTRC, Seminar D. Reyck, B. D. and W. Herroelen (1996). "On the use of complexity index as a measure of complexity in activity networks," European Journal of operational research. Vol. 91, 2:347. Richter, M. K. and K-C. Wong (1996a). "Bounded Rationalities and Definable Economies," Discussion paper No. 295, December. Center for Economic Research, Department of Economics, University of Minnesota. Richter, M. K. and K-C. Wong (1996b). "Definable Utility in 0-Minimal Structures," Discussion paper No. 296, December. Center for Economic Research, Department of Economics, University of Minnesota. Richter, M. K. and K-C. Wong (1996c). "Computability of Preference, Utility, and Demand," Discussion paper No. 298, December. Center for Economic Research, Department of Economics, University of Minnesota. Rickert, M. and K. Nagel (1997). Experiences with a simplified microsimulation for the Dallas/Forth Worth area. Draft posted on the Santa Fe Institute web page, dated February 27, 1997.
582
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
Rickert, M. and P. Wagner (1996). Parallel Real-time Implementation of Large-scale, Routeplan-driven Traffic Simulation. Paper posted on Santa Fe Institute web page, dated April 1, 1996. Rust, J. (1996). "Dealing with the Complexity of Economic Calculations," Invited paper for Workshop on Fundamental Limits to Knowledge in Economics, Santa Fe Institute, JulySl-Augusts, 1996. Revisited draft dated October, 1996. Sadek, A. W., B. L. Smith and M. L. Demetsky (1997). Dynamic Traffic Assignment: A Genetic Algorithms Approach. Presented at Transportation Reserach Board 76* Annual Meeting, Washington, D.C. Shelling, T. C. (1971). "Dynamic Models of Segregation," Journal of Mathematical Sociology, Vol. 1, pp. 143-186. Shelling, T. C. (1978). Micromotives andMacrobehavior. New York, NY: Norton. Tesfatsion, L. (1997). "How economists can get a life," to appear in The Economy as a Complex Evolving System, 11, B. Arthur, S. Durlauf, and D. Lane (Eds.). The paper is also downloadable from his webpage http://www.econ.iastate.edu/tesfatsi/abe.htm.. Young, P., S. Parkinson et al. (1996). "Simplicity out of complexity in environmental modeling: Occam's razor rQwisited,'' Journal ofApplied Statistics, Vol. 23, 2-3:165-212. Waldorp, M. (1992). Complexity: the emerging science at the edge of order and chaos. New York, NY: Simon and Schuster. Weigend, A. S., B. A. Huberman and D. E. Rumelhart (1991). "Forecasting Chaotic Computational Ecosystems," in 1990 Lectures in Complex Systems, Santa Fe Institute Studies in the Sciences of Complexity, Lect Vol III, L. Nadel and D. Stein (Eds.), Reading, MA: Addison-Wesley. Wilson, C. (1996). Activity Pattern Analysis Using Sequence Alignment Methods. A paper prepared for the Conference of the International Association of Time Use Research, September 1996. Wu, F. (1996). A Linguistic Cellular Automata Simulation Approach for Sustainable Land Developmant in a Fast Growing Region. Computers, Environment and Urban Systems, Vol 20, 6:367-387. ^ See Santa Fe Institute yearly publications "Lectures on Theory of Computation and Complexity", from 1988 onwards, or Xerox Park website . ^ These examples are chosen just to show the large area of application fields. The cited works are not meant to represent the main publications in each field. ^ The descriptions of basic cellular automata and genetic algorithms are based on Hordijk (1997). "* The description is based on Hertz et al.(1992). ^ The presentation of basic principles of evolutionary programming, evolutionary strategy, and classifier systems are modified from Heitkotter and Beasley (1997). ^ For instance the inaugural edition of a new academic journal Macroeconomic Dynamics was dedicated to computation and estimation in economics and finance. ^ See Kitamura et al 1996 as an example to incorporate these aspects.
In Perpetual Motion: Travel Behavior Research Opportunities and Application Challenges H. S. Mahmassani (Editor) ©2002 Elsevier Science Ltd. All rights reserved.
28
MiCROSIMULATION: WORKSHOP REPORT
Kay Axhausen and Ram Pendyala
INTRODUCTION Microsimulation approaches are the potential w^orkhorses for the application of complex and dynamic models of travel and activity behaviour. The goal of the workshop was to take stock of how close we are to this goal and what issues need to be addressed on the way to this goal. The discussion was based on the resource paper by Miller and Salvini (2001), which set the scene and defined microsimulation in the context of travel behaviour research as an approach for exercising a set of disaggregate behavioural models over time.This implies that the status of the disaggregate entities, e.g. persons, is updated and their actions performed, resulting in a consistent evolution of the system state and in individual trajectories/profiles for each entity. The authors highlighted that current technologies bring us closer to the promise of microsimulation as a standard tool for analysis and forecasting, but that there are still substantial difficulties, which need to be addressed.
STATUS IN SIMULATION MODELS OF TRAVEL BEHAVIOUR RESEARCH The discussion of the workshop highlighted that a number of new technologies (in particular agent-based computing, non-linear system theory or the advances in computing power) have improved the ease of development and application of microsimulation models, but that they still have not found widespread application in their full form in the modelling of travel behaviour, which is in marked contrast to traffic flow simulation were such models are dominant (see excursus below).
584
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
The sample-enumeration tools, associated with disaggregate modelling systems, are the most widely applied tools, although their lack of real-time and activity-location interaction between the simulated units and their lack of an explicit time axis puts them outside the definition provided above. Well known examples are the Dutch National Model (Gunn, van der Horn and Daly, 1988 or Van der Hoom and van Hoek, 1989), the work of Zumkeller (1989) or the tool VISEM (Fellendorf, Haupt, Heidi and Scherr, 1997). In recent years, models have been formulated, which try to integrate the time axis and the realtime interactions in a consistent way, e.g. ORIENT/RV (Axhausen, 1990), Eurotopp (Axhausen and Goodwin, 1991) or DYNASMART (Mahmassani, Hu and Peeta, 1994) among others. Perhaps the best known of these is TRANSIMS (Barrett et al, 1995), which is intended as a replacement of the standard UTPS-approach in the USA. TRANSIMS provides an iterative scheme to ensure the macroscopic consistency between the demand for travel and the performance of the network, which is evaluated using a microscopic flow simulation. This leaves the possibility for microscopic inconsistencies, but makes the approach applicable for large scale planning studies. Housing market models have been able to achieve this consistency in a slightly less complex environment than the activity-travel-regime (for a review see Wegener, 1994).
Excursus: Simulation of Traffic Flow The problems of operating and designing multi-lane single-direction carriageways of motorways stimulated early on an interest in the simulation of motorway traffic flow (May, 1990) first in the US, then in Europe and elsewhere (Leutzbach, 1988). The purpose of this excursus is neither a historical review (for this see May, 1990 or Leutzbach, 1988) nor an overview of currently available models (for recent reviews see Algers, Bemauer, Boero, Breherent, Di Taranto, Dougherty, Fox and Gabard, 1997 or Reiter, 1997), but a discussion of their general structures and problem areas, in particular in relation to the state-of-the art in simulation models of travel behaviour. The motorway context allowed a relatively simple conceptualisation of the drivers, their vehicles and their environment due to the absence of other types of road users, an uniform and mostly disturbance-free environment custom-made for the motor vehicle and few flow interruptions. The core of the conceptualisation was and is the driver-vehicle unit and a set of rules and models describing the choice of the driving speed and of lateral position, implying rules for acceleration and deceleration, overtaking and interaction with surrounding vehicles. For simplicity the rules are generally specified:
Microsimulation: Workshop Report
• • •
585
Only for the immediate neighbours of the driver-vehicle unit considered at any one time (cars in front and behind). The models are time-stepped treating each driver-vehicle-unit in a given order of precedence. Time steps of one second are the norm. The models assume that there are only n positions available in the cross-section, with n the number of lanes. There are no interactions with on-coming and crossing traffic and with merging or demerging traffic only at on- and off-ramps.
This framework was then transferred to other environments, where some, but not all of these assumptions had to be relaxed: multi-lane rural roads with on-coming traffic influencing overtaking (e.g. Brilon, 1976; Leutzbach, Brannolte and Schmidt, 1989; Okura and Matsumoto, 1990; Pursula and Siimes, 1993; Botha, Zeng and Sullivan, 1993), urban traffic with crossing traffic and signal control (Leutzbach and Wiedemann, 1986 or Khasnabis, Kamati and Rudraraju, 1996) interaction with guidance and control systems, e.g. radio-based direction guidance, in-car dynamic route guidance, variable speed control etc (among many others Emmerink, Axhausen, Nijkamp and Riedveld, 1995; Hu and Mahmassani, 1995 or Reiter, 1994). The modelling idea was also transferred to other types of vehicles, such as bicycles (e.g. Wiedemann and Zhang, 1989), planes and trains (e.g. among many others VISION (British Rail), TRANSIT (Siemens) or SIMU VII (TU Braunschweig), or applications such as Reid, Sicking and Paulsen, 1995; Venglar, Fambro and Bauer, 1995 or Buck, 1992), as well as to pedestrians. The core of each model remains the description of the choice of speed by each simulated unit and the choice of lateral position in conjunction with the rules specifying the response to the various control systems modelled. Within the speed-choice context, four main modelling traditions can be distinguished in the literature: • Implementation of continuous car-following models, which describe the desired distance between any two vehicles as a function of their distance, their relative speeds and, sometimes, their relative accelerations. Assuming that a desired maximum speed is known the acceleration/speed for the next time step is then calculated to fit the desired distance. A well known example among many is Gipps' car-following model (Gipps, 1981), whioch has also been implemented widely in operational models. Well known implementations of such approaches are the FHWA models FRESIM, and NETSIM or AIMSUM2 (Barcelo and Ferrer, 1994) • Implementations of the psycho-physical spacing model, which relates acceleration and deceleration to the speed- and distance-differences detectable by the driver (Leutzbach and Wiedemann, 1986; Fellendorf, 1993, Benz, 1994)
586
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges Cellular automata, which simplify the description further providing only discrete positions along the roadway axis and across the cross-section and only discrete speeds, trading simplicity for speed of calculation (e.g. Nagel and Schreckenberg, 1992). Nagel (1995), for example, specifies his model of single-lane flow completely as follows: 1) Acceleration: If the velocity v of a vehicle is lower then (uniform) Vmax and if there is enough room ahead (v < gap -1), then speed is increased by one. 2)
Slowing down (due to other cars): If the vehicle ahead is to close (v > 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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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
In Perpetual Motion: Travel Behaviour Research Opportunities and Challenges
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