Modelling Strategic Behaviour in Anticipation of Congestion PROEFSCHRIFT ter verkrijging van de graad van doctor aan de Technische Universiteit Eindhoven, op gezag van de Rector Magnificus, prof.dr.ir. C.J. van Duijn, voor een commissie aangewezen door het College voor Promoties in het openbaar te verdedigen op woensdag 22 maart 2006 om 16.00 uur door Qi Han geboren te Nanjing, China Dit proefschrift is goedgekeurd door de promotoren: prof.dr. H.J.P. Timmermans en prof.dr. W.F. van Raaij Copromotor: prof.dr.ir. B.G.C. Dellaert Copyright © 2006 Q. Han Technische Universiteit Eindhoven, Faculteit Bouwkunde, capaciteitsgroep Stedebouw Cover design: Tekenstudio, Faculteit Bouwkunde Printed by the Technology University of Eindhoven Press Facilities CIP-DATA KONINKLIJKE BIBLIOTHEEK, DEN HAAG ISBN 90-6814-598-3 Preface In this thesis, a approach is developed to explore and describe strategic choice behaviour in anticipation of congestion, based on interaction and interdependencies between individuals and between individuals and service providers. It formulates the process of how people would arrive at choices in a congested environment and how information or guidance would become effective towards serving a certain goal. My research interest in this area emerged during my master study. When I first saw the research proposal of this Ph.D. project a few years ago, I knew that I found something that would suit me. The research project was sponsored by SOBU: a collaboration between Tilburg University and Eindhoven University of Technology. The original proposal was developed by Harry Timmermans and Benedict Dellaert when he still was in Tilburg University. Soon after the proposal was funded, he became a professor at Maastricht University and Fred van Raaij joined the team. However, Benedict stayed heavily involved in the project, and hence I was lucky to have such an active scholarly environment. The available travel grant provided me also great opportunities to attend international conferences and meet other researchers. Besides, many people have supported me in conducting the research reported in this thesis. I wish to acknowledge all the help that they given. There are some that I would like to mention specifically for their contributions. iii Professor Harry Timmermans, my first supervisor, has supported me in many ways. With his special style of open guidance he knows how to encourage people to work independently, to develop their own expertise and to pursue their research goals. This and his strategic insights in the world of science have helped me in not only developing the line of my research interest, but also offering me a wonderful opportunity to learn how to conduct research and publish in international journals. Besides, his kind and careful comments on the draft version of my thesis were very important. Many thanks also go to Professor Fred van Raaij, my second supervisor. He succeeded Professor Benedict Dellaert and continued bridging the cooperation between the two universities on this project. He introduced me the educational credit system in Tilburg University that I could use to get sufficient respondents for the data collection. From preparing the recruit announcement, arranging the laboratory, to credit registration afterwards, he has given me invaluable support in two experimental data collections in my research project. Besides, his insightful reviews gave valuable contributions to my work. I also owe much to Professor Benedict Dellaert, my co-promotor. He was my second supervisor for the first half year before he took the job position at Maastricht University. His contribution to my research however never stopped. His ability to find creative solutions for complicated problems and his efforts to discuss my work at the most detailed level are much appreciated. His suggestions and helpful comments on my work have improved the quality of my research significantly. During the years, my colleagues in the urban planning group supported me in many different ways. I learned many things from working with them and I feel honoured simply to be surrounded by them. Although it is difficult to put in words, I want to take this opportunity to express my appreciation to all of them, in particular to Aloys Borgers, Theo Arentze, Peter van der Waerden and Astrid Kemperman for helping me out in explaining models and software, Leo van Veghel for providing me access to valuable references, and both Mandy van de Sande – van Kasteren and Anja Janssen for their excellent secretarial assistance and care. Also I enjoyed a lot the discussions with former group members Manon van Middelkoop, Chang-Hyeon Joh, Maarten Ponje and current Ph.D. candidates. Among them, a special word of thanks goes to Dick Saarloos, who shared the office with me for the last few years, for the interesting talks on research and cultural difference, and for his help in times of need. Finally, I would like to thank my husband, Hai. His support has been a constant encouragement. Last but not least, I also like to thank my parents, who have always been there for me with their unconditional care. Qi Han Eindhoven, December 2005 iv Contents Preface iii List of figures ix List of tables x 1 Introduction 1 2 Modelling spatial choice behaviour in urban planning 5 2.1 Introduction 5 2.2 Individual choice models 6 2.2.1 Spatial interaction modelling 6 2.2.2 Discrete choice modelling 7 2.2.3 Activity-based modelling 9 2.2.4 Evaluation 9 2.3 Choice behaviour in response to information 10 2.3.1 Information use and provision 10 2.3.2 Evaluation 11 2.4 Choice behaviour in response to other individuals’ behaviour 12 2.4.1 Interdependence within multi-person households 12 (intra household) 2.4.2 Interdependence within the social-spatial network 12 (inter household) 2.4.3 Game theory and interactive decision making 14 2.4.4 Game theory and discrete choice models 15 2.4.5 Evaluation 15 2.5 Conclusion and discussion 16 3 Theory 17 3.1 Introduction 17 3.2 Models of strategic choice behaviour based on interaction and 18 interdependencies between individuals v 3.2.1 N-player J-option game formulation 18 3.2.2 Strategic discrete choice model 22 3.2.2.1 Specification of the exogenous component 22 3.2.2.2 Specification of the endogenous component 23 3.2.2.3 Specification of the disturbance 24 3.2.3 Equilibrium conception: consistence condition 24 3.2.3.1 Bayesian Nash equilibrium 25 3.2.3.2 Quantal response equilibrium 27 3.2.3.3 Correspondence between QRE and BNE 28 3.2.4 The logit approach 30 3.2.4.1 The error parameter 31 3.2.4.2 Estimation issues 33 3.2.5 Inconsistence: imperfect expectation in one-shot situation 35 3.2.6 Interpretation, application 37 3.2.6.1 Understanding mixed-motives 37 3.2.6.2 Application issues: user, domain and data 37 3.3 Models of strategic choice behaviour based on interaction and 38 interdependencies between individuals and information providers 3.3.1 Two-player two-stage game formulation 39 3.3.2 Interactive decisions between individuals and information 41 providers 3.3.3 Update of individual’s expectation given information providers’ 42 recommendation 3.3.4 Information provider’s objective 44 3.4 Conclusion and discussion 46 4 Numerical simulation 49 4.1 Introduction 49 4.2 Interactive decisions between individuals 50 4.2.1 Algorithm 51 4.2.2 Simulation results 51 4.2.2.1 Sensitivity to choice alternative facets 52 4.2.2.2 Sensitivity to interaction 53 4.2.2.3 Sensitivity to uncertainty of conjectures 54 4.3 Interactive decisions between individuals and an information 55 provider vi 4.3.1 Algorithm 56 4.3.2 Simulation results 57 4.3.2.1 Effects of different objectives of the information 58 provider 4.3.2.2 Effects of different levels of individuals’ conjecture 60 ability 4.4 Conclusion and discussion 63 5 Timing choice 65 5.1 Introduction 65 5.2 Hypothesis 66 5.3 Timing choice data 67 5.3.1 Data collection 67 5.3.2 Dinner timing choice 68 5.3.3 Participation timing choice 69 5.3.4 Structure of the experimental design 70 5.4 Analysis of timing choice data 71 5.4.1 Dinner timing choice model 71 5.4.2 Participation timing choice model 77 5.4.3 Non-strategic and strategic behaviour 79 5.4.4 Test of hypothesis 84 5.5 Conclusion and discussion 88 6 Repeated strategic choice 89 6.1 Introduction 89 6.2 Research question 90 6.3 Repeated strategic choice data 92 6.3.1 Data collection 92 6.3.2 Structure of the experimental design 93 6.3.3 Implementation 94 6.3.4 Process 95 6.4 Analysis of repeated strategic choice data 98 6.4.1 General results: performance 98 6.4.2 General results: choice frequencies 101 6.4.3 Regression model estimation 103 vii 6.4.4 Choice model estimation 106 6.4.4.1 Strategic choice model 106 6.4.4.2 Simulation for projecting the anticipated congestion 108 6.4.4.3 Results for one-shot situations 108 6.4.4.4 Results for repeated situations 111 6.4.4.5 Results about learning in situations without 113 recommendation 6.4.4.6 Results about learning in situations with 115 recommendation 6.5 Conclusion and discussion 118 7 Conclusion and discussion 121 7.1 Introduction 121 7.2 Short summary of the study 122 7.3 Discussion of the proposed modelling approach 124 7.4 Avenues for future research 125 Bibliography 129 Appendix A-1 Announcement of timing choice experiment 137 Appendix A-2 Questionnaire of timing choice experiment 138 Appendix B-1 Announcement of repeated choice experiment 147 Appendix B-2 Overview of the tasks 148 Appendix B-3 Screen shot of the instruction 149 Appendix B-4 Screen shot of the tips 150 Appendix B-5 A screen shot after subjects made a choice 151 Author index 153 Subject index 157 Samenvatting (Dutch summary) 161 Curriculum vitae 163 viii List of figures Figure 3-1 Game 1 formulation 20 Figure 3-2 Game 1 extensive form 21 Figure 3-3 Relation between type and strategy 29 Figure 3-4 Error parameter effect 32 Figure 3-5 Game 2 formulation 40 Figure 3-6 Game 2 extensive form 40 Figure 3-7 Effect of self-expectation on compliance rate 44 Figure 4-1 Choice alternative impact 53 Figure 4-2 Interaction impact (θ) 54 Figure 4-3 Uncertainty impact 55 Figure 4-4 Results for the Nash equilibrium 58 Figure 4-5 Results for the quantal response equilibrium 59 Figure 4-6 Results for the noisy introspection model 60 Figure 4-7 Results for the equal usage objective 61 Figure 4-8 Results for the welfare maximum objective 61 Figure 4-9 Results for the user equity objective 62 Figure 5-1 Structure of timing choice experimental design 71 Figure 5-2 Simulations of participation timing choice 86 Figure 6-1 Structure of two dimensional effects 91 Figure 6-2 A screen shot of choice task 97 Figure 6-3 Average performance with and without recommendation 98 Figure 6-4 Average performance with recommendation for different 99 objectives Figure 6-5 Average performances for the certainty of others’ preference 100 Figure 6-6 Recommendation effect in choice frequencies 101 Figure 6-7 Preference certainty effect in choice frequencies 102 Figure 6-8 Learning effect in the no recommendation situation 114 Figure 6-9 Learning effect in the with recommendation situation 115 Figure 6-10 Learning effect for the marginal disutility of congestion 117 Figure 6-11 Learning effect for the certainty of others’ preference 118 ix List of tables Table 5-1 Overview of registration for timing choice experiment 67 Table 5-2 Non-strategic dinner timing choice estimates (1-A) 72 Table 5-3 Strategic dinner timing choice estimates (1-B) 74 Table 5-4 Dinner timing choice model estimates (1-A & 1-B) 76 Table 5-5 Participation timing choice estimates (2-B & 2-C) 78 Table 5-6 Observed participation timing choice frequencies 78 Table 5-7 Non-strategic behaviour estimates (1-A & 2-B) 79 Table 5-8 Non-strategic behaviour contrast parameter estimates (1-A & 2-B) 80 Table 5-9 Strategy behaviour estimates (1-B & 2-C) 81 Table 5-10 Strategy behaviour contrast parameter estimates (1-B & 2-C) 82 Table 5-11 Information impacts (2-C) 83 Table 5-12 Simulations of participation timing choice 86 Table 6-1 Overview of registration for repeated strategic choice 92 Table 6-2a Regression model estimates (1st round vs. 13th round) 104 Table 6-2b Test statistics for the classical regression model 104 Table 6-3 Contrast parameter estimates (1st round vs. 13th round) 105 Table 6-4 One-shot situation strategic choice model estimates (1st round) 109 Table 6-5 One-shot situation MNL model log likelihoods (1st round data) 110 Table 6-6 Repeated situation strategic choice model estimates (13th round) 112 Table 6-7 Repeated situation MNL model log likelihoods (13th round data) 113 x
Description: