modal shift forecasting models for transit service planning · 2014-01-30 · second deals with...
TRANSCRIPT
Modal Shift Forecasting Models
for Transit Service Planning
By
Ahmed Osman Idris
A thesis submitted in conformity with the requirements
for the degree of Doctor of Philosophy
Graduate Department of Civil Engineering
University of Toronto
© Copyright by Ahmed Osman Idris, 2013
ii
Modal Shift Forecasting Models for Transit Service Planning
Ahmed Osman Idris
Doctor of Philosophy
Department of Civil Engineering
University of Toronto
2013
ABSTRACT
This research aims at developing a better understanding of commuters’ preferences and mode
switching behaviour towards local transit for work trips. The proposed methodological
approach incorporates three main stages. The first introduces a conceptual framework for
modal shift maximized transit route design model that extends the use of demand models
beyond forecasting transit ridership to the operational extent of transit route design. The
second deals with designing and implementing a socio-psychometric COmmuting Survey for
MOde Shift (COSMOS). Finally, the third stage focuses on developing econometric choice
models of mode switching behaviour towards public transit.
Advanced mode shift models are developed using state-of-the-art methodology of combining
Revealed Preference (RP) and Stated Preference (SP) information. The results enriched our
understanding of mode switching behaviour and revealed some interesting findings. Some
socio-psychological variables have shown to have strong influence on mode shift and
improved the models in terms of fitness and statistical significance. In an indication of the
superiority of the car among other travel options, strong car use habit formation was realized
for car drivers, making it hard to persuade them to switch to public transit. Further, unlike
conventional choice models, the developed mode shift models showed that travel cost and in-
vehicle travel time are of lower importance compared to other transit Level of Service (LOS)
attributes such as waiting time, service reliability, number of transfers, transit technology, and
iii
crowding level. The results also showed that passengers are more likely to shift to rail-based
modes (e.g. LRT and subway) than rubber-tyred modes (e.g. BRT). On the other hand, the
availability of park-and-ride facilities as well as both schedule and real-time information
provision did not appear to be significant for mode switching to public transit for work trips.
This research provides evidence that mode shift is a complex process which involves socio-
psychological variables beside common socio-demographic and modal attributes. The
developed mode switching models present a new methodologically sound tool for evaluating
the impacts of alternative transit service designs on travel behaviour. Such tool is more
desirable for transit service planning than the traditional ones and can aid in precisely
estimating transit ridership.
iv
ACKNOWLEDGEMENT
Time flew by and suddenly I realized that the long journey of my doctoral research has come
to an end. To me, as a strong believer in public transit, I feel as if I were travelling by bus in a
morning peak learning trip! The bus made several stops along the way where some
passengers alighted and others boarded. Now it seems to be my turn to get off my ride, after
reaching the terminal point of that feeder route, to make a transfer towards another future
stop.
Throughout my learning journey, I have been guided and supported by many people without
whom this research would not have been successfully completed. Although it is not an easy
task to show my gratitude to everyone in a few words, I will try my best to do so in the
following lines. However, my appreciation extends to everyone who helped me in any
capacity and apologies to those I did not mention by name.
First, I would like to sincerely thank my thesis supervisors: Prof. Amer Shalaby and Prof.
Khandker Nurul Habib for their valuable guidance and genuine moral support. In fact, their
supervision and mentoring have set for me an example that I would like to follow with my
students. It was a great honour for me to get to know and learn from them over the past five
years.
Second, many thanks to the members of my examination committee: Prof. Baher Abdulhai
and Prof. Matthew Roorda for their insightful comments on my thesis and for the
knowledge I gained from them during the course of my studies. I am also thankful to Prof.
Martin Trépanier, my external examiner, for spending his valuable time reading my thesis,
providing constructive feedback, and attending my final thesis defence.
Third, lots of gratitude to all my colleagues and friends within our transportation group at the
University of Toronto. I greatly appreciate the contribution of all of them to this research
either directly (by providing help) or indirectly (by providing support and encouragement).
Special thanks to Zhong Yi Wan, Mohamed Elshenawy, Tamer Abdulazim, Keith
Cochrane, and Rinaldo Cavalcante for helping me during different phases of developing
my COmmuting Survey for MOde Shift (COSMOS).
v
Fourth, heartfelt thanks to my Father and Sister for their overwhelming support,
encouragement, and love. Actually, they contributed to this work a lot with their continuous
support that was a key factor in successfully completing this stage of my life. Finally, I would
like to dedicate this thesis to my Father and the memory of my Mother who is no longer
with us on this earth.
Yet, I still believe that a bad day of fishing is better than a good day at the office!
Take the Bus,
Ahmed Osman Idris
August 2013
vi
The following chapters of this dissertation have been reproduced with modifications from my
previously published and presented material:
CHAPTER 3 - MODAL SHIFT MAXIMIZED TRANSIT ROUTE DESIGN MODEL:
Idris, A. O., Shalaby, A. and Nurul Habib, K. M. (2012) “Towards a Learning-based Mode
Shift Model: A Conceptual Framework.” Transportation Letters: The International Journal of
Transportation Research, Vol. 4, No. 1, pp. 15-27.
Osman, A. O. and Shalaby, A. (2010) “A Modal Shift Optimized Transit Route Design
Model.” Paper presented at the 2nd
Annual Conference on Transportation and Logistics
(TRANSLOG), Hamilton, Ontario, Canada, June 15th
- 16th
, 2010.
CHAPTER 4 - INVESTIGATING THE EFFECTS OF PSYCHOLOGICAL FACTORS
ON MODE CHOICE BEHAVIOUR:
Idris, A. O., Nurul Habib, K.M., Tudela, A., and Shalaby, A., (2013). “Investigating the
Effects of Psychological Factors on Commuting Mode Choice Preferences.” Paper under
review for the Journal of Transportation Planning and Technologies.
Idris, A. O., Nurul Habib, K. M., Tudela, A. and Shalaby, A. (2012) “Investigating the
Effects of Psychological Factors on Commuting Mode Choice Behaviour.” Paper presented at
the 91st TRB Annual Meeting, January 22
nd - 26
th, 2012, Washington D.C.
Osman, A. O., Nurul Habib, K. M., Tudela, A. and Shalaby, A. (2011) “Investigating the
Effects of Psychological Factors Measured in a Semantic Scale on Commuting Mode Choice
Behaviour.” Paper Presented at the 12th
International Conference on Computers in Urban
Planning and Urban Management (CUPUM 2011), Lake Louise, Alberta, Canada, July 5th
-
8th
, 2011.
CHAPTER 5 - COMMUTING SURVEY FOR MODE SHIFT (COSMOS):
Idris, A. O., Nurul Habib, K. M. and Shalaby, A. (2012) “Modal Shift Forecasting Model for
Transit Service Planning: Survey Instrument Design.” Paper presented at the 12th
International Conference on Advanced Systems for Public Transport (CASPT), July 23rd
-
27th
, 2012, The Ritz – Carlton Santiago, Chile.
CHAPTER 6 - SURVEY IMPLEMENTATION, DATA COLLECTION AND
DESCRIPTION:
Idris, A. O., Nurul Habib, K. M. and Shalaby, A. (2013) “Joint RP/SP Mode Switch
Forecasting Model for Transit Service Planning.” Presented at the 92nd
TRB Annual Meeting,
January 13th
- 17th
, 2013, Washington D.C.
CHAPTER 7 - MODE CHOICE/MODAL SHIFT MODELLING:
Idris, A. O., Nurul Habib, K. M. and Shalaby, A. (2013) “Dissecting the Role of Transit
Service Attributes in Attracting Commuters.” Working paper.
vii
TABLE OF CONTENTS
ABSTRACT ............................................................................................................................... ii
ACKNOWLEDGEMENT ........................................................................................................ iv
TABLE OF CONTENTS ......................................................................................................... vii
LIST OF TABLES .................................................................................................................... xi
LIST OF FIGURES ................................................................................................................. xii
GLOSSARY ........................................................................................................................... xiv
1 INTRODUCTION ............................................................................................................. 1
1.1 Chapter Overview ....................................................................................................... 1
1.2 Problem Statement ...................................................................................................... 1
1.3 Motivation ................................................................................................................... 3
1.4 Research Goal and Objectives ..................................................................................... 4
1.5 Methodology ............................................................................................................... 5
1.6 Thesis Layout .............................................................................................................. 7
2 LITERATURE REVIEW ................................................................................................ 10
2.1 Chapter Overview ..................................................................................................... 10
2.2 Transit Planning and Route Design ........................................................................... 10
2.2.1 Current Practice in Transit Route Design .......................................................... 12
2.2.1.1 Mathematical Approaches .............................................................................. 12
2.2.1.2 Heuristic and Evolutionary Approaches ........................................................ 13
2.2.2 Limitations of Current Practice in Transit Route Design .................................. 13
2.2.2.1 Model Practicality .......................................................................................... 14
2.2.2.2 Objective Function ......................................................................................... 14
2.2.2.3 Demand Treatment ......................................................................................... 14
2.2.2.4 Model Realism ............................................................................................... 14
2.3 Current Practice in Mode Choice Modelling ............................................................ 15
2.4 Incorporating Behavioural Factors in Mode Choice Models .................................... 18
2.5 Current Practice in Survey Design ............................................................................ 22
2.6 Chapter Summary ...................................................................................................... 27
3 MODAL SHIFT MAXIMIZED TRANSIT ROUTE DESIGN MODEL ....................... 28
3.1 Chapter Overview ..................................................................................................... 28
3.2 The Conceptual Framework ...................................................................................... 28
3.3 Towards a Learning-based Mode Shift Model .......................................................... 32
viii
3.3.1 Modelling the Formation of Habits in terms of the Step Size Parameter (α) .... 34
3.3.1.1 Estimating the Step Size Parameter (α) .......................................................... 35
3.3.2 Modelling the Awareness Level in terms of the Temperature Parameter (τ) .... 36
3.3.2.1 Estimating the Temperature Parameter (τ) ..................................................... 37
3.3.3 Modelling the level of Information Provision in terms of the Updating Rules . 37
3.3.4 Numerical Simulation ........................................................................................ 41
3.3.4.1 Simulation Results.......................................................................................... 42
3.3.4.1.1 Traditional Mode Choice Model .............................................................. 42
3.3.4.1.2 Learning-based Mode Shift Model ........................................................... 43
3.3.4.1.2.1 Partial Information (Belief-based Updating Rule) ............................. 43
3.3.4.1.2.2 Partial Information (Reinforcement Learning-based Updating Rule) 44
3.3.4.1.2.3 Perfect Information............................................................................. 46
3.3.5 PRACTICAL IMPLICATIONS ........................................................................ 47
3.4 Chapter Summary ...................................................................................................... 49
4 INVESTIGATING THE EFFECTS OF PSYCHOLOGICAL FACTORS ON MODE
CHOICE BEHAVIOUR .......................................................................................................... 50
4.1 Chapter Overview ..................................................................................................... 50
4.2 Reasons for the Investigation .................................................................................... 50
4.3 Structural Equation Models (SEMs) ......................................................................... 51
4.4 Understanding Mode Choice Behaviour ................................................................... 53
4.5 Data Description ........................................................................................................ 56
4.6 Structural Equation Modelling .................................................................................. 57
4.6.1 SEM Measurement Models ................................................................................ 57
4.6.2 SEM with Latent Variables ................................................................................ 60
4.7 Investigation Outcomes ............................................................................................. 62
5 COMMUTING SURVEY FOR MODE SHIFT (COSMOS) .......................................... 64
5.1 Chapter Overview ..................................................................................................... 64
5.2 Study and Survey Objectives .................................................................................... 64
5.3 Study Area ................................................................................................................. 65
5.3.1 The Census Metropolitan Area (CMA) of Toronto ........................................... 66
5.3.2 The City of Toronto ........................................................................................... 68
5.4 Survey Sample Design .............................................................................................. 70
5.4.1 Target and Survey Populations .......................................................................... 71
5.4.2 Sampling Method ............................................................................................... 78
ix
5.4.3 Sample Size Determination ................................................................................ 80
5.4.4 Sample Allocation Method ................................................................................ 82
5.5 Survey Instrument Design ......................................................................................... 84
5.6 Chapter Summary .................................................................................................... 109
6 SURVEY IMPLEMENTATION, DATA COLLECTION and DESCRIPTION .......... 110
6.1 Chapter Overview ................................................................................................... 110
6.2 General Sample Descriptive Statistics .................................................................... 110
6.3 General Revealed Preference (RP) Information Statistics ...................................... 114
6.4 General Stated Preference (SP) Information Statistics ........................................... 121
6.5 General Psychological Information Statistics ......................................................... 123
6.6 Chapter Summary .................................................................................................... 129
7 MODE CHOICE/MODAL SHIFT MODELLING ....................................................... 130
7.1 Chapter Overview ................................................................................................... 130
7.2 Fundamental Definitions and Assumptions ............................................................ 131
7.2.1 Unit of Travel Demand .................................................................................... 131
7.2.2 Trip Purpose ..................................................................................................... 131
7.2.3 Trip Time ......................................................................................................... 131
7.2.4 Study Area ....................................................................................................... 132
7.3 Modes of Travel ...................................................................................................... 133
7.3.1 Auto Option ..................................................................................................... 133
7.3.2 Public Transit Option ....................................................................................... 134
7.3.3 Non-Motorized Transport (NMT) Option ....................................................... 135
7.4 Generating Level of Service Attributes ................................................................... 138
7.5 Modelling Commuting Work Trip Mode Choice ................................................... 138
7.5.1 General Model Specification ........................................................................... 138
7.5.2 Empirical Analysis ........................................................................................... 139
7.6 Modelling Commuting Work Trip Mode Choice with Latent Variables ................ 142
7.6.1 General Model Specification ........................................................................... 142
7.6.2 Empirical Analysis ........................................................................................... 146
7.7 Modelling Commuting Work Trip Mode Shift ....................................................... 151
7.7.1 Modelling Mode Shift for Car Users ............................................................... 151
7.7.2 Modelling Mode Shift for Transit Riders ........................................................ 159
7.7.3 Modelling Mode Shift for Non-Motorized Transport (NMT) Users ............... 162
x
7.8 Models Validation and Policy Analysis .................................................................. 165
7.9 Chapter Summary .................................................................................................... 170
8 CONCLUSIONS AND RECOMMENDATIONS ........................................................ 172
8.1 Chapter Overview ................................................................................................... 172
8.2 Research Summary .................................................................................................. 172
8.3 Research Contributions ........................................................................................... 178
8.4 Future Research ....................................................................................................... 180
9 REFERENCES .............................................................................................................. 183
Appendix: COmmuting Survey for MOde Shift (COSMOS)................................................ 192
xi
LIST OF TABLES
Table 5-1 Toronto CMA, Census Subdivisions, Population Change, 2006 to 2011 ............... 67
Table 5-2 Toronto CMA, 2006 Commuting Work Trip Breakdown by Gender ..................... 71
Table 5-3 Toronto CMA, 2006 Commuting Work Trip Percentage by Gender ...................... 71
Table 5-4 Toronto CMA, 2006 Commuting Work Trip Percentage by Mode ........................ 71
Table 5-5 City of Toronto, 2006 Commuting Work Trip Breakdown by Gender ................... 75
Table 5-6 City of Toronto, 2006 Commuting Work Trip Percentage by Gender .................... 75
Table 5-7 City of Toronto, 2006 Commuting Work Trip Percentage by Mode ...................... 75
Table 5-8 Toronto CMA, N-Proportional Sample Allocation ................................................. 83
Table 5-9 Survey Population Breakdown ................................................................................ 84
Table 5-10 City of Toronto, N-Proportional Sample Allocation ............................................. 84
Table 5-11 Average Operating Speeds for Various Transit Technologies .............................. 95
Table 5-12 Percentage of Change in Operating Speed for Various Transit Technologies ...... 96
Table 5-13 Travel Time Conversion Factors for Various Transit Technologies ..................... 96
Table 5-14 Equivalent In-Vehicle Travel Time for Various Transit Technologies ................. 96
Table 5-15 Factors and Factor Levels Used in the SP Experiment ......................................... 98
Table 5-16 D-Efficient Experimental Design (72 Choice Tasks blocked into 12 blocks) ...... 99
Table 6-1 Toronto CMA, N-Proportional Sample Allocation ............................................... 111
Table 6-2 Survey Sample Breakdown ................................................................................... 111
Table 6-3 City of Toronto, N-Proportional Sample Allocation ............................................. 112
Table 6-4 Actual and Theoretical Sampled Work Trips for the Toronto CMA .................... 113
Table 6-5 Actual and Theoretical Sampled Work Trips for the City of Toronto .................. 113
Table 6-6 Toronto CMA Sample Descriptive Statistics ........................................................ 114
Table 7-1 CMA 2006 Transit Work Trips by Access Mode ................................................. 134
Table 7-2 RP Mode Choice Model ........................................................................................ 140
Table 7-3 RP Mode Choice with Latent Variables ................................................................ 148
Table 7-4 Mode Shift Models for Car Drivers....................................................................... 155
Table 7-5 Mode Shift Models for Car Passengers and Carpoolers ........................................ 157
Table 7-6 Mode Shift Model for Transit Riders (All Access Modes) ................................... 161
Table 7-7 Mode Shift Model for Non-Motorized Transport Users ....................................... 164
Table 7-8 Forecasting Performance using a Subset of 239 Car Drivers ................................ 166
Table 7-9 Forecasting Performance using Expanded Subset of 1407 Car Drivers ................ 167
xii
LIST OF FIGURES
Figure 1-1 Thesis Organization Chart ........................................................................................ 9
Figure 2-1 Experimental Design and Final Questionnaire ...................................................... 24
Figure 3-1 Modal Shift Maximized Transit Route Design Model ........................................... 30
Figure 3-2 Agents Adjusting their Choices based on their Experience with the System ........ 33
Figure 3-3 Learning-based Mode Shift Model ........................................................................ 40
Figure 3-4 Learning-based Mode Shift Model, Partial Information, Belief-based Rule ......... 43
Figure 3-5 Learning-based Mode Shift Model, Partial Information, RL-based Rule .............. 45
Figure 3-6 Learning-based Mode Shift Model, Perfect Information ....................................... 46
Figure 4-1 Measurement and Structural Models ..................................................................... 52
Figure 4-2 The Theory of Interpersonal Behaviour (TIB) ....................................................... 54
Figure 4-3 Path Diagram Inspired by the Theory of Interpersonal Behaviour ........................ 56
Figure 4-4 Path Diagram for the Measurement Model of Car Users ....................................... 58
Figure 4-5 Path Diagram for the Measurement Model of Transit Riders ................................ 60
Figure 4-6 Path Diagram for the SEM with Latent Variables ................................................. 61
Figure 5-1 GTA and Toronto CMA Boundaries...................................................................... 65
Figure 5-2 The Census Metropolitan Area (CMA) of Toronto ............................................... 66
Figure 5-3 The City of Toronto ............................................................................................... 68
Figure 5-4 TTC Network ......................................................................................................... 69
Figure 5-5 Toronto CMA, 2006 Commuting Work Trips Mode Split .................................... 72
Figure 5-6 Toronto CMA, 2006 Commuting Work Trips Gender Split by Mode .................. 73
Figure 5-7 Toronto CMA, 2006 Commuting Work Trips Mode Split by Gender .................. 73
Figure 5-8 City of Toronto, 2006 Commuting Work Trips Mode Split .................................. 76
Figure 5-9 City of Toronto, 2006 Commuting Work Trips Gender Split by Mode ................ 77
Figure 5-10 City of Toronto, 2006 Commuting Work Trips Mode Split by Gender .............. 77
Figure 5-11 Stratification by Geography, Gender, and Mode Split ......................................... 79
Figure 5-12 Multi-Instrument COmmuting Survey for MOde Shift (COSMOS) ................... 86
Figure 5-13 Daily Commuting Work Trips and Current Travel Options ................................ 87
Figure 5-14 Stated Preference (SP) Experiment for Car Users ............................................... 90
Figure 5-15 Stated Preference (SP) Experiment for Transit Users .......................................... 91
Figure 5-16 Stated Preference (SP) Experiment for Active Mode Users ................................ 92
Figure 5-17 Habitual Behaviour ............................................................................................ 105
Figure 5-18 Affective Appraisal Dimensions of the Chosen Mode ...................................... 106
Figure 5-19 Affective Appraisal Dimensions of Public Transit ............................................ 107
Figure 5-20 Personal Attitude ................................................................................................ 107
Figure 5-21 Socioeconomic and Demographic Questions ..................................................... 108
Figure 6-1 Gender .................................................................................................................. 116
Figure 6-2 Age Distribution ................................................................................................... 117
Figure 6-3 Occupation (According to the NOC of Canada) .................................................. 117
Figure 6-4 Marital Status ....................................................................................................... 118
Figure 6-5 Dwelling Type ...................................................................................................... 118
Figure 6-6 Household Size (18 years old and above) ............................................................ 119
Figure 6-7 Household Size (below 18 years old) .................................................................. 119
xiii
Figure 6-8 Driving License Holding ...................................................................................... 120
Figure 6-9 No of Vehicles in the Household ......................................................................... 120
Figure 6-10 Personal Income Distribution ............................................................................. 121
Figure 6-11 Proportions of SP Mode Switching Behaviour .................................................. 122
Figure 6-12 Degree of Compliance to the SP Choice ............................................................ 123
Figure 6-13 Proportions of Habitual Behaviour .................................................................... 125
Figure 6-14 Emotional Response towards Primary Chosen Mode ........................................ 126
Figure 6-15 Emotional Response towards Public Transit ...................................................... 127
Figure 6-16 Proportions of Personal Attitude ........................................................................ 128
Figure 7-1 Mode Shares by Trip Length................................................................................ 136
Figure 7-2 Trip CDF by Trip Length ..................................................................................... 137
Figure 7-3 RP Mode Choice Model Structure ....................................................................... 139
Figure 7-4 RP Mode Choice Model with Latent Variables ................................................... 142
Figure 7-5 Transit Ridership Estimation ................................................................................ 169
Figure 7-6 Car Driver Mode Split Estimation ....................................................................... 169
xiv
GLOSSARY
ABC Affect, Behaviour and Cognition
ANN Artificial Neural Networks
ATIS Advanced Traveller Information Systems
AVC Asymptotic Variance-Covariance
BFGS Broyden-Fletcher-Goldfarb-Shanno
BRT Bus Rapid Transit
CAA Canadian Automobile Association
CFI Comparative Fit Index
CMA Census Metropolitan Area
COTS Customer Oriented Transit Service
COSMOS COmmuting Survey for MOde Shift
CSD Census Subdivision
DEFF Design Effect
FPM Forecasting Performance Measure
GA Genetic Algorithms
GTA Greater Toronto Area
HOV High Occupancy Vehicles
IID Independently and Identically Distributed
ITS Intelligent Transportation Systems
LISREL LInear Structural RELation
LOS Level of Service
LRT Light Rail Transit
MDP Markovian Decision Process
MILATRAS Microsimulation Learning-based Approach for TRansit Assignment
MNL Multinomial Logit
NFI Normed Fit Index
xv
NL Nested Logit
NMT Non-Motorized Transport
NOC National Occupational Classification
OD Origin-Destination
OR Operation Research
PD Planning District
RL Reinforcement Learning
RMSEA Root Mean Square Error of Approximation
ROW Right-of-Way
RP Revealed Preference
RUM Random Utility Maximization
SC Stated Choice
SEM Structural Equation Model
SP Stated Preference
SRS Simple Random Sample
SRSWOR Simple Random Sample Without Replacement
TDM Travel Demand Management
TIB Theory of Interpersonal Behaviour
TOD Transit Oriented Development
TPB Theory of Planned Behaviour
TTC Toronto Transit Commission
TTS Transportation Tomorrow Survey
1
1 INTRODUCTION
1.1 Chapter Overview
This chapter starts with a discussion of the problem statement investigated by this thesis in
Section 1.2. This is followed by an explicit presentation of the motivation of this research in
Section 1.3. Subsequently, Section 1.4 highlights the goal and main objectives of this thesis.
Then, Section 1.5 provides an overview of the research methodology. Finally, Section 1.6
presents a graphical layout of the dissertation.
1.2 Problem Statement
The growth of nations and the associated need for better mobility have remarkably
accelerated urban motorization and increased the reliance on the private automobile as the
primary mode of travel. In turn, such growth in auto dependency has provided unprecedented
levels of mobility and liberty to motorists. However, the adverse effects associated with the
use of private cars in urban areas cannot be overstated. Obviously, the unlimited use of single
occupancy vehicle has raised concerns about resource consumption, traffic congestion, and
emissions. Further, it reduced the economic, social and environmental viabilities of urban
communities (Garber and Hoel 2002; Litman and Laube 2002).
Consequently, aiming at providing a better quality movement of people, the focus of
transport planners has been directed, since the early 1970’s, towards managing the increasing
travel demand rather than boosting supply, which is known as Travel Demand Management
(TDM). In general, numerous TDM policies (e.g. congestion pricing, parking management,
and public transit provision) have been adopted to attain better mobility by changing
individuals’ travel behaviour from extensive automobile usage towards the use of more
sustainable means of transport (Meyer 1999; Ogilvie et al. 2004; Nurdden et al. 2007).
Of the TDM policies, increasing transit provision might be an effective strategy that is
capable of addressing many traffic and environmental problems. In general, public transit is a
generic term involving a large family of conventional and innovative technologies
complementing each other to provide system-wide mobility in urban and rural areas. Public
transit enables high capacity, energy efficient and low emission movement of people. In
addition, it provides auto owners who do not want to drive with an attractive travel
alternative, and it represents an essential service for those who lack access to private vehicles
2
as well as students, senior citizens and others who may be economically or physically
disadvantaged (Garber and Hoel 2002; Vuchic 2005).
In such context, there has been a growing interest in promoting sustainable communities that
incorporate compact, mixed-use development and pedestrian-friendly street network design
to support high-quality transit services. Such form of development is commonly referred to as
Transit Oriented Development (TOD). While TOD helps to support high-quality transit, it is
insufficient alone to achieve this goal, since elements of the transit service itself play a key
role in defining transit quality. Recently, the concept of Customer Oriented Transit Service
(COTS) has been promoted to further support high quality transit, with the ultimate goal of
attracting auto users to transit and maintaining acceptable levels of transit ridership (Hale
2009). COTS is characterized by fast and reliable transit service, passenger information
systems, attractive vehicle design (both interior and exterior), distinctive and attractive station
design, electronic fare collection, etc.
As noted above, the main objective of COTS is to attract and retain transit ridership while
making transit a viable competitor to auto driving. COTS is now considered an integral part
of sustainable transportation and community development programs. However, planning
sustainable communities and designing COTS are not very straightforward. The success of
any sustainable community planning and COTS design relies on how the policies and design
elements affect peoples’ travel choices and behaviour. Hence, without proper analytical tools
of evaluating the impacts of alternative sustainable transportation policies (such as TDM
policies, transit-oriented land use policies, etc.) and COTS elements (some of which are
qualitative) on travel behaviour, it is difficult to assess and develop effectively successful
TOD plans and COTS designs.
Unfortunately, classical methods of sustainable community development and transit service
planning tools are plagued with many problems. They are generally aggregate, hence more
appropriate for regional planning than community/neighbourhood planning. Moreover,
conventional mode choice models often overestimate mode shift to transit and are insensitive
to customer-oriented service elements (e.g. passenger information provision, Intelligent
Transportation Systems (ITS) technologies that improve reliability, rail vs. bus attraction,
etc.) (Winston 2000; Beimborn et al. 2003; Flyvbjerg et al. 2005; Quentin and Hong 2005;
Cantillo et al. 2007; Domarchi et al. 2008). Nevertheless, recent research advancements in
3
travel demand modelling provide a new dimension for improving current practice in
sustainable community development and transit service planning.
1.3 Motivation
Over the decades, research has continuously improved mode choice models on an analytical
viewpoint in an effort to make them better explain modal split. Nevertheless, traditional mode
choice models are criticized for their poor characterization of human behaviour and weakness
of their assumptions. Such models do not only imply rational passenger behaviour, but also
complete knowledge of the transportation system and perfect information about all the
available alternatives and their choice consequences (Barff et al. 1982; Chorus and
Timmermans 2009).
In fact, the rationality of passengers is bounded by the information they could have, the
cognitive capacity of their minds and the terminable amount of time available to them to
make decisions (Simon 1957; Barros 2010). Thus, passengers lack the ability and resources
to find an optimal solution, and they instead apply their rationality only after simplifying the
available travel choices. Hence, a passenger usually seeks a satisfactory solution rather than
the optimal one (Bamberg et al. 2003; Chorus and Timmermans 2009).
Numerous research efforts have attributed the lack of searching and processing of
information to some behavioural factors of sub-optimal characteristics that could lead to the
domination of a specific mode even in cases where the rational choice favours another
(Banister 1978; Johansson et al. 2006; Cantillo et al. 2007).
Further, evidence in the literature shows that traditional mode choice models fail to forecast
modal shift in response to new improvements in the transit service (Winston 2000; Beimborn
et al. 2003; Flyvbjerg et al. 2005; Forsey et al. 2011). The previous drawback is generally
attributed to the lack of tools that can adequately forecast the behaviour of potential transit
ridership. In particular, traditional mode choice models tend to overestimate the attractiveness
of transit for choice users which leads to over predicting transit ridership. Such models are
criticized for their weak characterization of several behavioural aspects, contributing in part
to their misleading modal shift estimation (Quentin and Hong 2005; Domarchi et al. 2008).
Furthermore, more recent research explicitly attributed the reluctance to mode switch to the
effect of some psychological aspects (Domarchi et al. 2008; Behrens and Mistro 2010).
4
Hence, conventional mode choice models may result in misleading modal split estimation in
cases where those psychological factors exist. This in turn induces a poor knowledge of the
demand for the new transit service and a subsequent difficulty in designing an economically
sustainable system.
Moreover, it is often difficult to accommodate COTS elements as well as attributes of
emerging systems and technologies, such as passenger information systems, ITS technologies
that improve reliability, and new transit technologies (e.g. LRT and BRT) in conventional
mode choice models because detailed information of such attributes are often missing in
traditional cross-sectional household-based RP travel survey data. This is a critical issue in
transit service planning where improving service to facilitate modal shift in favour of transit
is targeted.
1.4 Research Goal and Objectives
In an attempt to overcome the above mentioned limitations in current practice in sustainable
community development and transit service planning, this thesis is intended to provide a
better understanding of commuters’ preferences and mode switching behaviour. Such goal is
achieved through the completion of the following main objectives:
1. Developing a conceptual framework for a modal shift maximized transit route design
model that extends the capabilities of the existing MIcrosimulation Learning-based
Approach for TRansit ASsignment (MILATRAS) to tackle the route design and mode
shift problems.
2. Designing and implementing a multi-instrument socio-psychometric survey to collect
detailed information for mode shift modelling.
3. Developing enhanced ridership forecasting tools for improved transit service
planning.
The goal and set of main objectives of this research stem from the following facts. First, the
decision making process a passenger has to undertake while shifting to an alternative mode of
travel is informed and guided by information on the service levels of alternative modes that
have to be considered in the modelling process. Second, mode shift decisions are affected by
5
some behavioural factors in which passengers are more (less) inclined to choose (change) the
modes they are already accustomed to, which are usually overlooked in traditional choice
models.
1.5 Methodology
This research is intended to overcome the previously mentioned gaps in both mode choice
modelling and transit service planning. In particular, proper analytical tools are developed to
aid the transit service planning process by adopting the following threefold approach:
The first stage of the proposed approach is to develop a conceptual framework for a modal
shift maximized transit route design model that extends the use of traditional models beyond
forecasting transit ridership (demand) to the operational extent of transit route design
(supply). By explicitly considering the multi-objective nature of the transit route design
problem, the developed framework represents a practical transit route design tool that is more
desirable for transit planners. The proposed framework is intended to generate transit route
designs that maximize demand attraction. The framework builds upon and extends the
capabilities of the existing MIcrosimulation Learning-based Approach for TRansit
ASsignment (MILATRAS) (Wahba and Shalaby 2005; Wahba and Shalaby 2009a), to tackle
both the route design and mode shift problems. MILATRAS currently models transit
assignment given a fixed set of transit routes and transit demand (Wahba 2009; Wahba and
Shalaby 2009b). The presented framework adds a mode shift module to MILATRAS in order
to find operationally implementable transit route(s) that can provide alternative design
concepts corresponding to different service requirements. Further, modal shift barriers (e.g.
habit formation) are captured in the model by specifying a threshold or inertia against shifting
between modes. Transit demand variability among both modes and routes is considered at the
microscopic level by running the joint mode shift and route choice models of MILATRAS,
allowing for consistency between the supplied service level and passenger demand (Osman
and Shalaby 2010; Idris et al. 2012a). This thesis describes the different elements of the
conceptual framework then gives explicit attention to the development of the mode shift
module, while jointly running both components (route choice and mode shift) of MILATRAS
is left for future research.
The second stage is concerned with the design and implementation of a multi-instrument
COmmuting Survey for MOde Shift (COSMOS). COSMOS is responsible for gathering
6
Revealed Preference (RP) and Stated Preference (SP) travel data along with psychological
information of travellers associated with different modes of travel (Osman et al. 2011; Idris et
al. 2012b). The developed survey is conducted online among a representative sample of
Toronto commuters who are asked about their willingness to shift to different transit
technologies of varying characteristics. In addition to collecting common socioeconomic,
demographic and modal attributes, the survey gathered data on the revealed mode choice
behaviour as well as the stated mode switching preferences to public transit considering some
important preference attributes such as advance information provision, ITS technologies and
rail vs. bus attraction. Moreover, the survey gathered psychological information regarding
habit of auto driving, affective appraisal (emotional response), and personal attitudes
associated with different travel options. Different psychometric tools are used to capture
psychological factors affecting mode choice. Further, the survey set up a SP experiment
based on efficient experimental design techniques to maximize the information gained while
minimizing the number of hypothetical scenarios required. In the SP experiment, survey
respondents are asked to identify their propensity to perform their work trip by a non-existing
transit service in the future. In an attempt to use practical attribute level ranges in the SP
experiment, best practices in transit service planning are utilized in terms of service
accessibility standards, service frequency and headway standards, as well as service
reliability standards (Idris et al. 2012c).
The third stage is to develop enhanced ridership forecasting tools for improved transit service
planning. Econometric demand models of mode switching behaviour are estimated to
evaluate transit investments that usually target car users. As opposed to traditional mode
choice models based on RP data, adequate mode shift models are developed using state-of-
the-art methodology of combining Revealed Preference (RP) and Stated Preference (SP)
information to accurately forecast transit ridership (Idris et al. 2013).
In light of the aforementioned, the proposed research approach represents a significant step
towards a better understanding of commuters’ preferences and mode switching behaviour that
enrich the transit service design toolbox for delivering more efficient and attractive services.
7
1.6 Thesis Layout
This thesis is organized as follows:
CHAPTER 2 - LITERATURE REVIEW: This chapter starts with an overview of the
overall transit planning process, with more focus on the current practice and limitations in
transit route design, in Section 2.2. Then, Section 2.3 discusses the current practice in mode
choice modelling and its drawbacks. Subsequently, Section 2.4 highlights recent research
efforts that account for the inclusion of behavioural factors in mode choice models in an
attempt to overcome some of their limitations. Next, Section 2.5 reviews the literature
concerning the use of Stated Preference (SP) methods as a recent advancement in quantifying
people’s choices. Finally, a chapter summary is provided in Section 2.6.
CHAPTER 3 - MODAL SHIFT MAXIMIZED TRANSIT ROUTE DESIGN MODEL:
This chapter starts with a full documentation of the proposed conceptual framework for
modal shift maximized transit route design model in Section 3.2. This is followed by a
presentation of the evaluation component of the framework, where a learning-based mode
shift model is introduced as an alternative way to mode shift modelling, in Section 3.3.
Finally, a chapter summary is provided in Section 3.4.
CHAPTER 4 - INVESTIGATING THE EFFECTS OF PSYCHOLOGICAL FACTORS
ON MODE CHOICE BEHAVIOUR: This chapter starts with a discussion about the
reasons behind the presented investigation in Section 4.2. This is followed by a description of
the methodology used in the analysis in Section 4.3, and a review of the important
psychological theories that study the relationship between different aspects affecting the
decision making process underlying mode choice in Section 4.4. Then, Section 4.5 provides a
brief description of the dataset used in this investigation. Subsequently, Section 4.6 presents
the developed models, and finally Section 4.7 documents the outcomes of this investigation
and its effects on the following chapters.
CHAPTER 5 - MULTI-INSTRUMENT SURVEY DESIGN: This chapter presents an
overview of the activities involved in conducting the developed survey, with details provided
on the study and survey objectives in Section 5.2, study area in Section 5.3, survey sample
design in Section 5.4, and survey instrument design, in Section 5.5. Finally, a chapter
summary is provided in Section 5.6.
8
CHAPTER 6 - SURVEY IMPLEMENTATION, DATA COLLECTION &
DESCRIPTION: This chapter highlights the general sample descriptive statistics in Section
6.2. This is followed by presenting general Revealed Preference (RP) information statistics in
Section 6.3, and general Stated Preference (SP) information statistics in Section 6.4. Finally,
a chapter summary is provided in Section 6.5.
CHAPTER 7 - MODE CHOICE/MODAL SHIFT MODELLING: This chapter
documents the fundamental definitions and assumptions upon which the models are built in
Section 7.2. In addition, Section 7.3 provides a detailed description of the modes of travel
considered in the choice set. Then, level of service attributes generation is discussed in
Section 7.4. Further, Sections 7.5, 7.6, and 7.7 present the modelling efforts with respect to
commuting work trip mode choice, commuting work trip mode choice with latent variables,
and commuting work trip mode shift, respectively. Subsequently, Section 7.8 provides model
validation and policy analysis. Finally, Section 7.9 provides a chapter summary.
CHAPTER 8 - CONCLUSIONS AND RECOMMENDATIONS: This chapter starts with
a summary of the presented research in Section 8.2. Then, Section 8.3 highlights the main
contributions of this thesis. Finally, Section 8.4 provides ideas for future continuation of this
research.
9
Figure 1-1 Thesis Organization Chart
CHAPTER 2: LITERATURE REVIEW
This chapter includes: transit planning and route design, current practice in mode choice
modelling, incorporating behavioural factors in mode choice models, and Stated Preference (SP)
experimental design.
CHAPTER 1: INTRODUCTION
This chapter includes: problem statement, motivation, research goal and objectives, methodology,
and thesis layout.
CHAPTER 3: MODAL SHIFT MAXIMIZED TRANSIT ROUTE DESIGN MODEL
This chapter includes: the conceptual framework, and towards a learning-based mode shift model.
CHAPTER 4: INVESTIGATING THE EFFECTS OF PSYCHOLOGICAL FACTORS ON
MODE CHOICE BEHAVIOUR
This chapter includes: reasons behind the investigation, Structural Equation Models (SEMs),
understanding mode choice behaviour, data description, structural equation modelling, and
investigation outcomes.
CHAPTER 5: COMMUTING SURVEY FOR MODE SHIFT (COSMOS)
This chapter includes: study and survey objectives, study area, survey sample design, and survey
instrument design.
CHAPTER 6: SURVEY IMPLEMENTATION, DATA COLLECTION AND DESCRIPTION
This chapter includes: general sample descriptive statistics, general Revealed Preference (RP)
information statistics, and general Stated Preference (SP) information statistics.
CHAPTER 7: MODE CHOICE/MODAL SHIFT MODELLING
This chapter includes: fundamental definitions and assumptions, modes of travel, generating level
of service attributes, modelling commuting work trip mode choice, modelling commuting work
trip mode choice with latent variables, modelling commuting work trip mode shift, and models
validation and policy analysis.
CHAPTER 8: CONCLUSIONS AND RECOMMENDATIONS
This chapter includes: research summary, research contributions, and future research.
10
2 LITERATURE REVIEW
2.1 Chapter Overview
This chapter starts with an overview of the overall transit planning process, with more focus
on the current practice and limitations in transit route design, in Section 2.2. Then, Section
2.3 discusses the current practice in mode choice modelling and its drawbacks. Subsequently,
Section 2.4 highlights recent research efforts that account for the inclusion of behavioural
factors in mode choice models in an attempt to overcome some of their limitations. Next,
Section 2.5 reviews the literature concerning the use of Stated Preference (SP) methods as a
recent advancement in quantifying people’s choices. Finally, a chapter summary is provided
in Section 2.6.
2.2 Transit Planning and Route Design
As presented in the introduction of this thesis, the concept of Customer Oriented Transit
Service (COTS) has been recently promoted, with the ultimate goal of increasing mode shift
from auto towards public transit. Automobile users might consider shifting to transit if they
have an affordable and a good quality service available. Thus, transit providers attempt to
maintain attractive alternatives by improving the Level of Service (LOS) in terms of
frequency, speed, reliability, information provision, vehicle design, station design, fare
collection, etc., while minimizing its associated cost. This trade-off between quality and cost
turns the transit planning process into a multi-objective problem where passengers’ and
operator’s interests are in conflict (Kepaptsoglou and Karlaftis 2009).
In general, the transit planning process consists of three main steps. First, strategic planning
which deals with routes design; second, tactical planning which involves both frequency
setting and timetabling; and finally, operational planning including both transit unit
scheduling and crew scheduling (Ceder and Wilson 1986). In the first two steps, all
information needed by passengers is determined. Treating all those steps simultaneously
ensures the interaction and feedback between them. However, this approach is intractable in
practice due to the extreme complexity of the process which requires huge computational
effort. As a result, numerous approaches have been proposed to deal with sub-problems of the
main transit planning process in an effort to solve it in a sequential manner instead of a
simultaneous one. Such approaches usually yield sub-optimal solutions with no guarantee of
global optimality (Guihaire and Hao 2008).
11
Given the constrained fleet size and crew resources of transit agencies due to the costly
transit operations, it is essential to improve various transit network elements and enhance the
quality of the offered service in order to attain the maximum transit ridership. The transit
route design problem is a sub-problem of the main transit planning process. Several
objectives and constraints might exist for the route design problem depending on the policies
of the transit agency, but in general cost minimization is considered a main objective. On the
other hand, and from the passengers’ perspective, the transit route network should meet the
demand by providing affordable, fast, frequent, accessible and direct service. Thus, the main
challenge in route design is to achieve an acceptable trade-off between these conflicting
objectives (Guihaire and Hao 2008; Kepaptsoglou and Karlaftis 2009).
Route design aims at defining various design elements which reflect the system performance
requirements and resource limitations in order to serve the demand within a particular area. A
primary data requirement to solve the problem is the route topology which can be defined by
the roadway network and the potential locations for the transit stops, terminals, depots and
transfer zones. In addition, the origin-destination (OD) trip matrix is required to represent the
level of demand that needs to be served (Guihaire and Hao 2008; Kepaptsoglou and Karlaftis
2009). However, serving the whole transit demand is unrealistic since transit units cannot
stop at every point along a regular transit route, but rather at pre-specified stop locations only.
Thus, the service coverage is measured based on estimating the actual demand that can be
served by public transit within a reasonable walking threshold from the designated stops. In
practice, up to 400 m between either passenger’s origin or destination and the nearest transit
stop is considered an acceptable access/egress distance (Murray and Wu 2003).
In general, transit route design depends largely on the experience of the planner aided with a
set of service standards and practical guidelines that specify the minimum acceptable level of
service. That is succeeded by generating and examining a number of design scenarios based
on different combinations of design elements in order to select the best alternative. Such
approach is criticized for yielding suboptimal designs; for example in terms of attracting auto
users to transit and maintaining acceptable levels of transit ridership (the main objective of
COTS). Another limitation is being unable to capture the effect of the proposed design on the
existing demand along other transit routes and whether the attracted demand is resulting from
a mode shift or a transit route shift.
12
2.2.1 Current Practice in Transit Route Design
Route design is the first step in the transit planning process which greatly affects all
subsequent planning steps, namely frequency setting and timetabling as well as transit unit
scheduling and crew scheduling (Ceder and Wilson 1986). Transit Route design is described
as the process of determining a transit route consisting of two terminals and a sequence of
intermediate stops. Given topological characteristics and trip demand distribution, the transit
route should achieve the desired objectives of both passengers and the operator subject to a
set of constraints such as total operating cost and fleet size. Unfortunately, passengers’ and
operator’s objectives not only vary but rather conflict. From the passengers’ point of view,
the transit route should maximize service coverage, accessibility, trip directness and demand
satisfaction. On the other hand, the operator’s point of view is to ensure keeping the route
length under a certain bound so as to reduce the operating costs (Guihaire and Hao 2008;
Kepaptsoglou and Karlaftis 2009).
Numerous mathematical, heuristic and evolutionary solution methods are developed to deal
with various aspects of the transit route design such as route configuration, service
frequencies and/or other related design parameters, in either a sequential or a joint manner. In
terms of optimization strategies, analytical optimization or exact search methods such as
linear programming and some forms of integer and mixed integer programming are used
when the targeted problem can be formulated with a known mathematical model.
2.2.1.1 Mathematical Approaches
Research efforts have been focused on the applications of Operations Research (OR) on
transit route design problems. Instead of determining both the route structure and design
parameters simultaneously, these analytical optimization models were applied to determine
one or several design parameters in sequence on a predetermined transit route structure, such
as stop spacing, bus size and service frequencies (Wirasinghe and Ghoneim 1981; Oldfield
and Bly 1988; Furth and Rahbee 2000; Van Nes and Bovy 2000; Saka 2001; Murray and Wu
2003).
Although the small tested instances permit the proposed models to attain optimality, the
previous models are not applicable to larger networks as they are only effective in solving
problems of small size networks or with one or two decision variables. Therefore, they have
been criticized for their limitations in solving more complex real-world instances of the route
13
design problem in which many parameters need to be determined. Thus, transit planners
deem optimal design methods as overly theoretical and lacking simplicity, flexibility and
practical realism for real-world applications and hence they are rarely used in practice. That
in turn has directed recent research efforts at developing heuristic and evolutionary
approaches in order to overcome the limitations on hand and to find an operationally
acceptable designs, although global optimality is no longer guaranteed (Guihaire and Hao
2008).
2.2.1.2 Heuristic and Evolutionary Approaches
Heuristic approaches can deal with the transit route design problem and the determination of
its associated service frequencies. One of the first research efforts to tackle the transit route
design using heuristics was (Lampkin and Saalmans 1967). The authors treated each of route
design and frequency setting separately, then tackled them in a sequential manner instead of a
simultaneous one. Further, Mandl (1980) proposed a two-stage heuristic algorithm to define a
transit network given an empty route network and a constant frequency on all routes.
Over the past few decades, the interest in biologically motivated approaches like Artificial
Neural Networks (ANN) and Genetic Algorithms (GA) for solving optimization problems
has emerged as a new research trend. Various research efforts used genetic algorithms, which
is based on natural genetics and selection as a high-level simulation of a biologically
motivated adaptive system, in order to solve the transit route design problem (Xiong and
Schneider 1992; Pattnaik et al. 1998; Guihaire and Hao 2008).
Generally, heuristic and evolutionary approaches have showed better efficiency and less
computational effort than traditional exact mathematical optimization techniques especially
with more complex transit route design problems. An additional advantage of such methods
is the fact that they are not designed for a particular problem formulation, but rather they
define very general search frameworks and can fit to almost any form of constraints and
objectives.
2.2.2 Limitations of Current Practice in Transit Route Design
In general, route design involves the determination of both physical and operational design
elements. Although numerous mathematical, heuristic and evolutionary solution methods are
developed to tackle different aspects of the transit route design problem, several limitations
still exist in the current practice. Such limitations can be summarized in terms of model
14
practicality, objective function, demand treatment and model realism (Guihaire and Hao
2008; Kepaptsoglou and Karlaftis 2009).
2.2.2.1 Model Practicality
Many of the previous studies focus on theoretical problems as a way of examining the
performance of the proposed solution rather than finding a practical one. In addition, they fail
to incorporate practical service planning guidelines that match the operators’ needs. As a
result, the proposed designs are sometimes operationally infeasible. Hence, more practical
guidelines should be integrated in the design to improve the quality, efficiency and
applicability of the solutions.
2.2.2.2 Objective Function
In general, the most widely used objective function is either minimizing the total generalized
cost/time or maximizing user benefits. However, maximizing modal shift from private cars to
public transit, which is considered as an important desirable objective for insuring system-
wide mobility, is not precisely addressed.
2.2.2.3 Demand Treatment
For simplicity, most of the previous approaches assumed a fixed transit demand matrix which
is unresponsive to the route alignment and service quality. However, considering variable
demand is more desirable in route design, since transit demand is largely dependent on the
transit route alignment and its associated frequencies. In addition, even for those researches
that accounted for demand variability, they failed to capture the effect of the proposed design
on the existing demand along the adjacent transit routes and whether the attracted demand is
resulting from a mode shift or a transit route-shift. Further, public demand is usually
aggregated and considered as a single point demand in the centroids of zones or in other
distribution nodes, although transit demand is actually scattered and distributed around transit
routes and stops.
2.2.2.4 Model Realism
The previous research efforts lack realism by not accounting for some behavioural factors in
terms of attitudes and habit formation that could act as barriers precluding switching between
modes. In addition, most of them ignore some essential aspects of the problem when
computing the total travel time, as they primarily focus on the total in-vehicle travel time
without proper attention to access time, waiting time, transfer time and egress time. Further,
both single path transit assignment and deterministic arrival and running times of transit units
are problematic assumptions. Hence, using multiple paths transit assignments and
15
stochasticity in the arrival and running times of transit units are more realistic and can help
securing transfer possibilities, which can greatly influence transit service planning.
2.3 Current Practice in Mode Choice Modelling
Selecting the mode of transport, or “mode choice” in short, is the third step in the four-stage
(sequential demand forecasting) model. Mode choice models are traditionally based on the
Random Utility Maximization (RUM) framework originating in microeconomics, assuming
that utilities (measure of satisfaction) are random to the modeller while choice strategies are
deterministic from the decision maker’s perspective. Choice decisions can be conceptualized
under such framework where a number of travel options are available to a passenger;
according to his/her preferences, the passenger assigns weights to the different attributes
characterizing each of the competing alternatives and finally selects the travel option that
maximizes her/his utility (with a higher choice probability) considering his/her socio-
economic and demographic characteristics as well as the relative attractiveness of the
available alternatives (McFadden 1974; Banister 1978; Ben-Akiva and Boccara 1995).
Given that utilities are random (contain unknown aspects) to the modeller, they are
decomposed into two parts: a deterministic component which is calculated based on observed
characteristics, and a stochastic error term which is unobserved. Voluminous Random Utility
Maximization-based mode choice models have been developed with various types and
mathematical formulations according to different assumptions for the correlation between
random residuals. The simplest form of which is the Multinomial Logit (MNL) model,
considering error terms to be Independently and Identically Distributed (IID) following the
double exponential (Gumbel Type I extreme value) distribution with homogeneous matrix of
variance-covariance across all alternatives (Ben-Akiva and Bierlaire 1999; Chih-Wen 2005).
The assumption of independence in the MNL model implies that the error terms are
uncorrelated and have the same variance for all alternatives which gives rise to the
Independence from Irrelevant Alternatives (IIA) property. The IIA property means that the
ratio of choice probabilities between any two alternatives is unaffected by the presence of a
third alternative. While providing a very convenient form for the choice probability, the
previous assumption can be inappropriate in some situations where the unobserved attributes
related to one alternative are similar to those related to another alternative. Hence, more
complex models are developed to overcome the simplified assumptions of the MNL model.
16
For example, Generalized Extreme Value (GEV) models, Probit models, and Mixed Logit
models (McFadden 1986).
GEV models are based on the generalization of the extreme value distribution to permit
correlation between the error terms over alternatives. However, in case this correlation is
zero, a GEV model collapses to a Logit model. The Nested Logit (NL) model is a GEV
model which groups the alternatives into several nests, with error terms having the same
correlation for all alternatives within a nest and no correlation for alternatives in different
nests. Probit models are based on the assumption that the random components of utility are
normally distributed, accommodating any pattern of correlation and heteroskedasticity given
its full covariance matrix. Mixed Logit models are fully general statistical models that can
approximate any discrete choice model by allowing the random component of utility to
follow any distribution. In specific, the error term in Mixed Logit models is decomposed into
two parts. First, a part that follows any distribution specified by the observer and contains all
the correlation and heteroskedasticity. Second, another part that is IID distributed. On the
other hand, other discrete choice models have been specified by modellers by combining
concepts from different models for specific purposes. The result is a set of models that are
capable to state the probability of choosing a travel alternative under a given set of
circumstances. Mode choice models have been widely used by transport planners to predict
and derive results that describe the mode choice decisions made by passengers in response to
policy changes in the transport system (Train 2003).
In light of the above, mode choice has commonly been assumed a rational reasoning process
that is related to some socioeconomic and demographic aspects of the decision maker (e.g.
age, and gender) and other factors representing the relative attractiveness of the available
options (e.g. travel cost, and travel time) (Eriksson et al. 2008). Such treatment implies that
travellers have complete knowledge and perfect information about the available options as
well as full awareness of the changes occurring in the transport system once they occur (Barff
et al. 1982; Chorus and Timmermans 2009).
In fact, the previous assumption and implications are in conflict with research on bounded
rationality which found out that travellers are boundedly rational by the cognitive limitations
of their minds, the information they could have, and the limited amount of time available to
them to make decisions. Therefore, travellers lack the ability and resources to find an optimal
17
solution (best choice), and they instead apply their rationality only after simplifying the
available travel options. Hence, travellers always seek satisfactory solutions rather than the
optimal ones (Bamberg et al. 2003; Chorus and Timmermans 2009). Although modellers
claim that the random component of utility within the conventional mode choice modelling
framework can accommodate not only the limitations of the observer but also imperfect
information and random variation in tastes on the part of the decision maker, treating mode
choice as a rational decision making process is still in question.
Moreover, several research efforts attributed the lack of both searching and processing of
information to the existence of some psychological aspects such as habits, beliefs, values,
emotions and attitudes that have some sub-optimal characteristics and could result in the
domination of a specific travel option even in cases where the rational choice favours another
(Banister 1978; Johansson et al. 2006; Cantillo et al. 2007; Domarchi et al. 2008). Although
rational reasoning may have been the origin of many daily-based decisions, research showed
that individuals do not go through such deliberate decision making process when they repeat
the same decisions over a long period of time, as it becomes habitual (Ronis et al. 1989; Aarts
et al. 1997). Further, a more recent research added that choosing a mode of travel may often
be about overcoming negative emotions, even more than about maximizing the level of utility
(Chorus et al. 2006).
In addition to the previous fundamental limitations, the traditional approach of mode choice
modelling has been mainly considering various attributes related to the decision maker and
others related to the travel alternatives. Of the decision maker characteristics, on the one
hand, car ownership and availability are usually considered the major determinants of mode
choice. On the other hand, travel cost and time play a bigger role in determining mode choice
than others that characterize the attractiveness of the competing modes (Quarmby 1967;
Williams 1978; Barff et al. 1982). However, recent research has shown that passengers do not
usually choose a travel alternative given only marginal gains in cost or time. Instead, it was
found that certain behavioural factors may help reinforce the attractiveness of a specific
travel alternative relative to other options. Those behavioural aspects imply paying additional
psychological cost as a result of exploring and trying out different alternatives (Cantillo et al.
2007; Chorus et al. 2009).
18
As such, conventional mode choice models have been criticized for their weak
characterization of human behaviour which reduce their ability to accurately forecast
passengers’ choices (Ben-Akiva et al. 2002). Such inadequate behavioural representation
leads traditional mode choice models to overestimate the attractiveness of public transit for
choice users, and subsequently to over predict transit ridership (Winston 2000; Beimborn et
al. 2003; Flyvbjerg et al. 2005). This is a critical issue in transit planning where improving
service to facilitate modal shift to transit is targeted.
In turn, previous research recommended that mode choice modelling should be more
sensitive to some underlying behavioural aspects in order to precisely describe travel demand
(Mackett 2003; Chorus and Timmermans 2009). Such behavioural factors could resist
changing individuals’ choices, such that the same choice may prevail even after a significant
change in the transport system (Ajzen 1991; Aarts et al. 1997; Gärling et al. 1998; Fujii and
Kitamura 2003; Gärling and Axhausen 2003).
2.4 Incorporating Behavioural Factors in Mode Choice Models
As an indication of the effect of habit formation on mode choice, Sheth (1976) stated that
people tend to stay with the mode they are already accustomed to even though other modes
may be more appropriate for them. The previous argument was further supported by
Goodwin (1977) who showed that habits may prevail even in cases where the more deliberate
choice favours another mode. In addition, Aarts et al. (1997) argued that mode choice
decisions like many other routine behaviours are supposed to be often made in a habitual
mindless fashion. In a study about the moderating effect of habits on the final observed
behaviour within established commuting contexts, Gardner (2009) showed that habits usually
dominate behavioural outcomes. Furthermore, Chorus et al. (2009) found that even travellers
that make rational decisions exhibit inertia during a series of risky choices, such that
choosing the same alternative from an initial set of equally risky alternatives repeatedly is a
rewarding strategy under the essence of risk aversion.
Hence, routine-based choice can describe the reason behind the domination of car as a mode
of travel which is hard to be altered even after a policy change which favours transit.
Research has showed that a number of psychological and sociological variables help
reinforce the relative attractiveness of car as a travel option, such that the superiority of the
auto mode can be related to the strong habits towards the car. In an indication of the
19
domination of auto as a mode of travel, Gwilliam and Banister (1977) showed that auto
passengers and second drivers in one car owning households adjust their trips to allow for the
lack of car availability during the day. Other research findings implied that car is a superior
mode which will be used whenever available once the initial investment has been made
(Lucarotti 1977; Bailey 1984). Further, Ory and Mokhtarian (2005) have argued that the
increase in car use might not always be a result of cost and time savings, but rather as a result
of other behavioural factors.
Numerous research attempts have been made to incorporate socio-psychological factors in
mode choice. Two different perspectives to incorporate socio-psychological factors in the
mode choice decision were realized in the literature. First, modelling mode choice as a
learning process. Second, introducing socio-psychological factors in terms of explanatory
variables in the choice models.
An early attempt to introduce the effect of habit formation to mode choice decisions was
presented in Banister (1978). The author outlined a conceptual structure of a sequential modal
split model based on learning theory and habit formation. Starting from an important aspect
which is that travel patterns are based on decisions which are strongly influenced by habits,
the author postulated that an individual is likely to consider his previous experience while
taking his decision in the following day. Each subsequent decision is then influenced by
changes in the system and experience gained from the previous trips. Based on that, the study
suggested a four-stage decision making framework as an alternative way of looking at modal
choice.
The results showed that after a learning period, decisions may be a function of the formed
habits in terms of satisfactory outcomes from previous trips, which means that passengers do
not really choose their mode of travel, but rather personal mobility has become more
dependent on car ownership and availability while being less dependent upon the competition
between alternative modes. Strong evidence for habit formation was shown in the presented
model which implied that future choices can be predicted with high accuracy if habits are
identified. However, an important issue with this model is in modelling satisfaction,
specifically, how to define, identify and measure personal satisfaction. The approach was
rather dependent only on whether a car is owned and whether it is available to model
individual decisions.
20
Other research attempts are made to capture the intricacies of the decision making process by
incorporating socio-psychological factors within the traditional mode choice models, besides
conventional personal and modal level of service attributes, as a way to overcome their
limitations (Johansson et al. 2006; Cantillo et al. 2007; Domarchi et al. 2008; Habib et al.
2010).
Johansson et al. (2006) hypothesized that the differences in people’s attitudes and personality
traits can be revealed in their transport mode choice. The authors postulated the existence of
both safety and environmental personality traits that affect mode choice decisions. Aiming at
addressing the unobservable preferences in mode choice models, indicators of attitudes and
personality traits were used to produce latent variables with environmental propensities and
individual preferences for flexibility and safety, before including them in a conventional
discrete choice model. Having a choice set size of three modes, the authors modelled five
individual specific latent variables postulated to be important for mode choice, namely
environmental preferences, safety, comfort, convenience and flexibility. The results showed
that modal travel time and cost are significant for mode choice. In addition, latent variables in
terms of flexibility and comfort were very important and enriched the choice model. In
general, the previous research provides evidence that attitudes and personality traits are
important and should be considered in mode choice modelling.
Cantillo et al. (2007) related the reluctance to change passengers’ travel behaviour to the
formation of habits or inertia and serial correlation between the choices made by the same
passenger over time, which act as barriers to change her/his travel behaviour. The authors
incorporated randomly distributed inertia thresholds and serial correlation in a general
discrete choice model framework, assuming that passengers will only shift to an alternative
mode when the difference between utilities favours the new alternative by a threshold
reflecting the reluctance to change or inertia effect. In addition, inertia was estimated as a
function of the previous valuation of alternatives which allows for serial correlation, and
inertia thresholds were postulated to be normally distributed among individuals as a function
of their socio-economic characteristics and choice conditions. The results showed that it is
necessary to consider inertia and serial correlation effects on mode choice models in order to
avoid an unrealistic model which might lead to bias in coefficient estimates and produce
significant response errors, especially in the case of large policy changes.
21
Another research has argued that not only socio-economic factors, but also socio-
psychological factors affect mode choice decisions. In an attempt to account for the
underlying psychological factors on mode choice, Domarchi et al. (2008) used the Attitudinal
Theory in order to capture psychological factors and add them as explanatory variables into
the conventional discrete choice modelling framework.
According to the ABC-Model, an attitudinal response is formed of three basic and correlated
components, namely Affect, Behaviour and Cognition. Attitudes are defined as the result of
either direct or indirect experience with the environment. The affect is the emotional response
of the decision maker that represents her/his degree of preference for a specific good or
service. The behaviour is the verbal representation or behavioural tendency of the decision
maker, while the cognition is the evaluation of the good or service based on the decision
maker’s beliefs and knowledge about the good or service.
The authors measured passengers’ attitude, habit and affective appraisal towards their modes
of travel using ad-hoc instruments applied to a random sample of university staff members
through a questionnaire regarding work trips. The authors further constructed a revealed
preference (RP) database and added the effect of attitudinal factors through dummy variables
in the linear-in-parameters utility functions of the estimated simple MNL models. The results
showed that behaviour is not developed until car is used for individuals with strong car use
habit. After that, it is possible for travellers to develop a positive attitude towards car that
could make them have positive emotions related to that alternative and then habits are
developed and strengthened. Hence, car use becomes a vicious circle which is hard to break,
because car use habit is not based upon a rationalization of the problem and does not always
involve informed choices.
A more recent research effort to capture the unobserved latent variables in defining
perceptions and attitudes towards transit was presented by Habib et al. (2010). The authors
investigated the reasons for using transit in terms of people’s perceptions and attitudes
towards transit service quality in the oil-rich Canadian City of Calgary, Canada. A
multinomial Logit model combined with latent variable models is estimated based on the
Calgary transit customer satisfaction data survey conducted in 2007. The developed model
tested the significance of two individual specific latent variables, namely the perceptions of
‘reliability and convenience’ and ‘ride comfort’. The results showed that Calgarians value
22
‘reliability and convenience’ more than ‘ride comfort’ which imply that improving the
reliability and convenience of the transit service would effectively increase transit ridership.
In general, the previous research efforts provide evidence that mode choice is a complex
process which not only involves socio-economic factors, but also socio-psychological
variables that have shown to have strong influence on mode choice and improved the
developed models in terms of fitness and statistical significance. While being a step forward
to better explain modal split, the previous attempts still experience the following major
drawbacks. First, the majority of the discussed models accounted for indicators of some
behavioural factors without having an underlying theoretical foundation to explain the
relationship between such factors. Second, even in research that used a theoretical
background as a foundation for the analysis, the inclusion of behvioural factors through
dummy variables in the estimated models is problematic since behavioural factors are not
directly observable, but rather they are unobservable latent constructs that greatly influence
individuals’ choices. The treatment of latent variables in choice models has been deemed
necessary by behavioral researchers for long, but is often either ignored or introduced in a
sub-optimal model structure in statistical models. Third, the previous models relied mainly on
cross-sectional household-based Revealed Preference (RP) data which often suffer from
many problems. Evidence in the literature shows that cross-sectional RP data based mode
choice models fail to accurately forecast modal shift in response to new improvements in the
transit services. This is due to the weak representation of various emerging transit
technologies and Customer Oriented Transit Service (COTS) elements that are difficult to
capture in RP surveys, yet have attributes affecting travellers’ perceptions and their
subsequent mode shift.
2.5 Current Practice in Survey Design
Modelling discrete choice behaviour relies on travel data collection to elicit people
preferences. In principle, two methodologies are commonly utilized for quantifying people
choices, namely Revealed Preference (RP) and Stated Preference (SP) or Stated Choice (SC)
methods (Ben-Akiva et al. 1994). The revealed preference approach uses information
collected about actual choices made by individuals to estimate statistical demand models.
Accordingly, such approach is limited to analyzing the effect of existing factors in the
transport system (Gunn et al. 1992). Obviously, collecting RP data from the field is
challenging if the factors to be analyzed do not exist, or if they are not well known by
23
potential users (e.g. introducing a new transit service that has never been used before) (Diana
2010). In such cases, SP experiments where respondents are directly asked about their
preferences for hypothetical options may be more efficient. (Louviere et al. 2000; Arasan and
Vedagiri 2011). Hence, SP methods are capable to extend the implementation of choice
models beyond the limit provided by RP-based methods.
The design of SP experiments, originated at Marketing and Economics, have lately received
increasing attention in the transportation field. In general, the main purpose of conducting SP
experiments is to determine the independent influence of design attributes (variables or
factors) such as transit service design characteristics on an observed outcome (e.g. mode
shift) made by sampled respondents undertaking the experiment (Louviere and Hensher
1983; Louviere and Woodworth 1983).
In a typical SP survey, a number of choice tasks (hypothetical scenarios) are presented to
each respondent where he/she is asked to select one or more alternatives from amongst a
finite set of options. Such alternatives are defined by a number of different factors described
by pre-specified factor levels that are drawn from some underlying experimental design.
Conceptually, an experimental design might be thought of as a matrix of values that represent
factor levels, where the rows and columns of the matrix represent factors and choice
situations corresponding to different alternatives, respectively, as shown in Figure 2-1.
Nevertheless, the way the levels of the design factors are distributed within the experiment
plays a major role in determining the independent contribution of each attribute to the
observed choices. Moreover, the allocation of the different factor levels within the
experimental design may also affect the statistical power of the experiment and its ability to
detect statistical relationships that may exist within data (Rose and Bliemer 2009; Cooper et
al. 2011).
In light of the aforementioned, the issue of how to allocate attribute levels to the design
matrix is crucial to SP experimental designs. Over the years, research has relied upon
orthogonal experimental designs to generate the hypothetical choice tasks shown to
respondents. In general, orthogonal designs relate to the correlation structure between design
attributes such that all correlations between attributes are equal to zero (Louviere et al. 2000;
Bliemer et al. 2008; Bliemer and Rose 2011). Recently, researchers put the relevance of
orthogonal design-based SP experiments in question, claiming that orthogonality is unrelated
24
to the desirable properties of the econometric models used to analyse SP data (e.g. Logit and
probit models) (Huber and Zwerina 1996; Kanninen 2002; Kessels et al. 2006). The previous
claim was further supported by Train (2003) who argued that whilst orthogonality is an
important criterion to determine independent effects in linear models, discrete choice models
are not linear. In fact, the correlation structure between the attributes is not what matters in
models of discrete choice, but rather the correlations of the differences in the attributes (Train
2003; Bliemer et al. 2008).
Figure 2-1 Experimental Design and Final Questionnaire
The previous findings led to the emergence of a class of designs, known as efficient or
optimal designs, that is considered a recent advancement in SP experimental designs.
Efficient or optimal designs have been considered by researchers as the current best practice
of designing SP experiments. Unlike orthogonal designs, efficient designs do not merely try
to minimize the correlation between the attribute levels in the choice situations, but rather
aim at finding statistically efficient designs in terms of estimating parameters with the
smallest asymptotic standard errors. Accordingly, such designs would either improve the
reliability of the parameters estimated from SP data at a fixed sample size or reduce the
sample size required to produce a fixed level of reliability in the parameter estimates (Huber
and Zwerina 1996).
25
Given that the standard error is calculated as the root of the diagonal of the Asymptotic
Variance-Covariance (AVC) matrix of the parameters, prior information about the parameter
estimates is required in order to generate efficient designs. Such prior parameter information
can be estimated from similar studies or pilot tests. An efficient experimental design yields
data that enables parameter estimation with the lowest possible standard errors. Generally, the
efficiency of an experimental design can be derived from the AVC matrix. However, instead
of assessing a whole AVC matrix, it is easier to assess a design based on a single value.
Hence, numerous efficiency measures have been developed in order to calculate such
efficiency value representing an efficiency error that should be minimized (Rose et al. 2008).
By taking the determinant of the AVC matrix based only on a single respondent, the D-error
is the most widely used efficiency measure in the literature. Although it is hard to find in
practice, the design with the lowest D-error is called D-optimal design. Alternatively, it is
more common to look for a design with a sufficiently low D-error, or in other words the D-
efficient design. Depending on the available information on the prior parameters β, different
types of D-error have been developed such as the Dz-error (‘z’ from ‘zero’) where no
information is available (not even the sign of the parameters, β=0); the Dp-error (‘p’ from
‘priors’) where relatively accurate information is available with good approximations of β;
and Db-error (‘b’ from ‘Bayesian’) where information is available with uncertainty about the
approximations of β (Hensher et al. 2009).
Generally, efficient designs always outperform orthogonal designs when prior information
about the parameters (even only the sign) is available. Unfortunately, such information is not
usually available before estimating the parameters of the specified model. However, given
that some attributes (e.g. transit fare) are typically negatively perceived while others (e.g.
transit frequency) are positively perceived, it should be always possible to obtain some
information on the parameters (at least the signs), even without estimating them relying on
reasoning alone. Further, prior parameter estimates can be obtained by referring to similar
surveys. Otherwise, conducting a small pilot study might be useful to get an initial idea about
the parameter values (Rose and Bliemer 2009).
In efficient designs, prior parameter estimates are required to compute utilities that are
essential to obtain more information from each choice task. Maximizing information gained
from each choice situation is achieved by optimizing utility balance (i.e. avoiding situations
26
where alternatives are clearly dominating the choice set). For example, consider a choice
situation between two unlabelled transport alternatives. The first option has both a lower
travel time as well as a lower travel cost, making it clearly the preferred alternative. The first
option therefore clearly dominates in this choice situation, and therefore no information will
be gained. In contrast, a different choice situation where there is no clear dominant alternative
(i.e. the respondent has to make a clear tradeoff between travel time and cost), will provide
useful information. As illustrated in the example above, balancing the utilities of alternatives
is a desirable property of efficient designs (Huber and Zwerina 1996).
In addition to maintaining utility balance, research has shown that many features of the SP
experiments can influence the efficiency of the resulting parameter estimates. Of which,
number of attribute levels, attribute level ranges, and the number of choice tasks provided to
each respondent are of importance. Transport researchers have been questioning the ability of
respondents to comprehend and respond to complex designs that involve many alternatives,
attributes, and choice situations (treatments). In general, the lesser the number of attributes
and attribute levels, the more convenient for the respondent the design is. Commonly, the
number of attribute levels depends on the model specification. If a certain attribute is
expected to have nonlinear effects, then more than two levels are needed for this attribute to
be able to capture these nonlinearities. However, if dummy attributes are included, then the
number of levels required for these attributes is predetermined. Further, the number of
attribute levels used impact the resulting number of choice situations such that the more
levels used, the higher the number of choice situations is. Moreover, mixing the number of
attribute levels for different attributes is not desirable as it may also yield a higher number of
choice situations in order to maintain attribute level balance (Rose and Bliemer 2009).
Furthermore, the wider the attribute level ranges, the higher the efficiency of the design is.
Research have shown that having wide attribute level range (e.g. waiting time= 2 min – 12
min) is statistically preferable to having a narrow range (e.g. waiting time= 1.5 min – 2 min)
as this will lead to better parameter estimates (i.e. smaller asymptotic standard error).
However, using extremely wide ranges might result in choice tasks with dominant
alternatives which in turn would affect the choice probabilities obtained from the design.
Moreover, using too narrow attribute level ranges will result in alternatives which are largely
indistinguishable. Hence, there should be a trade-off between the statistical preference for a
wide range and the practical limitations that may limit the range while maintaining attribute
27
levels within limits that make sense to the respondents. Another important property that
substantially affects the efficiency of the design is maintaining attribute level balance (i.e. all
attribute levels appear equally in the dataset). Although imposing attribute level balance may
result in sub-optimal designs, it is generally considered a desirable property. Balancing
attribute levels ensures that the parameters are estimated on the whole range of levels, instead
of having data points at few of the attribute levels, and hence provides a good basis for
estimation (Caussade et al. 2005; Scarpa and Rose 2008; Bliemer and Rose 2009).
In terms of the number of choice situations, research did not provide evidence of any
systematic relationship between the value of the design parameter and the number of
treatments (Hensher 2001b). It has been shown that generally efficient designs with a small
number of treatments perform just as good (or even better) than a more complex design
(Bliemer and Rose 2011).
2.6 Chapter Summary
Chapter 2 overviewed the overall transit planning process and the current practice in transit
route design and its drawbacks. In addition, this chapter discussed the current practice in
mode choice modelling, its limitations, and highlighted the recent research efforts that
accounted for the inclusion of behavioural factors in mode choice models in an attempt to
overcome some of their limitations. Further, this chapter presented the Stated Preference (SP)
methods as a recent advancement in quantifying people’s choices that is capable to extend the
implementation of choice models beyond the limit provide by Revealed Preference (RP)
based methods.
28
3 MODAL SHIFT MAXIMIZED TRANSIT ROUTE DESIGN MODEL
3.1 Chapter Overview
This chapter proposes a conceptual framework for generating transit route designs that
maximize demand attraction. The framework builds upon and extends the capabilities of the
existing MIcrosimulation Learning-based Approach for TRansit ASsignment (MILATRAS)
to tackle both the route design and mode shift problems. Several psychological aspects that
act as modal shift barriers are captured in the framework as well as different Customer
Oriented Transit Service (COTS) attributes that are of importance to mode shift. Moreover,
the last section of this chapter introduces a learning-based mode shift model as a major
component of the proposed framework. The presented learning-based model is capable to
model the mode switching mechanism while being consistent with the intuition behind
bounded rationality.
The following sections of this chapter are arranged as follows: a full documentation of the
proposed conceptual framework is provided in Section 3.2. This is followed by a presentation
of the evaluation component of the framework, where a learning-based mode shift model is
introduced as an alternative way to mode shift modelling, in Section 3.3. Finally, a chapter
summary is provided in Section 3.4.
3.2 The Conceptual Framework
In general, the proposed framework for modal shift maximized transit route design is
intended to fill some of the current gaps in the route design literature with respect to model
practicality, objective function, demand treatment, and model realism. The presented
conceptual framework builds upon and extends the capabilities of the existing MILATRAS
(Wahba and Shalaby 2005; Wahba and Shalaby 2009a), as a component of an integrated
transit service planning framework. MILATRAS currently models transit assignment given a
fixed set of transit routes and transit demand (Wahba 2009; Wahba and Shalaby 2009b). The
proposed framework adds a mode shift module, based on the models developed later in
Chapter 7, to MILATRAS to enable evaluating the impact of transit investments that usually
target car users. The added mode shift module allows MILATRAS to capture the variability
of demand among both modes and routes at the microscopic level, by running its joint mode
switching and multiple-path transit assignment models. Modal shift barriers (e.g. habit
formation) are captured in the framework. Further, the framework considers different transit
Level of Service (LOS) attributes that are of importance to mode shift modelling.
29
The proposed approach is divided into two main parts: a design tool and an evaluation
component. On the one hand, the design tool deals with generating transit route design(s)
based on shortest path algorithms, service guidelines, and constraints regarding several
design aspects such as minimum stop spacing and maximum route length. The evaluation
component, on the other hand, is concerned with the assessment of the generated route
design(s) considering transit demand variability among both modes and routes. Integrating
both components together using a feedback loop results in a modal shift maximized transit
route design model that is capable to select the optimum transit route alignment and design
characteristics with the ultimate goal of maximizing transit ridership. However, this thesis
describes the whole conceptual framework and reports only on the evaluation component
(mode shift module), while the design tool is left for future research. The proposed
framework is more desirable for transit service planning than the previous approaches as it
explicitly considers the multi-objective nature of the transit route design problem from the
points of view of both the transit user and the transit operator.
Figure 3-1 presents the conceptual framework of the proposed model with its two main
elements that deal with service design and evaluation. Given the socio-demographic and
psychological information, as well as modal attributes of the anticipated travel demand, the
framework adopts the mode choice model with latent variables developed later in Section 7.6
to estimate the modal share of each mode of travel. The estimated transit demand is then used
as an input for the design tool to generate transit route design(s) and allocate transit stops
according to practical guidelines, service standards, and subject to a set of constraints.
Subsequently, a critical component deals with the determination of route frequency and fleet
size given the estimated transit demand and subject to a set of constraints. At that stage, the
designed transit service is ready to be assessed in terms of mode/route shift using the
evaluation component of the framework.
The evaluation component utilizes the mode shift models developed later in Section 7.7 to
examine mode shift in response to the changes in the transit network. Separate mode
switching models for different mode users (e.g. car drivers and car passengers) may be used
to estimate their propensity to shift to public transit. The route shift component of
MILATRAS is then used to assign the estimated transit demand among different transit
routes given the updated transit network. This process is repeated iteratively while revisiting
the route design until reaching a state of choice stabilization among both modes and routes.
30
Route Choice
Using MILATRAS
Yes
No
Transit Route Generation
Transit Stops Allocation
Route Shift Modelling
(MILATRAS)
Current Transit DemandL
ea
rnin
g P
roc
es
sT
ran
sit R
ou
te D
es
ign
(Stra
teg
ic P
lan
nin
g)
Total Demand
Including
Personal and
Modal Attributes
Stop
Route Topology,
Demand
Distribution,
Service
Standards and
Constraints
Update
Transit Demand
Frequency Setting
Updated Transit Network
Fre
qu
en
cy
Se
tting
(Ta
ctic
al P
lan
nin
g)
Learning-based Mode Shift Model
To
tal T
rave
l De
ma
nd
Mode Choice Model
with Latent Variables
Transit
Demand
Stability?
De
sig
n T
oo
lE
va
lua
tion
Co
mp
on
en
t
Mode Shift Modelling
Transit Demand
Figure 3-1 Modal Shift Maximized Transit Route Design Model
As presented, the proposed approach provides an integrated framework for designing transit
routes that maximize demand attraction. Elements of the demand side interact with
components of the supply side using a feedback loop until equilibrium is reached. This
treatment ensures the consistency between transit level of service and transit demand, which
is more desirable for transit service planning. Further, the presented approach enables
capturing whether the attracted demand is a result of mode shift or route shift.
Moreover, a learning-based approach for mode shift modelling is presented in the following
section. The developed approach is capable to model the mode switching mechanism while
being consistent with the intuition behind bounded rationality. The proposed learning-based
mode shift model is built on top of the mode shift models developed later in Section 7.7. The
learning process, however, ensures modelling personal behaviour at the individual level
based on personal experience and evaluation of the transportation system in a more dynamic
31
fashion which is more compatible with MILATRAS. Further, the learning process models the
mode switching mechanism while simultaneously accounting for habitual inertia against
shifting modes, different levels of information provision and awareness limitations. What is
unique to the proposed approach is that it models the insights of the decision making process
and the period of time required to reap the benefits of the proposed policy changes.
In fact, the need for a learning-based mode shift model is supported by the following facts.
First, the decision process a passenger has to undertake while shifting to a mode of travel is
informed and guided by information on the service levels of alternative modes. Such
knowledge is usually gained through various means (including travel experience) over time.
Second, mode shift decisions are affected by some behavioural factors such as habit
formation, in which passengers are more (less) inclined to choose (change) the modes they
are already accustomed to. Third, the stochastic and time-dependent nature of the
transportation system most likely gives rise to adaptive mode choice decisions by passengers,
in which they may learn their choice decisions over time by updating their expected utilities
for each mode of travel based on previous experience (Wahba and Shalaby 2005).
The proposed approach uses methods of learning to model travel behaviour (i.e. mode
switching), representing individual travellers as agents. The underlying hypothesis is that
individual passengers are expected to adjust their choice behaviour according to their
experience with the transport system performance and their previous valuation of the
available alternatives as stored in a “mental model”. Mode shift, therefore, is modelled as a
dynamic process of repetitively making decisions and updating perceptions, according to a
long term adaptive learning process. By iteratively making a decision, an individual acquires
knowledge about his/her environment and thereby forms expectations about attributes of the
environment. Individuals may make different choices over time and thus learn which of these
alternatives is more effective in achieving particular goals.
As opposed to traditional discrete choice models, the decision making process is modelled
using the concepts of Markovian Decision Process (MDP), which represents a stochastic
process, where mode shift decisions are rewarded (or penalized) and consequently
passenger’s optimal policies can be estimated and updated under either perfect or partial
information availability using either reinforcement learning or belief-based updating rules.
Hence, for example, the proposed research is innovative in dealing with the issue of service
32
reliability and the way it affects mode choice decisions. The dynamic feedback, using
learning and adaptation, is unique to the proposed framework and presents a more
behaviourally sound approach to mode shift analysis.
In general, the presented framework can be conceptualized under both the Reinforcement
Learning (RL) concepts and the Random Utility Maximization (RUM) Theory. On the one
hand, the framework employs the principles of RL to account for the long-term accumulation
of rewards while considering some behavioural aspects that affect the learning process.
Specifically, the formation of habits, the level of awareness of the changes in the transport
system and finally different updating rules are used to represent different cases of information
provision. On the other hand, the framework also employs the principles of RUM to measure
passengers’ satisfaction in an immediate sense in terms of the short-term reward within the
learning process.
Some behavioural aspects that act as modal shift barriers are captured in the model by
specifying a threshold or inertia against shifting between modes. Such thresholds are
estimated using different updating rules in the learning process to account for individuals’
previous valuation of alternatives, information availability and choice conditions. Knowing
that information regarding transit network conditions can be made available to car drivers,
and traffic conditions can be provided to transit riders, the proposed approach is more
desirable for evaluating the effects of information provision in terms of the impacts of ITS
deployments on service reliability. Further, with its microscopic representation level of transit
network supply and transit demand, this approach is suitable for the analysis of Bus Rapid
Transit (BRT) and Light Rail Transit (LRT) initiatives where design details and behavioural
aspects combine together to drive the choice decision. As a result, some hidden aspects of the
choice can be addressed, such as the higher tendency of passengers to shift towards LRT than
other means of travel (also known as the rail effect), this phenomenon can be postulated to
have some socio-psychological aspects rather than only marginal gains.
3.3 Towards a Learning-based Mode Shift Model
In general, learning-based models have been widely used in a number of various fields of
research. A major contribution of this thesis is modelling mode shift decisions as a learning-
based process which involves learning by interaction with the transportation system in a
dynamic context. The proposed approach can be conceptualized under both the
33
Reinforcement Learning (RL) concepts and the Random Utility Maximization (RUM)
framework. Within the proposed approach, passengers’ choices depend on examining the
system and updating their perceptions of modal utilities while being influenced by some
behavioural aspects that act against learning new knowledge (e.g. formation of habits). In
particular, mode shift is modelled as a long term decision process which involves learning
over a period of time until reaching a state of habit stabilization.
Within the reinforcement learning concepts, passengers are assumed to be goal-directed
agents that apply an optimal policy to choose the best alternative. At each episode, agents
perceive the state of the system and choose a mode of travel accordingly while considering
their past experiences. Based on earning positive or negative rewards, agents adjust their
choices (e.g. mode choice) while seeking to maximize the total return received in the long-
term in terms of a value function considering travel time, cost, and other factors that affect
the choice decision (Wahba and Shalaby 2009b). For instance, according to Barto and Sutton
(1998), if a state (st) is visited at time (t), the reinforcement learner updates its long-term
estimate V(st+1) based on the immediate reward gained after that visit R(st), in addition to
what happened before that visit V(st), using a simple reinforcement learning updating rule as
follows:
V(st+1) ← V(st) + α [R(st) - V(st)], (3-1)
where:
α: Step size parameter
Figure 3-2 Agents Adjusting their Choices based on their Experience with the System
34
In an attempt to comply with the context of bounded rationality, the proposed approach
employs the principles of reinforcement learning to account for the long-term accumulation
of rewards while considering some behavioural aspects that affect the learning process.
Specifically, the formation of habits is modelled in terms of the step size parameter (α), the
level of awareness of the changes in the transport system is considered in terms of the
temperature parameter (τ), and finally different updating rules are used to represent different
cases of information provision. On the other hand, the proposed approach also employs the
principles of the Random Utility Maximization (RUM) Theory to measure passengers’
satisfaction in an immediate sense in terms of the short-term reward within the learning
process.
3.3.1 ModellingtheFormationofHabitsintermsoftheStepSizeParameter(α)
The step size parameter (α) is a small positive fraction (0 ≤ α ≤ 1) which is commonly used in
reinforcement learning methods to influence the learning rate, such that the higher the step
size is, the more the agent learns from recent experience. The step size parameter is generally
reduced over time within the learning process as the agent tends to rely more on what it has
already learnt. From a behavioural perspective, this adaptive learning mechanism can
describe mode shift barriers where travellers (more specifically commuters) become more
systematic with respect to their chosen mode and insensitive to changes in the transport
system once habits are formed towards a specific mode of travel. Once the learning process
has been completed, small scale economic policies may be of little effect due to habit
formation, as indicated by Lucarotti (1977) and further supported by Banister (1978) who
showed that commuters may not change their chosen mode until a certain threshold of the
corresponding utility has been reached. Consequently, future choices can be predicted with a
high degree of accuracy if habits are identified, which is possible knowing that habits are
characterized by their invariability, repetition and persistence (Golledge and Brown 1967).
The previous findings provide evidence that habits act against learning new knowledge,
which is opposite to the function of the step size parameter in the reinforcement learning
process. Thus, habits can be modelled in terms of the value of the step size parameter such
that the strength of formed habits is inversely proportional with the step size towards learning
new knowledge.
35
3.3.1.1 Estimating the Step Size Parameter (α)
Over the past few decades, the conventional choice rule for modelling choice decisions has
been the Logit or exponential rule. Logit models are discrete choice models that attempt to
explain the behavior of individuals making choices between a finite number of alternatives.
In the Logit model, actions with higher propensities are chosen with higher probabilities
(Hopkins 2007).
(3-2)
where:
Pim : Probability that decision maker (i) selects alternative (m)
Vim : Systematic utility that decision maker (i) obtains from alternative (m);
i= 1, ...,I; m, n= 1,…,N
Research findings showed that traditional mode choice models do not only provide
information about the probabilities of mode selection in a stochastic manner, but also the
explanatory variables of the models imply some behavioural aspects. As an early indication
of strong habits towards the auto mode, Banister (1978) found that the car is almost
invariably preferred whenever available and argued that personal mobility is dependent on
car ownership, licence holding and car availability and less dependent upon the competitive
attributes of alternative modes. This finding has been further supported by Domarchi et al.
(2008) who showed that habitual frequency of car use is positively correlated with car
availability, which means that the inclusion of auto ownership as an explanatory variable in
traditional mode choice models can act as an indirect measure of car use habit.
In light of the above, the proposed approach hypothesizes that the previous choice probability
of a particular mode can be considered as an indicator of habitual inertia towards that mode.
This hypothesis implies that the higher the previous choice probability, the stronger the
formed habits towards that choice. In addition, knowing that habits act against learning new
knowledge, the step size parameter is postulated to be an inversely proportional function of
the previous dominating mode choice probability, for example:
e.g. αi = 1 - Pid, (3-3)
,P
1
im
N
n
V
V
im
im
e
e
36
where:
αi: Step size parameter associated with decision maker (i)
Pid: Previous mode choice probability that decision maker (i) selects the dominating
alternative (d)
Within such treatment, previous dominating mode choice probabilities represent agents’
willingness to switch modes after a change in the system such that if the value of Pid is close
to one (i.e. αi closer to zero) then there is a strong inertia towards the previous choice and the
previous mode prevails (i.e. the agent will not learn much from recent experience).
3.3.2 ModellingtheAwarenessLevelintermsoftheTemperatureParameter(τ)
In general, balancing exploration and exploitation is an issue in reinforcement learning
approaches. Exploiting the actions estimated (through agent learning) to be best is usually
insufficient, because many relevant state-action pairs in the reinforcement learning
framework may never be visited by the agent. Excessive exploration on the other hand will
make it hard to learn the good actions to take at different states. Therefore, maintaining a
balance between exploration and exploitation is necessary to ensure that the agent is really
learning to take the optimal decisions (Barto and Sutton 1998).
One popular technique of exploration is the ε-greedy method, where a learner behaves
greedily most of the time but every once in a while it selects an action at random with small
probability ε. The disadvantage of this method is that it chooses among all actions with equal
probability, irrespective of the estimated reward value of each action. An alternative is to use
the softmax action selection method, where actions with higher estimated rewards are chosen
with greater priority than actions with lesser estimated rewards (Abdulhai and Kattan 2003).
The most common softmax method relies on the Boltzmann distribution where an action (a)
is selected using the following probability:
(3-4)
where:
Qr (a): Value of action (a)
τ : Temperature parameter
,P
1
/)(
/)(
a
n
b
bQ
aQ
t
t
e
e
37
The temperature parameter (0 < τ ≤ ∞) is a positive parameter controlling the degree to which
actions with higher values are favoured in selection. In general, high temperatures cause all
actions to have nearly the same probability of selection, whereas low temperatures increase
the difference in the action selection probability. In other words, as τ tends to 0+, softmax
action selection becomes the same as greedy action selection (Barto and Sutton 1998).
3.3.2.1 Estimating the Temperature Parameter (τ)
From a behavioural viewpoint, strong habit formation can act against exploring new
alternatives and consequently against being aware of recent changes in the transport system.
For example, travellers might be unaware of the changes in the transit service due to the lack
of exploring as a result of strong habits towards driving. In other words, the formation of
habits might put passengers in a state of limited awareness of system changes.
In such context, the exploration rate can be maintained to address passengers’ awareness of
changes in the transport system which in turn is affected by the strength of the formed habits.
In particular, the temperature parameter (τ) is assumed to be inversely proportional with the
previous dominating mode choice probability which acts as an indicator of habitual inertia,
for example:
τi= 1 - Pid, (3-5)
where:
τi: Temperature parameter associated with decision maker (i)
Pid: Previous mode choice probability that decision maker (i) selects the dominating
alternative (d)
This assumption implies that the higher the previous choice probability of a specific mode,
the stronger the formed habits towards that mode, and consequently the lower the temperature
parameter (i.e. the agent tends to exploit greedily). This treatment maintains the balance
between exploration and exploitation as a function of the previous dominating mode choice
probabilities such that if the value of Pid is close to one (i.e. τi closer to zero) then there is a
strong inertia towards the previous choice and the agent tends to exploit and vice versa.
3.3.3 Modelling the level of Information Provision in terms of the Updating Rules
One of the drawbacks of the traditional mode choice models is being unable to precisely
address the effect of information provision on mode choice decisions, specifically the
variation in the perceived transport service attributes and the way it could affect the choice
38
behaviour among passengers over time (Quentin and Hong 2005; Wahba and Shalaby
2009a).
In order to describe the choice behaviour, adaptive learning models assume that consumers
learn about the relative quality of products adaptively using learning rules. In this context,
Hopkins (2007) showed that small differences in the learning rules between belief-based and
reinforcement learning behaviour can have large effects on market outcomes. In addition, the
results showed that even simple adaptive learning models can help explain actual choice
behaviour at the micro decision making level.
In general, two commonly used assumptions about available information can be identified
while updating agent propensities, according to Hopkins (2007). The first corresponds to a
state of partial information, in which an agent can only observe the reward resulting from the
implemented action. The second corresponds to a state of full (perfect) information, in which
an agent can observe the return of all possible actions including the rewards of actions that
were not taken.
An important aspect of mode choice decisions under the assumptions of bounded rationality
is that an agent is considered in a situation of partial information while choosing a travel
option, as it (the agent) can only perceive the reward from the alternative that is actually
chosen, while information about unselected modes is unavailable. Therefore, updating rules
under partial information are more desirable while dealing with mode choice.
Two updating rules can be described under the states of partial information, a belief-based
learning and a stimulus-response type learning (reinforcement learning) rule. The belief-
based learning rule can be described as follows. At time (t), a passenger (i) perceives a utility
(Rim(t)) after choosing mode (m). He/she updates his/her utilities as follows:
Vim(t+1)= Vim(t) + αi [Rim(t) –Vim(t)], Vin(t+1) = Vin(t), for all n ≠ m, (3-6)
where:
Vim(t) : Utility that decision maker (i) obtained from mode (m) till time step (t)
Vin(t) : Utility that decision maker (i) obtained from mode (n≠m) till time step (t)
Rim(t) : Immediate utility that decision maker (i) obtains from mode (m) at time step (t)
Vim(t+1): Updated utility that decision maker (i) obtains from mode (m) at time step (t+1)
Vin(t+1) : Updated utility that decision maker (i) obtains from mode (n≠m) at time step (t+1)
αi : Step size parameter (0 ≤ αi ≤ 1), associated with decision maker (i)
39
If values of αi are closer to zero, then agent’s experience from long ago still have a significant
effect on current beliefs, while values of αi closer to one means that only the very recent
experience is remembered. Within the previous model, the propensity towards the selected
mode potentially incorporates the reward of each action, while the utility of each unselected
alternative remains unaltered, as there is no new information about the alterative with which
to update its value. In this model, the agent is assumed to have adaptively formed beliefs
about the quality of each of the competing modes.
The reinforcement type learning, on the other hand, could be conceptualized as follows. At
time (t), a passenger (i) perceives a utility (Rim(t)) after choosing mode (m). Upon perceiving
the utility, the passenger updates his/her utilities as follows:
Vim(t+1)= Vim(t) + αi [Rim(t) –Vim(t)], Vin(t+1) = αi Vin(t), for all n ≠ m, (3-7)
Within the previous model, the propensity towards the selected mode potentially incorporates
an accumulation of positive feelings (e.g. familiarity or recognition) such that the utility of
the unselected modes decreases naturally as familiarity with those alternatives declines
relative to the selected mode.
Importantly, both updating rules respond only to the experienced utilities of the selected
mode (Rim(t)), while information on utilities of the unselected modes is not utilized because
they were not observed/experienced at that time. That can be interpreted as being in a state of
partial information, or in other words, the agent is boundedly rational (Simon 1957; Barros
2010). However, such rules might be inadequate for evaluating the effects of the emerging
information technologies in terms of the impacts of Intelligent Transportation Systems (ITS)
deployments on service reliability and real-time information provision capabilities, which in
turn affect passengers’ behaviour.
Although a state of perfect information might not practically exist, it could be argued that the
new Advanced Traveller Information Systems (ATIS) can supply travellers with information
on the alternative modes as well as the selected option. In other words, passengers may
receive real-time information on the travel times of different modes through the network.
Hence, an updating rule under the state of perfect information can be used to update
simultaneously all utilities as follows:
40
Vim(t+1)= Vim(t) + αi [Rim(t) –Vim(t)], for m= 1 to M (3-8)
The proposed approach is the first towards a learning-based mode shift model. The
underlying hypothesis is that passengers are expected to adjust their choices according to
their experience with the performance of the transport system and their previous valuation of
the available alternatives, while being subject to awareness limitations and habit formation
that might have been formed towards a specific mode of travel.
In this research, individual passengers are represented as agents that are endowed with
different propensities associated with each of the possible choices in the choice set. As utility
maximizers, the agents’ policy is to choose the travel option that maximizes their satisfaction
on long-term basis. However, balancing exploration and exploitation is used to ensure that
agents can make different choices over time and thus learn which of these alternatives is
more effective in achieving the desired goals. Further, agents can examine their choices by
interacting with the transport system through a microsimulation model which represents the
agents’ environment. The outlined learning mechanism is iterated until the agents learn their
choices and achieve a state of choice stabilization, as shown in Figure 3-3.
Agents, endowed with
different propensities
and formed habits
Decision to Explore/Exploit
Mode Choice
Examining the Choice
Through Microsimulation
Estimate
Immediate Reward
Update
Long-Term Value
Total Demand with
Indicators of
Habit Formation and
Awareness Limitations
Stop
Yes
No
Le
arn
ing
-ba
se
d M
od
e S
hift M
od
el
Learning Process
Mode
Choice
Stability?
Figure 3-3 Learning-based Mode Shift Model
41
Under such framework, mode shift is modelled as a dynamic process of repetitively making
decisions and updating perceptions according to a long-term adaptive learning process. This
dynamic feedback, using learning and adaptation, is unique to the proposed framework and
presents a more behaviourally sound mode shift approach that is suitable to perform mode
shift analysis after a policy change. Moreover, the presented approach can be looked at as a
simultaneous way of modelling both mode choice/shift and network assignment as opposed
to the traditional way of sequential modelling.
3.3.4 Numerical Simulation
This section provides numerical simulation results and compares different choice strategies
under both the traditional and learning-based mode shift modelling frameworks. Various
updating rules are examined under different states of information provision, specifically the
states of partial and perfect information and how they could affect the mode switching
behaviour. Two updating rules are modelled under the state of partial information, namely a
belief-based rule which considers the accumulation of former beliefs about the quality of
each alternative and a reinforcement learning-based rule which considers natural decay of
beliefs where familiarity with those alternatives decline. On the other hand, one updating rule
is modelled under the state of perfect information which assumes the availability of
information regarding the unselected modes to the decision maker.
The modelling scenario considers a hypothetical mode choice situation, in which 100
passengers face a daily mode choice between auto, transit and walk options. Under the
traditional mode choice framework, a simple conventional Logit model is used to estimate the
choice probabilities based only on travel time and cost without dealing with the different
transit travel time components (access/egress time, wait time, in vehicle time and transfer
time), in order to simplify the calculations. The learning-based mode shift model on the other
hand involves examining the system and updating long-term experience throughout twenty
learning episodes during which the decision makers choose a travel alternative, earn an
immediate reward and accordingly update long-term estimates. The specifications of the used
Logit model are as follows:
V(Auto) = 1.0 - 0.1 * Auto In-Vehicle Travel Time (min) - 0.05 * Auto Travel Cost ($)
V(Transit)= - 0.1 * Transit In-Vehicle Travel Time (min) - 0.05 * Transit Fare ($)
V(Walk) = -0.5 - 0.1 * Walk Travel Time (min)
42
The simulation scenario assumes that on episode one the attributes corresponding to each
alternative are as follows: Auto In-Vehicle Travel Time= 15 min, Transit In-Vehicle Travel
Time= 15 min, Walk Travel Time= 30 min, Auto Travel Cost= $1.6 and Transit Fare= $1.5.
Between episode one and episode two, the transit travel time is reduced due to a significant
change that favours the transit option such that Transit In-Vehicle Travel Time= 4 min.
In order to illustrate the evolution of habits with respect to the change in the awareness level,
balancing exploration and exploitation ensures continual exploration after the fifth episode at
which the decision maker becomes aware of the changes in transit mode by means of direct
experience. In other words, the reinforcement learner will act greedily (i.e. exploit its choice)
by choosing the most favourable mode based on what it has learnt at the beginning of the
simulation. At the fifth episode, the agent becomes aware of the changes in the system and
the assumption of exploring starts (i.e. explore transit), and continues until the termination of
the simulation. Importantly, note that the choice situations, the values of the attributes and the
model parameters are chosen arbitrarily; hence, the outcomes presented in this section should
be considered merely an illustration of the model, not a case study.
3.3.4.1 Simulation Results
3.3.4.1.1 Traditional Mode Choice Model
Based on the conventional mode choice framework, the auto mode was the most attractive
alternative with the highest choice probability (70.2%) on episode one. However, after
reaching the steady state conditions following the reduction in transit travel time on episode
two, the transit option took the lead of the choice where 51.3% of the passengers use transit
and 46% use auto.
Obviously, conventional mode choice models have always been cross-sectional models under
steady state conditions before and after the change. Hence, they might be useful to describe
mode shares but they do not capture the time-dependent processes underlying a possible
mode shift which involves breaking the previously formed habits and building new ones. In
other words, conventional models cannot provide information about the period of time
required to reap the benefits of a proposed policy scenario. This drawback of conventional
mode choice models in policy analysis further supports the need for a learning-based mode
shift model, which is presented in the following sections.
43
3.3.4.1.2 Learning-based Mode Shift Model
In order to utilize the effectiveness of traditional mode choice models in describing current
mode split, the outcomes of the above mode choice model in terms of utilities and choice
probabilities for each of the competing modes were used as the initial values of the learning
process before the policy change. However, different updating rules were used to model the
evolution of the agent’s experience throughout the simulation with respect to different cases
of information provision.
3.3.4.1.2.1 Partial Information (Belief-based Updating Rule)
Vim(t+1)= Vim(t) + αi [Rim(t) –Vim(t)], Vin(t+1) = Vin(t), for all n ≠ m (3-6)
Although the transit option had the higher choice probability after the reduction in its travel
time, the superiority of transit is realized on episode ten after passing a period of unawareness
which is followed by a period of reformation of habits. The simulation results based on the
belief-based model are illustrated in Figure 3-4.
Figure 3-4 Learning-based Mode Shift Model, Partial Information, Belief-based Rule
Initially, the same values of choice probabilities across modes were used such that the car
was the superior mode. After the change on the second episode, an increase in the observed
utility of transit has been achieved favouring transit over other modes in the choice set.
44
However, and knowing that the agent is still exploiting its favourite choice (i.e. auto mode),
this increment in transit utility was not yet observed by the decision maker. In other words,
the agent was unaware of the changes in the system till the fifth episode when it started to
explore and directly examine the transit mode. Based on that, the values of choice
probabilities remained unaltered during the first five episodes as shown in Figure 3-4, which
could be interpreted as being in a state of unawareness.
During the unawareness zone, the previously formed habits remained stable. However, when
the agent became aware of the changes at the fifth episode, habits started to reform according
to the new experience with the system until reaching another stabilization zone on episode
sixteen.
It can be also noticed that the superiority of transit has not been realized immediately after
being aware of the changes, but rather after a transition period of habits reformation from the
fifth episode till the tenth episode, in which the long-term value of transit utility exceeded the
long-term value of the utility of auto and hence transit mode was more likely to be selected.
In this scenario, the length of the modal shift period only depends on the impact of the policy
change on the utility functions, regardless of how long the agent has been exploiting the auto
mode before exploring the transit mode.
Obviously, the belief-based updating rule is in line with the assumptions of bounded
rationality such that the propensity towards the selected mode potentially incorporates the
reward of each action, while the utility of each unselected alternative remains unaltered, as
there is no new information about the alterative with which to update its value.
Importantly, habits started to reform only after examining and being aware of the change in
the service. In this model, the agent is assumed to have adaptively formed beliefs about the
quality of each of the competing modes.
3.3.4.1.2.2 Partial Information (Reinforcement Learning-based Updating Rule)
Vim(t+1)= Vim(t) + αi [Rim(t) –Vim(t)], Vin(t+1) = αi Vin(t), for all n ≠ m (3-7)
Although the transit option had the higher choice probability after the reduction in its travel
time, the superiority of transit is realized on episode eleven after passing a period of
45
unawareness which is followed by a period of reformation of habits. The simulation results
based on the reinforcement learning-based model are illustrated in Figure 3-5.
Figure 3-5 Learning-based Mode Shift Model, Partial Information, RL-based Rule
Initially, the same values of choice probabilities across modes were used such that the car
was the favourable mode. After the change on episode two, an increase in the observed utility
of transit has been achieved promoting transit over other modes in the choice set. However,
and knowing that the agent is still exploiting the auto mode, this increment in transit utility
was not yet observed by the decision maker till the fifth episode as being in the unawareness
state.
Unlike the belief-based rule, the previously formed habits in addition to the modal choice
probabilities were always varying with respect to time such that the superiority of auto mode
was increasing with familiarity while being decaying for the other modes with unfamiliarity.
However, when the agent became aware of the changes and started to get familiar with the
transit mode, the attractiveness of the car started to decline while that of transit started to
develop until it became the superior mode which is more likely to be selected at the eleventh
episode.
46
Similar to the belief-based updating rule, the superiority of transit has not been realized
immediately after being aware of the changes, but rather after a transition period of habits
reformation from the fifth episode till the eleventh episode. However, under the
reinforcement learning-based updating rule, the length of the modal shift period is longer as it
depends on both the impact of the change on the utility functions and the period of time the
agent has exploited the auto mode (i.e. frequency of past use) before exploring public transit.
Obviously, the reinforcement learning-based updating rule is based on familiarity and
accumulation of positive feelings such that the utility of the unselected modes decreases
naturally as familiarity with those alternatives declines. Hence, habits were always varying
depending on the frequency of past choice regardless of being aware of the change or not.
3.3.4.1.2.3 Perfect Information
Vim(t+1)= Vim(t) + αi [Rim(t) –Vim(t)], for m= 1 to M (3-8)
Although the transit option had the higher choice probability after the reduction in its travel
time, and knowing that the decision maker is fully aware of the system characteristics, the
superiority of transit is realized on episode six after passing a period of reformation of habits.
The simulation results based on the assumption of perfect information are illustrated in
Figure 3-6.
Figure 3-6 Learning-based Mode Shift Model, Perfect Information
47
Initially, the same choice preferences across modes were used such that the car was the
superior mode. After the change on the second episode, an increase in the observed utility of
transit has been achieved favouring transit over other modes in the choice set.
Interestingly, and even though the agent was still exploiting its favourite choice (car option),
the increment in transit utility was observed by the decision maker under the assumption of
perfect information. In other words, the agent was fully aware of the changes in the system
without exploring and directly examining the transit mode. Based on that, the previously
formed habits and the modal choice probabilities started to evolve since episode two, until
reaching another stabilization zone on episode twelve.
As illustrated in Figure 3-6, the superiority of transit has not been realized immediately after
perceiving the reduction in transit travel time, but rather after a transition period of habits
reformation from the second episode till the sixth episode. In this scenario, the length of the
modal shift period only depends on the impact of the policy change on the utility functions,
regardless of how long the agent has been exploiting the car option before exploring transit.
Practically, the assumption of perfect information requires receiving continuous information
updates on each mode (selected as well as unselected). Adding to that the issue of
information reliability and how much the passenger trusts the supplied information, it can be
said that being in a state of perfect information might not practically exist. However the
previous updating rule presents the upper bound of information provision which would speed
up the learning process and the expected modal shift.
3.3.5 PRACTICAL IMPLICATIONS
Traditional mode choice models are based on static knowledge and lack to recognize that the
interaction with the environment generally leads to adaptation of behaviour through learning.
This section introduced a new approach for modelling mode shift as an adaptive learning
process which involves learning by interaction with the transportation system in a dynamic
context. Within the presented approach, passengers’ choices depend on updating their
perceptions of choice utility while taking into consideration some behavioural factors that
affect the choice at the individual level. The underlying hypothesis is that passengers are
expected to adjust their choices according to their experience with the performance of the
transport system and their previous valuation of the available alternatives, while being subject
48
to awareness limitations and habit formation that might have been formed towards a specific
mode of travel. Further, the presented approach models the insights into understanding
individuals' sensitivity to various policy scenarios and how long it would take to reap the
benefits of the proposed policies, which is typically not answered by traditional approaches
owing to their cross-sectional nature.
The learning-based mode shift model as presented and numerically illustrated in this section
tackles some of the behavioural limitations of the traditional models under the contexts of
bounded rationality and limited awareness. Importantly, the presented approach combines
both perspectives of habitual inertia and awareness limitations rather than substituting one for
the other as assumed by other models. Further, while traditional mode choice models
implicitly assume rational decision making, perfect information availability and full
awareness of the changes in the transport system, the proposed model is considered more
behaviourally realistic and incorporates a number of practical implications.
The presented approach implies that passengers’ awareness is limited and depends on their
direct experience with the transport system and the available level of information provision.
In addition, unlike other models considering only the formation of habits, this simulation
presents the time-dependent processes of change in behaviour that follows a change in the
transport system, during which passengers update their formerly formed habits and change
their choices accordingly.
In general, the illustrated learning behavioural patterns are in agreement with the prior
expectations which can be considered a first step towards the model’s validity. Nevertheless,
another strong point towards the model credibility is utilizing the effectiveness of the
traditional mode choice models in describing the choice situation at the initiation of the
learning process. In light of the above, this chapter introduced a promising approach that
states the art for a more behaviourally sound mode shift model. In addition, the presented
approach can be looked at as a simultaneous way of modelling both mode and route
choice/shift as opposed to the traditional way of sequential modelling. What is unique to the
proposed model is that it can explain the transitional process underlying the modal shift
mechanism which is important from a transit service design point of view.
49
Nevertheless, conducting controlled lab experiments of travel behaviour is suggested to
specify and test the learning-based mode shift process and estimate its parameters under
various assumptions and levels of information provision. It is also required to conduct ex-ante
and ex-post policy analyses at regular time intervals until the mode shares stabilize to validate
the proposed formulations and assumptions of habit formation, level of information provision
and awareness limitations. In addition, more research is suggested to test the spatial and
temporal transferability of the presented model.
3.4 Chapter Summary
This chapter proposed a conceptual framework for a modal shift maximized transit route
design model. The proposed model is comprised of two components. First, a design tool that
deals with generating transit route(s) based on service guidelines and standards. Second, an
evaluation component that concerns with the assessment of the generated route design
considering transit demand variability among both modes and routes. Further, the framework
builds upon and extends the capabilities of the existing MIcrosimulation Learning-based
Approach for TRansit ASsignment (MILATRAS) to tackle both the route design and mode
shift problems. MILATRAS currently models transit assignment given a fixed set of transit
routes and transit demand. The proposed framework adds a mode shift module to
MILATRAS to enable evaluating the impacts of transit investments that usually target auto
users. Modal shift barriers such as habit formation are captured in the framework. Moreover,
the framework considers different Customer Oriented Transit Service (COTS) attributes that
are of importance to mode shift modelling. Furthermore, a major contribution of this chapter
is modelling mode shift decisions as a learning-based process which involves learning by
interaction with the transportation system in a dynamic context. As mentioned earlier, this
research deals only with the evaluation component of the proposed framework. However,
proper attention is given to mode shift modelling, as an essential element of the proposed
framework, in the subsequent chapters.
50
4 INVESTIGATING THE EFFECTS OF PSYCHOLOGICAL FACTORS ON
MODE CHOICE BEHAVIOUR
4.1 Chapter Overview
In an attempt to better understand the effects of psychological factors on commuting mode
choice behaviour, this chapter utilizes socio-psychometric data measured using ad hoc
surveys to investigate the influence of attitudes, affective appraisal and habit formation on
commuting mode choice. The dataset used in this analysis was collected in 2009-2010 in the
City of Edmonton, Alberta, Canada. The Structural Equation Modelling (SEM) approach is
used in this analysis. SEM captures the latent nature of psychological factors and uses a path
diagram to identify the directionality as well as intensity of the relationships. The Theory of
Interpersonal Behaviour (TIB) by Triandis (1977) is utilized as the theoretical foundation of
SEM Analysis. The analysis conducted in this chapter is considered a primary step towards
learning how mode choice decisions are made and deciding which behavioural factors are
relevant to mode shift modelling to be considered in the developed survey (one of the
primary objectives of this thesis).
The remainder of this chapter is arranged as follows: Section 4.2 discusses the reasons behind
the presented investigation. This is followed by a description of the methodology used in the
analysis in Section 4.3, and a review of the important psychological theories that study the
relationship between different aspects affecting the decision making process underlying mode
choice in Section 4.4. Then, Section 4.5 provides a brief description of the dataset used in this
investigation. Subsequently, Section 4.6 presents the developed models, and finally Section
4.7 documents the outcomes of this investigation and its effect on subsequent chapters.
4.2 Reasons for the Investigation
As stated earlier in the literature review, Random Utility Maximization (RUM)-based mode
choice models are extensively used to analyze the choice of an alternative mode from a set of
mutually exclusive options. Conventional mode choice models have been criticized for their
weak characterisation of some psychological constructs such as habit formation, personal
attitude and affective appraisal (Kenyon and Lyons 2003; Shannon et al. 2006). In most
cases, such psychological factors are not directly observable, but they greatly influence
individuals’ choices (Heinen et al. 2010). There have been compelling arguments to consider
behavioural psychological factors directly in the mode choice models (Banister 1978; Aarts et
51
al. 1997; Gärling et al. 1998; Fujii and Kitamura 2003; Gärling and Axhausen 2003; Mackett
2003). Numerous research attempts have been made to capture the intricacies of the choice
process by including socio-psychological aspects as explanatory variables within the
traditional mode choice models in addition to the conventional socioeconomic and service
variables (Johansson et al. 2006; Cantillo et al. 2007; Domarchi et al. 2008).
It is clear that mode choice is a complex process, which is strongly influenced by different
socio-psychological factors. It is also established that incorporating psychological factors in
the utility functions of the mode choice model improves its goodness of fit. Although a
number of attempts have been made to incorporate psychological factors directly within the
mode choice analyses, in most cases the direct effects of psychological variables are
incorporated through the inclusion of alternative-specific constants or dummy variables
without having a theoretical foundation to support the causal relationships between latent
variables (Johansson et al. 2006; Temme et al. 2008; Habib et al. 2010; Galdames et al.
2011). In order to address this critical issue, this chapter adopts a multivariate statistical
modelling approach to investigate the causal relationships between the underlying
psychological aspects affecting mode choice such as habit formation, personal attitude and
affective meaning. Further, the Theory of Interpersonal Behaviour (TIB), by Triandis 1977, is
utilized as the theoretical framework of the adopted approach.
4.3 Structural Equation Models (SEMs)
Structural equation models (SEMs), also known as simultaneous equation models, refer to a
statistical technique for linear-in-parameters multivariate (i.e. multi-equation) regression
models representing causal relationships between variables in the model. In addition, the
response variable in one regression equation may appear as a predictor in another equation.
Hence, variables in SEM can affect one another either directly or indirectly. Further, in
addition to the inclusion of observed exogenous and endogenous variables, a SEM can
incorporate unobservable latent variables, also called constructs or factors, that are not
measured directly but rather indirectly through their effects (indicators) or observable causes.
Such latent variables are modelled by specifying a measurement model and a structural
model. The measurement model (represented by dashed arrows) specifies the relationships
between the latent variables and their observed indicators, whereas the structural model
(represented by solid arrows) specifies the relationships amongst the latent variables
themselves, as shown in Figure 4-1. Furthermore, what differentiates SEM from other
52
conventional multivariate linear models is that it requires specification of a model in terms of
a system of unidirectional effects between variables based on theory and research. Therefore,
SEM is considered a confirmatory rather than exploratory method (Hoyle 1995; MacCallum
and Austin 2000).
Figure 4-1 Measurement and Structural Models
In general, a full SEM consists of three sub models, namely a measurement model for the
endogenous variables, a measurement model for the exogenous variables, and a structural
model for the latent variables. Nevertheless, one or both of the measurement models can be
eliminated in practice. Hence, SEM analyses can be classified in one of the following three
categories. An SEM with both measurement and structural models is called an SEM with
latent variables. On the other hand, an SEM with no measurement models is called an SEM
with observed variables, whereas a measurement model alone is typically a confirmatory
factor analysis (Golob 2003).
Within the structural equation modelling framework, cause and effect relationships are
commonly expressed in the form of a causal graph or a path diagram. Path diagrams provide
a graphical representation of the SEM such that circles or oval shapes enclose the
unobservable (latent) variables. Rectangular boxes, on the other hand, enclose directly
observed variables, whereas the disturbances (error terms) are not enclosed. Further,
unidirectional straight arrows are used to represent the structural parameters indicating a
linear impact of the exogenous variable at the base of the arrow on the endogenous variable at
the head of the arrow. Bidirectional curved arrows represent non-causal linear covariance
(correlation) between exogenous variables and also between disturbances/errors. Compared
to other modelling techniques, the SEM has major advantages in behaviour modelling given
its capabilities in dealing with latent variables with multiple indicators, modelling mediating
factors and dynamic phenomena such as habit and inertia in mode choice (Golob 2003).
53
In light of the aforementioned, the objective of this chapter is to identify the causal effects of
several psychological aspects on mode choice behaviour using the SEM approach. As a first
step for achieving this objective, the path diagram that represents the hypothesized
relationships between all latent and observed variables is specified using a psychological
theory as the theoretical framework describing the underlying interaction between latent
variables and final behaviour (mode choice/modal shift).
4.4 Understanding Mode Choice Behaviour
In general, research in social psychology has suggested that the decision making process
underlying mode choice can be better understood by modelling the relationship between
attitude and behaviour. Numerous psychological theories have studied such interaction
between attitude and behaviour such as the Theory of Interpersonal Behaviour (TIB) by
Triandis (1977), and the Theory of Planned Behaviour (TPB) by Ajzen (1985).
This section utilizes the elements of TIB, which has an advantage over other theories by
accounting for the role of the frequency of past behaviour (habits) in mediating the final
behaviour, to provide a better understanding of the mode choice decision making process
(Galdames et al. 2011; Zmud et al. 2013).
According to Triandis (1977), observed behaviour is generally assumed to succeed both
intention and habit that respectively represent the motivation to perform a specific action and
the past frequency of a specific behaviour, while being mediated by contextual facilitating
conditions, as shown in Figure 4-2.
The Theory of Interpersonal Behaviour assumes that intention is guided by three major
determinants, namely attitudinal, social and affective factors. First, attitudinal factors refer to
the degree to which an individual has a favourable or unfavourable appraisal of the behaviour
under consideration (Ajzen 1991). In other words, attitude is considered as the accumulated
evaluation of the choice which has a magnitude and a direction. Based on the Expectancy-
Value Theory, the magnitude of an attitude depends on two components which are the
expectations that an individual has regarding the results of the behaviour, and the values that
he/she assigns to these possible results. On the other hand, the direction of an attitude
represents whether the decision maker is for or against a specific behaviour (Triandis 1977;
Ajzen 1985; Gärling et al. 1998).
54
Facilitating
Conditions
Behaviour
Intention
Attitude
Social
Factors
Affective
Factors
Expectations
Values
Social Norm
Social Role
Emotions
Self Concept
Frequency of
Past BehaviourHabit
Figure 4-2 The Theory of Interpersonal Behaviour (TIB)
The second determinant of intention is the social factors which include social norm, social
role and self-concept. Social norms are the social rules about what should and should not be
done, whereas social roles are sets of behaviours that are considered appropriate for persons
holding particular status in a group. On the other hand, self-concept refers to the idea that an
individual has of his/herself, the goals that it is appropriate for the person to pursue or to
eschew, and the behaviours that the person does or does not engage in.
The third determinant of intention is the affective factors (affective appraisal) which refer to
the emotional response that an individual has towards or against a specific mode of travel.
Affect is more or less unconsciously evoked such that it is governed by instinctive
behavioural responses to particular situations. According to the Affect Control Theory,
emotions can be disaggregated into four fundamental dimensions, namely evaluation,
potential, activation, and control. Evaluation refers to feelings of goodness or badness elicited
by a concept, potential is associated with feelings of being strong and big as opposed to weak
or small, activation is related to whether the feeling induced by thinking about a concept is
lively or calm, and control refers to feelings of being simple or complex (Domarchi et al.
2008).
55
Finally, facilitating conditions (contextual factors) refer to the ease or difficulty of
performing the behaviour in terms of several attributes representing the socioeconomic and
demographic characteristics of the decision maker, and relative attractiveness of the
competing alternatives such as mode availability, level of service, travel time and cost.
Among such characteristics, auto ownership, auto availability, travel time and travel cost are
considered the major determinants of mode choice (Quarmby 1967; Williams 1978; Barff et
al. 1982).
As discussed, the TIB provides a detailed description of the decision making process starting
from the initial determinants of the behavioural response and moving forward till reaching
the final observed outcome. The theory indicates that attitude and behaviour are positively
correlated and can be described such that the more favourable the attitude, social and
affective factors, the stronger should be an individual’s intention to perform the behaviour in
question. Such intention interacts with habit (the past frequency of a specific attitude) and
contextual aspects producing the final behaviour (e.g. mode choice). Habitual behaviour is
hard to be changed and mostly yields sub-optimal decisions due to the lack of searching and
information processing (Bamberg et al. 2003).
Having the TIB as a theoretical framework, this research investigates the causal relationship
between the underlying psychological aspects affecting mode choice. Given that the
indicators do not have causal relationships that influence the final outcome, the dashed
arrows point from the latent variable to its indicators that are only used to measure the
underlying causal relationships. In other words, the proposed approach starts from the final
observed mode choice behaviour and moves backward till reaching the determinants of such
choice. For example, car drivers and transit users were identified before measuring their
personal attitudes, affective appraisal, and habit formation.
Figure 4-3 shows the path diagram of SEM analysis, indicating the relationship and
correlation between unobservable behavioural factors and their observable indicators.
56
Facilitating
Conditions
Mode Choice
BehaviourIntention
Attitude
Social
Factors
Affective
Factors
Expectations
Values
Social Norm
Social Role
Emotions
Self Concept
Frequency of
Past BehaviourHabit
Figure 4-3 Path Diagram Inspired by the Theory of Interpersonal Behaviour
As shown in Figure 4-3, intention is represented by a latent variable which in turn is affected
by a set of three constructs, namely attitudinal, social, and affective factors. Each of the three
factors is indirectly measured through its indicators as suggested by the Theory of
Interpersonal Behaviour. In addition, habit is represented by a latent variable which is
indirectly measured through the frequency of past behaviour as its effect. Finally, intention,
habit and facilitating conditions interact together to produce the observed mode choice
behaviour.
4.5 Data Description
A dataset gathered in 2009-2010 in Edmonton, Canada (Dogar 2010) was used in this work.
The dataset was oriented to investigate the behavioural factors affecting travel mode choice.
The survey followed an innovative procedure where habit, affective and attitudinal factors
were explicitly measured using different scales. The study was conducted using face-to-face
random intercept interviews at transit stops/stations, shopping malls and restaurants in the
central business district during the afternoon lunch period.
A total sample of 176 records was initially collected. This number was reduced to only 141
records with 88 records of car users and 53 records of transit riders that were available for the
model estimation after a process of cleaning the dataset. In addition, people walking or using
57
other means of transportation were excluded from the analysis. With respect to gender, 79.4%
were males and 20.6% were females. The average age was 37.8 years old, with a standard
deviation of 9.8 years.
4.6 Structural Equation Modelling
This section focuses on the interaction between the psychological precursors of the observed
mode choice behaviour, utilizing the Theory of Interpersonal Behaviour as the theoretical
foundation of the analysis. In particular, the analysis models the interaction between habitual
inertia and those aspects affecting intention, namely attitudinal and affective factors.
Although social factors are not studied in this research, it is suggested that they should be
considered in future work. Alternative SEM specifications were estimated and tested against
one another till reaching the final models. The covariance analysis method (method of
moments) is used to estimate the proposed models using the LInear Structural RELation
(LISREL) software version 8.80.
Furthermore, in order to determine the goodness of fit of the estimated models to the
observed data, several statistical tests were performed such as Chi-square statistics, Normed
Fit Index (NFI), Comparative Fit Index (CFI), and Root Mean Square Error of
Approximation (RMSEA) were examined. In practice, the recommended acceptance of a
good fit to a model requires that the obtained NFI and CFI value should be in range from 0 to
1, with higher values indicating better model fit and a recommended value of 0.90 or greater
for model acceptance. On the other hand, RMSEA values below 0.05 indicate good fit, while
those ranging from 0.08 to 0.10 indicate mediocre fit whereas those greater than 0.10 indicate
poor fit (Long et al. 2011). However, it is important to note that although model fit is
necessary, it is not a sufficient condition for the validity of the hypothesis or theory.
Goodness of fit within reasonable values implies only that the data under consideration
support the hypothesis. Nevertheless, a conceptual model that guides the specification
process, especially the paths between latent and observed variables, is required.
4.6.1 SEM Measurement Models
In this research, two separate SEM measurement models were built separately for car and
transit users since their choice behaviours were different. The developed models specified a
set of four latent variables (i.e. habit, affective factor, attitude towards car and attitude
towards transit), as linear functions of other observed exogenous indicators measured using
semantic scales through an ad hoc questionnaire. Such models are considered simultaneous
58
confirmatory factor analysis such that the measurement models contained the relationships
between four factors and their indicators. Importantly, neither social factors nor contextual
conditions were studied in this analysis. Path diagrams for car and public transit users,
including habit, affective and attitudinal factors, are shown in Figure 4-4 and Figure 4-5,
respectively.
Figure 4-4 Path Diagram for the Measurement Model of Car Users
In an indication that car users would use the car for almost every single trip, habit is stronger
and positive for car usage, being negative for transit. Car users will seldom use public
transport given their strong car use habit formation. On the other hand, the model shows that
car users give a stronger weight to the activation (lively vs. calm), control (simple vs.
complex), and evaluation (good vs. bad) dimensions of the affective factor compared to the
potential (big vs. small) one. This might be related to the sense of independence associated
with private transportation.
59
Further, the results show that car users give more importance to the value (important) rather
than the expectation (good) component of attitude for car; whereas they give more
importance to the expectation (good) rather than the value (important) component of attitude
for transit. This means that they might know that transit is a good alternative in general
(reduce congestion, emissions, etc.), although they do not perceive it as an important mode
for their work trips. It is also interesting to notice the negative correlation between attitudes
for transit and car, for auto user. Certainly, this affects the possibility of promoting the use of
transit between auto users.
The previous model has a chi-square value of 32.57 with 37 degrees of freedom, RMSEA=
0.00, NFI= 0.92, and CFI= 1.00. The goodness of fit statistics indicate that the model has a
good fit. Specifically, the NFI value of 0.92 and CFI value of 1.00 are considered within the
acceptable range of 0 to 1.
As shown in Figure 4-5, similar result was realized while examining the habitual behaviour
of transit riders; a negative relationship with the transit option and a positive relationship with
the car option. This might be interpreted as that transit riders are forced to use public transit
for work trips; however they might shift to the car option if it is available. The previous
finding corroborates the superiority of the car as a mode of travel. On the other hand, transit
riders give a stronger weight to the potential, control and evaluation dimensions of the
affective factor compared to the activation one; that is, there is a low motivation for using
public transport. Furthermore, in contrast to car users, transit riders give more weight to the
expectation (good) rather than the value (important) component of attitude for both car and
transit. In other words, they might know that transit is good and that is the reason why they
use it, although it is not important for them. There is a sort of detachment toward the transit.
The negative correlation between affection and both habit and attitude towards transit
reinforces what has been expressed before. People use transit because they have to, without
being attached to it.
The previous model has a chi-square value of 34.69 with 33 degrees of freedom, RMSEA=
0.031, NFI= 0.840, and CFI= 0.970. The goodness of fit statistics indicates that the model has
a good fit. Specifically, the RMSEA value of 0.031 and the CFI value of 0.970 are considered
60
within the acceptable ranges, although the NFI value of 0.840 is lower than the 0.90
threshold.
Figure 4-5 Path Diagram for the Measurement Model of Transit Riders
4.6.2 SEM with Latent Variables
A joint SEM with latent variables is estimated such that a structural model for the
relationship between the latent variables, and two measurement models for both the
endogenous (i.e. mode choice) and exogenous indicators of the psychological factors are
integrated. The Theory of Interpersonal Behaviour is utilized as the path diagram of the
corresponding SEM, as shown in Figure 4-6.
In general, the proposed model specifies the causal influences among the latent variables by
incorporating both a measurement model to deal with how indicators are related to the
factors, and the structural model to deal with the causal relationships among factors.
61
Figure 4-6 Path Diagram for the SEM with Latent Variables
As shown in Figure 4-6, intention is modelled as a latent variable which is indirectly
measured through three constructs, namely affective factor, attitude towards car and attitude
towards transit. Further, each of the three factors is indirectly measured through its effects as
indicated by the measurement models. In addition, habit is modelled as a latent variable
which is indirectly reflected by the frequency of past use. Finally, both intention and habit
affects the observed mode choice behaviour as suggested by Triandis (1977).
The SEM with latent variables shows similar relationships as that indicated by the
measurement model for car users. In an indication that car users would use the car for almost
every single trip, habit was found to be strong and positive for the car, while being negative
for public transit. On the other hand, the results show that Edmonton passengers give a
stronger weight to the activation and evaluation dimensions of the affective factor compared
to the potential and control ones. Further, users give more weight to the value (important)
rather than the expectation (good) component of attitude for car; whereas they give more
weight to the expectation (good) rather than the value (important) component of attitude for
transit.
62
In addition, the SEM with latent variables shows the causal relationship among factors such
that intention is reflected by affective and attitudinal factors towards car and transit.
Interestingly, it can be shown that Edmonton commuters give a strong positive weight to the
attitude towards transit whereas a negative sign is associated with the attitude towards car. It
seems that intention is guided by the attitude towards transit rather than the attitude to the car.
Further, both habit and intention integrate to influence the final observed mode choice
behaviour. In an indication of the superiority of the car as a mode of travel, the final mode
choice is associated with a negative habitual behaviour towards transit and a positive one
towards car usage. On the other hand, intention is associated with a negative sign for car and
positive sign for transit. This would be interpreted as that Edmonton travellers know the
importance of the transit service and might be motivated to use it, although the strong
frequency of car use does not allow that.
The previous model has a chi-square value of 104.36 with 51 degrees of freedom, RMSEA=
0.086, NFI= 0.920, and CFI= 0.950. The goodness of fit statistics indicates that this model
fits the data well, although the RMSEA value of 0.086 is higher than the 0.05 threshold.
4.7 Investigation Outcomes
In this chapter, the Structural Equation Modelling (SEM) approach is adopted to investigate
the cause and effect relationships between the underlying psychological aspects affecting
mode choice. The proposed approach focused on the psychological antecedents of mode
choice behaviour following the Theory of Interpersonal Behaviour by Triandis (1977) as the
theoretical foundation of the investigation. Different psychometric tools were used to
measure the effects of psychological factors such as habitual behaviour, attitudes and
affective factors. Although such psychological factors were measured using different
semantic scales, the SEM analysis allowed for the detection of correlation between latent
variables and the determination of the importance of each latent attribute.
Several structures were proposed and estimated using LISREL software for SEM analysis.
The results showed that the consideration of psychological attributes, namely personal
attitude, habit formation, and emotional response as latent variables helped explain mode
choice behaviour. In addition, it was shown that commuters have positive attitudes and
emotions towards their chosen mode. Further, the magnitude and sign associated with the
63
habitual factors provided evidence for the superiority of the car as a travel alternative such
that car users would use the car for almost every single trip. Although social factors were not
studied in this research, it is suggested that their effect on intention should be considered in
future work.
The impact of these findings on policy issues is a matter that should be kept in mind,
especially while modelling mode shift to transit (the main objective of this thesis). The strong
habit towards auto use, associated with a positive attitude and affection to it, and the lower
attitude and affection towards transit, certainly constitute a deterrent when trying to promote
the use of transit facilities. Actually, demand management schemes, such as promoting transit
provision, might not have the expected result given the found level of attachment to the car.
The analysis conducted in this chapter confirms the causal relationships between the
underlying psychological aspects affecting mode choice as indicated by the Theory of
Interpersonal Behavior. As such, and given the previously mentioned policy implications, the
survey proposed in Chapter 5 collects detailed information about habit formation, personal
attitude, and affective appraisal besides personal and modal attributes as major determinants
of the mode shift decision making process.
64
5 COMMUTING SURVEY FOR MODE SHIFT (COSMOS)
5.1 Chapter Overview
This chapter presents the design of a multi-instrument COmmuting Survey for MOde Shift
(COSMOS) that combines three types of instruments for collecting detailed information on
commuters’ mode switching behaviour. COSMOS exploits qualitative psychometric
questions on users’ perception along with Revealed Preference (RP) mode choice information
and Stated Preference (SP) mode switching experiments. The RP part of the survey collects
detailed information on recent commuting trips. The RP-pivoted SP choice experiments are
based on efficient experimental design technique (D-Efficient design), and measured
participants’ stated mode switching preferences in favour of public transit in response to
different policy changes. Based on the outcomes of Chapter 4, the psychometric instruments
are designed to collect information on habitual behaviour, affective appraisals and personal
attitudes. The survey was conducted in Toronto, Canada in 2012.
The following sections of this chapter presents an overview of the activities involved in
conducting the developed survey with details provided on study and survey objectives in
Section 5.2, study area in Section 5.3, survey sample design in Section 5.4, and survey
instrument design in Section 5.5. Finally, a chapter summary is provided in Section 5.6.
5.2 Study and Survey Objectives
In general, the development and completion process of a survey can be divided into several
interconnected phases starting with the planning phase; which is followed by the design and
development phase; then the implementation phase; and finally the revision and evaluation
phase for the entire survey process (Richardson et al. 1995).
The first step in planning a survey is to identify the study and survey objectives in order to
guide all subsequent survey tasks. As stated previously, the main objective of this research is
to develop a better understanding of commuters’ travel choice preferences and mode
switching behaviour towards public transit. Unlike traditional mode choice models, precise
mode shift models are to be developed to accurately forecast transit ridership. However,
extensive information about revealed mode choice as well as stated mode switching
behaviours is required for the development of such models. Unfortunately, existing travel
survey datasets, where no psychological information on the decision maker and insufficient
data about COTS elements exists, does not provide such information. Hence, as first task
65
towards mode shift modelling, the developed survey is designated to gather such information
from the population of interest.
5.3 Study Area
The Census Metropolitan Area (CMA) of Toronto is selected as a case study for the proposed
analysis. In general, a CMA is a Statistics Canada definition for a metropolitan region that
covers multiple municipalities. In other words, CMAs are more formal yet similar to the
unofficial designations for urban areas such as the Greater Toronto Area (GTA).
The Toronto CMA is the largest population centre in Canada1. It has similar, but not exactly
the same, geographic boundaries as the Greater Toronto Area (GTA), as shown in Figure 5-1.
The Toronto CMA consists of the City of Toronto in addition to the surrounding regional
municipalities of Durham, York, Peel, and Halton. However, on the one hand, some
municipalities (Burlington, Whitby, Oshawa, Clarington, Scugog, and Brock) that are
considered part of the GTA are not within the Toronto CMA. On the other hand, other
municipalities (Mono, New Tecumseth, and Bradford West Gwillimbury) that are considered
part of the Toronto CMA are not within the GTA.
Figure 5-1 GTA and Toronto CMA Boundaries2
1 http://www.statcan.gc.ca/tables-tableaux/sum-som/l01/cst01/demo05a-eng.htm 2 http://datalib.chass.utoronto.ca/caq/g1.htm
66
In addition to covering a very large area and having one of the longest commute distance (9.4
km) of any CMA in 2006, the selection of the Toronto CMA as a case study region for the
proposed analysis is influenced primarily by being clearly defined by Statistics Canada, the
main source of population statistics of this research. Moreover, studying this large area
allowed for developing and comparing separate mode switching models for two groups of
commuters since their mode shift behaviours are expected to be different. First, developing
mode switching models for choice users who reside and work within the City of Toronto,
where public transit is competitive to auto travel. Second, developing mode switching models
for commuters who reside and/or work (i.e. having at least one of their trip ends) within the
outskirts of the City of Toronto, where much lower transit coverage and usage exist.
5.3.1 The Census Metropolitan Area (CMA) of Toronto
Over a total land area of 5,905.71 km², the Toronto CMA contains 21 separate cities or
towns, two townships, and one Indian reserve, comprising a large urban core area and its
closely integrated urban fringes3, as shown in Figure 5-2.
Figure 5-2 The Census Metropolitan Area (CMA) of Toronto4
3 http://www12.statcan.gc.ca/census-recensement/2011/as-sa/fogs-spg/Facts-cma-
eng.cfm?Lang=Eng&TAB=1&GK=CMA&GC=535 4 http://urbantoronto.ca/forum/showthread.php/8084-Wikipedia-Toronto
67
The Toronto CMA is Canada’s most populous CMA (about 500,000 people fewer than the
GTA), with a population which rose from 5,113,149 to 5,583,064 persons between 2006 and
2011 respectively. This increase represents the highest percentage change of 9.2% compared
to a national growth of 5.9% and an average CMAs growth of 7.4%. Table 5-1 provides a
more complete breakdown of the Toronto CMA census subdivisions’ population change
between 2006 and 2011. Given its land area and population, the Toronto CMA has a
relatively high population density of 865.8 persons per square kilometre compared to a
national population density of 3.7 persons per square kilometre and an average CMAs
population density of 249.6 persons per square kilometre.
Table 5-1 Toronto CMA, Census Subdivisions, Population Change, 2006 to 20115
Census Subdivision (CSD) Name Type Population
2006 2011 % change
Toronto (PD 1 to 16) City 2,503,281 2,615,060 4.5
Mississauga (PD 36) City 668,599 713,443 6.7
Brampton (PD 35) City 433,806 523,911 20.8
Markham (PD 31) Town 261,573 301,709 15.3
Vaughan (PD 33) City 238,866 288,301 20.7
Richmond Hill (PD 29) Town 162,704 185,541 14
Oakville (PD 39) Town 165,613 182,520 10.2
Ajax (PD 21) Town 90,167 109,600 21.6
Pickering (PD 20) City 87,838 88,721 1
Milton (PD 38) Town 53,889 84,362 56.5
Newmarket (PD 27) Town 74,295 79,978 7.6
Caledon (PD 34) Town 57,050 59,460 4.2
Halton Hills (PD 37) Town 55,289 59,008 6.7
Aurora (PD 28) Town 47,629 53,203 11.7
Georgina (PD 25) Town 42,346 43,517 2.8
Whitchurch-Stouffville (PD 30) Town 24,390 37,628 54.3
New Tecumseth (PD 84) Town 27,701 30,234 9.1
Bradford West Gwillimbury (PD 83) Town 24,039 28,077 16.8
Orangeville (PD 80) Town 26,925 27,975 3.9
East Gwillimbury (PD 26) Town 21,069 22,473 6.7
Uxbridge (PD 18) Township 19,169 20,623 7.6
King (PD 32) Township 19,487 19,899 2.1
Mono (PD 144) Town 7,071 7,546 6.7
Chippewas of Georgina
Island First Nation Indian Reserve 353 275 -22.1
Toronto Census
Metropolitan Area (CMA) 5,113,149 5,583,064 9.2
5 http://www12.statcan.gc.ca/census-recensement/2011/as-sa/fogs-spg/Facts-cma-
eng.cfm?Lang=Eng&TAB=1&GK=CMA&GC=535
68
The Toronto CMA has an extensive multi-modal transportation network. It has the busiest
freeway network in Canada which comprises of both provincially funded and operated 400-
series freeways and municipal expressways. In addition, the CMA is covered by separate
transit services that include commuter rail and long-range bus routes, rapid transit lines, and a
multitude of street transit routes (e.g. streetcar and bus routes). Those services are operated
by local transit agencies such as Brampton Transit, Durham Region Transit, Mississauga
Transit, Oakville Transit, Toronto Transit Commission, and York Region Transit.
5.3.2 The City of Toronto
As the heart of the Toronto CMA, explicit attention is given to the City of Toronto where a
multimodal transit system and supportive land use make transit more competitive to auto
travel. According to Statistics Canada, 2006 census of population, the City of Toronto is
home to 2,503,281 people, making it the largest city in Canada, and one of the most populous
cities in North America6. Over a total land area of 630.18 km², and a population density of
3,972.4 persons per square kilometre, the City of Toronto consists of six main districts
(Etobicoke, York, Downtown Toronto, East York, North York, and Scarborough) according
to the 1998 amalgamation, as shown in Figure 5-3.
Figure 5-3 The City of Toronto7
6 http://www12.statcan.ca/census-recensement/2011/dp-
pd/prof/details/page.cfm?Lang=E&Geo1=CSD&Code1=3520005&Geo2=PR&Code2=35&Data=Count&Searc
hText=Toronto&SearchType=Begins&SearchPR=01&B1=All&GeoLevel=PR&GeoCode=3520005 7 http://wikitravel.org/en/Toronto
69
More than 160,000 of the City of Toronto residents live downtown, with a very high
proportion of them walking and biking, and using transit8. Further, what is unique to the City
of Toronto compared to the rest of the Toronto CMA is its transit provision. The City of
Toronto has an extensive multimodal transit network operated by the Toronto Transit
Commission (TTC). Having a transit fleet consisting of about 700 subway cars, 247 streetcars
(52 are higher-capacity articulated streetcars), and 1800 buses of various ages and types; the
TTC serves the City of Toronto using a north-south, east-west grid of routes conforming with
the grid of major arterial roads in the area. Such grid transit network comprises of four
subway lines, 11 streetcar routes, and more than 140 bus routes, where all surface routes feed
the grid of rapid transit lines allowing for high-speed trips into the downtown core and
throughout the network9, as shown in Figure 5-4.
Figure 5-4 TTC Network
8 http://www.toronto.ca/planning/pdf/living_downtown_nov1.pdf 9 http://www3.ttc.ca/Routes/General_Information/General_Information.jsp
70
Further, the TTC operates 13 bus routes into adjacent municipalities, and the neighbouring
transit agencies operate more than 30 bus routes which connect directly with the TTC subway
system or other surface routes. In general, the TTC ridership accounts for more than 80% of
all transit ridership in the Toronto CMA, where approximately 460 million customers are
carried per year, or about 1.5 million passengers on a typical weekday.
Moreover, transfer opportunities exist between several TTC services and the GO Transit
commuter rail services. Almost all TTC bus and streetcar routes operate all day, every day.
The TTC service coverage is largely unchanged for 18 operating hours per day, thus
providing transit services within a 5 to 7 minute walk of most areas within the City of
Toronto. Given its very simple and extremely successful fare system with a flat-fare structure
and free transfers between all services and modes, the TTC allows customers to travel an
unlimited distance per trip for one fixed price.
Furthermore, the TTC’s conventional services are planned so that the capacity is matched
with actual observed passenger demand in accordance with vehicle crowding standards (50
passengers per bus and 74 passengers per streetcar during peak periods). In turn, the majority
of the TTC services operate at peak intervals of 5-10 minutes, with some services as frequent
as every 2 minutes, and off-peak service every 5 to 20 minutes. On the other hand, the TTC
subways operate every 2 minutes 40 seconds during peak periods, and every 5 minutes or
better during off-peak. In addition to the fixed route services, the TTC operates a fully
accessible door-to-door specialized system, called Wheel-Trans, for people with substantial
mobility difficulties. The users of such service are required to book their trips one day in
advance, or reserve regular daily trips on a subscription basis. Wheel-Trans carries 1.5
million trips per year, or about 5000 trips on a typical weekday through a fleet consisting of
135 fully accessible buses, and contracted accessible and regular taxis.
5.4 Survey Sample Design
Given that this research is intended to measure commuters' travel choice preferences and
willingness to switch to public transit, the target population of this study is identified as the
total employed labour force, 15 years and over, in the Toronto CMA. However, due to the
difficulty of surveying people with no fixed workplace address (since the survey is concerned
with typical work trip), the survey population is identified as all individuals in the employed
71
labour force, 15 years and over, having a usual place of work in the Toronto CMA (excluding
those who work at home).
5.4.1 Target and Survey Populations
According to the Place of Work and Commuting to Work data released by Statistics Canada
in 2006, the survey population is estimated as 2,324,270 commuters (45.46% of the total
Toronto CMA population) distributed among different modes of travel10
. Table 5-2,
Table 5-3, and Table 5-4 provide a more complete breakdown of commuting work trip mode
choice in the Toronto CMA.
Table 5-2 Toronto CMA, 2006 Commuting Work Trip Breakdown by Gender
Mode of Transportation Male Female Total Modal Share
Car Driver 845,730 640,295 1,486,025
Shared Ride (Car Passenger and Carpooler) 54,600 113,715 168,315
Public Transit 206,360 312,340 518,700
Cycle 14,920 7,585 22,505
Walk 46,950 62,945 109,895
Other Modes (Taxicab, Motorcycle, etc.) 8,010 10,820 18,830
Total Commuting Trips per Gender 1,176,570 1,147,700 2,324,270
Table 5-3 Toronto CMA, 2006 Commuting Work Trip Percentage by Gender
Mode of Transportation Male Female Total Modal Share
Car Driver 56.91% 43.09% 63.94%
Shared Ride (Car Passenger and Carpooler) 32.44% 67.56% 7.24%
Public Transit 39.78% 60.22% 22.32%
Cycle 66.30% 33.70% 0.97%
Walk 42.72% 57.28% 4.73%
Other Modes (Taxicab, Motorcycle, etc.) 42.54% 57.46% 0.81%
Total Commuting Trips per Gender 50.62% 49.38% 100%
Table 5-4 Toronto CMA, 2006 Commuting Work Trip Percentage by Mode
Gender Car
Driver
Shared
Ride
Public
Transit Cycle Walk
Other
Modes
Total
Gender
Share
Male 71.88% 4.64% 17.54% 1.27% 3.99% 0.68% 50.62%
Female 55.79% 9.91% 27.21% 0.66% 5.48% 0.94% 49.38%
Total Commuting
Trips per mode 63.94% 7.24% 22.32% 0.97% 4.73% 0.81% 100.00%
10 http://www12.statcan.gc.ca/census-recensement/2006/dp-pd/tbt/Lp-
eng.cfm?LANG=E&APATH=3&DETAIL=0&DIM=0&FL=A&FREE=0&GC=0&GID=0&GK=0&GRP=1&P
ID=0&PRID=0&PTYPE=88971,97154&S=0&SHOWALL=0&SUB=0&Temporal=2006&THEME=76&VID=
0&VNAMEE=&VNAMEF
72
At 63.94%, car driver is clearly the dominant work trip mode throughout the Toronto CMA.
Combined with an additional 7.24% of trips made as car passenger or carpooler, 71.18% of
all work trips are made by car. On the other hand, the combined percentage of commuters
choosing public transit and active modes (walk and bike) for work trips is around 28% in the
Toronto CMA, which is considered one of the highest Canada wide. Figure 5-5 presents the
work trip modal split breakdown in the Toronto CMA.
Figure 5-5 Toronto CMA, 2006 Commuting Work Trips Mode Split
As shown in Figure 5-5, 22.32% of the total commuting trips in the Toronto CMA are made
by public transit, which is the highest rate of commuters using transit in any CMA in Canada.
Interestingly, it was found that females are more likely to use either the shared ride (car
passengers/carpoolers), public transit, or walk options while males tend to drive or cycle, as
shown in Figure 5-6 and Figure 5-7.
Car Driver, 63.94%
Car Passenger / Carpooler,
7.24%
Transit Rider, 22.32%
Cycle, 0.97%
Walk, 4.73% Other, 0.81%
73
Figure 5-6 Toronto CMA, 2006 Commuting Work Trips Gender Split by Mode
Figure 5-7 Toronto CMA, 2006 Commuting Work Trips Mode Split by Gender
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Car Driver CarPassenger /Carpooler
Transit Rider Cycle Walk Other
Male Female
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Male Female
Car Driver Car Passenger / Carpooler Transit Rider Cycle Walk Other
74
According to the Commuting Patterns and Places of Work of Canadians released by Statistics
Canada in 200611
, the number of workers in the study area rose faster in the peripheral
municipalities than in the central municipality between 2001 and 2006. In more specific
terms, the number of workers (based on their place of work) increased by 12.9% in the
peripheral municipalities as a whole (with the largest increase of +22.2% in Vaughan),
compared to only 0.7% in the City of Toronto. However, the clusters of workplaces in the
heart of the city centre continued to dominate despite the growth of the peripheral
municipalities.
Further, in terms of commute distance, residents of Toronto CMA experienced the second
highest average commute distance of 9.4 kilometres in 2006 (after Oshawa that had 11
Kilometres), with a slight increase of +0.2 Kilometres compared to 2001.
In 2006, the proportion of workers whose usual place of work was in the Toronto CMA who
used a sustainable mode of transportation (i.e. public transit, walking or biking) to get to
work, was much lower in the peripheral municipalities of the census metropolitan area (where
a sharp growth in employment attracted more commuters). Hence, this research gives explicit
consideration to the City of Toronto where multimodal transit system and supportive land use
make transit more competitive to auto travel.
According to the Place of Work and Commuting to Work data released by Statistics Canada
in 2006, the number of individuals in the employed labour force, 15 years and over, having a
usual place of work in the City of Toronto (excluding those who work at home) is estimated
as 1,251,070 commuters distributed among different modes of travel12
. This number
constitutes 49.98% of the total population in the City of Toronto and 53.83% of the survey
population. Table 5-5, Table 5-6, and Table 5-7 provide a more complete breakdown of work
trip mode choice in the City of Toronto.
11 http://www12.statcan.gc.ca/census-recensement/2006/as-sa/97-561/p33-eng.cfm 12 http://www12.statcan.gc.ca/census-recensement/2006/dp-pd/tbt/Rp-
eng.cfm?TABID=1&LANG=E&APATH=3&DETAIL=0&DIM=0&FL=A&FREE=0&GC=0&GID=858074&
GK=0&GRP=1&PID=95839&PRID=0&PTYPE=88971,97154&S=0&SHOWALL=0&SUB=0&Temporal=20
06&THEME=76&VID=0&VNAMEE=&VNAMEF=&D1=0&D2=0&D3=0&D4=0&D5=0&D6=0
75
Table 5-5 City of Toronto, 2006 Commuting Work Trip Breakdown by Gender
Mode of Transportation Male Female Total Modal Share
Car Driver 360,345 271,225 631,570
Shared Ride (Car Passenger and Carpooler) 19,475 51,705 71,180
Public Transit 173,590 267,630 441,220
Cycle 11,400 6,545 17,945
Walk 34,610 44,355 78,965
Other Modes (Taxicab, Motorcycle, etc.) 4,505 5,685 10,190
Total Commuting Trips per Gender 603,925 647,145 1,251,070
Table 5-6 City of Toronto, 2006 Commuting Work Trip Percentage by Gender
Mode of Transportation Male Female Total Modal Share
Car Driver 57.06% 42.94% 50.48%
Shared Ride (Car Passenger and Carpooler) 27.36% 72.64% 5.69%
Public Transit 39.34% 60.66% 35.27%
Cycle 63.53% 36.47% 1.43%
Walk 43.83% 56.17% 6.31%
Other Modes (Taxicab, Motorcycle, etc.) 44.21% 55.79% 0.81%
Total Commuting Trips per Gender 48.27% 51.73% 100%
Table 5-7 City of Toronto, 2006 Commuting Work Trip Percentage by Mode
Gender Car
Driver
Shared
Ride
Public
Transit Cycle Walk
Other
Modes
Total
Gender
Share
Male 59.67% 3.22% 28.74% 1.89% 5.73% 0.75% 48.27%
Female 41.91% 7.99% 41.36% 1.01% 6.85% 0.88% 51.73%
Total Commuting
Trips per mode 50.48% 5.69% 35.27% 1.43% 6.31% 0.81% 100.00%
Similar to the main trend of mode split in the Toronto CMA, car driver is the dominant mode
for work trips throughout the City of Toronto. However, while the car is still the mostly used
mode of transportation, the combined percentage of commuters choosing public transit and
active modes (walk and bike) in the City of Toronto is estimated as 43%, which is much
higher than the same category in the Toronto CMA. At this high percentage of transit and
active mode usage, the City of Toronto comes second after the City of Montreal that had a
value of 46% for the same category of commuters. Figure 5-8 presents the work trip mode
split breakdown in the City of Toronto.
76
It should be clear that in the City of Toronto, with its extensive multi-modal transit system,
public transit is the major alternative to auto driving. This is unlike the case in smaller cities
where auto-passenger, walk, and bike options are the major competitors to car driving.
As shown in Figure 5-8, 35.27% of the total commuting trips in the City of Toronto are made
by public transit, which is the highest rate of commuters using transit of any major city in
Canada. Furthermore, similar to the general trend within the Toronto CMA, females tend to
take either the shared ride (car passengers/carpoolers), public transit, or walk options
compared to males who are more likely to drive or cycle, as shown in Figure 5-9 and
Figure 5-10.
Figure 5-8 City of Toronto, 2006 Commuting Work Trips Mode Split
Car Driver, 50.48%
Car Passenger / Carpooler, 5.69%
Transit Rider, 35.27%
Cycle, 1.43% Walk, 6.31%
Other, 0.81%
77
Figure 5-9 City of Toronto, 2006 Commuting Work Trips Gender Split by Mode
Figure 5-10 City of Toronto, 2006 Commuting Work Trips Mode Split by Gender
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Car Driver Car Passenger/ Carpooler
Transit Rider Cycle Walk Other
Male Female
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Male Female
Car Driver Car Passenger / Carpooler Transit Rider Cycle Walk Other
78
5.4.2 Sampling Method
As discussed in the previous section, the survey population of this study is estimated as
2,324,270 commuters distributed among different modes of travel, representing all
individuals in the employed labour force, 15 years and over, having a usual place of work in
the Toronto CMA. In addition, explicit sampling consideration is given to the City of
Toronto.
Given this large population, it is clear that conducting a census to gather the required
information is not a feasible option. Therefore, a sample survey is more desirable. As such,
proper attention is given to the sampling design as a fundamental part that affects the quality
of the collected data and all subsequent steps.
Depending on whether reliable inferences are to be made about the population, two types of
sampling techniques are available. While the non-probability sampling provides fast, easy
and inexpensive way of selecting a sample using a subjective (i.e. non-random) method for
selecting units from a population, it does not ensure having representative sample of the
population (i.e. may result in large biases and reduce the variability of the population).
Probability sampling, on the other hand, is more complex, time consuming and costly than
non-probability sampling. Probability sampling involves the selection of units from a
population randomly based on their inclusion probability and therefore avoids any selection
bias. Hence, it is possible to generalize the results from the sample to the population and
produce reliable parameter estimates along with estimates of the sampling error (Franklin et
al. 2003). In light of the above, the probability sampling technique is adopted in this research
in order to ensure a representative sample that allows for making reliable inferences about the
population based on observations from the sample.
In general, there exist numerous types of probability sample designs that fit in different
situations such as simple random sampling, systematic sampling, probability-proportional-to-
size sampling, cluster sampling, stratified sampling, multi-stage sampling, multi-phase
sampling and replicated sampling.
In order to maintain an efficient adequate size sample, this study uses the Simple Stratified
Random Sampling method where survey population is divided into homogeneous mutually
exclusive strata based on geography, gender, and mode split, with interlocking between all
79
strata. Then, independent Simple Random Samples are selected without Replacement
(SRSWOR) from each stratum (i.e. once a unit has been selected, it cannot be selected again).
This ensures that every possible sample of size (n) has an equal chance of being selected from
the survey population (N). Consequently, each unit in the sample has the same inclusion
probability (π= n/N). For example, consider a sample population of five people (A, B, C, D
and E) and suppose that a sample of three is to be selected (SRSWOR). Then, there are ten
possible samples of three people: (A, B, C), (A, B, D), (A, B, E), (A, C, D), (A, C, E), (A, D,
E), (B, C, D), (B, C, E), (B, D, E), and (C, D, E). Each of these samples has an equal chance
of being selected and each individual is selected in 6 out of the 10 possible samples, thus each
individual has an inclusion probability of π= n/N= 3/5 (Franklin et al. 2003).
Figure 5-11 Stratification by Geography, Gender, and Mode Split
In this research, the survey population is stratified into homogeneous mutually exclusive
subpopulations based on geography (the City of Toronto and the rest of the CMA), gender
(Male and Female), and mode split (Car Driver, Shared Ride, Public Transit, Cycle, Walk,
and Other), as shown in Figure 5-11. Then, independent Simple Random Samples (SRS) are
80
selected without replacement from each stratum assuming that individuals from the same
stratum exhibit similar mode shift behaviour.
5.4.3 Sample Size Determination
In general, sample size determination attempts to control sampling and nonresponse errors
that occur randomly. The determination of sample size is crucial to the precision of the
survey estimates. In other words, the greater the precision required of the estimates, the larger
the sample size needed, given that sampling variance decreases as the sample size increases.
Hence, the appropriate sample size depends on the desired precision of the survey estimates
expressed in terms of one or more of the following terms: the allowable standard error, the
margin of error, and/or the coefficient of variation (Richardson et al. 1995).
It is important to decide on the appropriate level of precision for the survey estimates in terms
of the margin of error that can be tolerated. Further, given the effect of the variability of the
characteristic of interest in the survey population on the sample size, it is also important to
specify how big the sampling variance is relative to the survey estimate. Moreover, the size
of the survey population (N) should be taken into account as it plays a major role in sample
size determination for small populations, a moderately important role for medium size
populations and a minor role for large populations. Another factor that alters the precision of
the survey estimates is the sampling strategy. Accordingly, the sample size required to satisfy
a given level of precision should be multiplied by the design effect (DEFF).
In general, DEFF is a factor used to adjust the sample size based on the sampling strategy
being used. Commonly, DEFF = 1 for a simple random sample design, DEFF ≤ 1 for a
stratified sample design, and DEFF ≥ 1 for a cluster sample design. An estimate of the design
effect can usually be obtained from a pilot or similar previous survey. If a stratified sample
design is used where no suitable prior estimate of the design effect is available, DEFF = 1 can
be used to calculate the sample size (i.e. assume SRS). The resulting precision of the survey
estimates should be no worse than that obtained with a simple random sample (i.e. ensuring
better precision). However, deciding on the value of the design effect is not an easy task
when a cluster sample design is used due to the lack of prior information about the effect of
clustering on the sampling variance. In such case, a design effect of at least 2 might be used,
although the design effect of a highly clustered design may reach as high as 6 or 7.
81
In addition to the previous terms, adjusting the sample size for the anticipated response rate
(r) is required to achieve the desired precision for the survey estimates. This adjustment is
usually done by selecting a larger sample based on an expected response rate estimated from
similar surveys or a pilot survey on the same population.
In this research, typical values for travel surveys are used for the previous terms. A more
conservative sample is produced by assuming a maximum population variability of P= 0.5.
Further, the sample size is determined to maintain a margin of error e= 0.05 at a 95%
confidence level for estimates of the true value of the characteristics of interest of the whole
population (i.e. estimates are not required for individual strata). In other words, there is a 5%
chance of getting a sample that produces an estimate outside the range P±e (i.e. z= 1.96). In
addition, a typical travel surveys response rate value of 20% (i.e. r= 0.2) is expected. The
following steps show the sample size determination process in more details.
1. Calculate the initial sample size, n1:
(5-1)
2. Adjust the sample size to account for the size of the population, n2:
(5-2)
3. Adjust the sample size for the effect of the sample design, n3:
(5-3)
Note that for stratified random sampling, DEFF < 1 is usually used. However, selecting
sampling frame based on e-mail addresses may impose clustering effect (i.e. clustering based
on e-mail availability before stratifying based on geography, gender, and mode choice).
2
2
1
)1(n
e
PPz
384)05.0(
)5.01(5.0)96.1(n
2
2
1
1
12nnN
Nn
384384 2,324,270
2,324,270384n2
23n nDEFF
350,1344,13845.3n3
82
Given that no estimate of DEFF is available in this study, setting DEFF > 1 (3.5 is used as an
average DEFF value) should have the effect of producing a more conservative (i.e. larger)
sample size estimate.
4. Adjust for response to determine the final sample size, n:
(5-4)
Then, the total required sample size is estimated as 6,750 observations.
5.4.4 Sample Allocation Method
An important consideration in determining the efficiency of stratified sampling is the way in
which the total sample size (n) is allocated to each stratum. This section discusses how the
total sample is allocated among different strata.
In this study, the fixed sample size criterion is adopted to allocate the total sample size (n)
among different strata. In the fixed sample size allocation method, the proportion of the
sample allocated to the hth
stratum is denoted as ah= nh/n, where each ah is between 0 and 1
inclusively (i.e. 0 ≤ ah ≤ 1) and the sum of the ah’s is equal to 1 (i.e. Σ ah= 1). Therefore, for
each stratum h, the sample size nh is equal to the product of the total sample size n and the
proportion ah of the sample coming from that particular stratum: nh= n × ah.
Under this allocation criterion, since the overall sample size (n) is already known, the sample
size nh for each stratum can be calculated as soon as the value of ah is determined for each
stratum. In this research, the N-proportional allocation method is used for the choice of (ah)
for each stratum. In the N-proportional method, the allocation factor (ah) for each stratum is
equal to the ratio of the population size (Nh) in the stratum to the entire population size (N).
(5-5)
Similarly, the sample size (nh) in each stratum is proportional to the ratio of the population
size (Nh) of the stratum to the to the entire population size (N) (i.e. larger strata receive more
of the sample and smaller strata receive less of the sample).
r
n3n
750,620.0
350,1n
N
Nhha
83
(5-6)
This results in the sampling fraction, fh= nh/Nh, being the same in each stratum and equal to
the overall sampling fraction f= n/N. In light of the above, the calculated sample size n=
1,350 is allocated to each of the six strata using the N-proportional allocation for a fixed
sample, considering the entire population in the Toronto CMA. The results are summarised in
Table 5-8 below.
Table 5-8 Toronto CMA, N-Proportional Sample Allocation
h Stratum Gender Nh ah nh fh
(Mode of Transportation) (Trips) (Nh/N) (nxah) (nh/Nh)
1 Car Driver Male 845,730 0.3639 491.22 0.00058
Female 640,295 0.2755 371.90 0.00058
2 Car Passenger / Carpooler Male 54,600 0.0235 31.71 0.00058
Female 113,715 0.0489 66.05 0.00058
3 Public Transit Male 206,360 0.0888 119.86 0.00058
Female 312,340 0.1344 181.42 0.00058
4 Cycle Male 14,920 0.0064 8.67 0.00058
Female 7,585 0.0033 4.41 0.00058
5 Walk Male 46,950 0.0202 27.27 0.00058
Female 62,945 0.0271 36.56 0.00058
6 Other Male 8,010 0.0034 4.65 0.00058
Female 10,820 0.0047 6.28 0.00058
Total - Mode of Transportation (N) 2,324,270 1.0000 1,350.00 0.00058
As can be seen in Table 5-8, the majority of the sample is allocated to the larger strata, Car
Driver and Transit Rider, where 863.12 and 301.28 commuters are sampled respectively. The
smaller stratum, the other modes, received a small portion of the entire sample with a sample
of only 10.93 commuters. In addition, Table 5-8 also shows that the N-proportional allocation
method produces a self-weighting design because the sampling fraction, fh, is equal to
0.00058 in all six strata. In other words, the previous sample design is self-weighting since
the N-proportional allocation is used (i.e. all units have the same inclusion probability (π=
0.00058) and hence the same design weight, 1/π= 1/0.00058= 1,724).
The previous sample allocation is maintained with a good distribution among the Toronto
CMA municipalities. However, proper attention is given to the City of Toronto since the
majority of the survey population (53.83%) lies within, as shown in Table 5-9.
N
Nn hhn
84
Table 5-9 Survey Population Breakdown
H Stratum (Geographic Boundaries) Population
(Nh)
Percentage Sample Size
per Stratum
1 Census Metropolitan Area of Toronto (N) 2,324,270 100% 1,350
2 City of Toronto 1,251,070 53.83% 726.65≈ 727
3 Toronto CMA - City of Toronto 1,073,200 46.17% 623.34≈ 623
As shown in Table 5-9, a subsample of size 727 observations out of the total sample of size
1,350 is allocated to each of the six strata using the N-proportional allocation for a fixed
sample, considering the survey population statistics in the City of Toronto, as shown in
Table 5-10.
Table 5-10 City of Toronto, N-Proportional Sample Allocation
h Stratum Gender Nh ah nh fh
(Mode of Transportation) (Trips) (Nh/N) (nxah) (nh/Nh)
1 Car Driver Male 360,345 0.2880 209.40 0.00058
Female 271,225 0.2168 157.61 0.00058
2 Car Passenger / Carpooler Male 19,475 0.0156 11.32 0.00058
Female 51,705 0.0413 30.05 0.00058
3 Public Transit Male 173,590 0.1388 100.87 0.00058
Female 267,630 0.2139 155.52 0.00058
4 Cycle Male 11,400 0.0091 6.62 0.00058
Female 6,545 0.0052 3.80 0.00058
5 Walk Male 34,610 0.0277 20.11 0.00058
Female 44,355 0.0355 25.77 0.00058
6 Other Male 4,505 0.0036 2.62 0.00058
Female 5,685 0.0045 3.30 0.00058
Total - Mode of Transportation (N) 1,251,070 1.0000 727.00 0.00058
As can be seen in Table 5-10, the majority of the sample is allocated to the Car Driver and
Transit Rider strata where 367.01 and 256.39 commuters are sampled respectively; whereas
the other modes stratum received a small portion of the entire sample of only 5.92
commuters.
5.5 Survey Instrument Design
This research combines both RP and SP data in order to take advantage of their strengths and
minimize their individual drawbacks. On the one hand, it is established that RP data may
have substantial amount of noise that result from many factors such as measurement error.
For example, an individual self-report of an actually made choice is likely to be uncertain.
Such uncertainty probably increases as the time between the actual choice and the report of
85
that choice increases. On the other hand, SP experiments are usually generated by some
systematic and planned design process in which the attributes and their levels are pre-defined
without measurement error and varied to create preference or choice alternatives.
Nevertheless, SP responses are stated and not actual, and hence are uncertain because
individuals may not actually choose the alternatives that they select during the experiment.
Hence, both methods may have potential for error. Therefore, mixing RP and SP data may be
more beneficial (Morikawa 1994; Hensher and King 2001; Dosman and Adamowicz 2006;
Hensher and Rose 2007).
Within the context of mode switching, the focus of the developed survey in this study is to
gather socioeconomic and demographic characteristics of respondents, their factual as well as
their stated experiences with travel mode choice and other psychological aspects that reflect
their tendency to mode switch. In particular, the survey collected information related to the
trip maker (e.g. age, gender, income, auto ownership and availability), the competing travel
alternatives (e.g. travel cost, parking cost, travel time, and waiting time), in addition to some
psychological factors (habit formation, personal attitude, and affective meaning) that have
shown to have a great effect on the human decision making process. Figure 5-12 presents the
four sections of the questionnaire and the information collected in each section.
The web-based data collection method is adopted in this research, where each of the recruited
participants received an invitation via email and assigned a unique code to access the
questionnaire. Although it suffers from low response rate, the online survey offered sufficient
time and cost savings as well as tailor-made interviews for each individual participant based
on his/her earlier responses in the questionnaire (Cobanoglu et al. 2001; Kwak and Radler
2002; Kaplowitz et al. 2004). In general, the questionnaire is divided into four sections.
Section A gathered revealed information regarding daily commuting work trips and current
travel options. In particular, this section asked questions about trip origin and destination, trip
start time, and primary mean of commuting as one of the following options: car driver, car
passenger, carpool, public transit, cycle, walk or other. Further, transit users were asked to
provide explicit information about their access mode as one of the following options: ride-all-
way, park-and-ride, kiss-and-ride, carpool-and-ride, or cycle-and-ride.
86
Info
rma
tion
ab
ou
t Da
ily C
om
mu
ting
Wo
rk T
rips
EMME/2
Network
Se
ctio
n A
Gender
Primary Mode
Car Driver
Travel Time
Travel Cost
Parking Cost
Car Type
Car Make
Car Model
Car Year
Transmission Type
Secondary Choice
Perception about Public Transit
Access Time
Waiting Time
In-Vehicle Time
Egress Time
Transit Fare
Transit Technology
Frequency of Past Use
Start
Car Passenger
Travel Time
Travel Cost
Secondary Choice
Carpool
Travel Time
Number of
Carpoolers
Travel Cost
Secondary Choice
Public Transit
Access Mode
Technology (Worst)
Payment Method
Reimbursement
Number of Transfers
Access Time
Waiting Time
In-Vehicle Time
Egress Time
Fare
Secondary Choice
Cycle
Travel Time
Months of Year
Secondary Choice
Walk
Travel Time
Months of Year
Secondary Choice
Other
Mode
Travel Time
Travel Cost
Parking Cost
Months of Year
Secondary Choice
Trip Start Time
Trip Origin and Destination
Full Address (Optional)
Postal Code
City
Modes Description
Sta
ted
Pre
fere
nc
e (S
P)
Ex
pe
rime
nt
Willingness to Comply to the SP Choice
Choice Tasks
In each of the six presented hypothetical scenarios, select the travel alternative that you would use to make your work trip
based on the given situation, mode features, and LOS attributes
Se
ctio
n B
Habitual Behaviour
Be
ha
vio
ura
l Info
rma
tion
Affective Appraisal (Emotional Response) for both the Chosen Mode and Public Transit Explicitly
Personal Attitude
Age
Marital Status
Occupation
Dwelling Unit Type
Household Occupancy (Older than 18 and under 18 Explicitly)
Car Ownership
Driver’s License Holding
Personal Annual Income
End
D-Efficient
Design
(72 Scenarios)
So
cio
ec
on
om
ic a
nd
De
mo
gra
ph
ic In
form
atio
n
Se
ctio
n C
Se
ctio
n D
Figure 5-12 Multi-Instrument COmmuting Survey for MOde Shift (COSMOS)
After identifying the primary mode of travel, additional mode-specific information was
gathered. Car drivers were asked about travel time, travel cost, parking cost, car type, make,
model, year and either conventional, hybrid or electric. Information about travel time and
87
travel cost was collected from both car passengers and carpoolers, in addition to the number
of passengers in the carpool for the latter. Public transit users were asked about the number of
transfers they made. In addition, detailed data about their modal combination was collected
by allowing them to choose between streetcar, bus and subway. Also, transit users were asked
about transit fare, payment method, and whether it is paid by their employer. Moreover,
special consideration was given to each of the transit trip time components by asking explicit
questions about access, waiting, in-vehicle, transfer, and egress times. Finally those who use
non-motorized (active) modes (i.e. walk and bike) were asked about travel time as well as the
months of year they tend to use this option. After that, the survey collected information about
secondary means of commuting that is used in case of unavailability of the primary option to
have a clearer idea about the hierarchies within the choice set. Finally, the last part in this
section gathered information regarding non transit users’ perceptions about public transit
service in terms of transit fare, access, waiting, in-vehicle, and egress time as well as
technological preferences (e.g. rail vs. bus attraction) and frequency of past use. Figure 5-13
shows a snapshot of Section A.
Figure 5-13 Daily Commuting Work Trips and Current Travel Options
88
The first Section of the survey allowed for gathering factual experiences and current travel
options for the trip under investigation. Moreover, the web-based data collection method
allowed for customizing the SP experiment based on earlier responses entered by the
participants. For example, gathering non transit users’ perceptions about the transit service
helped generating reasonable attribute levels for each respondent in the SP scenarios
presented in Section B of the questionnaire. In case survey respondents were unaware of the
transit service attributes, they were allowed to skip such set of questions. However, an
EMME/2 origin-destination matrix for the study area was residing in the background of the
survey and was used to estimate such missing information (given the origin and destination
postal codes of the respondent) if those questions were skipped.
Section B set up a SP experiment which is considered a key component of the developed
survey. The D-efficient design is adopted in this research to develop the stated choice
experiment. The Ngene13
software package was used to generate the design that maintains the
utility balance and maximizes the information gained from each hypothetical scenario while
minimizing the Dp-error. In order to ensure more reliable parameter estimates, a small-scale
pilot survey was conducted among a random sample of students and staff members of the
University of Toronto, Canada, based on orthogonal design. Such pilot survey was then used
to obtain prior parameter estimates for the actual experimental design.
Based on the number of attributes and their levels, the SP experimental design generated 72
scenarios that maintain attribute level balance. Obviously, it was too large to give all the 72
choice situations to a single respondent. Hence, the orthogonal design was blocked into 12
blocks of 6 choice tasks each, defining block 1 as the first 6 rows of the design, block 2 as the
second 6 rows, and so on. Importantly, each of the 12 blocks is not orthogonal by itself, but
rather the combination of all blocks is orthogonal. As such, each respondent will be faced
with a random block of 6 choice tasks instead of 72. In particular, a block (b) is randomly
drawn from blocks 1, 2, 3, …, and 12 and assigned to respondent 1. Then the rest of blocks
are assigned as follows: block [(b mod 12) + 1] to respondent 2, block [((b+1) mod 12) + 1]
to respondent 3, …, block [((b+10) mod 12) + 1] to respondent 12. We then go to block 1 for
the next set of 12 respondents. For example, if the first respondent faces block 11 of the
design, the next respondents will receive blocks 12, 1 and 2 and so on. Once all blocks are
13 http://www.choice-metrics.com/features.html
89
assigned, a number from 1 to 12 is drawn and the block sequence is repeated again. The
advantage of the previous procedure is that as long as the number of respondents is a multiple
of 12, we will have a symmetrical representation of each block (having exactly the same
number of respondents in each block) and yet a complete orthogonality in model estimation is
guaranteed (Hensher 2001a). Furthermore, in order to eliminate the order effect in the SP
experiment, the 6 choice tasks within the same block are assigned to each respondent at
random.
The designed experiment measured participants’ stated mode switching preferences in favour
of public transit given some policy changes. The stated choice experiment asked respondents
to rate their propensity to perform the same trip (their work trip) by a non-existing/modified
transit service in the future. Given that the resulting mode shift model specification has
alternatives with alternative-specific parameters, respondents were asked to choose between
labelled alternatives in the experiment (e.g. car driver, streetcar, subway). Six hypothetical
scenarios were presented to each respondent where he/she was asked to choose between
his/her primary option that was revealed earlier in the questionnaire (after some change in
factor levels), shift to a new hypothetical option or shift to other alternative that is identified
by the respondent, as shown in Figure 5-14, Figure 5-15, and Figure 5-16.
In contrast to common SP surveys, and since it is hard for respondents to make a clear choice
between the mode they are already accustomed to and a new alternative that has not been
experienced before, respondents were asked to express their degree of compliance to the
choice they stated in the experiment using a five-point Likert scale. Such scale is used later to
decrease the measurement error of the responses (Diana 2010).
Factors such as travel time, travel cost and parking cost for the car option are considered in
the experiment. Further, different components of the transit trip travel time (access, waiting,
transfer, in-vehicle, and egress time) were included as well as transit fare for the public transit
alternative. In addition, various Customer Oriented Transit Service (COTS) design factors
were considered in the experiment such as service accessibility in terms of access/egress to
public transit stops/stations as well as park-and-ride availability; service frequency and
headway in terms of the expected waiting time; trip directness in terms of number of
transfers; and service reliability standards in terms of transit schedule delay (on-time
performance). Moreover, the experiment is sensitive to loading standards in terms of
90
crowding levels and some important preference attributes such as advance information
provision, ITS technologies and rail vs. bus attraction. The previous factors are important
design parameters routinely analyzed in the service planning process. Table 5-15 shows all
factors along with their levels that were used in the SP experiment.
Figure 5-14 Stated Preference (SP) Experiment for Car Users
91
Figure 5-15 Stated Preference (SP) Experiment for Transit Users
92
Figure 5-16 Stated Preference (SP) Experiment for Active Mode Users
93
In order to ensure practical attribute level ranges, previous research and current practices in
transit service design were consulted. According to (Mistretta et al. 2009), service design
standards refer to specific goals, objectives and policies that a transit agency sets for itself in
various areas of transit service design to maintain an acceptable balance between operating
cost and service quality. In general, service design standards deal with all facets of a transit
system that affects both the passengers and the operator. In this research, more attention is
given to service design standards that affect mode shift towards public transit from the
passenger’s viewpoint. In particular, the proposed SP experiment considers factors such as
service accessibility, frequency and headway, directness and reliability.
Service accessibility standards ensure a reasonable passenger utilization of the transit service.
In general, standards for service accessibility address several aspects of the transit system that
affect the utilization of the service such as service coverage, route layout and design, stop
location and spacing. As an important measure of service accessibility, service coverage
identifies the extent to which the defined service area is being served. Service coverage is
commonly measured by the percentage of the population that resides within a suitable access
distance from a transit stop. Typically, physical access to a transit stop is achieved by
walking, riding a bicycle or driving a short distance in an automobile. Based on assumed
average walking speed of about 1.3 m/s, 400 meters (5 minutes) walk is often considered
reasonable for local transit service, which can be increased up to 800 meters for express or
rapid transit service (Murray et al. 1998; Murray 2003; Murray and Wu 2003). Another
important measure of service accessibility involves the availability of park-and-ride facilities
which extend the use of the transit system to include automobile users. Commonly, park-
and-ride facilities should be provided at appropriate stops on rapid and express services to
serve transit users from medium and low density residential areas. Sufficient off-street auto
parking should be provided at park-and-ride facilities to accommodate the total parking
demand. Park-and-ride facilities may be provided at any suitable location which can be
shown to attract 200 autos per day for express service and 150 autos per day for limited stop
service (Highway et al. 2004; Deakin et al. 2006).
Service frequency and headway are often used interchangeably to provide guidance on the
schedule design functions of a transit system. Generally, service frequency refers to how
often transit units arrive at a particular stop/station, whereas headway refers to the time
interval between the arrivals/departures of two successive transit units at a transit stop/station.
94
The common practice in service design is to have a more frequent service during peak
periods and less frequent service during off-peak periods. However, headways are not usually
allowed to exceed a specified threshold or a policy headway that defines the transit system
policy and represents the minimum level of service with respect to time of day or day of the
week. In general, policy service levels are identified as a compromise between economic
efficiency and the functionality of the system. Given that service levels below 30 minutes are
generally unacceptable from the passenger’s perspective and are not enough to develop a
solid and a consistent base of ridership, a widely used policy headway is 30 minutes during
peak hours and can reach 60 minutes during off-peak hours. Moreover, headways for night,
Saturday, and Sunday service usually match the off-peak headways or may be even longer. In
addition, policy headways can also be altered according to the offered service technology.
For example, Bus Rapid Transit (BRT) should combine a much higher service frequency by
utilizing advanced technologies such as transit signal priority, off-board payment, and queue-
jump lanes to increase the speed of the service (Vuchic 2005).
Transit travel should be as competitive as possible with private auto travel in order to provide
attractive and convenient service. One measure of such competitiveness is service directness
which refers to the degree to which a route deviates from the shortest path between the origin
and destination points of the route. In practice, agencies measure service directness using
different methods. One measure of service directness is the number of transfers required for
a passenger to reach his/her final destination. Obviously, the more transfers required in a
system, the longer total travel time will be and consequently the less desirable the service is.
Service reliability, also known as punctuality, involves the direct impact of the transit
service’s on-time performance on the passengers and the way they perceive it. In general, the
transit system should be designed and operated to maximize schedule adherence. On‐time
performance in the transit industry is defined as the percentage of trips that arrive/depart
within a specified timeframe at a specific scheduled time point. The majority of the systems
define a route as being late if it is late over 5 minutes, whereas they define a route as early
even if it is 1 minute early. In other words, some standards define “on-time” as arriving from
one minute early to five minutes late.
Crowding effects can be expressed in terms of the loading standards that are created to
maintain acceptable passenger loads on transit units. In practice, the load factor indicates the
95
extent of crowding or the need for additional transit units/vehicles. It is expressed as the ratio
of passengers actually carried versus the total seating capacity of a transit unit/vehicle (Katz
and Rahman 2010; Li and Hensher 2011)
In light of the above, best practices in transit service planning14
and TTC service planning
standards were utilized in the design to maintain reasonable attribute levels in the SP
experiment. Accordingly, different attribute level ranges are set up as upper and lower
bounds of service characteristics based on the technological differences between the current
and the proposed transit services. Moreover, SCOOT15
, the adaptive traffic control system
that is used in the City of Toronto, was consulted to come up with reasonable in-vehicle
travel time ranges for both the car and the street transit (e.g. buses, and streetcars) options
(according to SCOOT, an average reduction in journey time of 8% is achieved in the City of
Toronto). Further, designated in-vehicle travel time values were estimated based on the
difference in average operating speed between various transit technologies. Table 5-11 shows
average operating speeds for various transit technologies (Vuchic 2005).
Table 5-11 Average Operating Speeds for Various Transit Technologies
Transit
Technology
Operating Speed
km/h
Average
km/h
Bus, ROW (C) 8-12 10
Streetcar, ROW (C) 8-14 11
BRT, ROW (B) 16-20 18
LRT, ROW (B) 18-30 24
BRT, ROW (A) 22-40 31
Subway, ROW (A) 24-40 32
Based on the average operating speeds in Table 5-11, percentage of change in operating
speed is estimated for various transit technologies, as shown in Table 5-12. Then, designated
in-vehicle travel time values are estimated based on the conversion factors shown in
Table 5-13. Such designated in-vehicle travel time values insure that a subway option will be
of higher (lower) speed (in-vehicle travel time) than a streetcar one, for example.
14 http://www.nctr.usf.edu/pdf/77720.pdf 15 http://www.scoot-utc.com/WhatIsSCOOT.php?menu=Overview
96
Table 5-12 Percentage of Change in Operating Speed for Various Transit Technologies
Transit
Technology
Bus
ROW (C)
Streetcar
ROW (C)
BRT
ROW (B)
LRT
ROW (B)
BRT
ROW (A)
Subway
ROW (A)
Bus, ROW (C) 0% 10% 80% 140% 210% 220%
Streetcar, ROW (C) -9% 0% 64% 118% 182% 191%
BRT, ROW (B) -44% -39% 0% 33% 72% 78%
LRT, ROW (B) -58% -54% -25% 0% 29% 33%
BRT, ROW (A) -68% -65% -42% -23% 0% 3%
Subway, ROW (A) -69% -66% -44% -25% -3% 0%
Table 5-13 Travel Time Conversion Factors for Various Transit Technologies
Transit
Technology
Bus
ROW (C)
Streetcar
ROW (C)
BRT
ROW (B)
LRT
ROW (B)
BRT
ROW (A)
Subway
ROW (A)
Bus, ROW (C) *1 /1.1 /1.8 /2.4 /3.1 /3.2
Streetcar, ROW (C) /0.91 *1 /1.64 /2.18 /2.82 /2.91
BRT, ROW (B) /0.56 /0.61 *1 /1.33 /1.72 /1.78
LRT, ROW (B) /0.42 /0.46 /0.75 *1 /1.29 /1.33
BRT, ROW (A) /0.32 /0.35 /0.58 /0.77 *1 /1.03
Subway, ROW (A) /0.31 /0.34 /0.56 /0.75 /0.97 *1
Table 5-14 shows a numerical example of equivalent in-vehicle travel time for various transit
technologies based on a base in-vehicle travel time of 30 min.
Table 5-14 Equivalent In-Vehicle Travel Time for Various Transit Technologies
Transit
Technology
Bus
ROW (C)
Streetcar
ROW (C)
BRT
ROW (B)
LRT
ROW (B)
BRT
ROW (A)
Subway
ROW (A)
Bus, ROW (C) 30.00 27.27 16.67 12.50 9.68 9.38
Streetcar, ROW
(C) 32.97 30.00 18.29 13.76 10.64 10.31
BRT, ROW (B) 53.57 49.18 30.00 22.56 17.44 16.85
LRT, ROW (B) 71.43 65.22 40.00 30.00 23.26 22.56
BRT, ROW (A) 93.75 85.71 51.72 38.96 30.00 29.13
Subway, ROW (A) 96.77 88.24 53.57 40.00 30.93 30.00
In general, the assignment of levels to each SP attribute conditional on the RP levels was
straightforward. Except for some fixed values, the attribute levels are set as proportions
relative to those associated with a current trip identified earlier in the RP prior to the
application of the SP experiment, or designated values estimated based on the difference in
average operating speed between various transit technologies, as shown in Table 5-15.
However, if the RP trip had a zero level for an attribute, which is possible for one or more
factors (e.g. parking cost), suitable values were estimated based on the origin-destination
matrix running in the background of the survey.
97
Given the mode specific information indicated by the respondent earlier in the questionnaire,
auto in-vehicle travel time was decreased by 10% then increased by 50% and 75%. Transit
in-vehicle travel time on the other hand was first estimated based on the offered transit
service, based on the difference in average speed between transit technologies, before being
decreased by 10%, 20%, then increased by 10%. A combination of six transit technologies
and right-of-ways was considered in the experiment as follows: streetcar-ROW C, bus-ROW
C, Bus Rapid Transit (BRT)-ROW B, Light Rail Transit (LRT)-ROW B, Bus Rapid Transit
(BRT)-ROW A, and subway-ROW A. Travel and parking costs for car were increased by
25%, 50% and 75%, whereas transit fare was increased by 10%, 20% and 30% (following the
10% increment policy typically applied by the TTC). Given 5 min as a typical standard,
Access/Egress times were decreased by 50% (2.5 min) and increased by 100% (10 min). It
should be clear that access time corresponds to the walking, cycling, or time spent in a car
depending on the participant’s access mode (ride-all-way, cycle-and-ride, carpool-and-ride,
kiss-and-ride, or park-and-ride, etc.).
As for waiting and transfer times, both were decreased by 50% then increased by 50%. The
number of transfers was altered as 0, 1, and 2 or more. Three factor levels were used to
indicate the crowding level as uncrowded (seats available), moderately crowded (no seats
available), and overcrowded (wait for next transit unit). Similarly, three factor levels were
used to represent the schedule delay as early (more than 1 min early), on-time (between 1 min
early & 5 min late), and late (more than 5 min late). Finally, the availability of park-and-ride
facilities, schedule information, and real-time information was considered as Yes/No
attributes. Table 5-16 shows detailed information about the 72 choice tasks used in the SP
experiment after blocking into 12 blocks of 6 hypothetical scenarios each using Ngene16
.
16 http://www.choice-metrics.com/features.html
98
Table 5-15 Factors and Factor Levels Used in the SP Experiment
No. Factor Levels Attribute Car
Option Public Transit Option
1 A 4 Travel Cost/Fare Current Current
car.a ($/One-Way Trip) +25% +10%
transit.a
+50% +20%
+75% +30%
2 B 4 Parking Cost Current ---
car.b ($/One-Way Trip) +25% ---
+50% ---
+75% ---
3 D 3 Access Time --- -50%
(min/One-Way Trip) --- Typical
transit.d
--- +100%
4 J 3 Waiting & Transfer Time --- -50%
(min/One-Way Trip) --- Current
transit.j
--- +50%
5 C 4 In-Vehicle Travel Time -10% -20%
car.c (min/One-Way Trip) Current -10%
transit.c
+50% Designated Value
+75% +10%
6 E 3 Egress Time --- -50%
(min/One-Way Trip) --- Typical
transit.e
--- +100%
7 F 6 Transit Technology --- Streetcar, ROW C
(Rubber-Tyred, Rail) --- Bus, ROW C
transit.f
--- Bus Rapid Transit (BRT), ROW B
--- Light Rail Transit (LRT), ROW B
--- Bus Rapid Transit (BRT), ROW A
--- Subway, ROW A
8 H 2 Park-and-Ride Availability --- Yes
transit.h (Yes, No) --- No
9 G 3 Crowding Level --- Uncrowded (Seats available)
(Low, Medium, High) --- Moderately Crowded (No seats available)
transit.g
--- Overcrowded (Wait for next vehicle)
10 I 3 Number of Transfers --- 0
(0, 1 , 2 or more) --- 1
transit.i
--- 2 or more
11 K 3 Schedule Delay --- Early (More than 1 min early)
(min/One-Way Trip) --- On Time (Between 1 min early & 5 min late)
transit.k
--- Late (More than 5 min late)
12 L 2 Schedule Information --- Yes
transit.l (Yes, No) --- No
13 M 2 Real-Time Information --- Yes
transit.m (Yes, No) --- No
99
Table 5-16 D-Efficient Experimental Design (72 Choice Tasks blocked into 12 blocks)
Blo
ck N
um
ber
Ch
oic
e S
itu
ati
on
car.
a
Tra
vel
Cost
car.
b
Pa
rkin
g C
ost
car.
c
Tra
vel
Tim
e
tran
sit.
a
Fa
re
tran
sit.
c
Tra
vel
Tim
e
tran
sit.
d
Acc
ess
Tim
e
tran
sit.
e
Egre
ss T
ime
tran
sit.
f
Tec
hn
olo
gy
tran
sit.
g
Cro
wd
ing L
evel
tran
sit.
h
Pa
rk-a
nd
-Rid
e
tran
sit.
i
Tra
nsf
ers
tran
sit.
j
Wait
ing
Tim
e
tran
sit.
k
Sch
edu
le D
elay
tran
sit.
l
Sch
edu
le I
nfo
rmati
on
tran
sit.
m
Rea
l-T
ime
Info
rmati
on
1 1 Current Current -(10%) Current -(20%) -(50%) -(50%) Streetcar,
ROW C Uncrowded Yes 0 -(50%) Early Yes Yes
1 2 +(50%) +(50%) +(75%) +(30%) Designated
Value +(100%) +(100%)
BRT,
ROW B Overcrowded Yes
2 or
more +(50%) Late Yes No
1 3 +(50%) Current Current +(20%) -(10%) Typical Typical Bus,
ROW C
Moderately
Crowded No 1 Current On-Time No No
1 70 +(25%) +(75%) +(50%) +(10%) Designated
Value Typical Typical
BRT,
ROW A
Moderately
Crowded Yes 1 Current On-Time Yes Yes
1 71 +(25%) +(25%) -(10%) Current -(10%) +(100%) +(100%) Subway,
ROW A Overcrowded No
2 or
more +(50%) Late No Yes
1 72 +(75%) +(75%) +(75%) +(30%) +(10%) -(50%) -(50%) LRT,
ROW B Uncrowded No 0 -(50%) Early No No
2 19 +(75%) +(50%) +(75%) +(10%) -(20%) -(50%) Typical BRT,
ROW B
Moderately
Crowded No 1 Current Early No Yes
2 20 +(25%) +(50%) +(50%) +(20%) +(10%) Typical +(100%) Streetcar,
ROW C Overcrowded No
2 or
more +(50%) On-Time Yes No
2 21 +(25%) Current -(10%) +(10%) +(10%) +(100%) -(50%) Bus,
ROW C Uncrowded Yes 0 -(50%) Late No Yes
2 52 +(50%) +(75%) +(75%) +(20%) -(20%) +(100%) -(50%) BRT,
ROW A Uncrowded No 0 -(50%) Late Yes No
2 53 +(50%) +(25%) Current +(10%) -(20%) Typical +(100%) LRT,
ROW B Overcrowded Yes
2 or
more +(50%) On-Time No Yes
100
2 54 Current +(25%) -(10%) +(20%) +(10%) -(50%) Typical Subway,
ROW A
Moderately
Crowded Yes 1 Current Early Yes No
3 31 +(75%) Current Current +(20%) Designated
Value -(50%) +(100%)
BRT,
ROW A Uncrowded No
2 or
more -(50%) Early No Yes
3 32 +(25%) +(50%) -(10%) +(30%) -(20%) +(100%) Typical LRT,
ROW B Overcrowded Yes 1 +(50%) Late Yes No
3 33 +(25%) +(50%) +(75%) Current -(10%) Typical -(50%) Subway,
ROW A
Moderately
Crowded No 0 Current On-Time Yes Yes
3 40 +(50%) +(25%) -(10%) +(30%) Designated
Value Typical -(50%)
BRT,
ROW B
Moderately
Crowded Yes 0 Current On-Time No No
3 41 +(50%) +(25%) +(75%) Current +(10%) +(100%) Typical Streetcar,
ROW C Overcrowded No 1 +(50%) Late No Yes
3 42 Current +(75%) +(50%) +(10%) -(10%) -(50%) +(100%) Bus,
ROW C Uncrowded Yes
2 or
more -(50%) Early Yes No
4 25 Current Current +(50%) Current Designated
Value +(100%) Typical
Subway,
ROW A Uncrowded No 1 -(50%) Late No No
4 26 +(50%) +(50%) +(50%) +(30%) -(10%) Typical -(50%) BRT,
ROW A Overcrowded Yes 0 +(50%) On-Time No Yes
4 27 +(50%) Current Current +(10%) +(10%) -(50%) +(100%) LRT,
ROW B
Moderately
Crowded Yes
2 or
more Current Early Yes No
4 46 +(25%) +(75%) +(50%) +(20%) -(20%) -(50%) +(100%) Streetcar,
ROW C
Moderately
Crowded No
2 or
more Current Early No Yes
4 47 +(25%) +(25%) Current Current Designated
Value Typical -(50%)
Bus,
ROW C Overcrowded No 0 +(50%) On-Time Yes No
4 48 +(75%) +(75%) Current +(30%) -(10%) +(100%) Typical BRT,
ROW B Uncrowded Yes 1 -(50%) Late Yes Yes
5 4 +(50%) +(75%) +(50%) +(10%) Designated
Value Typical Typical
Subway,
ROW A Uncrowded No 0 +(50%) Early Yes No
5 8 +(75%) +(75%) Current Current Designated
Value +(100%) +(100%)
Streetcar,
ROW C
Moderately
Crowded Yes 1 -(50%) On-Time Yes Yes
5 9 +(25%) +(75%) Current +(30%) -(10%) -(50%) -(50%) Bus,
ROW C Overcrowded No
2 or
more Current Late Yes Yes
5 64 +(50%) Current +(50%) Current Designated
Value -(50%) -(50%)
BRT,
ROW A Overcrowded Yes
2 or
more Current Late No No
101
5 65 Current Current +(50%) +(30%) -(10%) +(100%) +(100%) LRT,
ROW B
Moderately
Crowded No 1 -(50%) On-Time No No
5 69 +(25%) Current Current +(20%) -(10%) Typical Typical BRT,
ROW B Uncrowded Yes 0 +(50%) Early No Yes
6 16 +(75%) Current -(10%) +(30%) Designated
Value +(100%) +(100%)
Bus,
ROW C
Moderately
Crowded No 0 +(50%) Early No Yes
6 18 +(75%) Current +(50%) +(10%) -(10%) Typical Typical Streetcar,
ROW C Uncrowded Yes
2 or
more Current Late Yes No
6 30 Current +(75%) -(10%) +(10%) +(10%) -(50%) -(50%) BRT,
ROW B Overcrowded No 1 -(50%) On-Time No No
6 43 +(75%) Current +(75%) +(20%) -(20%) -(50%) -(50%) Subway,
ROW A Overcrowded Yes 1 -(50%) On-Time Yes Yes
6 55 Current +(75%) Current +(20%) Designated
Value Typical Typical
LRT,
ROW B Uncrowded No
2 or
more Current Late No Yes
6 57 Current +(75%) +(75%) Current -(10%) +(100%) +(100%) BRT,
ROW A
Moderately
Crowded Yes 0 +(50%) Early Yes No
7 13 Current +(50%) Current +(20%) Designated
Value Typical -(50%)
Subway,
ROW A
Moderately
Crowded Yes
2 or
more -(50%) Late No No
7 22 +(25%) +(25%) +(75%) +(20%) -(20%) +(100%) Typical LRT,
ROW B Overcrowded Yes 0 Current Early Yes No
7 24 +(75%) +(25%) -(10%) +(20%) +(10%) -(50%) +(100%) BRT,
ROW A Uncrowded Yes 1 +(50%) On-Time Yes Yes
7 49 Current +(50%) +(75%) +(10%) -(20%) -(50%) +(100%) Bus,
ROW C Uncrowded No 1 +(50%) On-Time No No
7 51 +(50%) +(50%) -(10%) +(10%) +(10%) +(100%) Typical Streetcar,
ROW C Overcrowded No 0 Current Early No Yes
7 60 +(75%) +(25%) +(50%) +(10%) -(10%) Typical -(50%) BRT,
ROW B
Moderately
Crowded No
2 or
more -(50%) Late Yes Yes
8 11 +(75%) +(25%) +(50%) +(30%) -(10%) +(100%) +(100%) Subway,
ROW A Uncrowded Yes 0 Current On-Time No No
8 35 Current +(25%) +(75%) +(30%) Designated
Value Typical Typical
Bus,
ROW C Overcrowded Yes
2 or
more -(50%) Early Yes Yes
8 36 +(50%) +(75%) Current +(20%) -(10%) -(50%) -(50%) Streetcar,
ROW C
Moderately
Crowded Yes 1 +(50%) Late No No
102
8 37 +(25%) Current +(50%) +(10%) Designated
Value -(50%) -(50%)
LRT,
ROW B
Moderately
Crowded No 1 +(50%) Late Yes Yes
8 38 +(75%) +(50%) -(10%) Current -(10%) Typical Typical BRT,
ROW A Overcrowded No
2 or
more -(50%) Early No No
8 62 Current +(50%) Current Current Designated
Value +(100%) +(100%)
BRT,
ROW B Uncrowded No 0 Current On-Time Yes Yes
9 5 Current +(75%) -(10%) Current -(10%) Typical -(50%) LRT,
ROW B Uncrowded Yes 1 +(50%) Early No Yes
9 6 Current +(25%) +(75%) +(30%) +(10%) +(100%) Typical BRT,
ROW A
Moderately
Crowded No
2 or
more -(50%) On-Time No Yes
9 7 +(25%) +(25%) +(50%) +(20%) -(20%) -(50%) +(100%) BRT,
ROW B Overcrowded No 0 Current Late No No
9 66 +(50%) +(50%) Current +(10%) +(10%) -(50%) +(100%) Subway,
ROW A Overcrowded Yes 0 Current Late Yes Yes
9 67 +(75%) +(50%) -(10%) Current -(20%) +(100%) Typical Bus,
ROW C
Moderately
Crowded Yes
2 or
more -(50%) On-Time Yes No
9 68 +(75%) Current +(75%) +(30%) Designated
Value Typical -(50%)
Streetcar,
ROW C Uncrowded No 1 +(50%) Early Yes No
10 17 +(25%) +(50%) +(75%) Current +(10%) +(100%) -(50%) BRT,
ROW B
Moderately
Crowded Yes
2 or
more +(50%) Early No No
10 28 Current +(25%) +(75%) +(10%) -(20%) -(50%) Typical Streetcar,
ROW C Overcrowded Yes 0 -(50%) On-Time No Yes
10 29 +(50%) +(75%) +(50%) +(20%) +(10%) Typical +(100%) Bus,
ROW C Uncrowded No 1 Current Late Yes Yes
10 44 +(25%) Current Current +(10%) -(20%) Typical +(100%) BRT,
ROW A Uncrowded Yes 1 Current Late No No
10 45 +(75%) +(50%) -(10%) +(20%) +(10%) -(50%) Typical LRT,
ROW B Overcrowded No 0 -(50%) On-Time Yes No
10 56 +(50%) +(25%) -(10%) +(30%) -(20%) +(100%) -(50%) Subway,
ROW A
Moderately
Crowded No
2 or
more +(50%) Early Yes Yes
11 14 Current +(50%) -(10%) +(30%) -(20%) -(50%) Typical BRT,
ROW A
Moderately
Crowded No 0 +(50%) Late Yes Yes
11 15 +(50%) Current +(75%) Current -(10%) +(100%) -(50%) LRT,
ROW B Uncrowded No
2 or
more Current On-Time Yes Yes
103
11 23 +(75%) +(75%) Current +(10%) -(20%) Typical +(100%) Subway,
ROW A Overcrowded No 1 -(50%) Early No No
11 50 Current Current +(50%) +(20%) +(10%) Typical +(100%) BRT,
ROW B Overcrowded Yes 1 -(50%) Early Yes Yes
11 58 +(25%) +(75%) -(10%) +(30%) Designated
Value +(100%) -(50%)
Streetcar,
ROW C Uncrowded Yes
2 or
more Current On-Time No No
11 59 +(75%) +(25%) +(75%) Current +(10%) -(50%) Typical Bus,
ROW C
Moderately
Crowded Yes 0 +(50%) Late No No
12 10 +(75%) +(75%) +(50%) Current Designated
Value Typical +(100%)
LRT,
ROW B
Moderately
Crowded Yes 0 -(50%) Late No Yes
12 12 +(25%) +(25%) Current +(10%) +(10%) +(100%) -(50%) BRT,
ROW A Overcrowded No 1 Current Early Yes No
12 34 +(50%) +(25%) -(10%) Current -(20%) -(50%) Typical BRT,
ROW B Uncrowded No
2 or
more +(50%) On-Time Yes No
12 39 +(25%) +(50%) +(75%) +(30%) +(10%) -(50%) Typical Subway,
ROW A Uncrowded Yes
2 or
more +(50%) On-Time No Yes
12 61 +(50%) +(50%) +(50%) +(20%) -(20%) +(100%) -(50%) Bus,
ROW C Overcrowded Yes 1 Current Early No Yes
12 63 Current Current Current +(30%) -(10%) Typical +(100%) Streetcar,
ROW C
Moderately
Crowded No 0 -(50%) Late Yes No
104
In addition to the detailed information about the trip and the selected travel option, the survey
also gathered some latent constructs that affect respondents’ choices. Section C collected
unobservable psychological information regarding habit formation, affective appraisal and
personal attitudes. Such information allowed for matching factual experiences with personal
views concerning the trip under investigation. Different psychometric ad hoc instruments
were added to the questionnaire to measure psychological factors affecting mode choice.
Social psychologists claim that habitual behaviour can be identified given its invariability,
repetition and persistence (Golledge and Brown 1967; Banister 1978). One way of measuring
habitual behaviour is through the response-frequency measure (Verplanken and Aarts 1999).
The response-frequency measure of habit presents participants with a number of habit related
situations (e.g. to go to work, to go shopping), and asks them to respond as quickly as
possible to generate the mode of travel they associate with that situation (e.g. car, public
transit). The proportion of these responses serves then as a measure of habit formation
(Verplanken et al. 1997; Verplanken et al. 1998).
In this research, Verplanken’s response-frequency measure of habit is used for measuring
habitual frequency. A list of 9 non-working activities (e.g. to go shopping, to go to a park,
etc.) were given to the respondents, who were asked to provide the mode they would
eventually use among the following options (car driver, car passenger, carpool, streetcar, bus,
subway, cycle, walk or other) in order to accomplish those activities, as shown in
Figure 5-17. A 9-point car use habit index is then computed by counting how many times the
respondent had mentioned each mode to develop different activities. In order to take
advantage of the expected context independence of habitual behaviour, work-related
activities were excluded from the response-frequency questionnaire.
Further, affective appraisal is related to the unconsciously emotional response assigned to an
action such as shifting to a mode of transport. According to the Affect Control Theory,
emotions can be disaggregated into four fundamental dimensions, namely evaluation,
potential, activation, and control. Evaluation refers to feelings of goodness or badness elicited
by a concept, potential is associated with feelings of being strong and big as opposed to weak
or small, activation is related to whether the feeling induced by thinking about a concept is
lively or calm, and control refers to feelings of being simple or complex (Domarchi et al.
2008).
105
Figure 5-17 Habitual Behaviour
In that context, affective appraisal can be assessed indirectly using the Osgood's semantic
differential (Osgood et al. 1975). The semantic differential allows one to assign a metric to
each dimension of such feelings using a bipolar graphic rating scale with opposite adjectives
at each end that captures the connotative meaning of a concept.
In this research, the four dimensions of the semantic differential (evaluation, potential,
activation, and control) were used to measure the emotional response. A set of two-end
semantic scales was prepared to capture the latent meaning of the concept associated with
both the chosen mode and public transit. Each dimension was described by a number of
semantic scales ranging from -3 to +3, each one with words conveniently chosen to be perfect
antonyms (e.g. good/bad, fast/slow, etc.). Respondents were asked to point out quickly the
location in each semantic scale of the concept under analysis (chosen mode). In this study,
respondents faced 16 semantic scales that may describe the mode of transport they usually
take to work, in addition to 8 semantic scales that may describe public transit.
As for the chosen mode, on the one hand, evaluation was described by being good/bad,
comfortable/uncomfortable, pleasant/unpleasant, and clean/dirty. Potential was described by
being strong/weak, big/small, great/little, and flexible/inflexible. Activation was described by
106
being fast/slow, active/inactive, noisy/quiet, and crowded/empty. Finally, control was
described by being complex/simple, safe/unsafe, clear/unclear, and popular/unpopular, as
shown in Figure 5-18. On the other hand, the evaluation dimension of public transit was
described by being convenient/inconvenient, and bright/dark. Potential was described by
being significant/insignificant, and efficient/inefficient. Activation was described by being
frequent/infrequent, and cheap/expensive. At last, control was described by being
organized/disorganized, and reliable/unreliable, as shown in Figure 5-19.
Figure 5-18 Affective Appraisal Dimensions of the Chosen Mode
Furthermore, Personal attitude is associated with the expectancy (goodness) and value
(importance) related to an attitudinal object, as indicated by the Expectancy-Value Theory
(Reeve 2005). Attitudes towards car and transit were measured using five-point Likert scales
for all respondents (i.e. regardless of being a user or not). Following the Expectancy-Value
Theory, personal attitudes were measured as a combination of expectation (e.g. in general,
public transport is a good mode for work trips), and value (e.g. for me, public transit service
is important for work trips). Appropriate questions were prepared for auto users. Respondents
were asked to state whether they agree or disagree with these sentences using designated
scales. Cognitive attitude was then computed as the product of two scores (one for
expectancy and one for value), thus giving two attitudinal indices (one for car and one for
public transport), ranking from 1 to 25. Figure 5-20 shows a screenshot for the Likert scale
used for measuring attitude.
107
Figure 5-19 Affective Appraisal Dimensions of Public Transit
Figure 5-20 Personal Attitude
108
Finally, Section D collected information regarding common socioeconomic and demographic
characteristics such as gender, age, marital status, occupation, dwelling unit type, number of
persons in the household, number of cars in the household, driver’s license availability, and
annual income, as shown in Figure 5-21.
Figure 5-21 Socioeconomic and Demographic Questions
109
5.6 Chapter Summary
Emerging technologies, such as passenger information systems, ITS technologies and new
transit modes (e.g. LRT and BRT) have attributes affecting the perceptions of travellers
which are difficult to capture in RP surveys. This is a critical issue for transit service planning
where improving service to facilitate modal shift in favour of transit is targeted. This chapter
presented the design of a multi-instrument COmmuting Survey for MOde Shift (COSMOS).
The developed survey is conducted in the City of Toronto, Canada between April and May
2012 and combines three types of instruments for collecting detailed information on
commuters’ mode switching behaviour. COSMOS exploits Revealed Preference (RP) mode
choice information and Stated Preference (SP) mode switching experiments, along with
qualitative psychometric questions on users’ perception of transit service quality. The RP part
of the survey collects detailed information on recent work trips. The RP-pivoted SP choice
experiments are based on efficient experimental design technique (D-Efficient design), and
measures participants’ stated mode switching preferences in favour of public transit in
response to different policy changes. The psychometric instruments are designed to gather
information on habit of auto driving, affective appraisals and personal attitudes. The collected
dataset provides rich information with the potential of enhancing the understanding of mode
switching behaviour for commuting trips. Further, the data collected through such novel
survey is used to develop hybrid discrete choice models, where revealed mode choice models
are combined with stated mode switching probability models. More information about the
survey implementation, data analysis and modelling results are presented in the subsequent
chapters.
110
6 SURVEY IMPLEMENTATION, DATA COLLECTION AND DESCRIPTION
6.1 Chapter Overview
This chapter presents descriptive statistics and preliminary analysis of the collected dataset in
an attempt to enhance understanding commuters’ mode switching behaviour. In general, the
preliminary analysis of the collected dataset provides rich information with the potential of
enhancing our understanding of commuters’ mode switching behaviour and enriching the
transit service design toolbox for delivering more efficient and attractive services.
The remainder of this chapter is arranged as follows: Section 6.2 highlights general sample
descriptive statistics. This is followed by presenting general Revealed Preference (RP)
information statistics in Section 6.3, and general Stated Preference (SP) information statistics
in Section 6.4. Finally, a chapter summary is provided in Section 6.5.
6.2 General Sample Descriptive Statistics
The developed survey was conducted in the Toronto CMA between April and May 2012. A
total of 62,652 fully opted-in panel of Canadians who have agreed to be compensated for the
participation in market research was used as a survey sample frame of this study. A total of
13,265 individuals (21.17% of the total panel size) were recruited and invited to participate in
the survey via email. A detailed description of the study and the survey process as well as
incentives was introduced to the potential survey participants. A total of 3,769 respondents
agreed to participate in the study by signing an online consent of participation. Panellists who
agreed to participate in the study were incentivized with Air Miles through a market research
company. A total of 2,380 complete entries (1,389 incomplete entries) were initially received,
with a response rate of 17.94% which is in line with the typical travel surveys’ response rate
of 20% (Richardson et al. 1995; Franklin et al. 2003). Finally, after a process of cleaning the
dataset, the collected sample size was reduced to 1,211 observations (139 observations were
lost out of the required sample size of 1,350 observations) to maintain appropriate sample
representation of the study area for each stratum. Accordingly, the real design effect (DEFF),
is calculated as follows:
where:
n2: Adjusted sample size to account for the size of the population
n3: Adjusted sample size to account for the effect of the sample design
154.3384
211,1n
2
3 n
deff
111
Accordingly, the collected sample size n= 1,211 is allocated to each of the six strata using the
N-proportional allocation for a fixed sample, as summarised in Table 6-1.
Table 6-1 Toronto CMA, N-Proportional Sample Allocation
h Stratum Gender Nh ah nh fh
(Mode of Transportation) (Trips) (Nh/N) (nxah) (nh/Nh)
1 Car Driver Male 845,730 0.3639 440.65 0.00052
Female 640,295 0.2755 333.61 0.00052
2 Car Passenger / Carpooler Male 54,600 0.0235 28.45 0.00052
Female 113,715 0.0489 59.25 0.00052
3 Public Transit Male 206,360 0.0888 107.52 0.00052
Female 312,340 0.1344 162.74 0.00052
4 Cycle Male 14,920 0.0064 7.77 0.00052
Female 7,585 0.0033 3.95 0.00052
5 Walk Male 46,950 0.0202 24.46 0.00052
Female 62,945 0.0271 32.80 0.00052
6 Other Male 8,010 0.0034 4.17 0.00052
Female 10,820 0.0047 5.64 0.00052
Total - Mode of Transportation (N) 2,324,270 1.0000 1,211.00 0.00052
As shown in Table 6-1, the majority of the sample is allocated to the larger strata, Car Driver
and Transit Rider, where 774.26 and 270.26 commuters are sampled respectively. The
smallest stratum, the other modes, receives a small portion of the entire sample consisting of
only 9.81 commuters. In addition, Table 6-1 also shows that the sampling fraction, fh, is equal
to 0.00052 in all six strata (i.e. all units have the same inclusion probability (π= 0.00052) and
hence the same design weight, 1/π= 1/0.00052= 1,923).
The previous sample allocation is maintained with a good distribution among the Toronto
CMA municipalities. However, proper attention is given to the City of Toronto since the
majority of the survey population (53.83%) lies within, as shown in Table 6-2.
Table 6-2 Survey Sample Breakdown
H Stratum (Geographic Boundaries) Population
(Nh)
Percentage Sample Size
per Stratum
1 Census Metropolitan Area of Toronto (N) 2,324,270 100% 1,211
2 City of Toronto 1,251,070 53.83% 651
3 Toronto CMA - City of Toronto 1,073,200 46.17% 560
112
As can be seen in Table 6-2, a subsample of size 651 observations out of the total sample of
size 1,211 is allocated to each of the six strata using the N-proportional allocation for a fixed
sample, considering the survey population statistics in the City of Toronto, as shown in
Table 6-3.
Table 6-3 City of Toronto, N-Proportional Sample Allocation
h Stratum Gender Nh ah nh fh
(Mode of Transportation) (Trips) (Nh/N) (nxah) (nh/Nh)
1 Car Driver Male 360,345 0.2880 187.51 0.00052
Female 271,225 0.2168 141.13 0.00052
2 Car Passenger / Carpooler Male 19,475 0.0156 10.13 0.00052
Female 51,705 0.0413 26.90 0.00052
3 Public Transit Male 173,590 0.1388 90.33 0.00052
Female 267,630 0.2139 139.26 0.00052
4 Cycle Male 11,400 0.0091 5.93 0.00052
Female 6,545 0.0052 3.41 0.00052
5 Walk Male 34,610 0.0277 18.01 0.00052
Female 44,355 0.0355 23.08 0.00052
6 Other Male 4,505 0.0036 2.34 0.00052
Female 5,685 0.0045 2.96 0.00052
Total - Mode of Transportation (N) 1,251,070 1.0000 651.00 0.00052
Table 6-3 depicts the N-proportional sample allocation for the City of Toronto. Similar to the
sample allocation for the Toronto CMA, the majority of the sample is allocated to the Car
Driver and Transit Rider strata, where 328.64 and 229.59 commuters are sampled
respectively. On the other hand, the other modes stratum receives a small portion of the entire
sample consisting of only 5.30 commuters.
As mentioned earlier, 139 complete records were lost out of the total required sample size of
1,350 observations. The reason for removing those records from the analysis is because some
respondents reported the same postal code for their origin and destination points and hence
made it impossible to generate mode specific LOS attributes, as described later in this
chapter. In addition, some respondents reported unrealistic socioeconomic and/or
demographic information. After the process of cleaning the dataset, 1,211 observations were
allocated to different strata, as shown in the previous steps. By comparing the actual to the
theoretical sampled work trips for both the Toronto CMA and the City of Toronto, as shown
in Table 6-4 and Table 6-5, the collected sample was found to match all strata with a
113
reasonable margin of error (e=0.05), except for male (female) car drivers who were slightly
underrepresented (overrepresented) in the City of Toronto by 7.21%.
Table 6-4 Actual and Theoretical Sampled Work Trips for the Toronto CMA
Mode of
Transportation
Actual Sampled Work Trips Theoretical Sampled Work Trips
Male Female Total
Share
Male Female Total
Share
Car Driver 416
(53.75%)
358
(46.25%)
774
(63.91%)
440.65
(56.91%)
333.61
(43.09%)
774.25
(63.93%)
Car Passenger/
Carpooler
33
(37.50%)
55
(62.50%)
88
(7.26%)
28.45
(32.44%)
59.25
(67.56%)
87.70
(7.24%)
Public Transit 107
(39.63%)
163
(60.37%)
270
(22.30%)
107.52
(39.78%)
162.74
(60.22%)
270.26
(22.32%)
Cycle 8
(66.67%)
4
(33.33%)
12
(0.99%)
7.77
(66.24%)
3.95
(33.67%)
11.73
(0.97%)
Walk 23
(40.35%)
34
(59.65%)
57
(4.71%)
24.46
(42.72%)
32.80
(57.28%)
57.26
(4.73%)
Other 4
(40.00%)
6
(60.00%)
10
(0.83%)
4.17
(42.51%)
5.64
(57.49%)
9.81
(0.81%)
Total Commuting
Trips by Gender
591
(48.80%)
620
(51.20%)
1211
(100%)
613.02
(50.62%)
597.98
(49.38%)
1211
(100%)
Table 6-5 Actual and Theoretical Sampled Work Trips for the City of Toronto
Mode of
Transportation
Actual Sampled Work Trips Theoretical Sampled Work Trips
Male Female Total
Share
Male Female Total
Share
Car Driver 164
(49.85%)
165
(50.15%)
329
(50.54%)
187.51
(57.06%)
141.13
(42.94%)
328.64
(50.48%)
Car Passenger/
Carpooler
10
(27.03%)
27
(72.97%)
37
(5.68%)
10.13
(27.35%)
26.90
(72.62%)
37.04
(5.69%)
Public Transit 91
(39.57%)
139
(60.43%)
230
(35.33%)
90.33
(39.34%)
139.26
(60.66%)
229.59
(35.27%)
Cycle 6
(66.67%)
3
(33.33%)
9
(1.38%)
5.93
(63.49%)
3.41
(36.51%)
9.34
(1.43%)
Walk 18
(43.90%)
23
(56.10%)
41
(6.30%)
18.01
(43.83%)
23.08
(56.17%)
41.09
(6.31%)
Other 2
(40.00%)
3
(60.00%)
5
(0.77%)
2.34
(44.15%)
2.96
(55.85%)
5.30
(0.81%)
Total Commuting
Trips by Gender
291
(44.70%)
360
(55.30%)
651
(100%)
314.26
(48.27%)
336.74
(51.73%)
651
(100%)
114
6.3 General Revealed Preference (RP) Information Statistics
Table 6-6 shows the recruitment and response rates as well as sample descriptive statistics.
Table 6-6 Toronto CMA Sample Descriptive Statistics
Recruitment and Response Rates
Total Panel Size: 62,652
Recruited/Invitations Sent: 13,265
Recruitment Rate (%): 21.17%
Total Responses: 3,769
Completes: 2,380
Incompletes: 1,389
Response Rate: 17.94%
Final Sample Size: 1,211
Sample Descriptive Statistics
Variable Value Sample Size Percentage
Gender
Male
Female
591
620
48.80%
51.20%
Age
(In years)
15 - 24
25 - 34
35 - 44
45 - 54
55 - 64
65 - 74
75 and over
68
431
249
259
173
29
2
5.62%
35.59%
20.56%
21.39%
14.29%
2.39%
0.17%
Marital Status
Single
Married
Divorced
Widowed
464
650
76
21
38.32%
53.67%
6.28%
1.73%
Occupation
(According to
the
National
Occupational
Classification
(NOC)
of Canada)
(A) Management
(B) Business, Finance, and Administration
(C) Natural and Applied Sciences
(D) Health
(E) Social Science, Education,
Government Service, and Religion
(F) Art, Culture, Recreation, and Sport
(G) Sales and Service
(H) Trades, Transport, and Equipment
197
291
33
103
152
29
119
43
16.27%
24.03%
2.73%
8.51%
12.55%
2.39%
9.83%
3.55%
115
Operators
(I) Primary Industry
(J) Processing, Manufacturing, and Utilities
(K) Other
19
18
207
1.57%
1.49%
17.09%
Dwelling Type House
Townhouse
Apartment
675
128
408
55.74%
10.57%
33.69%
Household
Size
(18 years old
and above)
1
2
3
4+
156
437
335
283
12.88%
36.09%
27.66%
23.37%
Household
Size
(below 18
years old)
0
1
2
3+
839
194
150
28
69.28%
16.02%
12.39%
2.31%
No. of
Vehicles
in the
Household
0
1
2
3+
118
507
447
139
9.74%
41.87%
36.91%
11.48%
Driving
License
Holding
No
Yes
61
1150
5.04%
94.96%
Personal
Income
Less than $10,000
$10,000 to $19,999
$20,000 to $29,999
$30,000 to $39,999
$40,000 to $49,999
$50,000 to $59,999
$60,000 to $79,999
$80,000 and over
33
36
64
104
170
179
280
345
2.73%
2.97%
5.28%
8.59%
14.04%
14.78%
23.12%
28.49%
116
In terms of gender, the preliminary analysis of the collected dataset shows a slight
underrepresentation of males in the sample (48.80% instead of 50.62% in the base
population), and subsequently a slight overrepresentation of females (51.20% instead of
49.38% in the base population) is realized, as shown in Figure 6-1.
Figure 6-1 Gender
The preliminary analysis of the collected sample shows a skewed age distribution in the
sample starting with few observations 15 to 24 years old, followed by high representation of
those aged 25 to 34 years old, and ending with very few observations aged 75 years and over.
This may in part be due to the fact that very young people are often unemployed while
elderly people are often retired and do not commute to work (which is the main trip purpose
of this study), as shown in Figure 6-2.
Further, at 24.03%, the business, finance and administration occupation category has the
highest percentage in the sample; while the processing, manufacturing and utilities
occupation category has the lowest percentage, as shown in Figure 6-3. Moreover, the sample
shows a high percentage of married respondents (53.67%) and a high tendency for living in
houses (66.31%) as opposed to living in apartments, as shown in Figure 6-4 and Figure 6-5,
respectively.
Males 48.80%
Females 51.20%
117
Figure 6-2 Age Distribution
Figure 6-3 Occupation (According to the NOC of Canada)
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
35.00%
40.00%
15 to 24years
25 to 34years
35 to 44years
45 to 54years
55 to 64years
65 to 74years
75 yearsand over
A 16.27%
B 24.03%
C 2.73%
D 8.51%
E 12.55%
F 2.39%
G 9.83%
H 3.55%
I 1.57%
J 1.49%
k 17.09%
A: Management
B: Business, Finance, etc.
C: Natural and Applied Sciences
D: Health
E: Social Science, Education, etc.
F: Art, Culture, etc.
G: Sales and Service
H: Trades, Transport, etc.
I: Primary Industry
J: Processing, Manufacturing, etc.
k: Other
118
Figure 6-4 Marital Status
Figure 6-5 Dwelling Type
Single 38.32%
Married 53.67%
Divorced 6.28%
Widowed 1.73%
House 55.74%
Townhouse 10.57%
Apartment 33.69%
119
The preliminary analysis of the collected dataset also shows that the majority (36.09%) of the
households has two persons (including the respondent) living in the same household, while
12.88% has only the respondent living by himself/herself, as shown in Figure 6-6. This is in
line with the previous finding that the majority of the respondents are married. In addition,
the results shown that the majority (69.28%) of the sampled respondents has no kids, as
shown in Figure 6-7.
Figure 6-6 Household Size (18 years old and above)
Figure 6-7 Household Size (below 18 years old)
1 12.88%
2 36.09%
3 27.66%
4+ 23.37%
0 69.28%
1 16.02%
2 12.39%
3+ 2.31%
120
As an early indication of strong car use within the study area, the sample shows a very high
percentage of driving licence holding and auto ownership, as shown in Figure 6-8 and
Figure 6-9, respectively. While only 9.74% of the sampled respondents reported no vehicle
ownership, 41.87% reported owning one car, 36.91% reported two cars, and 11.48% reported
three cars or more.
Figure 6-8 Driving License Holding
Figure 6-9 No of Vehicles in the Household
No 5.04%
Yes 94.96%
0 9.74%
1 41.87%
2 36.91%
3+ 11.48%
121
Finally, the personal income distribution shows a few observations having annual income
below $10,000, while the majority of the respondents have annual income of $80,000 and
over, as shown Figure 6-10. Such income distribution might describe the high levels of auto
ownership and tendency for living in houses in the sample.
Figure 6-10 Personal Income Distribution
6.4 General Stated Preference (SP) Information Statistics
Being a key component of the developed survey, the collected SP dataset is further
investigated to have a better idea about the stated mode switching preferences of respondents.
Figure 6-11 depicts the proportions of SP mode switching responses for various RP primary
choices. As an evidence of strong inertia against shifting between modes, the SP experiment
results show that travellers tend to stay with the mode they are already accustomed to,
regardless of policy changes that sometimes are in favour of the competing options.
However, examining the second best option clearly shows that car users may shift to public
transit in case a proper service is provided to them. Interestingly, with a mode shift
percentage of 32.53%, shared ride users (car passengers and carpoolers) are more willing to
switch to public transit than car drivers who have a mode shift percentage of 25.47%. This
might be related to the higher sense of independence and privacy associated with the car
option which is lesser for car passengers and carpoolers. That is why shared ride users might
not have a problem to share the transit unit with other riders.
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
30.00%
Lessthan
$10,000
$10,000to
$19,999
$20,000to
$29,999
$30,000to
$39,999
$40,000to
$49,999
$50,000to
$59,999
$60,000to
$79,999
$80,000and over
122
Figure 6-11 Proportions of SP Mode Switching Behaviour
The previous argument is further supported by the mode shift model for car passengers and
carpoolers, where the crowding effect was found to be insignificant attribute in the model. On
the other hand, transit and active mode users (cycle and walk) seem to be more loyal to their
travel alternatives, as a very small proportion of them decided to switch to another option. In
terms of policy implications, certainly the previous observations affect the possibility of
promoting the use of public transit between car users. While it might be possible to persuade
shared ride users to switch to public transit, convincing car drivers seems not to be an easy
task and more efforts need to be done in order to attract them.
123
Further, examining survey respondents’ degree of compliance to their stated choices in the SP
experiment shows that they are confident about their decisions. As shown in Figure 6-12, the
largest percentage of the sample (29.11%) indicated a strong propensity to adhere to their
stated choices for future work trips. This is followed by 25.54% and 27.86% for both
moderately strong and neutral degree of adherence, respectively. The previous finding
increases the confidence in the SP choices reported by the respondents and subsequently the
confidence in the developed mode shift models.
Figure 6-12 Degree of Compliance to the SP Choice
6.5 General Psychological Information Statistics
In addition to the collection of RP and SP information, COSMOS gathered detailed data
about various psychological factors affecting commuters’ choices. Figure 6-13 portrays the
proportions of the habitual behaviour of the survey respondents. In general, the results show
that car users have strong habits associated with their primary chosen modes (e.g. car drivers
have 73.62% of their formed habits associated with car use). Such level of habit formation
makes it hard to change the mode they are already accustomed to. In an indication that car
drivers could use the car for most of their trips, the proportion of habitual behaviour is large
for the car, and much smaller for transit (i.e. car drivers will seldom use public transport).
Interestingly, a similar trend is realized while examining the formation of habits of transit
riders, a small habitual behaviour proportion is associated with transit while a high proportion
is associated with the car. It seems that transit riders are unattached to public transit in terms
of habits, as they might still consider using the car for non-commute trips. The previous
finding is considered evidence for the superiority of the car, which is in line with the
outcomes of the SEM analysis conducted earlier in Chapter 4.
Very Weak, 9.69%
Moderately Weak, 7.80%
Neutral, 27.86% Moderately
Strong, 25.54%
Strong, 29.11%
124
Moreover, Figure 6-14 depicts the emotional response of travellers associated with their
primary mode choice in terms of the four dimensions of the semantic differential, namely
evaluation (good vs. bad), potential (big vs. small), activation (lively vs. calm), and control
(simple vs. complex). As defined earlier in Chapter 5, the evaluation dimension refers to
feelings of goodness or badness elicited by a concept, potential is associated with feelings of
being strong and big as opposed to weak or small, activation is related to whether the feeling
induced by thinking about a concept is lively or calm, and finally, control refers to feelings of
being simple or complex. In general, the charts show a common trend where travellers give
higher values to the activation and potential dimensions of the affective factor compared to
the evaluation and control ones. This can be related to the sense of familiarity associated with
the mode they are accustomed to. In a similar analysis, Figure 6-15 portrays the emotional
response of travellers associated with public transit in terms of the four dimensions of the
semantic differential, regardless of being a transit rider or not. Car users give higher values to
the evaluation and activation dimensions compared to the potential and control ones, while
transit riders give a higher weight to the activation and control ones.
Studying personal attitude shows that car users give more importance to the value (important)
and the expectation (good) components of attitude for car; whereas they give lesser
importance to those of transit. On the other hand, transit and active modes (i.e. walk and bike)
give more importance to the value (important) and the expectation (good) components of
attitude for transit, as shown in Figure 6-16. Importantly, transit riders give more weight to
the value (important) component of attitude for transit rather than the expectation (good)
component. In other words, they might know that transit is important and that is the reason
why they use it, although they do not perceive it as good option. There is a sort of detachment
from public transit.
125
Figure 6-13 Proportions of Habitual Behaviour
126
Figure 6-14 Emotional Response towards Primary Chosen Mode
127
Figure 6-15 Emotional Response towards Public Transit
128
Figure 6-16 Proportions of Personal Attitude
129
6.6 Chapter Summary
As presented in this chapter, the preliminary analysis of the dataset enriched the
understanding of mode switching behaviour. The collected dataset can generally answer two
main research questions. First, what are the perceived importance of service quality and cost
in the mode shifting process? Second, how do trip makers make tradeoffs among the previous
attributes when shifting to a new travel option? Further, the results show that mode shift
decisions are affected by some behavioural factors in which passengers are more (less)
inclined to choose (change) the modes they are already accustomed to.
The collected dataset is used in the next chapter to develop econometric models of mode
shift, with emphasis on capturing mode switching behaviour of respondents towards public
transit. The developed models will provide more insights regarding how trip makers actually
make tradeoffs among different COTS elements and LOS attributes.
130
7 MODE CHOICE/MODAL SHIFT MODELLING
7.1 Chapter Overview
This chapter is intended to explore the effects of behavioural factors and the relative
importance of different Customer Oriented Transit Service (COTS) attributes on mode
choice/modal shift decisions. In general, four types of models are presented in this chapter
based on the collected dataset. First, a traditional mode choice model is developed using the
Revealed Preference (RP) information, including only socio-demographic characteristics of
the decision maker (e.g. age, gender, etc.) as well as basic modal LOS attributes of the
available alternatives (e.g. travel cost, travel time, etc.). Second, more complex mode choice
models with latent variables are developed using the same dataset after adding various
psychological factors, namely personal attitude, habit formation, and emotional response.
Third, separate mode switching models are developed for different groups of commuters (car
drivers, shared ride users, transit riders, and active mode users) using the Stated Preference
(SP) information. Fourth, enhanced mode switching models are developed using joint
(RP/SP) data. Furthermore, a policy analysis is conducted at the end of this chapter to
examine the predictability of the developed models in an attempt to quantify the transit
ridership overestimation that traditional models suffer from.
The following sections of this chapter are organized as follows: Section 7.2 documents the
fundamental definitions and assumptions upon which the models are built. It provides
information and definitions about the unit of travel demand, choice of analysis time period,
definition of trip purpose to be modelled, and definition of the study area. In addition, Section
7.3 provides a detailed description for the modes of travel considered in the choice set. Then,
level of service attributes generation is discussed in Section 7.4. Further, Sections 7.5, 7.6,
and 7.7 present the modelling efforts with respect to commuting work trip mode choice;
commuting work trip mode choice with latent variables; and commuting work trip mode
shift; respectively. In general, the latter three sections present key characteristics of the
modelling systems; thorough understanding of what the models are capable of doing; major
strengths and weaknesses; complete description of the employed modelling methods and
procedures; and model parameter statistical estimation results. Subsequently, Section 7.8
examines the forecast ability of the developed models and quantifies the transit ridership
overestimation imposed by traditional models. Finally, Section 7.9 provides a chapter
summary.
131
7.2 Fundamental Definitions and Assumptions
In general, travel demand models are concerned with quantifying individual’s choices
regarding their mode of travel in a particular period of time (e.g. morning peak-period) for a
specific purpose (e.g. home-to-work, home-to-school, etc.) within a well-defined area (e.g.
traffic zone, planning district, etc.). This section provides fundamental definitions and
assumptions used in the modelling process, including definitions of the unit of travel demand,
time period, trip purposes, and study area (Miller 2001).
7.2.1 Unit of Travel Demand
The basic unit of travel demand in the developed models is the trip, which is the movement of
an individual from a single origin to a single destination for a single purpose.
7.2.2 Trip Purpose
Another key factor of mode choice modelling is the treatment of trip purpose. In general,
trips are made for a particular purpose such as work trips, school trips, or discretionary trips
(e.g. shopping, entertainment, etc.). In this research, the developed models focus on
commuting work trips for two reasons. First, work trips constitute an increasingly large
proportion of urban trips in the City of Toronto and, therefore, have a major impact on traffic
congestion and emissions. Second, since habits can be identified by their repetition and
persistence, the behavioural factors in question are likely to have greater effects on people’s
behaviour when commuting to work than they do for pursuing non-work trips.
7.2.3 Trip Time
The temporal distribution of trips within a given time period is essential to choice modelling.
In general, trips are made over the course of the day, with mode choice behaviour varying by
time of day, day of the week and season of the year. Therefore, travel demand models usually
deal with a specified day (e.g. typical weekday), with all trips being made within a given
period of time (e.g. morning peak-period), to ensure consistency in the modelled travel
behaviour.
The developed models in this research deal with the weekday morning-peak period from 6:00
to 8:59 a.m. There are two main reasons for the emphasis on weekday morning-peak period
as the trip time. First, morning peak-period is the period of highest demand throughout the
day that determines the capacity required for the transportation system. Second, being
dominated by the trip to work (the primary focus of this research), morning peak-period
132
travel is considered the easiest type of travel behaviour to understand and model. Therefore,
morning peak-period is the primary period of analysis for most regional transportation
planning purposes according to the common Canadian practice (Miller 2001).
7.2.4 Study Area
The spatial distribution of trips within a given area is essential to choice modelling. In
general, trips are assumed to be originating from and destined to a given area such as traffic
zone, planning district, etc. In this research, models are developed for the Toronto CMA with
explicit attention to the City of Toronto where a multimodal transit system and supportive
land use make transit more competitive to auto travel (as described earlier in Chapter 5).
The base year for the developed models is 2006 which represents the most recent year where
extensive travel behaviour data for the Toronto CMA is available from both the Place of
Work and Commuting to Work data released by Statistics Canada in 2006, and the 2006
Transportation Tomorrow Survey (TTS). It is worth noting that the 2011 census data was not
yet released by the time COSMOS was designed and implemented. Whereas the 2006 Place
of Work and Commuting to Work data released by Statistics Canada presents 2006 census
highlights on mode of transportation, place of work and commuting distance to work within
the study area; the 2006 TTS consists of a 5% sample of all households within the GTA and
its surrounding areas, including detailed household characteristics and trip records for all
members of the surveyed households.
Further, the data used in the models estimation process comes from the multi-instrument
socio-psychometric survey collected in the Toronto CMA in April 2012. The 2012 multi-
instrument survey consists of a 0.05% (1,211 observations) sample of all individuals in the
employed labour force, 15 years and over, having a usual place of work in the Toronto CMA
and excluding those who work at home. This population is estimated as 2,324,270
individuals. The collected dataset contains extensive travel behaviour information for the
Toronto CMA. The survey gathered qualitative psychometric questions on users’ perception
along with Revealed Preference (RP) mode choice information and Stated Preference (SP)
mode switching experiments.
133
7.3 Modes of Travel
According to the Place of Work and Commuting to Work data released by Statistics Canada
in 2006, car drive was clearly found to be the dominating mode of travel, with 63.94% of all
commuting work trips in the Toronto CMA. Adding a value of 7.24% of car passenger trips
increases the total percentage of work trips that are made by car to 71.18%. Public transit is
the second most used mode overall with 22.32% of all work trips. Active modes (walk and
cycle) have a combined percentage of 5.70% of all work trips in the Toronto CMA, as shown
in Figure 5-5, Chapter 5. All other modes, including taxicab and motorcycle (0.81%) are of
very minor importance.
Further, a similar mode split distribution was observed in the City of Toronto. Car drive was
also found to be the prevailing mode of travel with 50.48% of all work trips in the City of
Toronto. Combined with an additional 5.69% of trips made as car passengers, 56.17% of all
work trips are made by car. Public transit is the second most used mode overall with 35.27%
of all work trips. Active modes (walk and cycle) have a combined percentage of 7.74% of all
work trips in the City of Toronto as shown in Figure 5-8, Chapter 5. All other modes,
including taxicab and motorcycle (0.81%) are of lower importance.
In light of the modal split proportions above, the “other” mode category is excluded from the
models estimation since they represent a negligible percentage of the overall sample.
Furthermore, taxicab trips are treated as auto passenger trips, while motorcycle trips are
treated as car driver trips. The following section discusses the mode definitions adopted
within the model system, namely auto mode, public transit, and non-motorized modes.
7.3.1 Auto Option
In order to be in line with the logical and behavioural categorization of auto-person used in
common practice, auto trips are categorized into auto-driver and auto-passenger. Given the
similarity in the associated choice behaviour, carpooling trips are treated as auto passenger
trips in the model system.
Auto driver all way: that is, the trip-maker drives in a car for the entire length of the trip
from home to work. This mode is assumed to be available to travellers based on driver’s
licence holding and household car ownership.
134
Auto passenger all way: that is, the trip-maker is a passenger in a car for the entire length of
the trip from home to work. This mode is assumed to be available to all travellers (carpool
and taxicab trips are treated as auto passenger trips).
7.3.2 Public Transit Option
Public transit is considered a key component of the model system in this research since it has
a great policy importance to mode shift modelling. In general, public transit has a large
family consisting of a wide variety of transit services characterized by different levels of
service (e.g. travel time, headway, etc.), technologies, fare policies, and other qualitative
attributes such as comfort and safety. The local transit services in the Toronto CMA are
operated by different transit agencies (TTC, Durham, York, Peel, and Halton) and composed
mainly of subway lines, streetcar routes, and different bus services. Each transit service can
be treated either as a main or a feeder to another main service. However, travelling within the
transit network is considered a route choice problem rather than a mode choice one. GO
services, on the other hand, are qualitatively different from most local transit services in the
Toronto CMA given its relatively high speed, high cost, low frequency, high quality, long
distance inter-municipality service (Miller 2001). In addition, the percentage of transit trips
that are served by Go transit in the study area is comparatively small given that GO services
are mainly concerned with regional trips. Thus, GO transit is excluded from the analysis.
Another key factor for modelling the use of public transit as a mode of travel is that of transit
access mode. According to the 2006 Transportation Tomorrow Survey (TTS), the transit
work trips distribution by access mode is presented in Table 7-1.
Table 7-1 CMA 2006 Transit Work Trips by Access Mode
Primary Mode
Access Mode
Walk
Oth
er
Au
to P
ass
enger
Cycl
e
Sch
ool
Bu
s
Taxi
Pass
enger
Au
to D
riv
er
Moto
rcycl
e
Un
kn
ow
n
Tota
l
Number of transit trips
excluding GO transit
300,2
77
304
15,9
27
681
126
136
18,0
59
35
61
335,6
06
Percentage of transit trips
excluding GO transit 89.5
%
0.1
%
4.7
%
0.2
%
0.0
%
0.0
%
5.4
%
0.0
%
0.0
%
100.0
%
135
As shown in Table 7-1, around 90% of all transit users (excluding GO transit) walk or cycle
to reach the origin transit stop or station. Meanwhile, 10.1% of transit users (excluding GO
transit) use the car, as either a driver or a passenger, to reach the transit service. Thus, the
public transit option is split into transit with car access and transit with Non-Motorized
Transport (NMT) access, within the model system.
Transit with NMT access: that is, the trip-maker accesses the transit system by walking or
cycling to a bus stop or station. This mode is assumed to be available to all travellers given
the high transit coverage in the study area.
Transit with car access: that is, the trip-maker accesses the transit system by car as either a
driver or a passenger to a bus stop or station. This mode is assumed to be available to
travellers based on household car ownership.
7.3.3 Non-Motorized Transport (NMT) Option
Non-motorized transport refers to the walk and cycle modes of travel. According to the Place
of Work and Commuting to Work data released by Statistics Canada in 2006, non-motorized
transport constitutes only 5.7% (4.73% for walk and 0.97 for bicycle trips) of all work trips in
the Toronto CMA, as shown in Figure 5-5. Further, it constitutes only 7.74% (6.31% for walk
and 1.43% for bicycle trips) of all work trips in the City of Toronto, as shown in Figure 5-8.
The preliminary analysis of the collected dataset shows that the importance of non-motorized
trips increases for shorter trips. Figure 7-1 depicts the modal shares for work trips by origin-
destination walking distance. On the one hand, walk is the primary mode for very short trips
(2 km or less) and is a significant mode of travel for trips of up to around 5 km. On the other
hand, bicycle accounts for about 4.7% of all work trips of 5 km length or less. Further,
Figure 7-2 plots the cumulative number of non-motorized trips versus walking trip distance in
comparison with the cumulative total number of trips by all modes. As shown, 95% of all
non-motorized trips are about 4 km or less in length, while 98% are approximately 5 km or
less. In more specific terms, 92% of cycle trips are about 5 km or less in length, while 100%
are approximately 8 km or less. On the other hand, 98% of walk trips are about 4 km or less
in length, while 100% are approximately 5 km or less. Despite their low modal share
percentage, active modes (walk and cycle) are retained in the model system given their
increasing policy interest within the study area. The cycle option is assumed to be available
for all trips with walking distance of 8 km or less, while the walk option is assumed to be
available for all trips with walking distance of 5 km or less.
136
Figure 7-1 Mode Shares by Trip Length
-10%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 5 10 15 20 25 30
Pe
rce
nt
of
Tota
l Tri
ps
at t
his
Dis
tan
ce (
%)
Walking Trip Distance (km)
Car Driver Car Passenger Carpool Transit Rider Cycle Walk Other
137
Figure 7-2 Trip CDF by Trip Length
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
0 5 10 15 20 25 30
Pe
rce
nt
of
Tota
l Tri
ps
at t
his
Dis
tan
ce (
%)
Walking Trip Distance (km)
Cumulative Walk Trips (%) Cumulative Cycle Trips (%)
Cumulative NMT Trips (%) Cumulative Total Trips (%)
138
7.4 Generating Level of Service Attributes
Apart from respondents’ perceptions that were collected in the questionnaire, more accurate
Level of Service (LOS) attributes were generated for use in the model estimation process.
Given the trip origin and destination postal codes collected earlier in the survey, trip length
(in Kilometres) and duration (in minutes) were estimated for car, bike, and walk options
using the Google Directions API. The Google Directions API calculates directions between
static (known in advance) locations using HTTP requests. Further, public transit trip length
and duration were generated using both Google Maps Trip Planner API and HopStop API.
On the other hand, travel cost for the car was estimated based on the driving costs 2012
manual released by the Canadian Automobile Association (CAA)17
. The average operating
cost per Kilometre based on vehicle type was calculated for car driver, divided by 2 for car
passenger, and divided by the number of travelers for the carpool option.
7.5 Modelling Commuting Work Trip Mode Choice
This section investigated commuting work trip mode choice in the study area. A traditional
mode choice model is developed using RP information (observing the actual choice of
commuting mode), including only socio-demographic characteristics of the decision maker
(e.g. age, gender, etc.) as well as basic modal LOS attributes of the available alternatives (e.g.
travel cost, travel time, etc.). This model does not include any behavioural factors.
7.5.1 General Model Specification
After a series of specification tests, the RP dataset is used to develop disaggregate
Multinomial Logit (MNL) model (Habib et al. 2009). The developed model is derived from
the fundamental Random Utility Maximization (RUM) Theory, such that:
Uim= V(Xim; β) + εim, (7-1)
where:
Uim : Utility that individual (i) obtains from mode (m); i= 1, …, I; m, n= 1, …, N
V : Systematic component of utility
Xim : Vector of explanatory variables including attributes of individual (i) and mode (m)
β : Vector of parameters
εim : Random component of utility
17 http://www.caa.ca/
139
The current investigation considers six modes as follows: (1) car driver, (2) shared ride (car
passenger and carpooler), (3) public transit with car access, (4) public transit with NMT
access, (5) cycle, and (6) walk.
Figure 7-3 illustrates the overall model structure of the developed RP mode choice model.
RP
Mode Choice
Car
Driver
Shared
Ride
Transit
Car Access
Transit
NMT AccessCycle Walk
Figure 7-3 RP Mode Choice Model Structure
In order to formulate the probabilistic choice model, the distribution of the random
component of utility εm is assumed to be Independently and Identically Distributed (IID)
Extreme Value Type I. The previous assumption leads to the following closed form
probability of mode selection (Ben-Akiva and Lerman 1985):
(7-2)
where:
Pim : Probability that individual (i) selects mode (m)
Vim: Utility that individual (i) obtains from mode (m) (i= 1, ..., I ; m, n= 1, …, N)
Ci : Choice set of feasible alternative modes (N) for individual (i)
7.5.2 Empirical Analysis
As noted above, a Multinomial Logit (MNL) model is developed to investigate commuting
work trip mode choice in the study area. The developed model estimates the probability that
an individual trip-maker will choose any given mode from the set of feasible alternatives
described earlier in this section. After examining a set of alternative modelling structures, the
final model is estimated and reported in Table 7-2.
,Pim
i
im
im
Cm
V
V
e
e
140
Table 7-2 RP Mode Choice Model
Loglikelihood of RP Mode Choice -786.45
Loglikelihood of Null Model -1748.23
Rho-Squared Value 0.55
Variable Mode Parameter t test
Alternative Specific Constant Car Driver -0.0531 -0.135
Shared Ride (Car
Passenger/Carpool)
-1.7283 -4.350
Transit Rider (Car Access) 0.0496 0.093
Transit Rider (NMT Access) 1.9115 4.487
Cycle -1.1534 -2.566
Walk (Base) 0 ---
Travel Cost All Motorized Modes -0.0967 -1.873
In-Vehicle Travel Time All Motorized Modes -0.0436 -6.949
Trip Length All NMT Modes -0.1689 -2.213
Waiting Time Transit Rider (Car Access) -0.0653 -2.180
Transit Rider (NMT Access) -0.0820 -4.406
(Access + Egress) Time Transit Rider (Car Access) -0.0018 -0.831
Transit Rider (NMT Access) -0.0088 -2.948
Number of Vehicles
in the Household
Car Driver 0.1907 1.677
Free Parking Availability Car Driver 1.8801 10.630
Number of People
in the Household
Shared Ride (Car
Passenger/Carpool)
0.3358 3.851
Gender: Female Transit Rider (Car Access) 0.9558 2.624
Transit Rider (NMT Access) 0.4245 2.116
Cycle -1.1562 -1.832
Transit Technology: Transit Rider (Car Access) -1.5499 -4.050
Street Transit (Bus/Streetcar) Transit Rider (NMT Access) -0.7131 -3.510
Age: 18 - 35 years old Walk 0.6794 2.015
As shown in Table 7-2, the specification of the final model is derived based on the
accommodation of variables with proper signs and statistical significance. The critical value
(1.96) of the t-statistic with a 95% confidence limit is considered as the threshold value of
considering variables in the model. However, some parameters with t-statistics values lower
than 1.96 are retained in the model because the corresponding variables provide considerable
insight into the behavioural process.
Given the presented rho-squared value of 0.55, the model has an acceptable goodness of fit
and explanatory power. The examination of explanatory variables in the model shows that
travel cost, trip length (for non-motorized options), and the different trip time components
have correct (negative) signs that match expectations. By focusing on different trip time
141
components, it can be observed that the in-vehicle and waiting times are more relevant to
mode choice decisions than walking (access and egress) time. Such finding may be related to
the high level of transit access coverage in the study area. The modelling results also show
that waiting time coefficients are around (1.5 - 2.0) times higher than the in-vehicle travel
time, which is close to ratios reported in similar studies. Besides, the subjective value of the
in-vehicle travel time is around 27.1 $/hr. On the other hand, values of waiting time are equal
to 40.5 $/hr and 50.9 $/hr for transit rider with car access and transit rider with NMT access,
respectively.
As for the effect of transit technology, it is clearly shown that transit users have negative
perception for street transit (e.g. streetcar and bus) compared to the rapid transit option (e.g.
subway). Interestingly, transit riders with car access valuate street transit options even less
than transit riders with NMT access do (i.e. passengers who walk to transit are more willing
to take a bus or a streetcar than those who access transit using a car).
Moreover, the model also shows that females are more likely to take public transit and less
likely to cycle than males. Further, it is found that younger commuters (18 to 35 years) have
high propensity to walk. On the other and, auto ownership has a positive sign associated with
car use as expected. This might be considered an early indication of car use habit formation.
The presented RP mode choice model explicitly models auto drive and auto passenger trips as
separate modes of travel. This implies that this basic model structure is well suited to the
analysis of car driver and auto passenger related policies (e.g. toll, HOV lane
implementations, etc.). However, it must be noted that this model is very simple and may
suffer from the major drawbacks of traditional mode choice models. Therefore, the
implementation of this model will not be able to support detailed mode shift analysis.
The developed RP mode choice model represents an important first step towards a policy
sensitive model of shifting to public transit, yet much work remains to actually achieve full
policy sensitivity. In order to improve the capabilities of the developed model, behavioural
factors are introduced to the model structure as latent variables in the following section.
142
7.6 Modelling Commuting Work Trip Mode Choice with Latent Variables
This section adopts a general methodology for incorporating latent variables in mode choice
models (Walker 2001; Ben-Akiva et al. 2002). The methodological approach requires the
estimation of an integrated multi-equation model that consists of a discrete choice model and
a latent variable model of structural and measurement equations. The integrated model is
estimated simultaneously using a maximum likelihood estimator where the likelihood
function includes complex multi-dimensional integrals. Unlike the research presented in Ben-
Akiva et al. (2002) which focused on the use of psychometric data to explicitly model
attitudes and perceptions and their influences on choices, this thesis builds on the findings of
Chapter 4 and explicitly models personal attitudes, habit formation, and emotional response
as the three major determinants of mode choice (Tudela et al. 2011).
7.6.1 General Model Specification
As shown in Figure 7-4, the path diagram of the integrated model is comprised of two
components: an RP mode choice model and a latent variable model. Similar to what is
discussed earlier in Chapter 4, the terms in ellipses represent unobservable latent constructs,
while those in rectangles represent observable variables.
Latent
Variables
(Z)
RP Mode
Choice
(y)
Latent Utility
(U)
Indicators
(I)
η
ν
Latent Variable ModelRP Choice Model
Explanatory
Variables
(X)
ε
Figure 7-4 RP Mode Choice Model with Latent Variables
In the RP mode choice model, individual’s utility (U) for each travel alternative is considered
a latent variable, whereas the observable choices (y) are considered indicators of the
underlying utility. Similarly, in the latent variable model, the associated behavioural factors
(Z) are considered latent variables which are indicated by observable indicators (I) gathered
by ad-hoc questionnaire.
143
Both components of the integrated model consist of structural equations (i.e. cause-and-effect
relationships) that are represented by solid arrows, and measurement equations (i.e.
relationships between observable indicators and the underlying latent variables) that are
represented by dashed arrows (refer to Chapter 4 for more details about Structural Equation
Modelling (SEM)).
The structural equations, on the one hand, show the causal relationships that govern the
choice behaviour. In the latent variable model, for example, the structural equations link the
observable variable (X) to the latent variable (Z), for example:
Z= F(X; γ) + η and η ~ D (0, Ση) (7-3)
where:
Z: latent (unobservable) variables
X: observed variables
γ: unknown parameters
η: random error term
F: linear-in-parameters function
D: generic distribution
Further, for the RP choice model, structural equations link the observable variables (X) and
latent variables (Z) to the random utility (U), for example:
U= V(X, Z; β) + ε and ε ~ D (0, Σε) (7-4)
where:
U: random utility in terms of systematic utility and random error
V: systematic utility in terms of observable and latent variables
X: observed variables
Z: latent (unobservable) variables
β: unknown parameters
ε: random error term
V: linear-in-parameters function
D: generic distribution
144
On the other hand, the measurement equations link the unobservable utility (U) and latent
variable (Z) to their observable indicators (y) and (I), respectively, for example:
I= G(X, Z; α) + ν and ν ~ D (0, Σν) (7-5)
where:
I: Indicators of Z
X: observed variables
Z: latent (unobservable) variables
α: unknown parameters
ν: random error term
G: linear-in-parameters function
D: generic distribution
Finally, the choice model expresses individual choices, based on utility maximization, in
terms of modal utilities as follows:
otherwise0,
max,1
innim
im
UUify (7-6)
In general, the integrated model consists of two parts as shown by equations (7-3) to (7-6).
The latent variable model is described by equations (7-3) and (7-5), whereas the RP choice
model is described by equations (7-4) and (7-6). The combined model deals with individual’s
utility (U) for each travel alternative given the observable outcome (y), while explicitly
accounting for the behavioural process underlying the formation of the latent variables (Z)
using their observable indicators (I).
As mentioned earlier in Chapter 4, the indicators do not have causal relationships that
influence 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 (Walker 2001).
In light of the aforementioned, a hybrid discrete choice model that integrates psychological
factors (personal attitude, habit formation, and affective meaning) along with facilitating
conditions is developed to study the revealed mode choice responses. Psychological factors
are accommodated in the model through latent variables which are expressed as a function of
socioeconomic and personal attributes. Given that psychological factors are not directly
145
observed, latent variables are considered to have a distribution, rather than fixed values.
Mode choice probabilities are then modelled such that latent variables are considered as
explanatory variables in the models. Such treatment leaves the likelihood function of a Logit
mode choice model in a non-closed form expression where simulation estimation is required
(Tudela et al. 2011).
Accordingly, individual traveller’s utility (U) of choosing mode (m) is defined by the
following utility function, which combines both observed and latent variables:
N ..., 1,n m,;)()()( mhhhaaapppmmm HAPxU (7-7)
In the previous utility function, xm indicates a vector of observed variables (personal socio-
demographic attributes and modal level of service attributes). βm is the vector of utility
coefficients associated with observed variables, whereas P, A, and H are vectors of utility
coefficients associated with the following latent variables: personal attitude, affective
meaning, and habit formation, respectively. The latent variables are represented by random
coefficients across the population. In these random coefficients, ηp, ηa, and ηh, indicate mean
values of the corresponding coefficients and εp θp, εa θa, and εh θh, represent variances (θp, θa,
and θh) multiplied by standard normal variables (εp, εa, and εh). The error term εm represents a
random error component to capture the unobserved and random component of the utility
function of the corresponding alternative.
The following functions consider the latent factors (personal attitude, affective meaning, and
habit formation) as endogenous variables and quantify them in terms of observed variables to
allow for capturing their variability across the population.
ppp zP (7-8)
aaa zA (7-9)
hhhzH (7-10)
In the previous functions, zp, za, and zh indicate vectors of observed variables (measures of
personal attitude, affective meaning, and habit formation), whereas γp, γa, and γh indicate their
146
corresponding coefficients. It is considered that the random components πp, πa, and πh are
normal error terms with zero means and τp, τa, and τh variances, respectively.
Since the latent variables are assumed to be random in nature, incorporating them into the
mode choice utility leaves the likelihood functions as non-closed form as shown below.
Hence, the simulation likelihood technique is used for model estimation (Habib et al. 2010).
D
d
h
hh
ha
aa
ap
pp
p
N
n
ndhhdhaadappdpnn
mdhhdhaadappdpmm
m
zHzAzP
HAPx
HAPx
DL
1
1
111
))()()(exp(
))()()(exp(
1
(7-11)
In the previous likelihood function, D indicates the total number of iterations used in
simulation estimation where subscript d refers to individual iterations and ξmd corresponds to
the specific constant related to the mth
alternative. is a standard normal probability density
function. Values of certain parameters were restricted to ensure model identification and
reduce the number of estimated parameters. In particular, variances τ/, τ
//, and τ
/// are
restricted to unit value. The RP mode choice model with latent variables was estimated using
a GAUSS code for simulated likelihood function, making use of the Broyden-Fletcher-
Goldfarb-Shanno (BFGS) gradient search algorithm. In order to ensure stable parameter
estimates, Halton sequence of 1000 iterations is used for generating random numbers (Habib
et al. 2011).
7.6.2 Empirical Analysis
After a series of specification tests, the final hybrid RP choice models with latent variables
are developed and reported in Table 7-3. The developed models integrate cost, level of
service, socioeconomic attributes together with psychological information to explain mode
choice. As shown, Table 7-3 presents two alternative models. Model 1 presents the full model
of commuting mode choice with latent variables, considering personal attitude, affective
meaning, and habit formation as unobservable factors. On the other hand, model 2 presents a
reduced model of commuting mode choice with latent habit formation.
Clearly, it is difficult to compare the developed models with latent variable(s) to the
traditional mode choice model presented earlier in Table 7-2. Given that the likelihood
147
functions are different in each case, directly comparing the likelihood values across the
models is illogical. However, to calculate the loglikelihood function of the choice models
with latent variable(s), the partial information extraction method was used where information
is extracted using only the structure model while dropping the measurement model. That is
exactly the same procedure used for forecasting given that the indicators of latent variables
will not be available in the future (Walker 2001).
By examining the developed models with latent variable(s), it can be observed that the value
of some coefficients as well as their level of significance have been changed compared to
those obtained in the traditional mode choice model. It can be said that incorporating
psychological factors in the model allows for the detection of the real role of some of the
variables on the choice process. For instance, the cost parameter gets smaller and less
significant, whereas the in-vehicle travel and waiting time coefficients almost remained
unchanged. In contrast, the auto ownership coefficient increased in terms of magnitude and
significance.
In general, the presented model has a good fit and explanatory power given its rho-squared
value of 0.53, although t-test values are below the significance level at 95% confidence
interval for some parameters. Negative signs are associated with travel cost, trip length, and
different trip time components. However, cost has low explanatory power given its low
parameter value and level of significance. Further, a similar trend was found for the different
trip time components (e.g. in-vehicle, waiting, and walking time) to that observed in the RP
mode choice model. As for the effect of transit technology, it is clearly shown that transit
users (especially those with auto access) have negative perception for the streetcar and bus
options compared to the subway option. As for gender, females are more likely to take public
transit and less likely to cycle compared to males. Furthermore, younger commuters (18 to 35
years) have high potential to walk. On the other hand, it seems that once someone has got a
car, he/she will use it, given the positive sign associated with auto ownership for the car
driver option. With respect to the role of the psychological variables, personal attitude is
associated with positive sign for both the auto and the transit options (being stronger towards
public transit). On the other hand, an opposite relationship is realized for emotional response.
As for habit, the results show strong and positive habit formation towards the car while being
negative for the other options. In general, it seems that car choice decision is determined
according to emotions and habits, whereas selecting transit is more based on attitude.
148
Table 7-3 RP Mode Choice with Latent Variables
RP Mode Choice Model with Latent Variables Model 1
(Joint RP with
Latent Variables)
Model 2
(Joint RP with
Latent Habit)
Loglikelihood of Joint RP Mode Choice and Latent Variable(s) Model -11990.16 -1862.95
Loglikelihood of RP Mode Choice Only -856.74 -1056.34
Loglikelihood of Null Model -1748.23 -1748.23
Rho-Squared Value 0.51 0.40
Variable Mode Parameter t test Parameter t test
Alternative Specific Constant Car Driver -1.2860 -0.464 -2.5914 -3.073
Shared Ride (Car Passenger/Carpool) -3.2550 -1.186 -5.2731 -6.104
Transit Rider (Car Access) 0.9278 0.393 -0.4734 -0.599
Transit Rider (NMT Access) 2.9980 1.288 1.5579 2.258
Cycle -0.9937 -2.033 -0.9686 -2.015
Walk (Base) 0 --- 0 ---
Travel Cost All Motorized Modes -0.0657 -1.006 -0.0415 -0.463
In-Vehicle Travel Time All Motorized Modes -0.0494 -4.713 -0.0453 -3.273
Trip Length All NMT Modes -0.1637 -5.004 -0.1883 -4.460
Waiting Time Transit Rider (Car Access) -0.0745 -1.785 -0.0957 -1.759
Transit Rider (NMT Access) -0.0967 -6.277 -0.0944 -2.583
(Access + Egress) Time Transit Rider (Car Access) -0.0015 -0.272
Transit Rider (NMT Access) -0.0109 -2.231 -0.0130 -2.839
Number of Vehicles in Household Car Driver 0.2231 1.819 0.2878 2.096
Free Parking Availability Car Driver 1.2908 4.247
Number of People in Household Shared Ride (Car Passenger/Carpool) 0.3988 4.046 0.3735 3.426
Gender: Female Transit Rider (Car Access) 1.1118 2.166 1.1115 1.994
Transit Rider (NMT Access) 0.5214 1.333 0.6361 1.577
Cycle -1.2517 -1.907 -1.2347 -1.809
Transit Technology: Transit Rider (Car Access) -1.7816 -3.556 -1.3736 -2.395
149
Street Transit (Bus/Streetcar) Transit Rider (NMT Access) -0.9531 -2.814 -0.6184 -1.594
Age: 18 - 35 years old Walk 0.8250 1.904 0.9714 2.311
Latent Personal Attitude Car Driver & Shared Ride 0.0751 0.330
Transit Rider (All Access modes) 0.1998 1.013
Latent Affective Meaning Car Driver & Shared Ride -0.1175 -0.157
(Emotional Response) Transit Rider (All Access modes) -2.0909 -2.723
Latent Habit Formation Car Driver & Shared Ride 1.0362 2.388 5.7693 5.593
Transit Rider (All Access modes) -1.7731 -2.728 -0.9274 -1.322
All NMT Modes -0.6557 -0.752 -1.2212 -1.493
Structural Model
Personal Attitude: Constant Car Driver & Shared Ride 3.7647 1.764
Transit Rider (All Access modes) 4.1048 1.179
Gender: Female Car Driver & Shared Ride 0.5970 1.731
Transit Rider (All Access modes) 0.4440 0.710
Logarithm of Age Car Driver & Shared Ride 1.7096 2.922
Transit Rider (All Access modes) 1.6107 1.703
Affective Meaning: Constant All Modes -0.1223 -0.701
Gender: Female All Modes -0.1209 -4.584
Logarithm of Age All Modes 0.3902 8.283
Habit Formation: Constant Car Driver & Shared Ride 0.8156 1.829 0.5707 2.962
Transit Rider (All Access modes) -0.0352 -0.488 1.2638 4.296
All NMT Modes 0.0310 0.259 0.4959 0.647
Gender: Female Car Driver & Shared Ride 0.7303 1.539 -0.0213 -0.693
Transit Rider (All Access modes) 0.0741 0.984 0.0173 0.336
All NMT Modes -0.0123 -0.096 -0.0580 -0.433
Logarithm of Age Car Driver & Shared Ride 0.4034 0.667 -0.0097 -0.191
Transit Rider (All Access modes) 0.0669 0.687 -0.1657 -2.089
All NMT Modes -0.0405 -0.235 -0.0259 -0.122
150
Measurement Model
Personal Attitude Car Driver & Shared Ride 1.5763 43.799
Transit Rider (All Access modes) 1.5835 24.319
Affective Meaning: Activation All Modes -0.0701 -2.518
Potential All Modes -0.3136 -11.939
Control All Modes -0.3580 -11.742
Evaluation All Modes -0.0863 -2.827
Latent Habit Formation Car Driver & Shared Ride -0.5617 -11.006 -1.4516 -27.507
Transit Rider (All Access modes) -0.8779 -15.383 -1.4910 -18.721
All NMT Modes -1.7613 -10.050 -1.5109 -8.306
151
7.7 Modelling Commuting Work Trip Mode Shift
Costumer Oriented Transit Service (COTS) elements and emerging technologies, such as
passenger information systems, ITS technologies and advanced transit modes (e.g. LRT and
BRT) have attributes affecting the perceptions of travellers, which are usually overlooked in
traditional mode choice models. This is a critical issue for transit service planning where
improving service to facilitate modal shift in favour of transit is targeted. In this section,
separate mode shift models are estimated for different groups of commuters. Disaggregate
Multinomial Logit (MNL) models are estimated to model mode shift in response to transit
service improvements using RP, SP, and psychological data. The developed models estimate
the probability that an individual trip-maker will either stay with his/her current choice or
shift to another option.
7.7.1 Modelling Mode Shift for Car Users
In an attempt to investigate the effects of different transit investments that usually target auto
users, separate mode shift models for car drivers and shared ride users (car passengers and
carpoolers) are estimated and analyzed. The estimated models are sensitive to Level of
Service (LOS) attributes of the competing options as well as socio-demographic and
behavioural information of the decision makers.
Several mode shift models are estimated and compared to one another, in terms of model
specification and explanatory power, till reaching the final models. In general, each of the
developed models has the following three alternatives: stay with the current mode, shift to
public transit, or shift to another option indicated by the respondent.
Table 7-4 shows the estimation results of three mode shift models for car drivers. Model 1 is
a restricted model including only SP data, model 2 considers both RP and SP information,
whilst model 3 is a joint RP/SP model with latent habit formation. Given the presented rho-
squared values of 0.37, 0.41, and 0.49 for models 1, 2, and 3, respectively, it should be clear
that combining RP and SP information together with latent habit enhanced the goodness of fit
and explanatory power of the final model (model 3). Moreover, t-test values are above the
significance level at 95% confidence interval, except for some parameters. However, those
parameters are kept in the models because the corresponding variables provide considerable
insights into the behavioural process.
152
The examination of traditional variables shows that travel cost, parking cost, and different
trip time components have correct (negative) signs that match expectations. By examining the
relative importance of different cost components, it is found that motorists valuate parking
cost more than travel cost. Further, the primary investigation of the model parameters shows
that traditional attributes (e.g. travel cost and time) are of lower importance to mode
switching behaviour compared to other transit design factors and COTS technologies (e.g.
crowding level, number of transfers, schedule delay, and transit technology). The modelling
results also show that the coefficient of waiting time is around 2.0 times higher than that of
the in-vehicle travel time, which is close to ratios reported in similar studies and common
practice. As for the effect of transit technology, it is clearly shown that travellers are more
likely to shift to rapid (e.g. subway) and semi-rapid (e.g. LRT and BRT) transit options
compared to street transit ones (e.g. streetcar and bus). In addition, of the semi-rapid transit
options, the results show a slightly higher preference for LRT than BRT. Such finding is
considered evidence for the superiority of the rail-based modes to the rubber-tyred ones,
which is referred to as the “rail effect”.
Further, it is found that the tendency to mode shift decreases with higher crowding levels,
number of transfers, and schedule delays. Moreover, younger commuters (18 to 35 years)
have relatively high propensity to switch to public transit. However, car drivers who decided
not to shift to transit are more confident about their decisions. Furthermore, car drivers who
are unfamiliar with the transit service are unlikely to switch to public transit even though a
better transit service is offered to them. As expected, auto ownership is associated with a
positive sign with car use in the model and acts as a barrier to mode shift. In other words,
once someone makes the initial investment to own a car, (s)he is more likely to use it.
Nevertheless, residents of the City of Toronto, where a multimodal transit system and
supportive land use make transit more competitive to auto travel, are more likely to break the
mode shift barriers and switch to public transit.
Interestingly, it is clearly observed that the values and levels of significance of some
coefficients have been changed in the full model (model 3). Clearly, the incorporation of the
full information (RP, SP, and latent habit formation) allowed for the detection of the real role
of some of the variables in the mode shift process. For instance, the combined access and
egress (walking time) parameter gets smaller and less significant, likewise the in-vehicle
travel time and parking cost, whereas the magnitude and significance of other LOS
153
coefficients increased significantly. As for the effect of behavioural factors, habit formation
has also shown a greater importance to mode shift modelling than traditional attributes such
as travel cost and time. The results show strong and positive habit formation towards the car
while being negative for transit.
Similarly, Table 7-5 shows the estimation results of three mode shift models for shared ride
users (car passengers and carpoolers). Model 1 is a restricted model including only SP data,
model 2 considers both RP and SP information, whilst model 3 is a joint RP/SP model with
latent habit formation. Obviously, the inclusion of RP and SP information along with latent
habit has improved the goodness of fit and explanatory power of the final model (model 3).
The traditional variables, namely travel cost and in-vehicle time were found to have very low
parameter and t-test values. Moreover, several transit design factors and technologies (e.g.
crowding level, schedule delay, and number of transfers) are found to be insignificant and
thus removed from the models. As opposed to the previous models, the in-vehicle and
walking times are found to be more relevant to mode shift decisions than waiting time. The
modelling results also show that travellers’ perception of waiting time is almost the same as
their perception of in-vehicle travel time. Further, transit technology is found to have a
similar behavioural effect to that it has on car drivers. Shared ride users are more likely to
shift to rapid and semi-rapid transit options compared to street transit ones. In addition,
evidence for the rail effect is realized. Furthermore, it is found that elder commuters (36 to 50
years) have relatively high propensity to switch to public transit. In addition, residents of the
City of Toronto are more likely to switch to public transit. However, shared ride users who
decided not to shift to transit are more confident about their decisions. Moreover, shared ride
users who are unfamiliar with the transit service are unlikely to switch to public transit even
though a better transit service is offered to them. As for the role of habit formation, the results
show strong and positive habit formation towards the car while being negative for transit,
similar to what was found before.
In light of the above, the modelling results unravel the reason why conventional models,
lacking detailed COTS elements and psychological information, tend to overestimate mode
switch to public transit. The inclusion of various transit design factors and technologies in the
developed models enrich the understanding of how trip makers make tradeoffs among
different LOS attributes. In general, the developed mode switching models are more desirable
154
for evaluating transit investments that usually target auto users. Further, the policy
implications of the previous findings are very important especially when targeting different
market segments (e.g. car users) to increase transit ridership. Such findings provide transit
planners with detailed information about what COTS elements to change to attract a specific
group of users. Moreover, by noticing that the confidence of shred ride users who decided not
to shift to transit is less than that of car drivers, it seems that promoting public transit among
shared ride users is a promising strategy. On the other hand, the strong car use habit
formation should be kept in mind. While enhancing transit service performance is essential to
increase modal shift, transport policies should also focus on breaking the strong habits
associated with the car (e.g. increasing parking cost).
155
Table 7-4 Mode Shift Models for Car Drivers
Mode Shift Model for Car Drivers Model 1
(SP Only Model)
Model 2
(Joint RP/SP Model)
Model 3
(Joint RP/SP Model
with Latent Habit)
Loglikelihood of Joint RP/SP Mode Shift
and Latent Variable Models
-2732.58
Loglikelihood of Mode Shift Model Only -3174.16 -2983.87
-2589.22
Loglikelihood of Null Model -5049.2221 -5049.2221
-5049.2221
Rho-Squared Value 0.37 0.41 0.49
Variable Mode Parameter t-test Parameter t-test Parameter t-test
Alternative Specific Constant Stay with Current Mode 3.4473 25.495 2.5637 13.331 0.4572 2.956
Shift to Public Transit 2.4987 14.501 2.3130 12.833 2.7035 14.611
Shift to Other (Base) 0 --- 0 --- 0 ---
Travel Cost All Motorized Modes -0.0025 -0.500 -0.0047 -0.925 -0.0080 -1.907
Parking Cost Stay with Current Mode -0.0221 -2.824 -0.0190 -2.349 -0.0179 -2.745
In-Vehicle Travel Time All Motorized Modes -0.0064 -6.689 -0.0066 -6.585 -0.0063 -7.869
Waiting Time Shift to Public Transit -0.0105 -5.257 -0.0081 -3.985 -0.0113 -5.838
(Access + Egress) Time Shift to Public Transit -0.0059 -0.630 -0.0057 -0.584 -0.0052 -0.432
Transit Technology: BRT Shift to Public Transit 0.1872 1.895 0.1931 1.886 0.3168 2.600
LRT Shift to Public Transit 0.2497 2.662 0.2623 2.246 0.3233 2.462
Subway Shift to Public Transit 0.2529 2.071 0.2675 2.104 0.4469 3.171
Crowding Level:
Wait for next vehicle Shift to Public Transit -0.3514 -4.181 -0.3408 -3.927 -0.4265 -4.122
Number of Transfers:
2 or more Shift to Public Transit -0.2333 -2.808 -0.2384 -2.764 -0.2716 -2.445
Schedule Delay: Late Shift to Public Transit -0.3349 -4.096 -0.3294 -3.896 -0.4135 -4.021
Age: 18 - 35 years old Shift to Public Transit 0 --- 0.2374 3.230 0.1103 1.774
No. of Vehicles in Household Stay with Current Mode 0 --- 0.1431 3.201 0.1357 3.790
Frequency of transit usage:
< once a month or never Stay with Current Mode 0 --- 0.3493 3.322 0.2990 4.084
156
Willingness to comply:
Very Strong Stay with Current Mode 0 --- 1.4240 15.649 1.6227 22.514
Living and Working in the
City of Toronto Stay with Current Mode 0 --- -0.2674 -3.539 0.0020 0.035
Latent Habit Formation Stay with Current Mode 2.5988 26.315
Shift to Public Transit -1.7606 -15.969
Shift to Other 0 ---
Structural Model
Habit Formation: Constant Stay with Current Mode 1.0422 56.262
Shift to Public Transit 0.2094 4.087
Shift to Other 0 ---
Gender: Female Stay with Current Mode -0.0268 -8.888
Shift to Public Transit -0.0556 -7.143
Shift to Other 0 ---
Logarithm of Age Stay with Current Mode -0.1246 -25.292
Shift to Public Transit 0.0422 3.069
Shift to Other 0 ---
Measurement Model
Latent Habit Formation Stay with Current Mode -1.7670 -263.906
Shift to Public Transit -1.5352 -113.227
Shift to Other 0 ---
157
Table 7-5 Mode Shift Models for Car Passengers and Carpoolers
Mode Shift Model for Car Passengers and Carpoolers Model 1
(SP Only Model)
Model 2
(Joint RP/SP Model)
Model 3
(Joint RP/SP Model
with Latent Habit)
Loglikelihood of Joint RP/SP Mode Shift
and Latent Variable Models
-259.26
Loglikelihood of Mode Shift Model Only -415.27 -379.55
-313.42
Loglikelihood of Null Model -580.06729 -580.06729
-580.06729
Rho-Squared Value 0.28 0.35 0.46
Variable Mode Parameter t-test Parameter t-test Parameter t-test
Alternative Specific Constant Stay with Current Mode 2.9316 12.166 2.5488 8.160 0.7547 1.994
Shift to Public Transit 2.8503 7.912 2.0135 4.180 2.3949 4.949
Shift to Other (Base) 0 --- 0 --- 0 ---
Travel Cost All Motorized Modes -0.0038 -0.401 -0.0098 -0.891 -0.0134 -1.354
In-Vehicle Travel Time All Motorized Modes -0.0105 -3.783 -0.0079 -2.295 -0.0120 -4.420
Waiting Time Shift to Public Transit -0.0104 -1.405 -0.0053 -0.688 -0.0116 -1.283
(Access + Egress) Time Shift to Public Transit -0.0534 -2.362 -0.0523 -2.171 -0.0704 -2.061
Transit Technology: BRT Shift to Public Transit 0 --- 0.6018 2.404 0.8032 2.992
LRT Shift to Public Transit 0 --- 0.6187 1.852 1.0001 3.339
Subway Shift to Public Transit 0 --- 0.6647 1.911 1.0122 2.363
Age: 36 - 50 years old Shift to Public Transit 0 --- 0.5083 2.351 0.6818 3.739
Frequency of transit usage:
< once a month or never Stay with Current Mode 0 --- 0.4866 2.271 0.3021 1.591
Willingness to comply:
Very Strong Stay with Current Mode 0 --- 1.4977 5.572 1.1981 4.984
Living and Working in the
City of Toronto Stay with Current Mode 0 --- -0.7631 -3.728 -0.2607 -1.545
Latent Habit Formation Stay with Current Mode 3.0508 8.243
Shift to Public Transit -1.0655 -3.976
Shift to Other 0 ---
158
Structural Model
Habit Formation: Constant Stay with Current Mode 0.8254 17.456
Shift to Public Transit 1.7397 14.539
Shift to Other 0 ---
Gender: Female Stay with Current Mode -0.2267 -22.598
Shift to Public Transit 0.1091 5.011
Shift to Other 0 ---
Logarithm of Age Stay with Current Mode -0.0150 -1.164
Shift to Public Transit -0.3929 -11.831
Shift to Other 0 ---
Measurement Model
Latent Habit Formation Stay with Current Mode -1.9823 -95.766
Shift to Public Transit -1.6843 -44.342
Shift to Other 0 ---
159
7.7.2 Modelling Mode Shift for Transit Riders
In an attempt to investigate the effects of different transit investments on current transit users,
mode shift models for transit riders are estimated and analyzed. The developed models
provide better understanding of the relative importance of different transit design factors and
technologies, as well as the way they influence mode shift decisions. The estimated models
are sensitive to different transit Level of Service (LOS) attributes, behavioural information, as
well as socio-demographic information of the decision makers.
Several mode shift models are estimated and compared to one another, in terms of model
specification and explanatory power, till reaching the final model. In general, each of the
developed models has the following two alternatives: stay with the current mode, or shift to
other option indicated by the respondent.
Table 7-6 shows the estimation results of two mode shift model for transit riders. Model 1
considers SP information, and model 2 is an SP model with latent habit. The RP data did not
show a good significance and therefore was discarded in all models. Given the presented
rho-squared values of 0.25, and 0.34 for models 1 and 2, respectively, it is clear that
combining SP information together with latent habit enhanced the goodness of fit and
explanatory power of the final model (model 2). Moreover, t-test values are above the
significance level at 95% confidence interval, except for some parameters that are retained in
the models to provide considerable insights into the behavioural process.
The primary investigation of model parameters shows that traditional attributes (e.g. in-
vehicle travel time and walking time) are of minor importance to mode switching behaviour
compared to other transit design factors and COTS elements (e.g. crowding level, schedule
delay, and transit technology). Interestingly, waiting time coefficients are much higher than
the in-vehicle travel time. As for the effect of transit technology, it is clearly shown that
travellers have higher preference to rapid transit options (e.g. subway) compared to semi-
rapid (e.g. LRT and BRT) and street transit ones (e.g. streetcar and bus). In addition,
evidence for the superiority of the rail-based modes to the rubber-tyred ones, which is
referred to as the “rail effect”, is observed. Further, it is found that the tendency to mode shift
from transit increases with higher crowding levels, and schedule delays. In addition, the
results show a negative habit formation associated with staying as a transit user.
160
The policy implications of the previous findings put the loyalty of transit riders in question.
In addition, the low habit formation associated with public transit shows that transit users are
not attached and might switch from public transit if crowding levels and schedule delays
increase. Transit agencies should be aware while planning their policies and do not take
transit ridership as for granted.
161
Table 7-6 Mode Shift Model for Transit Riders (All Access Modes)
Mode Shift Model for Transit Riders (All Access Modes)
Model 1
(SP Model)
Model 2
(Joint SP Model with Latent Habit)
Loglikelihood of Joint RP/SP Mode Shift and Latent Variable Models -489.60
Loglikelihood of Mode Shift Model Only -841.10 -746.47
Loglikelihood of Null Model -1122.8984 -1122.8984
Rho-Squared Value 0.25 0.34
Variable Mode Parameter t-test Parameter t-test
Alternative Specific Constant Stay with Current Mode 2.9789 6.169 3.4610 10.020
Shift to Other (Base) 0 --- 0 ---
Transit Fare Stay with Current Mode -0.8716 -2.250 -0.8261 -4.140
In-Vehicle Travel Time Stay with Current Mode -0.0051 -2.325 -0.0068 -4.752
Waiting Time Stay with Current Mode -0.5571 -2.043 -0.5912 -2.791
(Access + Egress) Time Stay with Current Mode -0.0045 -0.308 -0.0047 -0.309
Transit Technology: Subway Stay with Current Mode 0.4627 2.291 0.4585 2.048
Crowding Level: Wait for next vehicle Stay with Current Mode -0.5389 -4.270 -0.5531 -3.776
Schedule Delay: Late Stay with Current Mode -0.2391 -1.841 -0.2664 -1.694
Latent Habit Formation Stay with Current Mode -1.1317 -12.423
Shift to Other 0 ---
Structural Model
Habit Formation: Constant Stay with Current Mode 0.3295 16.568
Shift to Other 0 ---
Gender: Female Stay with Current Mode 0.0836 25.133
Shift to Other 0 ---
Logarithm of Age Stay with Current Mode 0.0958 17.379
Shift to Other 0 ---
Measurement Model
Latent Habit Formation Stay with Current Mode -2.0811 -269.121
Shift to Other 0 ---
162
7.7.3 Modelling Mode Shift for Non-Motorized Transport (NMT) Users
This section attempts to investigate the effects of different transit investments on Non-
Motorized Transport (NMT) users, a mode shift model for NMT users is estimated and
analyzed. The developed model allows for better understanding of the relative importance of
different transit design factors and technologies, as well as the way they influence mode shift
decisions. The estimated model is sensitive to different transit Level of Service (LOS)
attributes as well as socio-demographic information of the decision makers. In general, the
developed model has the following three alternatives: stay with the current mode, shift to
public transit, or shift to other option indicated by the respondent.
Table 7-7 shows the estimation results of two mode shift models for NMT users. The
presented models consider both RP and SP information. Model 1 is a restricted model
including only SP data, whilst model 2 considers both RP and SP information. The
psychological data did not show a good significance and therefore was discarded in all
models. Given the presented rho-squared values of 0.66, and 0.69 for models 1, and 2
respectively, it is clear that the inclusions of RP together with SP information enhanced the
goodness of fit and explanatory power of the final model (model 2).
The primary investigation of model parameters shows that travel cost, as well as different trip
time components have the correct sign. However, waiting and walking time are of minor
importance to mode switching behaviour compared to other transit design factors and
technologies (e.g. crowding level). Further, the in-vehicle time was found to be more relevant
to mode shift decisions than waiting and walking (access and egress) times. The modelling
results also show that waiting time coefficients are much lower than the in-vehicle travel
time.
Interestingly, NMT users did not show any preferences for different transit technologies.
However, crowding level is found to have a very high impact to mode shift towards public
transit especially when seats are available. Moreover, younger commuters (18 to 35 years)
have high potential to walk/cycle rather than switching to public transit. Furthermore, it
seems that those who use active modes of transport (walk and bike) are loyal to their chosen
option given the high coefficient value associated with the willingness to comply attribute.
163
It is also important to notice that the availability of park-and-ride facilities as well as real time
and schedule information were not found to be relevant to mode shift to local transit. Clearly,
the modelling results provided in this section allow for a better understanding of the relative
importance of different transit design factors and technologies, as well as the way they
influence mode shift decisions.
164
Table 7-7 Mode Shift Model for Non-Motorized Transport Users
Mode Shift Model for
Non-Motorized Transport Users
Model 1
(SP Only Model)
Model 2
(Joint RP/SP Model)
Loglikelihood of Mode Shift Model Only -152.65 -142.55
Loglikelihood of Null Model -454.82549 -454.82549
Rho-Squared Value 0.66 0.69
Variable Mode Parameter t-test Parameter t-test
Alternative Specific Constant Stay with Current Mode 4.9455 11.344 4.0550 8.989
Shift to Public Transit 4.0600 4.025 3.6642 2.261
Shift to Other (Base) 0 --- 0 ---
Transit Fare Shift to Public Transit -0.4453 -2.623 -0.3492 -2.063
Distance Stay with Current Mode -0.4879 -5.010 -0.3743 -4.123
In-Vehicle Travel Time Shift to Public Transit -0.7253 -1.673 -0.3184 -0.688
Waiting Time Shift to Public Transit -0.0012 -0.104 -0.0736 -0.056
(Access + Egress) Time Shift to Public Transit -0.0530 -1.255 -0.0432 -1.027
Crowding Level: Seats Available Shift to Public Transit 0.6747 1.819 0.8266 2.197
Age: 18 - 35 years old Shift to Public Transit 0 --- -0.7428 -2.033
Willingness to comply: Very Strong Stay with Current Mode 0 --- 1.5978 3.689
165
7.8 Models Validation and Policy Analysis
The previous modelling efforts have clearly shown that attracting and retaining transit
ridership depend largely on the service performance of the transit system, and the behavioural
characteristics of the travellers. Changes in such attributes are likely to lead to changes in
people’s mode choice preferences and transport mode switching over time. This section is
intended to investigate the forecasting performance and sensitivity of the developed mode
choice/modal shift models to changes in transit service design and behavioural attributes.
The data used in this analysis belongs to an independent subset, of the collected dataset, that
was not used in the models estimation process given that it presented excess in some quotas
(mainly car drivers) within the boundaries of the study area (Toronto CMA). The subset
consists of 239 car drivers who have an observed mode shift of 183 car drivers and 56 transit
riders, as stated in the SP experiment. In addition, a similar investigation is performed on an
expanded subset of 1407 observations after considering six stated choice scenarios for each
respondent (239 x 6= 1434) and excluding 27 observations that shifted to “other” option in
the SP experiment.
In an attempt to quantify the transit ridership overestimation and the forecasting performance
of the different types of models developed, the models are used to predict mode shift to
transit using the independent subsets of 239 car drivers and the expanded subset of 1407 car
drivers, as shown in Table 7-8 and Table 7-9, respectively.
Table 7-8 and Table 7-9 present the observed mode choices and the predicted mode shift, as
well as the overestimation percentage and Forecasting Performance Measure (FPM) of each
model, where FPM= ∑ [ ) (7-12)
Given that the FPM depends on the difference between the predicted (Pm) and observed (Om)
trips for each mode (m), the smaller the FPM, the smaller the aggregate forecasting errors of
the corresponding model, and subsequently, the better the forecasting performance of the
model (Habib et al. 2012).
166
Table 7-8 Forecasting Performance using a Subset of 239 Car Drivers
Model Type Observed Predicted Mode Shift Difference Overestimation FPM
Mode
Choice
Mode
Shift
Mode
Shift Predicted - Observed Percentage
RP Mode Choice
Auto Driver 239 183 108.02 -74.98 -40.97% 1.960
Transit Rider 0 56 130.98 74.98 133.89%
RP Mode Choice with Latent Habit
Auto Driver 239 183 147.73 -35.27 -19.27% 0.434
Transit Rider 0 56 91.27 35.27 62.98%
SP Mode Shift
Auto Driver 239 183 177.38 -5.62 -3.07% 0.011
Transit Rider 0 56 61.62 5.62 10.04%
Joint RP/SP Mode Shift
Auto Driver 239 183 180.09 -2.91 -1.59% 0.003
Transit Rider 0 56 58.91 2.91 5.20%
Joint RP/SP Mode Shift with Latent Habit
Auto Driver 239 183 146.17 -36.83 -20.13% 0.473
Transit Rider 0 56 92.83 36.83 65.78%
167
Table 7-9 Forecasting Performance using Expanded Subset of 1407 Car Drivers
Model Type Observed Predicted Mode Shift Difference Overestimation FPM
Mode
Choice
Mode
Shift
Mode
Shift Predicted - Observed Percentage
RP Mode Choice
Auto Driver 1407 1005 636.72 -368.28 -36.64% 0.974
Transit Rider 0 402 770.28 368.28 91.61%
RP Mode Choice with Latent Habit
Auto Driver 1407 1005 849.07 -155.93 -15.52% 0.175
Transit Rider 0 402 557.93 155.93 38.79%
SP Mode Shift
Auto Driver 1407 1005 1040.33 35.33 3.52% 0.009
Transit Rider 0 402 366.67 -35.33 -8.79%
Joint RP/SP Mode Shift
Auto Driver 1407 1005 1057.12 52.12 5.19% 0.019
Transit Rider 0 402 349.88 -52.12 -12.96%
Joint RP/SP Mode Shift with Latent Habit
Auto Driver 1407 1005 865.46 -139.54 -13.88% 0.140
Transit Rider 0 402 541.54 139.54 34.71%
168
In general, the developed modes can be classified into three groups in terms of their transit
ridership overestimation and forecasting performance, namely traditional mode choice model,
models with latent habit, and mode shift models without latent habit.
Examining the FPM of the developed models showed that the models with latent habit do not
perform as well as the SP and the joint RP/SP mode shift models. However, all of the four
models are better than the traditional RP mode choice model which showed the poorest
forecasting performance.
Further, Figure 7-5 and Figure 7-6 present estimates for transit ridership and car mode split,
respectively. In general, the models with latent habit showed a better performance than the
traditional mode choice model, while being outperformed by the mode shift models without
latent habit. The previous trend is observed for both subsets (239 and 1407 observations),
although the tendency to overestimate transit ridership decreases with increasing the number
of observations. A closer perusal of the two figures shows that the traditional RP mode choice
model, on the one hand, has a high tendency to over-predict transit ridership on the expense
of the car driver option. This may partly be due to the lack of behavioural as well as
Customer Oriented Transit Service (COTS) elements in the traditional model. On the other
hand, both the SP mode shift model and the joint RP/SP mode shift model showed the lowest
transit ridership overestimation. It is also interesting to notice that the models without latent
habit perform better than the models with latent habit. This may be in part due to the
redundancy between latent habit and some traditional attributes that act as indicators for habit
formation in the model (e.g. auto ownership and driving licence holding). The previous
findings reinforce the main claim raised by this research. In particular, the results confirm the
prior hypothesis of this research that RP models tend to overestimate mode shift to transit. It
is clear that the SP data complemented the RP information and resulted in improved
forecasting performance and less transit ridership overestimation.
In light of the above, the developed models provide a better understanding of commuters’
preferences and mode switching behaviour. Given that transit service planning is mainly
concerned with enhancing existing transit routes/lines by altering various LOS attributes (e.g.
accessibility, frequency, trip directness, reliability, crowding levels, and rail vs. bus
attraction), the presented models are more appropriate for transit planners being sensitive to
such elements.
169
Figure 7-5 Transit Ridership Estimation
Figure 7-6 Car Driver Mode Split Estimation
RP ModeChoice
RP ModeChoice withLatent Habit
SP Mode ShiftJoint RP/SPMode Shift
Joint RP/SPMode Shiftwith Latent
Habit
239 Observations 133.89% 62.98% 10.04% 5.20% 65.78%
1407 Observations 91.61% 38.79% -8.79% -12.96% 34.71%
133.89%
62.98%
10.04% 5.20%
65.78%
91.61%
38.79%
-8.79% -12.96%
34.71%
-20.00%
0.00%
20.00%
40.00%
60.00%
80.00%
100.00%
120.00%
140.00%
RP ModeChoice
RP ModeChoice withLatent Habit
SP Mode ShiftJoint RP/SPMode Shift
Joint RP/SPMode Shiftwith Latent
Habit
239 Observations -40.97% -19.27% -3.07% -1.59% -20.13%
1407 Observations -36.64% -15.52% 3.52% 5.19% -13.88%
-40.97%
-19.27%
-3.07% -1.59%
-20.13%
-36.64%
-15.52%
3.52% 5.19%
-13.88%
-50.00%
-40.00%
-30.00%
-20.00%
-10.00%
0.00%
10.00%
170
A general look at the effect of different modal characteristics on attracting commuters to
transit shows that both crowding level and schedule delay have a substantial influence on
mode switching decisions, followed by transit technology and number of transfers. Moreover,
the developed models provide useful information to transit planners on the relative propensity
of different mode users to switch to transit in response to changes in transit LOS attributes. In
particular, the models show that car users are more sensitive to crowding level, schedule
delay, transit technology, and number of transfers; whereas, shared ride users (car passenger
and carpoolers) are sensitive only to transit technology, and NMT users (walk and bike) are
sensitive only to crowding level. On the other hand, the models show that transit riders might
shift away from transit as a result of crowding level, and/or schedule delay increase.
Interestingly, park-and-ride availability and cost as well as schedule and real-time
information provision did not appear to be significant for mode switching to local transit. It
might be argued that the latter factors are more important for short-term route shift rather
than mode shift decisions.
The previous finings have strong policy implications with respect to targeting potential
markets for increasing transit ridership. For example, attracting shared ride users (car
passengers and carpoolers) that have shown, earlier in Section 6.4, to be more willing to
switch to transit if a proper service is provided to them. Knowing the preferences of shared
ride users certainly allows transit planners to decide on what LOS attributes and COTS
elements to alter.
7.9 Chapter Summary
Separate mode shift models for car drivers and shared ride users are estimated and analyzed.
The inclusion of various transit design factors and technologies in the developed models
explains the way trip makers make tradeoffs among different transit Level of Service (LOS)
attributes. In general, the developed mode switching models enrich the transit service design
toolbox for delivering more efficient and attractive services, and therefore, they are more
desirable for transit service planning especially for evaluating transit investments that usually
target auto users.
Interesting findings are observed by analyzing the developed models, leaving transit planners
with important takeaways. Comparing the traditional mode choice model to the mode choice
models with latent variables showed that the inclusion of behavioural factors (especially habit
171
formation) has improved the goodness-of-fit and explanatory power of the estimated models.
Moreover, the reasons why traditional mode choice models tend to over predict transit
ridership were unravelled by revealing the role played by different transit LOS attributes and
their relative importance to mode switching decisions. Such overestimation may partly be due
to the lack of behavioural as well as Customer Oriented Transit Service (COTS) elements in
the traditional model. On the other hand, both the SP mode shift model and the joint RP/SP
mode shift model showed the lowest transit ridership overestimation. It is clear that the SP
data complements the RP information and results in improved forecasting performance and
less transit ridership overestimation.
172
8 CONCLUSIONS AND RECOMMENDATIONS
8.1 Chapter Overview
This chapter starts with a summary of the presented research in Section 8.2. Then, Section 8.3
highlights the main contributions of this thesis. Finally, Section 8.4 provides ideas for future
continuation of this research.
8.2 Research Summary
Increasing modal shift from single occupancy vehicles towards public transit is a desirable
objective for addressing many traffic and environmental problems (Ogilvie et al. 2004;
Vedagiri and Arasan 2009; Hamer 2010). However, an effective implementation of mode
switching strategies requires proper evaluation of the proposed policy on changing travel
behaviour prior to the actual application (Tanadtang et al. 2005; Nurdden et al. 2007;
Vermote and Hens 2009). Traditionally, the classical four-stage demand forecasting model
has been developed to predict the number of trips made within an urban area (Meyer and
Miller 2001). In this context, the standard practice in studying passengers’ choices in terms of
their mode of travel makes use of the mode choice concept (McFadden 1974).
Research into mode choice modelling has shown that considering socioeconomic and
demographic aspects of the decision maker and other factors representing the relative
attractiveness of the available options usually increase the explanatory power of the
developed models (Eriksson et al. 2008). Of the decision maker characteristics, car ownership
and availability are usually considered the major determinants of mode choice (Williams
1978; Barff et al. 1982). On the other hand, travel time and cost play a bigger role in
determining mode choice than other factors characterizing the attractiveness of the competing
modes (Quarmby 1967; Williams 1978).
Over the decades, research has continuously improved mode choice models on an analytical
viewpoint in an effort to make them better explain modal split. While being useful and
insightful, traditional mode choice models often suffer from many problems. Evidence in the
literature shows that traditional choice models fail to accurately forecast modal shift in
response to new improvements in the transit services (Winston 2000; Beimborn et al. 2003;
Flyvbjerg et al. 2005; Quentin and Hong 2005). Such failures are generally attributed to the
lack of tools that can adequately forecast the behaviour of potential transit ridership (Cantillo
et al. 2007; Domarchi et al. 2008). This in turn induces a poor knowledge of the demand for
173
new transit services and a subsequent difficulty in designing an economically sustainable
transit system.
In specific, conventional mode choice models based only on Revealed Preference (RP) data
tend to overestimate the attractiveness of transit for choice users which leads to over
predicting transit ridership (Winston 2000; Beimborn et al. 2003; Flyvbjerg et al. 2005;
Forsey et al. 2011). In addition, such models are criticized for their weak characterization of
several behavioural aspects, contributing in part to their misleading modal shift estimation
(Quentin and Hong 2005; Cantillo et al. 2007; Domarchi et al. 2008). Further, it is often
difficult to accommodate Costumer Oriented Transit Service (COTS) elements and attributes
of emerging systems, such as passenger information systems, ITS technologies that improve
reliability, etc. in conventional mode choice models because detailed information of such
attributes are often missing in traditional household based RP travel survey data. This is a
critical issue in transit service design where improving service to facilitate modal shift
towards transit is targeted. Therefore, the key point towards developing adequate tools to
forecast transit ridership seems to be more the study of changing behaviours and less that of
the choices among alternatives (Cantillo et al. 2007; Behrens and Mistro 2010).
Since the attractiveness of any transit service relies on how the design factors affect peoples’
travel choices, behaviour and subsequently mode switching, this research aims at developing
a better understanding of commuters’ preferences and mode switching behaviour towards
public transit in response to changes in transit service design attributes. As opposed to
traditional RP-based mode choice models, mode shift models are developed using state-of-
the-art methodology of combining Revealed Preference (RP) and Stated Preference (SP)
information. The proposed methodological approach incorporates three main stages. The first
introduces a conceptual framework for modal shift maximized transit route design model that
extends the use of the developed models beyond forecasting transit ridership (demand) to the
operational extent of transit route design (supply). The second stage deals with designing and
implementing a socio-psychometric survey about personal attitudes and habit formation of
Toronto commuters regarding shifting to different transit technologies of varying
characteristics. The third stage focuses on developing econometric choice models of mode-
switching behaviour towards public transit.
174
The developed conceptual framework for modal shift maximized transit route design model
represents a practical transit route design tool that is more desirable for transit planners. The
proposed framework is intended to generate transit route designs that maximize demand
attraction. The framework builds upon and extends the capabilities of the existing
MIcrosimulation Learning-based Approach for TRansit ASsignment (MILATRAS) (Wahba
and Shalaby 2005; Wahba and Shalaby 2009a), to tackle both the route design and mode shift
problems. MILATRAS currently models transit assignment given a fixed set of transit routes
and transit demand (Wahba 2009; Wahba and Shalaby 2009b). The presented framework
adds a mode shift module to MILATRAS in order to find operationally implementable transit
route(s) that can provide alternative design concepts corresponding to different service
requirements. Further, modal shift barriers (e.g. habit formation) are captured in the model by
specifying a threshold or inertia against shifting between modes. Transit demand variability
among both modes and routes is considered at the microscopic level by running the joint
mode shift and route choice models of MILATRAS, allowing for consistency between the
supplied service level and passenger demand (Osman and Shalaby 2010; Idris et al. 2012a).
This thesis describes all elements of the conceptual framework then gives explicit attention to
the development of the mode shift module, while jointly running both components (route
choice and mode shift) of MILATRAS is left for future research.
As a primary step towards learning how mode choice decisions are made and deciding which
behavioural factors are relevant to mode shift modelling to be considered in the developed
survey, this thesis utilized the Structural Equation Modelling (SEM) approach to investigate
the effects of psychological factors on mode choice behaviour considering the Theory of
Interpersonal Behaviour (TIB) as a theoretical foundation. The dataset used in this analysis
was collected in 2009-2010 in the City of Edmonton, Alberta, Canada. The analysis
conducted in this chapter confirms the causal relationships between the underlying
psychological aspects affecting mode choice as indicated by the Theory of Interpersonal
Behavior. The results showed that the consideration of psychological attributes, namely
personal attitude, habit formation, and emotional response as latent variables helped explain
mode choice behaviour. As such, and given the previously mentioned policy implications, the
proposed survey collected detailed information about habit formation, personal attitude, and
affective appraisal besides personal and modal attributes to be considered in the mode shift
modelling process (Osman et al. 2011; Idris et al. 2012b).
175
As a key component of the mode shift module, an innovative COmmuting Survey for MOde
Shift (COSMOS), combining three types of instruments for collecting detailed information on
commuters’ mode switching behaviour, was developed and implemented in the City of
Toronto, Canada in 2012. In general, the survey exploits qualitative psychometric questions
on users’ perception along with Revealed Preference (RP) mode choice information and
Stated Preference (SP) mode switching experiments. The survey is divided into four sections.
The first section gathers revealed information concerning daily commuting work trips and
current travel options. The second section sets up an RP-pivoted SP choice experiment based
on efficient experimental design technique (D-Efficient design) that maximizes the
information gained from different hypothetical scenarios. The stated choice experiment
measures participants’ stated mode switching preferences towards public transit given some
policy changes. A total of six hypothetical scenarios were presented to each respondent as a
combination of transit services that can be easily figured out such as Streetcar, Bus and/or
Subway, and new services and/or technologies that are more innovative on a technological
point of view and have little chance of having been experienced before such as Bus Rapid
Transit (BRT) and/or Light Rail Transit (LRT). Factors such as travel time, travel cost and
parking cost for the car option were considered in the experiment. Further, different
components of the transit trip travel time were included as well as transit fare for the public
transit alternative. In addition, various transit service design factors were considered such as
service accessibility in terms of access/egress to public transit stops/stations as well as park-
and-ride availability; service frequency and headway in terms of the expected waiting time;
and service reliability standards in terms of transit schedule delay. Moreover, the experiment
was sensitive to some important preference attributes such as advance information provision,
ITS technologies and rail vs. bus attraction. Furthermore, in order to ensure practical attribute
level ranges, best practices in transit service planning were utilized in the design. The third
section of the survey gathered psychological information regarding habit of auto driving,
affective appraisal and personal attitudes. Different psychometric tools were used to capture
psychological factors affecting mode choice. Habitual behaviour was measured using
Verplanken’s response-frequency questionnaire. Affective appraisal was indirectly estimated
using the Osgood's semantic differential scale. A five-point Likert scale was used to measure
attitude. Finally, the last section of the survey collected information regarding common
socioeconomic and demographic characteristics (Idris et al. 2012c; Idris et al. 2013).
176
The developed survey was conducted in the City of Toronto, Canada between April and May
2012. A total of 62,652 fully opted-in panel of Canadians who have agreed to be
compensated for the participation in market research was used as a survey frame of this
study. A total of 13,265 (21.17% of the total panel size) was recruited and invited to
participate in the survey via email. A detailed description of the study and the survey process
as well as incentives was introduced to the potential survey participants. A total of 3,769
participants agreed to participate in the study and had to sign an online consent of
participation. A total of 2,380 complete entries (1,389 incomplete entries) were initially
received, with a response rate of 17.94% which is in line with the typical travel surveys’
response rate of 20% (Richardson et al. 1995; Franklin et al. 2003). Finally, after a process of
cleaning the dataset, the collected sample size was reduced to 1,211 observations (139
observations were lost out of the required sample size of 1,350 observations) to maintain
appropriate sample representation of the study area for each stratum.
The data collected through such novel survey is then used to develop econometric models of
mode switching behaviour towards public transit, with emphasis on capturing psychological
factors and Customer Oriented Transit Service (COTS) elements. Joint discrete mode
switching models, where revealed mode choice model is combined with a stated mode
switching probability model, are developed. The modelling results enrich our understanding
of mode switching behaviour and reveal some interesting findings. Socio-psychological
variables (mainly habit formation) have shown to have strong influence on mode shift and
improved the models in terms of fitness and statistical significance. In an indication for the
superiority of the car among other travel options, strong car use habit formation was realized
for car drivers, making it hard for them to switch to public transit. Further, unlike traditional
mode choice models, the developed mode shift models show that travel cost and time are of
relatively minor importance compared to other transit Level of Service (LOS) attributes such
as waiting time, service frequency, system reliability, number of transfers, transit technology,
and crowding level. It was also shown that passengers are more likely to switch to rail-based
modes (e.g. LRT and subway) than rubber-tyred modes (e.g. BRT). On the other hand, the
availability of park-and-ride facilities and parking cost as well as both schedule and real-time
information provision did not appear to be significant for mode shift decisions. Accordingly,
it might be argued that the latter factors are more important for short-term route shift rather
than mode shift decisions. The previous findings unravel the reason why conventional mode
choice models (based only on common socioeconomic and demographic characteristics of the
177
decision maker and basic mode-related attributes, and lacking psychological factors) tend to
overestimate mode switching to public transit.
Moreover, examining the Forecasting Performance Measure (FPM) of the developed models
showed that traditional RP mode choice models have the poorest forecast ability, whereas the
SP and the joint RP/SP mode shift models have the best performance. In particular,
traditional RP mode choice models have shown a very high tendency to over-predict transit
ridership, reaching a value of 133.89%. Such transit ridership overestimation can be
attributed to the lack of behavioural as well as Customer Oriented Transit Service (COTS)
elements (e.g. passenger information provision, ITS technologies that improve reliability, and
rail vs. bus attraction) in traditional models. On the contrary, both the SP and the joint RP/SP
mode shift models had the lowest transit ridership overestimation.
Interestingly, the previous observations confirm the initial hypothesis of this research that RP
models tend to overestimate mode shift to transit. It should be clear that the SP data
complemented the RP information and resulted in improved forecasting performance and less
transit ridership overestimation. In light of the above, the developed models provide a better
understanding of commuters’ preferences and mode switching behaviour.
In conclusion, this research provides evidence that mode shift is a complex process which
involves socio-psychological variables beside common socio-demographic and modal
attributes. The developed mode switch models present a new methodologically sound tool for
evaluating the impacts of alternative transit service designs on travel behaviour. Such tool is
more desirable for transit service planning than the traditional ones and can aid in precisely
estimating transit ridership. The presented models are also useful for evaluating alternative
emerging technologies, such as passenger information systems, ITS technologies and new
transit infrastructure development strategies (e.g. LRT and BRT).
178
8.3 Research Contributions
This dissertation presented a significant step towards a better understanding of commuters’
preferences and mode switching behaviour. In particular, the contribution of this thesis can be
divided into four main components. First, a conceptual framework for modal shift maximized
transit route design model is developed to extend the use of the developed models beyond
forecasting transit ridership (demand) to the operational extent of transit route design
(supply). The framework built upon and extended the capabilities of the existing
MIcrosimulation Learning-based Approach for TRansit ASsignment (MILATRAS) to tackle
both the route design and mode shift problems. The presented framework added a mode shift
module to MILATRAS in order to find operationally implementable transit route that can
provide alternative design concepts corresponding to different service requirements. Transit
demand variability among both modes and routes is considered at the microscopic level by
running the joint mode shift and route choice models of MILATRAS, allowing for
consistency between the supplied service level and passenger demand (Osman and Shalaby
2010; Idris et al. 2012a).
Second, this thesis introduced a learning-based mode shift model that compiles both adaptive
learning techniques in coincide with Random Utility Maximization (RUM) concepts as an
alternative way to mode shift modelling. The developed approach is capable to model the
mode switching mechanism while being consistent with the intuition behind bounded
rationality. The proposed learning-based mode shift model is built on top of the mode shift
models developed in Section 7.7. The learning process, however, ensures modelling personal
behaviour at the individual level based on personal experience and evaluation of the
transportation system in a more dynamic fashion which is more compatible with
MILATRAS. Further, the learning process models the mode switching mechanism while
simultaneously accounting for habitual inertia against shifting modes, different levels of
information provision and awareness limitations. What is unique to the proposed approach is
that it models the insights of the decision making process and the period of time required to
reap the benefits of the proposed policy changes.
Third, a multi-instrument socio-psychometric COmmuting Survey for MOde Shift
(COSMOS) is designed and implemented to gather Revealed Preference (RP) and Stated
Preference (SP) travel data along with psychological information such as personal attitudes,
emotional response and habit formation of travellers associated with different modes of travel
179
(Osman et al. 2011; Idris et al. 2012b). The developed survey is conducted online among a
representative sample of Toronto commuters who are asked about their willingness to shift to
different transit technologies of varying characteristics. In addition to collecting common
socioeconomic, demographic and modal attributes, the survey gathered data on the revealed
mode choice behaviour as well as the stated mode switching preferences to public transit
considering some important preference attributes such as advance information provision, ITS
technologies and rail vs. bus attraction. Moreover, the survey gathered psychological
information regarding habit of auto driving, affective appraisal and personal attitudes
associated with different travel options. Different psychometric tools are used to capture
psychological factors affecting mode choice. Further, the survey set up a stated choice
experiment based on efficient experimental design techniques to maximize the information
gained while minimizing the number of hypothetical scenarios required. The survey
respondents are asked to identify their propensity to perform their work trip by a non-existing
transit service in the future. In an attempt to maintain practical attribute level ranges in the
stated choice experiment, best practices in transit service planning are utilized in terms of
service accessibility standards, service frequency and headway standards, as well as service
reliability standards (Idris et al. 2012c).
Fourth, enhanced ridership forecasting tools for improved transit service planning are
developed. Econometric demand models of mode switching behaviour are estimated to
evaluate transit investments that usually target car users. As opposed to traditional mode
choice models based on RP data, adequate mode shift models are developed using state-of-
the-art methodology of combining Revealed Preference (RP) and Stated Preference (SP)
information to accurately forecast transit ridership (Idris et al. 2013). The developed models
showed that traditional RP mode choice models tend to over-predict transit ridership on the
expense of the car driver option due to the lack of behavioural as well as Customer Oriented
Transit Service (COTS) elements in traditional models. Further, the developed models
provide transit planners an idea about the power of modal characteristics to attract commuters
to transit. Moreover, the developed models provide useful information to transit planners on
the relative propensity of different mode users to switch to transit in response to changes in
transit LOS attributes. The policy implications of such findings should be kept in mind
especially when targeting potential markets (e.g. car drivers) for increasing transit ridership.
180
8.4 Future Research
While this thesis is considered a significant step towards a better understanding of
commuters’ preferences and mode switching behaviour, there are still a number of issues that
need to be addressed that can potentially be a motivation for future research. The following
are recommendation for some areas that have potential to be explored in future research:
First, a conceptual framework for modal shift maximized transit route design is presented in
Chapter 3. The presented model is comprised of two main parts: a design tool and an
evaluation component. However, it is clear that further work is required to make these ideas
practical and capable of implementation. In particular, more effort is required to develop the
following three main components of the design tool:
1- Developing a Transit Route Generation module capable of generating optimal transit
route/line designs that maximize demand attraction given topological characteristics
of the transportation network (e.g. roadway network) and the demand distribution
between two terminal points, and following a set of service standards and practical
guidelines.
2- Developing a Transit Stop Allocation module capable of allocating transit
stops/stations that maximize service coverage given potential locations for transit
stops/stations and/or transfer zones, in addition to trip demand distribution around the
transit route/line, and following a set of service standards and practical guidelines.
3- Developing a practical tool for Frequency Setting capable of calculating the number
of transit units and service frequency required based on passenger counts and cycle
time, given vehicle capacity, desired occupancy (loading standards) and policy
headway.
Second, Chapter 3 also introduced a learning-based mode shift model that compiles both
adaptive learning techniques in coincide with random utility maximization concepts as an
alternative way to mode shift modelling. However, conducting controlled lab experiments of
travel behaviour is suggested for further work to specify and test the learning-based mode
shift process and estimate its parameters (α, ε, τ, etc.) and convergence criteria under various
181
assumptions and levels of information provision. For example, given that previous choices
are not just based on habits, future work is required to determine the correct value of the step
size parameter (α) that mimics actual habit formation or decay. It is also suggested to collect
travel data after policy implementation at regular time intervals (e.g. every six months) until
the modal shares stabilize. The collected data can then be used to validate the proposed
formulations and assumptions of habit formation, level of information provision and
awareness limitations. In addition, such data can also be used to find mapping between the
time it takes the agent to learn and how long it takes to reap the benefits of the changes in real
life. Moreover, future efforts are suggested to test the forecasting performance of the model
(i.e. temporal transferability) as well as testing its transferability across space.
Third, future research is required for operating the proposed modal shift maximized transit
route design model. While the design component of the model deals with generating
operationally implementable transit route design(s), the evaluation component assesses the
generated route design(s) considering transit demand variability among both modes and
routes by jointly running both components (route choice and mode shift) of MILATRAS.
Integrating both components together using a feedback loop will allow for consistency
between the supplied service level and passenger demand. Such treatment will result in a
modal shift maximized transit route design model that is capable to select the optimum transit
route alignment and design characteristics with the ultimate goal of maximizing transit
ridership.
Fourth, the models developed in Chapter 7 showed that traditional RP mode choice models
tend to over-predict transit ridership on the expense of the car driver option due to the lack of
behavioural as well as Customer Oriented Transit Service (COTS) elements in traditional
models. The previous findings confirm the main claim raised by this research. However,
future research is suggested to precisely quantify such transit ridership overestimation. This
should be followed by developing a correction methodology to adjust traditional RP-based
mode choice models’ over-prediction of transit ridership. In addition, it is also suggested to
test the forecasting performance of the developed model across space and time (i.e.
spatiotemporal transferability).
182
Fifth, it is also suggested to study the relationship between travellers’ confidence in the stated
choice and their psychological attributes such as personal attitude, emotional response, and
habit formation. This might allow us to rely on the former as an indicator for the latter in
future models.
Sixth, future research is required to quantify the share of both contextual conditions and
psychological factors in the mode shift decision making process (e.g. 40% contextual and
60% psychological). In addition, it is suggested to study the relationship between
psychological factors and common socio-demographic attributes (e.g. relationship between
habit formation and car ownership).
183
9 REFERENCES
Aarts, H., B. Verplanken, et al. (1997). "Habit and information use in travel mode choices."
Acta Psychologica 96(1-2): 1-14.
Abdulhai, B. and L. Kattan (2003). "Reinforcement learning: Introduction to theory and
potential for transport applications." Canadian Journal of Civil Engineering 30(6):
981-991.
Ajzen, I. (1985). "From intentions to actions: A theory of planned behavior." Action-control:
From cognition to behavior 11: 39.
Ajzen, I. (1991). "The theory of planned behavior." Organizational behavior and human
decision processes 50(2): 179-211.
Arasan, V. and P. Vedagiri (2011). "Modelling Modal Shift from Personal Vehicles to Bus on
Introduction of Bus Priority Measure."
Bailey, J. (1984). "The meaning of car availability in mode choice decisions." Transportation
Planning and Technology 9(2): 125-134.
Bamberg, S., D. Rölle, et al. (2003). "Does habitual car use not lead to more resistance to
change of travel mode?" Transportation 30(1): 97-108.
Banister, D. (1978). "The influence of habit formation on modal choice: a Heuristic model."
Transportation 7(1): 5-33.
Barff, R., D. MacKay, et al. (1982). "A selective review of travel-mode choice models."
Journal of Consumer Research 8(4): 370-380.
Barros, G. (2010). "Herbert A. Simon and the concept of rationality: boundaries and
procedures." Revista de Economia Política 30(3): 455-472.
Barto, A. and R. Sutton (1998). Reinforcement Learning: an introduction, MIT Press
Cambridge, MA.
Behrens, R. and R. Mistro (2010). "Shocking Habits: Methodological Issues in Analyzing
Changing Personal Travel Behavior Over Time." International Journal of Sustainable
Transportation 4(5): 253-271.
Beimborn, E., M. Greenwald, et al. (2003). "Accessibility, connectivity, and captivity:
impacts on transit choice." Transportation Research Record: Journal of the
Transportation Research Board 1835(-1): 1-9.
Ben-Akiva, M. and M. Bierlaire (1999). Discrete choice methods and their applications to
short term travel decisions. Handbook of transportation science, Springer: 5-33.
Ben-Akiva, M. and B. Boccara (1995). "Discrete choice models with latent choice sets* 1."
International Journal of Research in Marketing 12(1): 9-24.
Ben-Akiva, M., M. Bradley, et al. (1994). "Combining revealed and stated preferences data."
Marketing Letters 5(4): 335-349.
Ben-Akiva, M. and S. R. Lerman (1985). "Discrete choice analysis: theory and application to
travel demand."
184
Ben-Akiva, M., J. Walker, et al. (2002). "Integration of choice and latent variable models." In
Perpetual Motion: Travel Behavior Research Opportunities and Application
Challenges, Oxford: 431-470.
Bliemer, M. C. J. and J. M. Rose (2009). Efficiency and sample size requirements for stated
choice experiments.
Bliemer, M. C. J. and J. M. Rose (2011). "Experimental design influences on stated choice
outputs: An empirical study in air travel choice." Transportation Research Part A:
Policy and Practice 45(1): 63-79.
Bliemer, M. C. J., J. M. Rose, et al. (2008). "Approximation of Bayesian efficiency in
experimental choice designs." Journal of Choice Modelling 1(1): 98-127.
Cantillo, V., J. de Dios Ortuzar, et al. (2007). "Modeling Discrete Choices in the Presence of
Inertia and Serial Correlation." Transportation Science 41(2): 195-205.
Caussade, S., J. D. Ortúzar, et al. (2005). "Assessing the influence of design dimensions on
stated choice experiment estimates." Transportation Research Part B: Methodological
39(7): 621-640.
Ceder, A. and N. Wilson (1986). "Bus network design." Transportation Research Part B:
Methodological 20(4): 331-344.
Chih-Wen, Y. (2005). "Modeling Multiple Sources of Heterogeneity in Mode Choice
Model." Journal of the Eastern Asia Society for Transportation Studies 6: 2031-2045.
Chorus, C., B. Dellaert, et al. (2009). "Inertia in travel choice: The role of risk aversion and
learning."
Chorus, C. G., E. J. E. Molin, et al. (2006). "Responses to Transit Information among Car-
drivers: Regret-based Models and Simulations." Transportation Planning and
Technology 29(4): 249 - 271.
Chorus, C. G. and H. J. P. Timmermans (2009). "Measuring user benefits of changes in the
transport system when traveler awareness is limited." Transportation Research Part A
43(5): 536-547.
Cobanoglu, C., B. Warde, et al. (2001). "A comparison of mail, fax and web-based survey
methods." International journal of market research 43(4): 441-452.
Cooper, B., J. Rose, et al. (2011). "Does anybody like water restrictions? Some observations
in Australian urban communities*." Australian Journal of Agricultural and Resource
Economics.
Deakin, E., M. Shirgaokar, et al. (2006). "System plan for California's bay area regional
express bus service." Transportation Research Record: Journal of the Transportation
Research Board 1986(-1): 190-197.
Diana, M. (2010). "From mode choice to modal diversion: a new behavioural paradigm and
an application to the study of the demand for innovative transport services."
Technological Forecasting and Social Change 77(3): 429-441.
185
Dogar, N. (2010). Socio-psychometric Survey to Measure Habit and Inertia in Commuting
Mode Choice. Department of Civil & Environmental Engineering, University of
Alberta. Master of Engineering.
Domarchi, C., A. Tudela, et al. (2008). "Effect of attitudes, habit and affective appraisal on
mode choice: an application to university workers." Transportation 35(5): 585-599.
Dosman, D. and W. Adamowicz (2006). "Combining stated and revealed preference data to
construct an empirical examination of intrahousehold bargaining." Review of
Economics of the Household 4(1): 15-34.
Eriksson, L., J. Garvill, et al. (2008). "Interrupting habitual car use: The importance of car
habit strength and moral motivation for personal car use reduction." Transportation
Research Part F: Psychology and Behaviour 11(1): 10-23.
Flyvbjerg, B., M. Holm, et al. (2005). "How (In) accurate Are Demand Forecasts in Public
Works Projects?: The Case of Transportation." Journal of the American Planning
Association 71(2): 131-146.
Forsey, D., K. M. N. Habib, et al. (2011). Evaluating Impact of VIVA Transit on Commuting
and School Trip Mode Choice in York Region. Working Paper and will be submitted
for presentation at 2012 TRB Annual Meeting.
Franklin, S., C. Walker, et al. (2003). Survey methods and practices, Statistics Canada, Social
Survey Methods Division.
Fujii, S. and R. Kitamura (2003). "What does a one-month free bus ticket do to habitual
drivers? An experimental analysis of habit and attitude change." Transportation 30(1):
81-95.
Furth, P. and A. Rahbee (2000). "Optimal bus stop spacing through dynamic programming
and geographic modeling." Transportation Research Record: Journal of the
Transportation Research Board 1731(-1): 15-22.
Galdames, C., A. Tudela, et al. (2011). Exploring the Role of Psychological Factors on Mode
Choice Models Using Latent Variable Approach. Transportation Research Record.
Forthcoming.
Garber, N. and L. Hoel (2002). Traffic and Highway Engineering, Thomson Learning, Inc.
Gardner, B. (2009). "Modelling motivation and habit in stable travel mode contexts."
Transportation Research Part F: Traffic Psychology and Behaviour 12(1): 68-76.
Gärling, T. and K. Axhausen (2003). "Introduction: Habitual travel choice." Transportation
30(1): 1-11.
Gärling, T., R. Gillholm, et al. (1998). "Reintroducing attitude theory in travel behavior
research: The validity of an interactive interview procedure to predict car use."
Transportation 25(2): 129-146.
Golledge, R. and L. Brown (1967). "Search, learning, and the market decision process."
Geografiska Annaler. Series B. Human Geography 49(2): 116-124.
186
Golob, T. (2003). "Structural equation modeling for travel behavior research." Transportation
Research Part B: Methodological 37(1): 1-25.
Goodwin, P. (1977). "Habit and hysteresis in mode choice." Urban studies 14(1): 95-98.
Guihaire, V. and J. Hao (2008). "Transit network design and scheduling: A global review."
Transportation Research Part A 42(10): 1251-1273.
Gunn, H. F., M. Bradley, et al. (1992). "High speed rail market projection: Survey design and
analysis." Transportation 19(2): 117-139.
Gwilliam, K. and D. Banister (1977). "Patterns of car usage and restraint modelling."
Transportation 6(4): 345-363.
Habib, K., L. Kattan, et al. (2010). "Model of personal attitudes towards transit service
quality." Journal of Advanced Transportation.
Habib, K. M., N. Day, et al. (2009). "An investigation of commuting trip timing and mode
choice in the Greater Toronto Area: application of a joint discrete-continuous model."
Transportation Research Part A: Policy and Practice 43(7): 639-653.
Habib, K. M. N., Y. Tian, et al. (2011). "Modelling commuting mode choice with explicit
consideration of carpool in the choice set formation." Transportation 38(4): 587-604.
Habib, K. N., J. Swait, et al. (2012). "Investigating Structural Changes in Commuting Mode
Choice Preferences with Repeated Cross-Sectional Travel Survey Data: The Contexts
of Greater Toronto and Hamilton (GTHA) Area."
Hale, C. (2009). Munich's central corridor: The mega-project as crux of integrated
metropolitan planning. European Transport Conference.
Hamer, P. (2010). "Analysing the Effectiveness of Park and Ride as a Generator of Public
Transport Mode Shift." Road & Transport Research: A Journal of Australian and New
Zealand Research and Practice 19(1): 51.
Heinen, E., K. Maat, et al. (2010). "The role of attitudes toward characteristics of bicycle
commuting on the choice to cycle to work over various distances." Transportation
Research Part D: Transport and Environment.
Hensher, D. A. (2001a). "The valuation of commuter travel time savings for car drivers:
evaluating alternative model specifications." Transportation 28(2): 101-118.
Hensher, D. A. (2001b). "Measurement of the valuation of travel time savings." Journal of
Transport Economics and Policy (JTEP) 35(1): 71-98.
Hensher, D. A. and J. King (2001). "Parking demand and responsiveness to supply, pricing
and location in the Sydney central business district." Transportation Research Part A:
Policy and Practice 35(3): 177-196.
Hensher, D. A. and J. M. Rose (2007). "Development of commuter and non-commuter mode
choice models for the assessment of new public transport infrastructure projects: A
case study." Transportation Research Part A 41(5): 428-443.
187
Hensher, D. A., J. M. Rose, et al. (2009). "Estimating the willingness to pay and value of risk
reduction for car occupants in the road environment." Transportation Research Part A:
Policy and Practice 43(7): 692-707.
Highway, A. A. o. S., T. Officials, et al. (2004). Guide for Park-and-ride Facilities, Amer
Assn of State Hwy.
Hopkins, E. (2007). "Adaptive learning models of consumer behavior." Journal of Economic
Behavior and Organization 64(3-4): 348-368.
Hoyle, R. (1995). Structural equation modeling: Concepts, issues, and applications, Sage
Publications, Inc.
Huber, J. and K. Zwerina (1996). "The importance of utility balance in efficient choice
designs." Journal of marketing research: 307-317.
Idris, A. O., K. M. Nurul Habib, et al. (2012c). Modal Shift Forecasting Model for Transit
Service Planning: Survey Instrument Design. 12th
International Conference on
Advanced Systems for Public Transport (CASPT 2012). Santiago, Chile.
Idris, A. O., K. M. Nurul Habib, et al. (2013). Joint RP/SP Mode Switch Forecasting Model
for Transit Service Planning. 92nd
TRB Annual Meeting (TRB 2013), Washington
D.C.
Idris, A. O., K. M. Nurul Habib, et al. (2012b). Investigating the Effects of Psychological
Factors on Commuting Mode Choice Behaviour. 91st TRB Annual Meeting (TRB
2012). Washington D.C.
Idris, A. O., A. Shalaby, et al. (2012a). "Towards a Learning-based Mode Shift Model: A
Conceptual Framework." Transportation Letters: The International Journal of
Transportation Research 4(1): 15-27.
Johansson, V. M., T. Heldt, et al. (2006). "The effects of attitudes and personality traits on
mode choice." Transportation Research Part A 40(6): 507-525.
Kanninen, B. J. (2002). "Optimal design for multinomial choice experiments." Journal of
marketing research 39(2): 214-227.
Kaplowitz, M. D., T. D. Hadlock, et al. (2004). "A comparison of web and mail survey
response rates." Public opinion quarterly 68(1): 94-101.
Katz, D. and M. M. Rahman (2010). "Levels of Overcrowding in Bus System of Dhaka,
Bangladesh." Transportation Research Record: Journal of the Transportation Research
Board 2143(-1): 85-91.
Kenyon, S. and G. Lyons (2003). "The value of integrated multimodal traveller information
and its potential contribution to modal change." Transportation Research Part F:
Traffic Psychology and Behaviour 6(1): 1-21.
Kepaptsoglou, K. and M. Karlaftis (2009). "Transit Route Network Design Problem:
Review." Journal of Transportation Engineering 135: 491.
Kessels, R., P. Goos, et al. (2006). "A comparison of criteria to design efficient choice
experiments." Journal of marketing research 43(3): 409-419.
188
Kwak, N. and B. Radler (2002). "A comparison between mail and web surveys: Response
pattern, respondent profile, and data quality." JOURNAL OF OFFICIAL
STATISTICS-STOCKHOLM- 18(2): 257-274.
Lampkin, W. and P. Saalmans (1967). "The design of routes, service frequencies, and
schedules for a municipal bus undertaking: a case study." OR 18(4): 375-397.
Li, Z. and D. A. Hensher (2011). "Crowding and public transport: A review of willingness to
pay evidence and its relevance in project appraisal." Transport Policy.
Litman, T. and F. Laube (2002). "Automobile dependency and economic development."
Victoria Transport Policy Institute, Canada.
Long, B., K. Choocharukul, et al. (2011). Psychological Factors Influencing Behavioral
Intention Toward Future Sky Train Usage in Phnom Penh, Cambodia. Transportation
Research Board.
Louviere, J. J. and D. A. Hensher (1983). "Using discrete choice models with experimental
design data to forecast consumer demand for a unique cultural event." Journal of
Consumer Research: 348-361.
Louviere, J. J., D. A. Hensher, et al. (2000). Stated choice methods: analysis and applications,
Cambridge Univ Pr.
Louviere, J. J. and G. Woodworth (1983). "Design and analysis of simulated consumer
choice or allocation experiments: an approach based on aggregate data." Journal of
marketing research: 350-367.
Lucarotti, P. (1977). "Car availability—the fundamental modal split." Transportation
Planning and Technology 3(4): 203-213.
MacCallum, R. and J. Austin (2000). "Applications of structural equation modeling in
psychological research." Annual review of psychology 51(1): 201-226.
Mackett, R. (2003). "Why do people use their cars for short trips?" Transportation 30(3):
329-349.
Mandl, C. (1980). "Evaluation and optimization of urban public transportation networks."
European Journal of Operational Research 5(6): 396-404.
McFadden, D. (1974). "Conditional logit analysis of qualitative choice behavior." Frontiers in
econometrics 8: 105–142.
McFadden, D. (1986). "The choice theory approach to market research." Marketing science
5(4): 275-297.
Meyer, M. (1999). "Demand management as an element of transportation policy: using
carrots and sticks to influence travel behavior." Transportation Research Part A:
Policy and Practice 33(7-8): 575-599.
Meyer, M. and E. Miller (2001). Urban transportation planning: a decision-oriented approach,
McGraw-Hill New York.
189
Miller, E. J. (2001). "The Greater Toronto Area Travel Demand Modelling System, Version
2.0, Volume I: Model Overview." Joint Program in Transportation, University of
Toronto, Toronto.
Mistretta, M., J. A. Goodwill, et al. (2009). "Best Practices in Transit Service Planning."
Morikawa, T. (1994). "Correcting state dependence and serial correlation in the RP/SP
combined estimation method." Transportation 21(2): 153-165.
Murray, A. (2003). "A coverage model for improving public transit system accessibility and
expanding access." Annals of Operations Research 123(1): 143-156.
Murray, A., R. Davis, et al. (1998). "Public transportation access." Transportation Research
Part D: Transport and Environment 3(5): 319-328.
Murray, A. and X. Wu (2003). "Accessibility tradeoffs in public transit planning." Journal of
Geographical Systems 5(1): 93-107.
Nurdden, A., R. Rahmat, et al. (2007). "Effect of transportation policies on modal shift from
private car to public transport in Malaysia." Journal of Applied Sciences 7(7): 1013-
1018.
Ogilvie, D., M. Egan, et al. (2004). "Promoting walking and cycling as an alternative to using
cars: systematic review." Bmj 329(7469): 763.
Oldfield, R. and P. Bly (1988). "An analytic investigation of optimal bus size."
Transportation Research Part B: Methodological 22(5): 319-337.
Ory, D. and P. Mokhtarian (2005). "When is getting there half the fun? Modeling the liking
for travel." Transportation Research Part A: Policy and Practice 39(2-3): 97-123.
Osgood, C. E., W. S. May, et al. (1975). Cross-cultural universals of affective meaning,
University of Illinois Press.
Osman, A. O., K. M. Nurul Habib, et al. (2011). Investigating the Effects of Psychological
Factors Measured in a Semantic Scale on Commuting Mode Choice Behaviour. 12th
International Conference on Computers in Urban Planning and Urban Management
(CUPUM 2011). Lake Louise, Alberta, Canada.
Osman, A. O. and A. Shalaby (2010). A Modal Shift Optimized Transit Route Design Model.
2nd
Annual Conference on Transportation and Logistics (TRANSLOG 2010).
Hamilton, Ontario, Canada.
Pattnaik, S., S. Mohan, et al. (1998). "Urban bus transit route network design using genetic
algorithm." Journal of Transportation Engineering 124(4): 368-375.
Quarmby, D. (1967). "Choice of travel mode for the journey to work: some findings." Journal
of Transport Economics and Policy: 273-314.
Quentin, K. and K. Hong (2005). "Effect of Attribute Perceptions on Mode Choice Behavior
in a Transit Market." Journal of the Eastern Asia Society for Transportation Studies 6:
1740-1750.
190
Reeve, J. (2005). "Understanding Motivation and Emotion, John Wiley & Sons." Inc., New
York.
Richardson, A. J., E. S. Ampt, et al. (1995). Survey methods for transport planning,
Eucalyptus Press Oakland, CA.
Ronis, D., J. Yates, et al. (1989). "Attitudes, decisions, and habits as determinants of repeated
behavior." Attitude structure and function: 213-239.
Rose, J. M. and M. C. J. Bliemer (2009). "Constructing efficient stated choice experimental
designs." Transport Reviews 29(5): 587-617.
Rose, J. M., M. C. J. Bliemer, et al. (2008). "Designing efficient stated choice experiments in
the presence of reference alternatives." Transportation Research Part B:
Methodological 42(4): 395-406.
Saka, A. (2001). "Model for Determining Optimum Bus-Stop Spacingin Urban Areas."
Journal of Transportation Engineering 127: 195.
Scarpa, R. and J. M. Rose (2008). "Design efficiency for non‐market valuation with choice
modelling: how to measure it, what to report and why*." Australian Journal of
Agricultural and Resource Economics 52(3): 253-282.
Shannon, T., B. Giles-Corti, et al. (2006). "Active commuting in a university setting:
Assessing commuting habits and potential for modal change." Transport Policy 13(3):
240-253.
Sheth, J. (1976). "A psychological model of travel mode selection." Advances in Consumer
Research 3: 425-430.
Simon, H. (1957). A behavioral model of rational choice, in (eds) Models of Man, Social and
Rational: Mathematical Essays on Rational Human Behavior in a Social Setting, New
York: Wiley.
Tanadtang, P., D. Park, et al. (2005). "Incorporating uncertain and incomplete subjective
judgments into the evaluation procedure of transportation demand management
alternatives." Transportation 32(6): 603-626.
Temme, D., M. Paulssen, et al. (2008). "Incorporating Latent Variables into Discrete Choice
Models-A Simultaneous Estimation Approach Using SEM Software." BuR –
Bussiness Research, Official Open Access Journal of VHB 1(2): 220-237.
Train, K. (2003). Discrete choice methods with simulation, Cambridge Univ Pr.
Triandis, H. (1977). Interpersonal behavior, Brooks/Cole Pub. Co.
Tudela, A., K. Habib, et al. (2011). Incorporating the explicit role of psychological factors on
mode choice: A hybrid mode choice model by using data from an innovative
psychometric survey. 2nd International Choice Modelling Conference, Leeds.
Van Nes, R. and P. Bovy (2000). "Importance of objectives in urban transit-network design."
Transportation Research Record: Journal of the Transportation Research Board 1735(-
1): 25-34.
191
Vedagiri, P. and V. Arasan (2009). "Estimating Modal Shift of Car Travelers to Bus on
Introduction of Bus Priority System." Journal of Transportation Systems Engineering
and Information Technology 9(6): 120-129.
Vermote, L. and L. Hens (2009). "Urban modal shift management in cities: An indicator
based approach." WIT Transactions on the Built Environment 107.
Verplanken, B. and H. Aarts (1999). "Habit, attitude, and planned behaviour: Is habit an
empty construct or an interesting case of goal-directed automaticity?" European
review of social psychology 10(1): 101-134.
Verplanken, B., H. Aarts, et al. (1998). "Habit versus planned behaviour: A field
experiment." British Journal of Social Psychology 37(1): 111-128.
Verplanken, B., H. Aarts, et al. (1997). "Habit, information acquisition, and the process of
making travel mode choices." European Journal of Social Psychology 27(5): 539-560.
Vuchic, V. (2005). Urban transit: operations, planning, and economics, John Wiley & Sons,
Hoboken, NJ.
Wahba, M. (2009). MILATRAS: MIcrosimulation Learning-based Approach to TRansit
ASsignment. Engineering & Computer Science, University of Toronto. Ph.D. thesis:
292.
Wahba, M. and A. Shalaby (2005). "Multiagent Learning-Based Approach to Transit
Assignment Problem: A Prototype." Transportation Research Record: Journal of the
Transportation Research Board 1926(-1): 96-105.
Wahba, M. and A. Shalaby (2009a). "Learning-based Departure Time and Path Choice
Modelling for Transit Assignment under Information Provision: A Theoretical
Framework." Transportation Research Part A 39: 325-344.
Wahba, M. and A. Shalaby (2009b). "MILATRAS: a new modeling framework for the transit
assignment problem." Schedule-Based Modeling of Transportation Networks: 1-24.
Walker, J. L. (2001). Extended discrete choice models: integrated framework, flexible error
structures, and latent variables, Massachusetts Institute of Technology.
Williams, M. (1978). "Factors affecting modal choice decisions in urban travel:: Some further
evidence." Transportation Research 12(2): 91-96.
Winston, C. (2000). "Government failure in urban transportation." Fiscal Studies 21(4): 403-
425.
Wirasinghe, S. and N. Ghoneim (1981). "Spacing of bus-stops for many to many travel
demand." Transportation Science 15(3): 210.
Xiong, Y. and J. Schneider (1992). "Transportation network design using a cumulative
genetic algorithm and neural network."
Zmud, J., M. Lee-Gosselin, et al. (2013). Transport Survey Methods: Best Practice for
Decision Making, Emerald Group Publishing.
COmmuting Survey for MOde Shift (COSMOS)
192
APPENDIX: COMMUTING SURVEY FOR MODE SHIFT (COSMOS)
Introduction
Dear Survey Respondent,
You have been randomly selected to participate in a research study conducted by the Department of Civil Engineering at
the University of Toronto. This study aims at achieving a better understanding of commuters' mode choice preferences
(e.g. drive, walk, bike, etc.) and willingness to use public transit in response to changes in transit service attributes.
We are contacting a random sample of commuters in the City of Toronto to gather information on their personal attitudes
and habits associated with daily commuting work trips.
The survey is divided into four sections: Section A will gather information on your daily commuting work trip and current
travel options; Section B will ask about your willingness to make your work trip using a new transit service; Section C will
gather information regarding habit formation, emotional response and personal attitudes towards different travel options;
and, Section D will collect socioeconomic and demographic characteristics.
We kindly ask you to participate in this web survey so that your opinion is represented in our study. This survey is
designed to be as short as possible and will take approximately 15 minutes to complete. Please answer every question in
each section in order to proceed to subsequent sections.
You may choose not to complete the survey at any time without any penalty. Keep in mind, however, that the responses
submitted in previous sections are not retrievable, and therefore will still be anonymously included in final survey results.
Please note that there is no related risk involved with your participation in this study. All the collected information will be
stored securely at the University and will be processed with the utmost confidentiality and for academic purposes only.
Your cooperation is highly appreciated.
Should you have any questions about the study, please feel free to e-mail us at [email protected]. For any questions
regarding your rights as a respondent in this survey, you are free to contact the office of Research Ethics, University of
Toronto, McMurrich Building, 2nd floor, 12 Queen's Park Crescent West Toronto, ON M5S 1S8, Tel: (416) 946-3273,
Fax: (416) 946-5763, Email: [email protected].
Consent of Participant
By pressing the "Login" button, you will indicate to us that you agree, of your own freewill, to voluntarily participate in
this study after carefully reading and fully understanding the information presented in the introductory section of the
survey.
COmmuting Survey for MOde Shift (COSMOS)
193
This survey should take approximately 15 minutes to complete
Section A. Your Trip to Work
In this section, the survey will gather information about your daily commuting work trip including trip origin and
destination, means of commuting, travel attributes and the reasons why you choose your mode.
Trip Origin & Destination
1. What is the location of your home?
Full Address: _________________________________________________________________________ (Optional)
Postal Code: ___________________ Without any space, e.g. M0A1B1
City: _________________
2. What is the location of your usual place of work?
Full Address: _________________________________________________________________________ (Optional)
Postal Code: ___________________ Without any space, e.g. M0A1B1
City: _________________
3. What is the start time of your typical home-to-work trip?
Trip start time: ______________________ 24-hour time format
Please enter in HH:MM format, e.g. 07:00, 08:30, 11:20
Means of Commuting
4. What transportation mode do you typically use to get to work? (Select one choice only under primary mode. For
transit users who make transfer(s), please choose the mode you have the worst experience with under transit
technology).
Note: Refer to definitions below for clarification of some modes of travel. Important: if you want to make change to
this question, press "Restart" button but please note that all answers after this point will be lost.
Primary Mode Transit Technology
Car Options:
□Car Driver
□Car Passenger
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
COmmuting Survey for MOde Shift (COSMOS)
194
□Cycle
□Walk
□Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
Section for Car Driver: If your answer to question number 4 is “CarDriver”, then answer the following questions.
If this does not apply to you please proceed to the Car Passenger section
5. What is your typical one-way travel time to work (door to door)?
Travel Time: _________________________ minutes/trip
6. What is your typical one-way travel cost per work trip (kindly include the cost of fuel and tolls, if any, but exclude
parking cost)?
Travel Cost: _________________________ $/trip
7. What is your typical one-way parking cost per work trip?
Parking Cost: _________________________ $/trip
8. What car do you use?
□Sedan □SUV □Coupe □Van □Truck
Make: _________, Model: _________, Year: _________, Type: □Conventional □Hybrid □Electric
9. In case of unavailability of the Car Driver option, what would be your second choice? (Select one choice only under
chosen mode. For transit users who make transfer(s), please choose the mode you use to travel the longest distance
under transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
Chosen Mode Transit Technology
Car Options:
□Car Passenger
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Cycle
□Walk
□Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
COmmuting Survey for MOde Shift (COSMOS)
195
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
10. If you were to take public transit, how long would it take to travel from home to the origin transit stop/station
(Access Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Access Time: _________________________ minutes
11. If you were to take public transit, how long would it take to wait/transfer at transit stop(s)/station(s)
(Waiting/Transferring Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Waiting Time: _________________________ minutes
12. If you were to take public transit, how long would it take from the origin transit stop/station to the destination
transit stop/station (In-Vehicle Travel Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
In-Vehicle Travel Time: _________________________ minutes
13. If you were to take public transit, how long would it take to travel from the destination transit stop/station to the final
workplace destination (Egress Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Egress Time: _________________________ minutes
14. If you were to take public transit, how much would it cost you per work trip (one-way)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Transit Travel Cost: _________________________ $/trip
51. If you were to take public transit, what mode would you use based on the travel time and cost you have just provided?
(If you were to combine more than one transit route/line, choose the mode you were to use to travel the longest
distance).
Public Transit Mode: □Streetcar □Bus □Subway
16. In the last 12 months, how often have you used public transit to commute to work?
□ More than once a week on average
□ Between once a month and once a week on average
□ Less than once a month on average
□ Never
COmmuting Survey for MOde Shift (COSMOS)
196
Section for Car Passenger: If your answer to question number 4 is “CarPassenger”, then answer the following
questions.
If this does not apply to you please proceed to the Carpool section
5. What is your typical one-way travel time to work (door to door)?
Travel Time: _________________________ minutes/trip
6. What is your typical one-way travel cost per work trip (including your share in fuel, parking and/or toll if any)?
Travel Cost: _________________________ $/trip
7. In case of unavailability of the Car Passenger option, what would be your second choice? (Select one choice only
under chosen mode. For transit users who make transfer(s), please choose the mode you use to travel the longest
distance under transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
Chosen Mode Transit Technology
Car Options:
□Car Driver
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Cycle
□Walk
□Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
8. If you were to take public transit, how long would it take to travel from home to the origin transit stop/station
(Access Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Access Time: _________________________ minutes
9. If you were to take public transit, how long would it take to wait/transfer at transit stop(s)/station(s)
(Waiting/Transferring Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Waiting Time: _________________________ minutes
COmmuting Survey for MOde Shift (COSMOS)
197
10. If you were to take public transit, how long would it take from the origin transit stop/station to the destination
transit stop/station (In-Vehicle Travel Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
In-Vehicle Travel Time: _________________________ minutes
11. If you were to take public transit, how long would it take to travel from the destination transit stop/station to the final
workplace destination (Egress Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Egress Time: _________________________ minutes
12. If you were to take public transit, how much would it cost you per work trip (one-way)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Transit Travel Cost: _________________________ $/trip
13. If you were to take public transit, what mode would you use based on the travel time and cost you have just provided?
(If you were to combine more than one transit route/line, choose the mode you were to use to travel the longest
distance).
Public Transit Mode: □Streetcar □Bus □Subway
14. In the last 12 months, how often have you used public transit to commute to work?
□ More than once a week on average
□ Between once a month and once a week on average
□ Less than once a month on average
□ Never
Section for Carpool: If your answer to question number 4 is “Carpool”, then answer the following questions.
If this does not apply to you please proceed to the Public Transit section
5. What is your typical one-way travel time to work (door to door)?
Travel Time: _________________________ minutes/trip
6. How many people do you typically Carpool with?
□2 people □3 people □ 4 people □ 5 or more people
7. What is your typical one-way travel cost per work trip (including your share in fuel, parking and/or toll if any)?
Travel Cost: _________________________ $/trip
8. In case of unavailability of the Carpool option, what would be your second choice? (Select one choice only under
chosen mode. For transit users who make transfer(s), please choose the mode you use to travel the longest distance
under transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
COmmuting Survey for MOde Shift (COSMOS)
198
Chosen Mode Transit Technology
Car Options:
□Car Driver
□Car Passenger Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Cycle
□Walk
□Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
9. If you were to take public transit, how long would it take to travel from home to the origin transit stop/station
(Access Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Access Time: _________________________ minutes
10. If you were to take public transit, how long would it take to wait/transfer at transit stop(s)/station(s)
(Waiting/Transferring Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Waiting Time: _________________________ minutes
11. If you were to take public transit, how long would it take from the origin transit stop/station to the destination
transit stop/station (In-Vehicle Travel Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
In-Vehicle Travel Time: _________________________ minutes
12. If you were to take public transit, how long would it take to travel from the destination transit stop/station to the final
workplace destination (Egress Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Egress Time: _________________________ minutes
13. If you were to take public transit, how much would it cost you per work trip (one-way)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Transit Travel Cost: _________________________ $/trip
COmmuting Survey for MOde Shift (COSMOS)
199
14. If you were to take public transit, what mode would you use based on the travel time and cost you have just provided?
(If you were to combine more than one transit route/line, choose the mode you were to use to travel the longest
distance).
Public Transit Mode: □Streetcar □Bus □Subway
15. In the last 12 months, how often have you used public transit to commute to work?
□ More than once a week on average
□ Between once a month and once a week on average
□ Less than once a month on average
□ Never
Section for Public Transit: If your answer to question number 4 is Ride all Way, Park & Ride, Kiss & Ride, Carpool &
Ride or Cycle & Ride, then answer the following questions.
If this does not apply to you please proceed to section of Cycle.
5. How do you typically pay your Public Transit fare?
□ Cash
□ Tickets or tokens
□ Transit pass
□ PRESTO card
6. Does your employer pay for your transit fares?
□No □Yes
7. How many times do you typically transfer when commuting by Public Transit? (One-way, enter 0 if none)
Number of Transfers: _________________________ Transfers
If your answer to question number 7 is 0, then answer the following questions.
If this does not apply to you please proceed to Question 8 below
8. How long does it typically take to travel from home to the origin transit stop/station (Access Time)? (Excluding any
stops you make (e.g. to pick up a coffee))
Access Time: _________________________ minutes
9. How long does it typically take to wait at the origin transit stop/station (Waiting Time)?
Waiting Time: _________________________ minutes
10. How long does it typically take to travel from the origin transit stop/station to the destination transit stop/station (In-
Vehicle Travel Time)?
In-Vehicle Travel Time: _________________________ minutes
COmmuting Survey for MOde Shift (COSMOS)
200
11. How long does it typically take to travel from the destination transit stop/station to the final workplace destination
(Egress Time)? (Excluding any stops you make (e.g. to pick up a coffee))
Egress Time: _________________________ minutes
If your answer to question number 37 is not 0, then answer the following questions.
If this does not apply to you please proceed to Question 16 below (varies based on number of transfers)
8. How long does it typically take to travel from home to the origin transit stop/station (Access Time)? (Excluding any
stops you make (e.g. to pick up a coffee))
Access Time: _________________________ minutes
9. How long does it typically take to wait at the origin transit stop/station (Waiting Time)?
Waiting Time: _________________________ minutes
10. How long does it typically take to travel from the origin transit stop/station to the destination transit stop/station (In-
Vehicle Travel Time)?
In-Vehicle Travel Time: _________________________ minutes
11. What transit mode do you typically take to travel from the origin transit stop/station to the following transfer
stop/station?
Public Transit Mode: □Streetcar □Bus □Subway
12. How long does it typically take to transfer between modes, including access and waiting times for the next public
transit unit (Transfer Time)?
Transfer Time: _________________________ minutes
13. How long does it typically take to travel from the transfer transit stop/station to the following transfer/destination
stop/station (In-Vehicle Travel Time)?
In-Vehicle Travel Time: _________________________ minutes
14. What transit mode do you typically take to travel from the transfer transit stop/station to the following
transfer/destination stop/station
Public Transit Mode: □Streetcar □Bus □Subway
15. How long does it typically take to travel from the destination transit stop/station to the final workplace destination
(Egress Time)? (Excluding any stops you make (e.g. to pick up a coffee))
Egress Time: _________________________ minutes
16. How much do you typically pay for your one-way transit trip to work? (If you pay for more than one ticket, please
sum)
Transit Fare: _________________________ $/trip
COmmuting Survey for MOde Shift (COSMOS)
201
17. In case of unavailability of the Ride All Way option, what would be your second choice? (Select one choice only
under chosen mode. For transit users who make transfer(s), please choose the mode you use to travel the longest
distance under transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
Chosen Mode Transit Technology
Car Options:
□Car Driver
□Car Passenger
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Cycle
□Walk
□Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
Section for Cycle: If your answer to question number 4 is Cycle, then answer the following questions.
If this does not apply to you please proceed to section of Walk.
5. What is your typical one-way travel time to work (door to door)?
Travel Time: _________________________ minutes
6. During which months do you typically cycle when commuting to work?
(Select all that apply)
□January □July
□February □August
□March □September
□April □October
□May □November
□June □December
7. In case of unavailability of the Cycle option, what would be your second choice? (Select one choice only under chosen
mode. For transit users who make transfer(s), please choose the mode you use to travel the longest distance under
transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
Chosen Mode Transit Technology
Car Options:
COmmuting Survey for MOde Shift (COSMOS)
202
□Car Driver
□Car Passenger
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Walk
□ Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
8. If you were to take public transit, how long would it take to travel from home to the origin transit stop/station
(Access Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Access Time: _________________________ minutes
9. If you were to take public transit, how long would it take to wait/transfer at transit stop(s)/station(s)
(Waiting/Transferring Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Waiting Time: _________________________ minutes
10. If you were to take public transit, how long would it take from the origin transit stop/station to the destination
transit stop/station (In-Vehicle Travel Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
In-Vehicle Travel Time: _________________________ minutes
11. If you were to take public transit, how long would it take to travel from the destination transit stop/station to the final
workplace destination (Egress Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Egress Time: _________________________ minutes
12. If you were to take public transit, how much would it cost you per work trip (one-way)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Transit Travel Cost: _________________________ $/trip
COmmuting Survey for MOde Shift (COSMOS)
203
13. If you were to take public transit, what mode would you use based on the travel time and cost you have just provided?
(If you were to combine more than one transit route/line, choose the mode you were to use to travel the longest
distance).
Public Transit Mode: □Streetcar □Bus □Subway
14. In the last 12 months, how often have you used public transit to commute to work?
□ More than once a week on average
□ Between once a month and once a week on average
□ Less than once a month on average
□ Never
Section for Walk: If your answer to question number 4 is Walk, then answer the following questions.
If this does not apply to you please proceed to section of Other
5. What is your typical one-way travel time to work (door to door)?
Travel Time: _________________________ minutes
6. During which months do you typically walk when commuting to work?
(Select all that apply)
□January □July
□February □August
□March □September
□April □October
□May □November
□June □December
7. In case of unavailability of the Walk option, what would be your second choice? (Select one choice only under chosen
mode. For transit users who make transfer(s), please choose the mode you use to travel the longest distance under
transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
Chosen Mode Transit Technology
Car Options:
□Car Driver
□Car Passenger
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Cycle
□Other, please specify:
COmmuting Survey for MOde Shift (COSMOS)
204
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
8. If you were to take public transit, how long would it take to travel from home to the origin transit stop/station
(Access Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Access Time: _________________________ minutes
9. If you were to take public transit, how long would it take to wait/transfer at transit stop(s)/station(s)
(Waiting/Transferring Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Waiting Time: _________________________ minutes
10. If you were to take public transit, how long would it take from the origin transit stop/station to the destination
transit stop/station (In-Vehicle Travel Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
In-Vehicle Travel Time: _________________________ minutes
11. If you were to take public transit, how long would it take to travel from the destination transit stop/station to the final
workplace destination (Egress Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Egress Time: _________________________ minutes
12. If you were to take public transit, how much would it cost you per work trip (one-way)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Transit Travel Cost: _________________________ $/trip
13. If you were to take public transit, what mode would you use based on the travel time and cost you have just provided?
(If you were to combine more than one transit route/line, choose the mode you were to use to travel the longest
distance).
Public Transit Mode: □Streetcar □Bus □Subway
14. In the last 12 months, how often have you used public transit to commute to work?
□ More than once a week on average
□ Between once a month and once a week on average
□ Less than once a month on average
□ Never
COmmuting Survey for MOde Shift (COSMOS)
205
Section for Other: If your answer to question number 4 is Other, then answer the following questions.
If this does not apply to you please proceed to Section B
5. What is your typical one-way travel time to work (door to door)?
Travel Time: _________________________ minutes
6. What is your typical travel cost per work trip (including fuel and toll if any, and excluding parking)?
Travel Cost: _________________________ $/trip
7. What is the cost of your parking space per work trip?
Parking Cost: _________________________ $/trip
8. During which months do you typically use the mode you entered when commuting to work? (Select all that apply)
□January □July
□February □August
□March □September
□April □October
□May □November
□June □December
9. In case of unavailability of the Other option, what would be your second choice? (Select one choice only under chosen
mode. For transit users who make transfer(s), please choose the mode you use to travel the longest distance under
transit technology). Those unavailable for selection, including your first choice, are disabled.
Note: refer to definitions below for clarification of some modes of travel.
Chosen Mode Transit Technology
Car Options:
□Car Driver
□Car Passenger
□Carpool Public Transit Options:
□Ride all way □Streetcar □Bus □Subway
□Park & Ride □Streetcar □Bus □Subway
□Kiss & Ride □Streetcar □Bus □Subway
□Carpool & Ride □Streetcar □Bus □Subway
□Cycle & Ride □Streetcar □Bus □Subway
Non-Motorized Options:
□Cycle
□Other, please specify:
Car Passenger: two or more adults sharing a single vehicle within the same household (Intra-household). Carpool:
two or more adults from different households sharing a single vehicle (Inter-household).
Park & Ride: the combination of car driver and using public transit.
Kiss & Ride: the combination of car passenger and using public transit.
Carpool & Ride: the combination of carpooling and using public transit.
Cycle & Ride: the combination of cycling and using public transit.
COmmuting Survey for MOde Shift (COSMOS)
206
10. If you were to take public transit, how long would it take to travel from home to the origin transit stop/station
(Access Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Access Time: _________________________ minutes
11. If you were to take public transit, how long would it take to wait/transfer at transit stop(s)/station(s)
(Waiting/Transferring Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Waiting Time: _________________________ minutes
12. If you were to take public transit, how long would it take from the origin transit stop/station to the destination
transit stop/station (In-Vehicle Travel Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
In-Vehicle Travel Time: _________________________ minutes
13. If you were to take public transit, how long would it take to travel from the destination transit stop/station to the final
workplace destination (Egress Time)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Egress Time: _________________________ minutes
14. If you were to take public transit, how much would it cost you per work trip (one-way)?
Please give your best estimate. Accuracy is not important. If you are unsure, press "I don't know" below the form.
For returning respondents, if you've already answered this question press "clear" to erase it from database.
Transit Travel Cost: _________________________ $/trip
15. If you were to take public transit, what mode would you use based on the travel time and cost you have just provided?
(If you were to combine more than one transit route/line, choose the mode you were to use to travel the longest
distance).
Public Transit Mode: □Streetcar □Bus □Subway
16. In the last 12 months, how often have you used public transit to commute to work?
□ More than once a week on average
□ Between once a month and once a week on average
□ Less than once a month on average
□ Never
COmmuting Survey for MOde Shift (COSMOS)
207
Section B. Stated Choice Experiment
In this section, you are provided with 6 hypothetical scenarios. In each scenario, you are making your usual trip from home
to work and are asked to choose which mode of travel you would use given different situations and transit service
attributes. Please take your time and consider each situation carefully.
Mode Descriptions
In addition to the regular Streetcar, Bus and Subway options, the survey considers two new rapid transit options: Bus
Rapid Transit (BRT) and Light Rail Transit (LRT).
Bus Rapid Transit (BRT) is a distinctive, frequent and limited-stop bus service that is designed and operated like a rail
line. BRT buses operate on regular roads with dedicated right-of-ways, transit priority at traffic signals and other enhanced
features such as improved passenger waiting areas and stops.
Light Rail Transit (LRT) is a modern electric railway system that falls somewhere in between subway and streetcar
systems in terms of performance. LRT is characterized by high capacity, spacious, quiet, and comfortable vehicles. LRT
can also be operated along a separate right-of-way without interference from other traffic either at ground level, elevated
on aerial structures or even in tunnels.
Note that Scarborough Rapid Transit (SRT) is considered part of the Subway category in this study.
COmmuting Survey for MOde Shift (COSMOS)
208
Useful Definitions
Transit Trip Travel Time is typically divided into the following components:
Access Time: Time taken to travel from home to the origin transit stop/station.
Waiting Time: Time taken to wait at a transit stop/station.
In-Vehicle Travel Time: Time taken to travel from the origin transit stop/station to the destination transit stop/station.
Transfer Time: Time taken to transfer between different transit units.
Egress Time: Time taken to travel from the destination transit stop/station to the final workplace destination.
Transit Right-of-Way (R.O.W.) is the physical space on which a transit line operates, and can be categorized as follows:
R.O.W. Category C (Shared R.O.W.): Transit routes/lines operated on a shared corridor with car traffic (e.g. Buses,
College Streetcar, Queen Streetcar).
R.O.W. Category B (Dedicated R.O.W.): Transit routes/lines operated on a dedicated longitudinal corridor; however, they
only share the intersections with car traffic (e.g. Spadina Streetcar, St. Clair Streetcar).
R.O.W. Category A (Exclusive R.O.W.): Transit routes/lines operated on an exclusive corridor without any interference
from car traffic (e.g. Bloor-Danforth Subway, Yonge University Spadina Subway).
COmmuting Survey for MOde Shift (COSMOS)
209
Choice Task #1/6 (Sample for non-transit users)
If the current choice is no longer available, please consider the following alternative choices and select the one that you
would use to make your current work trip based on mode features presented in the table below.
Factor Current
Choice
Alternative
Choice
Car
Option
Car Driver
Car
Option
Car Driver
Public Transit
Option
Bus Rapid Transit (BRT), R.O.W. B
Travel Cost/Fare
($/One-Way Trip)
Current +25% +10%
Auto Parking Cost
($/One-Way Trip)
Current Current ---
Access Time
(min/One-Way Trip)
--- --- -50% of Typical
Waiting & Transfer Time
(min/One-Way Trip)
--- --- - 50%
In-Vehicle Travel Time
(min/One-Way Trip)
Current +50% Current
Egress Time
(min/One-Way Trip)
--- --- Typical
Park & Ride Availability
(Yes/No)
--- --- No
Crowding Level
(Low, Medium, High)
--- --- Moderately Crowded
(No seats available)
Number of Transfers
(0, 1, 2 or more)
--- --- 1
On-Time Performance
(Early, On-Time, Late)
--- --- On Time
Schedule Information
(Yes/No)
--- --- Yes
Real-Time Information about Delays
(Yes/No)
--- --- Yes
Mode Shift Given the alternative modal characteristics presented above, which option would you choose?
Car option Shift to public transit option Shift to Other Mode,
(Car Driver) (Bus Rapid Transit (BRT), R.O.W. B) please specify: _____________
□ □ □
Willingness to Comply In the future, what would be your propensity to make your work trip using the option selected above?
Note: click on the button that matches your personal agreement, indicating how much you are willing to adhere to
your choice above.
Very
Weak
Moderately
Weak
Neutral Moderately
Strong
Very
Strong
Willingness to comply □ □ □ □ □
COmmuting Survey for MOde Shift (COSMOS)
210
Choice Task #1/6 (Sample for transit users)
If the current choice is no longer available, please consider the following alternative choices and select the one that you
would use to make your current work trip based on mode features presented in the table below.
Factor Current
Choice Alternative Choice
Public Transit
Option
Streetcar
Public Transit
Option
Light Rail Transit (LRT), R.O.W. A
Travel Cost/Fare
($/One-Way Trip)
Current +10%
Auto Parking Cost
($/One-Way Trip)
--- ---
Access Time
(min/One-Way Trip)
Current Typical
Waiting & Transfer Time
(min/One-Way Trip)
Current - 50%
In-Vehicle Travel Time
(min/One-Way Trip)
Current - 20%
Egress Time
(min/One-Way Trip)
Current - 50% of Typical
Park & Ride Availability
(Yes/No)
--- Yes
Crowding Level
(Low, Medium, High)
--- Uncrowded
(seats available)
Number of Transfers
(0, 1, 2 or more)
Current 0
On-Time Performance
(Early, On-Time, Late)
--- On Time
Schedule Information
(Yes/No)
--- Yes
Real-Time Information about Delays
(Yes/No)
--- Yes
Mode Shift Given the alternative modal characteristics presented above, which option would you choose?
Public transit option Shift to Other Mode,
(Light Rail Transit (LRT), R.O.W. A) please specify: _____________
□ □
Willingness to Comply In the future, what would be your propensity to make your work trip using the option selected above?
Note: click on the button that matches your personal agreement, indicating how much you are willing to adhere to
your choice above.
Very
Weak
Moderately
Weak
Neutral Moderately
Strong
Very
Strong
Willingness to comply □ □ □ □ □
COmmuting Survey for MOde Shift (COSMOS)
211
Section C. Behavioural Factors
In this section, behavioural factors regarding transportation mode choice will be collected in terms of habitual behaviour,
emotional response and personal attitude.
Habitual Behaviour
The following section will collect information regarding your typical travel behaviour related to mode choice.
The following is a list of activities that require some travel. Please choose the typical mode you use for each activity.
Note:
Select only one mode for each given activity. For trips with multiple modes, please choose the mode you use to travel
the longest distance.
Answer this section quickly and without giving it too much thought. Your first impression will best reflect typical
behaviour.
Activity Mode
Car
Driver
Car
Passenger Carpool Streetcar Bus Subway Cycle Walk Other
To visit friends □ □ □ □ □ □ □ □ □ To visit family □ □ □ □ □ □ □ □ □ To go shopping □ □ □ □ □ □ □ □ □ To go to dinner with
family at a restaurant □ □ □ □ □ □ □ □ □
To go to play sports □ □ □ □ □ □ □ □ □ To go to a park □ □ □ □ □ □ □ □ □ To go fishing on weekend □ □ □ □ □ □ □ □ □ To go to the movies □ □ □ □ □ □ □ □ □ To go to a party □ □ □ □ □ □ □ □ □
COmmuting Survey for MOde Shift (COSMOS)
212
Emotional Response
In this section you are required to rank different modes of transportation relative to two opposite words.
Note: the following is an EXAMPLE designed to demonstrate how questions in this section should be properly
answered.
For Example
If your transportation mode is closely related to one of the words
White :_X_: :___: :___: :___: :___: :___: :___: Black
Or
White :___: :___: :___: :___: :___: :___: :_X_: Black
If your transportation mode is generally related to one of the words
White :___: :_X_: :___: :___: :___: :___: :___: Black
Or
White :___: :___: :___: :___: :___: :_X_: :___: Black
If your transportation mode is somewhat related to one of the words
White :___: :___: :_X_: :___: :___: :___: :___: Black
Or
White :___: :___: :___: :___: :_X_: :___: :___: Black
If your transportation mode is not related to either word
White :___: :___: :___: :_X_: :___: :___: :___: Black
1. The following is a list of 16 adjectives that may describe the mode of transportation that you usually take to work.
Please describe your mode in terms of the adjectives and scale below.
Note: please do not look back and forth or try to remember what you have answered previously or change any
previous answers.
Good :___: :___: :___: :___: :___: :___: :___: Bad
Complex :___: :___: :___: :___: :___: :___: :___: Simple
Strong :___: :___: :___: :___: :___: :___: :___: Weak
Comfortable :___: :___: :___: :___: :___: :___: :___: Uncomfortable
Safe :___: :___: :___: :___: :___: :___: :___: Unsafe
Pleasant :___: :___: :___: :___: :___: :___: :___: Unpleasant
Flexible :___: :___: :___: :___: :___: :___: :___: Inflexible
Clean :___: :___: :___: :___: :___: :___: :___: Dirty
Noisy :___: :___: :___: :___: :___: :___: :___: Quite
Big :___: :___: :___: :___: :___: :___: :___: Small
Fast :___: :___: :___: :___: :___: :___: :___: Slow
Active :___: :___: :___: :___: :___: :___: :___: Inactive
Crowded :___: :___: :___: :___: :___: :___: :___: Empty
Clear :___: :___: :___: :___: :___: :___: :___: Unclear
Popular :___: :___: :___: :___: :___: :___: :___: Unpopular
Great :___: :___: :___: :___: :___: :___: :___: Little
COmmuting Survey for MOde Shift (COSMOS)
213
2. The following is a list of 8 adjectives that generally describe public transit.
Please describe public transit in terms of the adjectives and scale below.
Note: please do not look back and forth or try to remember what you have answered previously or change any
previous answers.
Personal Attitude
In this section, information regarding your personal attitude towards different transportation modes will be collected.
Note: click on the button that matches your personal agreement, indicating how much you agree or disagree.
Section D. Socioeconomic/Demographic Information
In this section, socioeconomic and demographic information about you will be collected.
1. What is your gender?
□ Male □ Female
2. What is your age?
Age: _________________________ Years
3. What is your marital status?
□Single □Married □Divorced □Widowed
4. What is your occupation?
Occupation: _________________________________
Reliable :___: :___: :___: :___: :___: :___: :___: Unreliable
Convenient :___: :___: :___: :___: :___: :___: :___: Inconvenient
Frequent :___: :___: :___: :___: :___: :___: :___: Infrequent
Efficient :___: :___: :___: :___: :___: :___: :___: Inefficient
Organized :___: :___: :___: :___: :___: :___: :___: Disorganized
Significant :___: :___: :___: :___: :___: :___: :___: Insignificant
Bright :___: :___: :___: :___: :___: :___: :___: Dark
Expensive :___: :___: :___: :___: :___: :___: :___: Cheap
Strongly
Disagree Disagree Neutral Agree
Strongly
Agree
In general, a car is a good mode for work trips. □ □ □ □ □ For me, it is important to use a car to get to work. □ □ □ □ □ In general, public transit is a good mode for work trips. □ □ □ □ □ For me, a public transit system is important to get to work. □ □ □ □ □
COmmuting Survey for MOde Shift (COSMOS)
214
5. What word best describes your home?
□House □Townhouse □Apartment
6. Besides yourself, how many people older than 18 live in your home?
Number of people: _________________________ People
7. How many people under 18 live in your home?
Number of people: _________________________ People
8. How many cars are there at your home?
Number of cars: ______________________ Cars
9. Do you have a driving’s licence?
□ Yes □ No
10. What is your total personal income range per year?
□Less than $10,000 □$60,000 to $69,999
□$10,000 to $19,999 □$70,000 to $79,999
□$20,000 to $29,999 □$80,000 to $89,999
□$30,000 to $39,999 □$90,000 to $99,999
□$40,000 to $49,999 □$100,000 and over
□$50,000 to $59,999
Closing
Thank you for completing the socio-psychometric survey. For more information, please contact Ahmed Osman
Idris at [email protected].