MODELLING COMMUTERS · 2018. 10. 10. · MODELLING COMMUTERS’ MODE CHOICE: INTEGRATING TRAVEL...

MODELLING COMMUTERS’

MODE CHOICE:

INTEGRATING TRAVEL BEHAVIOUR,

STATED PREFERENCES, PERCEPTION,

AND SOCIO-ECONOMIC PROFILE

Puteri Paramita

BSc.(Hons) Statistics and Mathematics, MSc. Transport Planning, AFHEA

Dr. MD Mazharul Haque, Professor Stephen Kajewski, Dr. Zuduo Zheng, Professor

Simon Washington, Professor Paul Hyland

Submitted in fulfilment of the requirements for the degree of

Doctor of Philosophy

School of Civil Engineering and the Built Environment

Science and Engineering Faculty

Queensland University of Technology

2018

i

Keywords

Binomial logit, bus, car, commuter, comparison study, discrete choice experiment

(DCE), fixed parameter, heterogeneity, mixed logit model, mode choice behaviour,

mode shift behaviours, multinomial logit, nested logit, ordered logit, perception,

policy intervention, qualitative assessment, quantitative assessment, random

parameter, Revealed Preference, socio-economic profiles, Stated Preference,

satisfaction, shift behaviour, train, train fare, travel behaviour, traveller, trip

characteristics, urban traveller.

ii

Abstract

This study focuses on identifying the factors that influence travellers’ transport

mode choice, and on understanding the impact of these factors on their decision-

making process. The relationship between travellers’ socio-economic profiles and trip

characteristics and their chosen mode are quantified using the data collected from

urban travellers in five Australian capital cities: Sydney, Melbourne, Brisbane,

Adelaide, and Perth. Specifically, this study comprises three interconnected sub-

studies in the following order: Urban travellers’ satisfaction with train fares in five

Australian cities, Consistency between perceptions and stated preferences data in a

nationwide mode choice experiment, and Policy interventions study to encourage

behavioural shift from car to public transport. The second sub-study is based on the

findings of the first, while the third incorporates the findings of the first and second

studies.

In the public transport industry, travellers’ perceived satisfaction is a key

element in understanding their evaluation of, and loyalty to ridership. Despite its

notable importance, studies of customer satisfaction are under-represented in the

literature, and the most previous studies are based on survey data collected from a

single city only. This does not allow a comparison across different transport systems.

To address this underrepresentation, the first sub-study is a comparative analysis of

user satisfaction with train fares for their most recent home-based train trip in five

Australian capital cities. Two data sources are used: a nationwide survey, and objective

information on the train fare structure in each of the targeted cities. In particular,

satisfaction with train fares is modelled as a function of socio-economic factors and

train trip characteristics, using a random parameters ordered logit model that accounts

for unobserved heterogeneity in the population.

Results of this first sub-study indicate that gender, city of origin, transport mode

from home to the train station, eligibility for a student or senior concession fare, one-

way cost, waiting time, and five diverse interaction variables between city of origin

and socio-economic factors are the key determinants of passenger satisfaction with

train fares. In particular, this sub-study reveals that female respondents tend to be less

satisfied with their train fare than their male counterparts. Interestingly, respondents

who take the bus to the train station tend to feel more satisfied with their fare compared

with the rest of the respondents. In addition, notable heterogeneity is detected across

iii

respondents’ perceived satisfaction with train fare, specifically with regard to the one-

way cost and the waiting time incurred. An intercity comparison reveals that a city’s

train fare structure also significantly affects a traveller’s perceived satisfaction with

their train fare. The findings of this study are significant for both policy makers and

transport operators, allowing them to understand traveller behaviours, and to

subsequently formulate effective transit policies.

The second study assesses the consistency between respondents’ perceptions and

their stated preferences in a stated preference (SP) experiment. A nationwide survey

of urban travellers in five Australian capital cities yielded their perceptions of various

service factors—representing respondent revealed preferences (RPs). These same

respondents also completed SP experiment. Two random parameter logit models were

developed to understand respondents’ mode choice behaviour: one using their

perceptions, and one using their responses to the SP experiment. The consistency

between these two models in explaining respondents’ mode choice behaviours are

assessed both qualitatively and quantitatively. At the qualitative level, the factor

mapping of the two models shows that respondents’ perceptions of four service factors

– such as train waiting time, on-board crowding, the availability of a laptop station,

and increased road congestion, and their corresponding attributes in SP experiment,

are well aligned. At the quantitative level, the marginal utilities of choosing the train

mode in these two models are tested through rigorous numerical simulations for the

same four service factors, and the corresponding estimated probabilities of choosing

the train mode from the SP model are found to be similar to those estimated from the

RP model.

The third and final sub-study focuses on the impact of transit policy interventions

on mode shift. By utilising datasets from urban travellers in Sydney, Melbourne, and

Brisbane, this third study determines the socio-economic factors and travel attributes

that significantly influence travellers’ mode choice in each city. A nested logit model

is estimated to identify targets for policy interventions that could influence travellers

to shift from car to public transport. The nested structure used consists of a public

transport branch, which includes bus and train alternatives, and a degenerative private

transport branch. Each best-fitted nested logit model consists of four utility functions.

Each of these functions is characterized by key travel attributes and socio-economic

factors. The travel attribute elements are useful in identifying policy intervention

targets, while the socio-economic factor elements are important to an understanding

iv

of the underlying profile of the travellers who are most affected by changes in transport

policies.

This final study also simulates and analyses over one hundred policy intervention

scenarios to optimise the mode change from car to public transport in each city. These

interventions are divided into two categories: those that encourage public transport

ridership by improving passenger service quality; and those that discourage regular car

usage by increasing the cost of that usage. Based on a comprehensive understanding

of target traveller profiles and travel behaviour, the simulation studies demonstrate that

the most efficient transport policy interventions are a combination of both intervention

categories. The successful replication of the study in three different cities provides

sufficient evidence that the overall framework could be applied to many other urban

traveller datasets.

v

Table of Contents

Keywords .................................................................................................................................. i

Abstract .................................................................................................................................... ii

Table of Contents ......................................................................................................................v

List of Figures ........................................................................................................................ vii

List of Tables .......................................................................................................................... ix

List of Abbreviations ................................................................................................................x

Statement of Original Authorship ........................................................................................... xi

Acknowledgements ................................................................................................................ xii

Chapter 1: Introduction ...................................................................................... 1

1.1 Background and context .................................................................................................1

1.2 Research question formulation .......................................................................................5

1.3 Study objectives, innovations, and contributions ...........................................................7

1.4 Research scopes ..............................................................................................................9

1.5 Thesis outline ................................................................................................................10

1.6 Publications from this study .........................................................................................11

Chapter 2: Literature Review ........................................................................... 13

2.1 Urban travellers’ satisfaction with train fares in five Australian cities.........................14

2.2 Consistency between perceptions and stated preferences data in a nationwide mode

choice experiment ...................................................................................................................17

2.3 Policy interventions study to encourage behavioural shift from car to public

transport ..................................................................................................................................20

Chapter 3: Dataset ............................................................................................. 31

3.1 Data collection ..............................................................................................................31

3.2 Mode choice experiment structure ................................................................................33

3.3 Data description ............................................................................................................35

3.4 Data utilisation ..............................................................................................................39

3.5 Train riders dataset .......................................................................................................41

3.6 Factor mapping of perceptions and attributes of mode choice experiment ..................46

3.7 Dataset for transit policy interventions study ...............................................................47

Chapter 4: Methodology .................................................................................... 49

4.1 Urban travellers’ satisfaction with train fares in five Australian cities.........................49

4.2 Consistency between perceptions and stated preferences data in a nationwide mode

choice experiment ...................................................................................................................51

4.3 Policy interventions study to encourage behavioural shift from car to public

transport ..................................................................................................................................56

vi

Chapter 5: Urban travellers’ satisfaction with train fares in five Australian

cities 59

5.1 Train fare structures in the five Australian cities ......................................................... 59

5.2 Modelling results .......................................................................................................... 62

5.3 Key socio-economic factors ......................................................................................... 64

5.4 Key Trip Characteristics .............................................................................................. 65

5.5 Interaction variables ..................................................................................................... 68

5.6 Heterogeneity ............................................................................................................... 71

5.7 Intercity comparison .................................................................................................... 77

5.8 Conclusions .................................................................................................................. 78

Chapter 6: Consistency between perceptions and stated preferences data in

a nationwide mode choice experiment .................................................................... 81

6.1 The random-parameters binomial logit model for perceptions ................................... 81

6.2 The mixed logit model for the SP data ......................................................................... 83

6.3 Qualitative assessment ................................................................................................. 85

6.4 Quantitative assessment ............................................................................................... 87

6.5 Conclusions .................................................................................................................. 98

Chapter 7: Policy interventions study to encourage behavioural shift from

car to public transport ........................................................................................... 101

7.1 Modelling results ........................................................................................................ 101

7.2 The goodness-of-fit test of the nested structure ......................................................... 103

7.3 The lower nest of the nested structure ........................................................................ 104

7.4 The upper nest of the nested structure ........................................................................ 107

7.5 The probability functions ........................................................................................... 109

7.6 The average profile of travellers for the baseline scenario ........................................ 110

7.7 The policy intervention scenario analysis .................................................................. 113

7.8 The discussions of combined policy intervention scenarios ...................................... 127

7.9 Conclusions ................................................................................................................ 128

Chapter 8: Conclusions.................................................................................... 131

8.1 Overarching conclusions ............................................................................................ 131

8.2 Contributions and policy implications ....................................................................... 134

8.3 Limitations of this study ............................................................................................ 135

8.4 Recommendations for future study ............................................................................ 136

Appendices .............................................................................................................. 137

Appendix A Urban Rail Travel Behaviour, Web-based Survey .......................................... 137

Bibliography ........................................................................................................... 139

vii

List of Figures

Figure 1.1 : The three interconnected sub-studies ....................................................... 7

Figure 2.1 : Body of literature of Travel Behaviour field. ......................................... 13

Figure 2.2 : Three guides underlying the theory of human behaviour (Ajzen,

1985) ............................................................................................................ 21

Figure 2.3 : The micro and macro process of the intervention method (Bamberg

& Schmidt, 1998) ......................................................................................... 25

Figure 3.1 : The diagram of the collected dataset ...................................................... 32

Figure 3.2 : The diagram of data utilisation ............................................................... 40

Figure 3.3 : Satisfaction with train fare in each city .................................................. 43

Figure 3.4 : The diagram of data utilisation in the transit policy interventions

study ............................................................................................................. 47

Figure 4.1 : The nested structure of the mode choice experiment ............................. 57

Figure 5.1 : The simulated marginal utility of one-way cost for 100 randomly

selected individuals ...................................................................................... 73

Figure 5.2 : The simulated marginal utility of waiting time for 100 randomly

selected individuals ...................................................................................... 75

Figure 6.1 : The simulated probability of individuals’ perception of train

running on schedule from the RP model ..................................................... 88

Figure 6.2 : The simulated probability of train waiting time from the SP model ...... 88

Figure 6.3 : The difference in probability value of the impact of a one unit

increase in individuals’ perception level of train running on schedule ....... 90

Figure 6.4 : The difference in probability value of the impact of one unit

increase in train waiting time in the SP experiment .................................... 90

Figure 6.5 : The simulated probability of individuals’ perception of the

probability of getting a seat from the RP model.......................................... 91

Figure 6.6 : The difference in probability value of the impact of a one unit

increase in individuals’ perception level of the probability of getting a

seat ............................................................................................................... 92

Figure 6.7 : The simulated probability of using a car for individuals

experiencing two different periods of car on-board time from the SP

model............................................................................................................ 95

Figure 6.8 : The corresponding simulated probability of taking the train for

individuals who experience two different periods of car on-board time from the SP model ....................................................................................... 95

Figure 6.9 : The difference in the probability value of using a car as the impact

of a one unit increase in car on-board time in the SP experiment ............... 96

Figure 6.10 : The difference in the probability value of taking the train as the

impact of a one unit increase in car on-board time in the SP

experiment.................................................................................................... 97

viii

Figure 7.1 : The probability of driving a car in Sydney as the result of policy

intervention related to bus and train waiting times and parking and toll

costs, while holding all other factors constant ........................................... 118

Figure 7.2 : The probability of taking public transport in Sydney as the result of

policy interventions in both bus and train waiting times and parking

and toll costs, holding all other factors constant ........................................ 118

Figure 7.3 : The probability of driving a car in Melbourne as the result of

policy interventions in bus and train waiting time and parking and toll

costs, holding all other factors constant ..................................................... 122

Figure 7.4 : The probability of taking public transport in Melbourne as the

result of policy interventions in bus and train waiting time and parking

and toll costs, holding all other factors constant ........................................ 122

Figure 7.5 : The probability of driving a car in Brisbane as the result of policy

interventions in bus and train waiting time and parking and toll cost,

holding all other factors constant ............................................................... 126

Figure 7.6 : The probability of taking public transport in Brisbane as the result

of policy interventions in bus and train waiting times and parking and

toll costs, holding all other factors constant ............................................... 126

ix

List of Tables

Table 3.1 The breakdown of train riders and non-riders across five cities .............. 33

Table 3.2 Attributes and their levels employed in the mode choice experiments ..... 34

Table 3.3 Socio-economic profiles of respondents from each city............................ 35

Table 3.4 Respondents’ perceptions of various service factors in each city ............ 36

Table 3.5 Trip characteristics of train riders from each city (Non-riders) .............. 37

Table 3.6 Socio-economic profiles of respondents from each city (Train riders) .... 43

Table 3.7 Trip characteristics of train riders from each city (Train riders) ............ 44

Table 3.8 Factor mapping of perceptions and attributes of mode choice

experiment .................................................................................................... 46

Table 5.1 Summary of the best fixed-parameter and random-parameter logit

models .......................................................................................................... 62

Table 5.2 Distribution of the random parameters .................................................... 63

Table 5.3 Heterogeneities in the random parameters ............................................... 64

Table 6.1 Summary of the best-fitted random-parameter binomial logit model ....... 81

Table 6.2 Summary of the best-fitted mixed logit model ........................................... 83

Table 6.3 Heterogeneities in car on-board time of car utility function within

best-fitted mixed logit model ........................................................................ 84

Table 6.4 Mapping of significant variables of the random-parameter binomial

logit (RP) model against the mixed logit (SP) model .................................. 85

Table 7.1 Summary of the best-fitted FIML of nested logit model .......................... 101

Table 7.2 The average profile of Sydney travellers for a baseline scenario........... 110

Table 7.3 The average profile of Melbourne travellers for a baseline scenario .... 111

Table 7.4 The average profile of Brisbane travellers for a baseline scenario ....... 112

Table 7.5 The probability values of driving a car and taking public transport

in Sydney: 176 different policy intervention scenarios .............................. 114


in Melbourne: 176 different policy intervention scenarios........................ 119


in Brisbane: 176 different policy intervention scenarios ........................... 123

x

List of Abbreviations

CEBE Civil Engineering and the Built Environment

CRC Cooperative Research Centres

CWA Cognitive Work Analysis

DCE Discrete Choice Experiment

FIML Full Information Maximum Likelihood

HDR Higher Degree Research

IV Inclusive Value

LOS Level of Service

MMNL Mixed Multinomial Logit

MNL Multinomial Logit

QUT Queensland University of Technology

RP Revealed Preference

SEF Science and Engineering Faculty

SP Stated Preference

TPB Theory of Planned Behaviour

QUT Verified Signature

xii

Acknowledgements

This journey would not be possible without the support of a number of people.

My deep appreciation to my supervisory team: Dr. Zuduo Zheng, Dr. MD Mazharul

Haque, Professor Stephen Kajewski, Professor Simon Washington, Professor Paul

Hyland. Zuduo and Shimul, I thank for believing in me, for your encouragement,

patience and understanding. You always provided me guidance and practical solutions

throughout the whole process. It is because of you I learned how to think critically and

analytically; I believe this has helped me tremendously to become a better researcher.

Professor Simon Washington and Professor Paul Hyland, I appreciate the valuable

advices and clear directions that you provided at the beginning of my study. Dr. Jake

Whitehead and Dr. Ashish Bhaskar, thank you for your supports and valuable

constructive feedbacks shared in my final seminar. Zuduo, Professor Simon

Washington and Professor Paul Hyland, my sincere gratitude for giving me the

permission to utilize the data gathered as part of Project R1.130 Understanding Urban

Rail Travel for Improved Patronage Forecasting funded by the CRC for Rail

Innovation (established and supported under the Australian Government's Cooperative

Research Centres program). A warm thank you to Professor Stephen Kajewski who

provided me with great supports from the School of Civil Engineering and the Built

Environment (CEBE). I also acknowledge and appreciate the assistance of the

professional editor (Ms. Denise Scott) for her timely supports in proofreading, editing,

and formatting of this thesis.

Thank you to supportive team from the Research Student Centre and Ms. Tiziana

La Mendola, a Higher Degree Research Support Officer for School of CEBE, it was

always pleasant to deal with all of you. This study is funded by Science and Engineering

Faculty (SEF). I gratefully acknowledge for the financial assistance. I wish to make a

special note of appreciation for the university counsellor (Mr. Samuel Zimmer),

university General Practitioner (Dr. Rhian Kenrick), my family and my long distance

best friends. Sam and Dr. Kenrick, I genuinely thank you, without your patience to

listening to me and continuous encouragements during my difficult time, the completion

of this study would not be possible. A warm thanks to my parents, sister and close

relatives, Andrés and Marie, who have been assisting a great deal in thriving my PhD

journey. I would also like to acknowledge all those who have provided feedback for this

xiii

work. I much appreciate your time, supports and interests in this study. Last but not the

least, I gracefully thank Jesus Christ and Mother Mary for their continuous supports every

single day.

Chapter 1: Introduction 1

Chapter 1: Introduction

This chapter provides a comprehensive introduction to the study of urban travel

behaviours in five Australian capital cities: Sydney, Melbourne, Brisbane, Adelaide,

and Perth. It begins with a detailed presentation of the background and context of the

study in Section 1.1. Section 1.2 formulates the study’s main research question, while

Section 1.3 presents its objectives, innovations, and contributions. Section 1.4 covers

the detailed inclusions and exclusions of the overall PhD study. Section 1.5 includes

an outline of the remaining chapters of the thesis. Finally, Section 1.6 summarizes the

publications extracted from this study.

1.1 BACKGROUND AND CONTEXT

Urban travellers’ transport mode choice decision has been widely discussed in

Australia over recent decades (Buys & Miller, 2011; Kamruzzaman, Baker,

Washington, & Turrell, 2014; McIntosh, Newman, & Glazebrook, 2013; McMillan,

2007; Soltani & Allan, 2006; Taplin & Qiu, 1997; Zheng, Washington, Hyland, Sloan,

& Liu, 2016; Zheng et al., 2013). In particular, it has been debated whether travellers’

travel experience and/or their socio-economic profile influence their daily mode choice

decision, especially for commuting purposes. In addition, other pressing questions

concern urban travellers’ level of satisfaction with their current mode choice, and

whether they would consider a mode shift in the future – especially from private to

public transport – if there were improved transit policies in place (Corpus, 2008;

Passenger Demand Forecasting Council (PDFC), 2013; Zheng et al., 2016; Zheng et

al., 2013).

A comprehensive understanding and knowledge of the factors that influence

urban travellers’ mode shift behaviour are critical for transport authorities to formulate

the most suitable policies to encourage mode shift. It is also important that transport

authorities provide sufficient public transport services to cater for increased public

demand as a result of shifting behaviours. Concurrently, the encouragement to shift

transport mode from private to public would ensure the sustainability of Australia’s

public transport systems across (Corpus, 2008; Zheng et al., 2016; Zheng et al., 2013).

2 Chapter 1: Introduction

This study aims to thoroughly investigate and address these public transport

issues. To this end, it is comprised of three interconnected sub-studies that address, in

turn: 1) urban traveller satisfaction; 2) consistency between urban travellers’ current

travel behaviour and their future preferences; and 3) policy interventions that could

influence mode shift. To facilitate the formulation of precise research questions, each

sub-study starts with an investigation of the background and context of its topic issue.

1.1.1 Urban traveller satisfaction

Satisfaction is an essential concept in the service industry and market research,

and an important factor in understanding customer behaviour (J. de Oña & de Oña,

2014; J. de Oña, de Oña, Eboli, & Mazzulla, 2013; R. de Oña, Machado, & de Oña,

2015; Eboli & Mazzulla, 2012; Fornell, 1992; Johnson & Gustafsson, 2006; Oliver,

2014). In public transport, user satisfaction is recognized as a key link between public

transport offerings and traveller reactions to these offerings (Fellesson & Friman,

2012). In general, knowledge of the level of user satisfaction with the existing public

transport system provides valuable information for both policy makers and public

transport operators, providing the basis for the development of effective strategies for

improving traveller experience and, in turn, increasing ridership.

Only a few studies in the literature focus on user satisfaction in the context of

public transport (Efthymiou, Antoniou, Tyrinopoulos, & Skaltsogianni, 2017;

Fellesson & Friman, 2012; Hensher, 2007; Thompson & Schofield, 2007). Available

studies identify that frequency, reliability, driver behaviour, information, cleanliness,

and comfort are typical factors that influence public transport users’ satisfaction with

services (J. Bates, Polak, Jones, & Cook, 2001; Beirão & Cabral, 2007; Eboli &

Mazzulla, 2012; Friman & Gärling, 2001). Despite its enormous importance, user

satisfaction studies are generally under-represented in the literature. In addition, the

fact that most previous studies are based on survey data collected from a single city

makes a comparative analysis across different public transport systems almost

impossible. Thus, the data used in many previous studies only partially captures the

potential factors that could be significantly linked to user satisfaction. An inter-city

comparison, supplemented by information on the characteristics of the current

transport system in each city, can be valuable in revealing underlying (or even causal)

factors that contribute to user satisfaction with public transport services.


1.1.2 Consistency between urban travellers’ current travel behaviours and their

future preferences

In the literature on travel behaviour, two types of data are commonly used:

revealed preferences (RP), and stated preferences (SP). Briefly speaking, RP are

respondents’ previously made mode-choice decisions, while SP are the hypothetical

mode-choice decisions that are made in a series of SP experiment. Each data type has

its pros and cons.

RP data emphasise the actual travel behaviours and recent trip characteristics of

travellers (Hensher, 1994). Thus, RP data are generally regarded as more reliable.

However, they have been criticized for having insufficient variation in explanatory

variables, high collinearity, and an inability to integrate new scenarios that differ

substantially from the current ones (Swait, Louviere, & Williams, 1994). On the other

hand, collecting SP data is usually more convenient and less expensive. SP data are

particularly useful for investigating respondents’ sensitivity to new transport modes or

to new attributes of an existing model (because relevant RP data would not yet exist).

However, SP are less reliable than RP data because information contained in the

former pertains to hypothetical scenarios (J. J. Louviere, Hensher, & Swait, 2000). For

example, personal constraints are not considered as constraints at the time of ‘choice’,

particularly when respondents do not take the SP task seriously. The task of the analyst

is, therefore, to make the hypothetical scenarios as realistic as possible (Hensher, Rose,

& Greene, 2005).

Hensher (1994) describes two broad categories of SP responses. In the first

category, a respondent indicates his or her preferences from a set of combinations of

attributes via a rating scale. The second category is a rating scale category that is

included when a respondent chooses only one of the combinations of attributes, and is

known as “first preference choice task”. To obtain meaningful interpretations, it is

crucial to ensure that respondents answer in a “rational” (i.e., internally consistent)

way (Miguel, Ryan, & Amaya‐Amaya, 2005). SP data is richer than RP data, and has

the ability to view the experiment as supplementary to RP data (Hensher, 1994;

Wardman, 1988). However, some SP responses might not reflect respondents’ current

preferences due to systematic bias, or to complications with the SP experiment

(Wardman, 1988). Despite the increasing usage of SP data in travel behaviour

research, there is only a small amount of empirical evidence to support the predictive


value of the hypothetical travel scenarios in similar real life situations (Lambooij et

al., 2015).

The persistent fundamental question is whether a carefully designed SP

experiment is able to elicit individuals’ true preferences, or whether the hypothetical

nature of the questions renders them irrelevant, regardless of the use of a truth-

revealing mechanism (Azevedo, Herriges, & Kling, 2003; Hensher, 1994; Wardman,

1988). Ideally, individuals’ SP responses can be compared to their observed choices

when the hypothetical scenarios are presented to them (Wardman, 1988). In practice,

however, such opportunities rarely occur, and are strictly constrained. Thus, travel

behaviour research is largely restricted to the comparison of SP related to a few unique

scenarios and the actual post-scenario behaviours (Chatterjee, Wegmann, &

McAdams, 1983; Couture & Dooley, 1981). As an alternative to these ‘before’ and

‘after’ studies, a validation test based on revealed travel behaviours and future travel

preferences can be carried out (Wardman, 1988).

Numerous validation studies comparing stated preferences and observed choices

have been undertaken in the past; these range from marketing studies (Green &

Srinivasan, 1978; Montgomery & Wittink, 1979; Parker & Srinivasan, 1976); to

healthcare studies (Lambooij et al., 2015) to transport studies. The comparison of RP

and SP models of travel behaviour research include implied values of time (Bates,

1984; Hensher, Li, & Ho, 2015; Hensher & Truong, 1985; Louviere et al., 1981), and

the comparison of the predicted (or individual choice) and real time market shares

(Benjamin & Sen, 1982; Kocur, Hyman, & Aunet, 1982; Lerman & Louviere, 1978;

J. J. Louviere & Hensher, 1982; J. J. Louviere & Kocur, 1983). Generally, findings

from these studies have encouraged on-going studies in a bid to establish a reliable

validation method (J. J. Bates, 1983; J. J. Bates & Roberts, 1986; Horowitz, 1985;

Leigh, MacKay, & Summers, 1984).

In addition, a group of researchers conducted comparison studies of SP and RP

data utilizing the willingness to pay (WTP) test (Blumenschein, Johannesson,

Yokoyama, & Freeman, 2001; Clarke, 2002; Ding, 2007; Lambooij et al., 2015; Ryan,

2004; Tselentis, Theofilatos, Yannis, & Konstantinopoulos, 2018). In summary, these

studies report that they overestimated the WTP factor in the SP experiment. Hence,

there are inconsistencies in the results of their comparative analysis.


1.1.3 Policy interventions to influence mode shift

In recent decades, trends in travel behaviour have been characterised by

increasing trip distances, and a mode shift from public transport services to private car

usage (Scheiner, 2010). A continuous increase in urban trips as the result of

urbanisation will ultimately lead to an immense increase in private car usage, and result

in both congestion and environmental concerns. In particular, an increasing level of air

pollution shows that transport is the fastest growing sector, and is responsible for

almost a quarter of all greenhouse gas emissions (Stanton et al., 2013). A shift to public

transport and other sustainable transport modes is crucial to slowing down, or even

reversing, this trend. Thus, a broad understanding of the factors that are influential in

transport mode choice and mode shift behaviour will support the ongoing promotion

of sustainable travel behaviour (Corpus, 2008). Concurrently, it will sustain public

transport services across Australia.

Numerous past studies of the travel behaviour context have investigated the

mode choice and mode shift behaviour of travellers from the perspective of various

psychological theories and intervention methods. Nevertheless, no study has yet used

a statistical model – estimated by using urban traveller datasets in an Australian capital

city – to identify key areas for targeted policy interventions, and to determine transit

policies that encourage mode shift. This study addresses this deficit by utilising urban

travellers’ mode choice responses in the three largest Australian capital cities: Sydney,

Melbourne, and Brisbane. Specifically, a nested multinomial logit (nested logit) model

is employed to estimate the utility functions of mode choice responses, and to identify

their related key travel attributes.

1.2 RESEARCH QUESTION FORMULATION

Having thoroughly identified the background and context of the three mode-

choice related issues, this study formulated the following main research question:

How do travellers’ socio-economic profiles, perceptions, and revealed travel

behaviour influence their choice of transport mode?

This question, in turn, determined the thesis title: Modelling Commuters’ Mode

Choice: Integrating Travel Behaviour, Stated Preferences, Perception, and Socio-

economic Profile. Throughout this thesis, the term ‘revealed travel behaviour’ has the


same definition as, and is used interchangeably with, the term ‘trip characteristics’.

Additionally, the term ‘mode choice experiment’ is used interchangeably with the term

‘discrete choice experiment (DCE)’ and with the term ‘SP experiment’, as well as the

term ‘commuters’ is used interchangeably with the term ‘travellers’.

To ensure the comprehensive nature of the research, an in-depth evaluation

process, the provision of significant theoretical and practical contributions, and to

ensure that all aspects of the research question were adequately addressed, the main

research question was further divided into three interconnected research sub-questions.

Each of the research sub-questions was addressed in a separate sub-study, as below:

1. What factors influence train riders’ satisfaction with the fare for their most

recent home-based train trip in five Australian capital cities: Sydney,

Melbourne, Brisbane, Adelaide, and Perth?

2. Is the carefully designed mode choice experiment able to present travellers’

true preferences? Are travellers’ perceptions of various service factors

constructively aligned, and consistent with their views of similar attributes

presented in the mode choice experiment? and

3. To what extent do the socio-economic factors and travel attributes in the mode

choice experiment influence the utility and probability values of travellers

taking the bus, train, and car, and encourage mode shift from car to public

transport service?

Based on the formulation of these sub-research questions, the three

interconnected sub-studies are:

1. Urban travellers’ satisfaction with train fares in five Australian cities

2. Consistency between perceptions and stated preferences data in a nationwide

mode choice experiment

3. Policy interventions study to encourage behavioural shift from car to public

transport

Figure 1.1 illustrates how these three sub-studies are interconnected and

cohesively work to address the main research question.


Figure 1.1 : The three interconnected sub-studies

1.3 STUDY OBJECTIVES, INNOVATIONS, AND CONTRIBUTIONS

This study focuses on identifying influential factors on mode choice and

understanding their impacts on travellers’ mode choice decision-making process

currently and in the future. The relationships between travellers’ socio-economic

profile and their trip characteristics versus their chosen mode are then quantified using

the data collected from urban travellers in five Australian capital cities: Sydney,

Melbourne, Brisbane, Perth and Adelaide. The consistency of travellers ‘revealed

travel behaviours is also assessed against their stated preferences. The detailed

objectives and uniqueness of each of the sub-study are discussed as below.


1.3.1 Urban travellers’ satisfaction with train fares in five Australian cities

To control confounding factors, this first sub-study, Urban travellers’

satisfaction with train fares in five Australian cities, specifically focuses on train

riders’ satisfaction with the paid train fare for their most recent home-based train trip

in five Australian capital cities: Sydney, Melbourne, Brisbane, Adelaide, and Perth. It

uses two main data sources: a nationwide survey; and objective information on the

characteristics of each city’s existing train system. A random-parameters ordered logit

model is developed to identify significant factors associated with train user satisfaction

levels. To gain more insights, underlying reasons behind differences in user

satisfaction across the five cities are traced back to relevant characteristics of the

existing train systems (and to fare structure, in particular) in these cities.

The practical contribution of this first sub-study is the provision of useful

knowledge for policy makers and authorities in formulating future transit policies to

increase travellers’ satisfaction level.

1.3.2 Consistency between perceptions and stated preferences data in a

nationwide mode choice experiment

This second sub-study, Consistency between perceptions and stated preferences

data in a nationwide mode choice experiment, aims to shed light on this ongoing

debate is whether a carefully designed SP experiment is able to provide individuals’

true preferences from a different perspective. Unlike previous studies, where RP data

were often used to assess the reliability of SP responses, this study employs another

valuable data source: travellers’ perceptions of the influence of various service factors.

In addition, sample sizes in previous studies were usually small, and respondents were

often from a particular user group that had been surveyed in previous studies. In

contrast, this study uses nationwide survey data relating to the most recent home-based

trip of urban travellers in five Australian capital cities (Sydney, Melbourne, Brisbane,

Adelaide, and Perth). Therefore, the sample size is large, and respondents come from

diverse backgrounds.

A distinctive model is estimated for each data type. More specifically, a random-

parameters binomial logit model that accounts for heterogeneity in the population is

estimated for the perceptions, while a mixed logit model is employed to model the

stated choice responses. To gain further understanding of the underlying reasons for


choosing a particular transport mode, the significant explanatory random parameters

are traced back to respondents’ diverse socio-economic backgrounds. Furthermore, the

consistency between these two models in terms of explaining respondents’ mode

choice behaviours are assessed both qualitatively and quantitatively.

The information consistency between the SP and RP datasets is assessed with a

comprehensive consistency assessment method. This method is a rich and significant

theoretical contribution to the SP literature.

1.3.3 Policy interventions study to encourage behavioural shift from car to

public transport

Having explored the theory of planned behaviour, cognitive work analysis, and

the intervention method, and having identified the research gaps, this third sub-study,

Policy interventions study to encourage behavioural shift from car to public transport,

aims to investigate the range of socio-economic factors and travel attributes that

influence the utility and probability values of travellers’ mode choice, and encourages

mode shift in each city. In addition, the magnitude of the model’s factor coefficients

is simulated to determine realistic policy interventions for mode shift behaviour.

Ultimately, this third sub-study contributes to the formulation of future transit

policies related to mode shift from car to public transport. Specifically: 1) It provides

useful knowledge for policy makers and authorities in their formulation of future

transit policies to encourage mode shift; 2) It provides an efficient and novel

framework for the analysis of mode shift, particularly, from car to public transport;

and 3) Its novel framework for mode shift analysis and policy intervention is a

theoretical contribution to the field of SP experiment.

1.4 RESEARCH SCOPES

This PhD study only investigates the research questions mentioned in Section

1.2 It is divided into the three interconnected sub-studies, namely Urban travellers’

satisfaction with train fares in five Australian cities, Consistency between perceptions

and stated preferences data in a nationwide mode choice experiment, and Policy

interventions study to encourage behavioural shift from car to public transport. The

overall study aims to achieve the stated objectives, innovations and contributions

(Section 1.3) within the duration of study (39 months). This study utilize the online


survey data collected by the project team of the “CRC for Rail Innovation Project

1.130: Urban Rail Travel Behaviour” from urban travellers in five Australian capital

cities: Sydney, Melbourne, Brisbane, Adelaide, and Perth (Zheng et al., 2013). The

data consists service demands and travellers’ experience from using private cars and

land-based public transport services, such as bus and train (excluding light rail and

tram) services, and private cars.

In overall, the study does not include the following details. Service demands and

travellers’ experience from water-based public transport, such as ferry services, and

active travel modes, such as walking and cycling, are not included in the study. The

transport card data are not available to serve as a comparison dataset. The original

project team did not conduct one-to-one interviews to the targeted urban travellers.

The train fare airport surcharge was not included in the study due to its irrelevancy

1.5 THESIS OUTLINE

The research questions are further investigated in Chapter 2: which focuses on a

review of the literature relevant to each sub-study, and identifies the main research

gaps. The last paragraph of each section of Chapter 2: summarises the research gaps

for each sub-study.

Chapter 3: explains the data collection process and the structure of the

questionnaire; provides a descriptive analysis of the socio-economic profiles,

perceptions and trip characteristics, as well as describes the utilisation of the collected

datasets (Section 3.1 to 3.4); and details the data preparation process for the first sub-

study (Section 3.5). Section 3.6 presents the factor mapping of perceptions and their

corresponding attributes of mode choice experiment for the second sub-study. Section

3.7 specifically illustrates the utilisation of each city dataset for the third sub-study.

Subsequently, Chapter 4: outlines and describes the modelling methodology of each

sub-study.

Chapter 5: begins with a summary of the characteristics of each city’s existing

train system in order to provide the study context, and information on the train fare

structure in each of the five capital cities. It then reports and analyses the modelling

results of the first sub-study (Urban travellers’ satisfaction with train fares in five

Australian cities). It also includes a discussion of heterogeneity, and the results of a

detailed comparison of train fare structures in the five Australian cities.


The beginning of Error! Reference source not found. reports and analyses the

modelling results of the second sub-study (Consistency between perceptions and

stated preferences data in a nationwide mode choice experiment). This is then

followed by a detailed qualitative and quantitative consistency assessment between

travellers’ perceptions and stated preferences in a nationwide mode choice survey.

The findings of the third sub-study (Policy interventions study to encourage

behavioural shift from car to public transport) are reported and analysed in Chapter

7:. In addition, Chapter 7: presents the simulation study to provide realistic illustrations

of how policy interventions are able to encourage mode shift behaviours.

Chapter 8: concludes the overall thesis, and discusses its contributions to wider

theoretical and practical contexts along with its policy implications. It also

acknowledges the limitations of the overall study, and provides potential directions for

future study.

1.6 PUBLICATIONS FROM THIS STUDY

This study has yielded three journal article manuscripts. One manuscript has

been published, one manuscript is under review, and another manuscript is being

prepared. These manuscripts are:

1. Paramita P, Zheng Z, Haque MM, Washington S, Hyland P. (2018). User

satisfaction with train fares: A comparative analysis in five Australian cities.

PLoS ONE 13(6): e0199449. https://doi.org/10.1371/journal.pone.0199449.

2. Paramita, P., Zheng, Z., Haque, M. M., & Washington, S. (2018). Consistency

between perceptions and stated preferences in a mode choice experiment:

Evidence from a national survey in Australia.

Manuscript under preparation to be submitted to Journal Transportation

Research Part A: Policy and Practice – Q1.

3. Paramita, P., Zheng, Z., Haque, M. M., Whitehead, J., & Washington, S.

(2018). Policy interventions to encourage behavioural shift from car to public

transport: A statistical approach.

Manuscript under preparation to be submitted to Journal Transportation

Research Part A: Policy and Practice – Q1.

https://doi.org/10.1371/journal.pone.0199449

Chapter 2: Literature Review 13

Chapter 2: Literature Review

The literature review chapter identifies and critically evaluates the relevant

previous literatures around each of the sub-study coherently. This study starts by

systematically reviewing wide range of aspects of travel behaviours literatures as

illustrated in Figure 2.1 below.

Figure 2.1 : Body of literature of Travel Behaviour field.

Following the formulation of the precise main research question and the three

interconnected sub research questions as elaborated in Section 1.2, this chapter focuses

on a review of the literature relevant to each sub-study, and identifies the main research

gaps. It consists of three sections. Each section demonstrates the synthesis and

integration of those reviews and arguments into one common theme of research gaps

for the respective sub-study. The last part of each section summarises the research gaps

for each sub-study.

14 Chapter 2: Literature Review

2.1 URBAN TRAVELLERS’ SATISFACTION WITH TRAIN FARES IN

FIVE AUSTRALIAN CITIES

2.1.1 Literature review

This section reviews notable studies on user satisfaction with public transport.

User satisfaction is a multi-dimensional concept (Oliver, 2014). Parasuraman et

al.(1985; 1994) and Zeithaml et al. (1988) identified five general dimensions of user

satisfaction, including assurance, reliability, empathy, tangibles, and responsiveness.

Friman et al.(2001) proposed a model to assess user satisfaction based on simplicity

of design and information, treatment by staff, and service reliability. As a slightly

different concept, a service quality model of user satisfaction consisting of functional

and technical service attributes was proposed by Grönroos (1984, 1990). Technical

service attributes are related to what services the customer receives, while functional

service attributes are related to how the customer receives those services. Later,

Fellesson and Friman (2012) labelled these two attributes ‘Factor A’ and ‘Factor B’,

with Factor A being a safety factor related to feeling secure at stations and on-board

vehicles, and Factor B being a system factor related to frequency, and travel and

waiting times. In addition, Fellesson and Friman (2012) defined three other factors:

Factor C, related to public transport comfort (e.g., in-vehicle cleanliness and level of

crowding); Factor D related to the behaviour, knowledge, and attitude of the staff; and

Factor E related to service delivery.

Satisfaction is also considered to be the foundation of consumer loyalty and

behaviour (McDougall & Levesque, 2000; Olsen, 2007), and has a strong connection

with perceived value and service quality (Chen, 2008; J. de Oña & de Oña, 2014; J. de

Oña et al., 2013; R. de Oña et al., 2015; Eboli & Mazzulla, 2012; Hitayezu, Wale, &

Ortmann, 2016; Jen & Hu, 2003). Travellers who have experienced a good quality

public transport services are inclining to have a higher level of perceived satisfaction,

and intending to continue using the same services. Satisfaction is also related to a

traveller’s evaluation of the quality of their entire trip experience (Chen, 2008;

Efthymiou & Antoniou, 2017; Fornell, 1992; Solvoll & Hanssen, 2017).

Beirão and Cabral (2007), Gadziński and Radzimski (2016), Tyrinopoulos and

Antoniou (2008), and Efthymiou, Antoniou, and Tyrinopoulos (2017) analysed the

behaviour of public transport travellers and their perceived satisfaction with services

in order to understand the underlying reasons for their preferred transport mode. They


found that travellers had a strong preference for a reliable and well-coordinated

transportation system. The cleanliness of vehicles, the condition of the waiting area,

service frequency, network coverage, and transfer distance were perceived as the most

important satisfaction attributes. This knowledge served as a foundation for policy

makers and transport operators to improve services by making operational adjustments

to frequency of services, transfer points, and network coverage (Brons, Givoni, &

Rietveld, 2009; Tyrinopoulos & Antoniou, 2008).

Travellers who have experienced unreliable services and long waiting time have

a low public transport user satisfaction level (Cantwell, Caulfield, & O’Mahony, 2009;

Dell’Olio, Ibeas, & Cecín, 2010). The importance of travel time reliability was

influenced by two factors: negative consequences for travellers arriving late at their

destinations, and the actual value that individual travellers placed on reliability of

transportation system (Bhat & Sardesai, 2006; Efthymiou & Antoniou, 2017). Lucas

and Heady (2002) discussed the concept of time urgency and assessed the difference

between travellers’ with a flexitime schedule and those without such flexibility. Time

urgency was found to be a personal concept relating to the individual’s perception of

time. Since flexitime scheduling greatly reduced commuting pressure, they argued that

flexitime travellers experienced less time urgency and more trip satisfaction(Ettema,

Friman, Gärling, Olsson, & Fujii, 2012).

Another essential factor in enhancing public transport service quality is the

reduction of on-board crowding (Cantwell et al., 2009). On-board crowding positively

contributes to the perceived walking and waiting time (Li & Hensher, 2011). In reality,

travel time is actually longer when vehicles are crowded because it takes longer to

board and alight when there are passengers standing in aisles. Li and Hensher (2011)

argued that a successful public transport system requires less crowding and increased

reliability at all phases of the service chain. Reliability is defined as the quality of

performing consistently well.

The availability of pre-departure travel information is vital for travellers, and

eventually affects their travel habits and satisfaction levels (Lyons, 2006). Pre-

departure travel information has a number of roles, such as providing an awareness of

the travel options for a particular trip; empowering travellers to make fully informed

travel choices; assisting travellers to successfully commence and complete a trip; and

reducing waiting travellers’ feelings of frustration stress (Cantwell et al., 2009;

Caulfield & O'Mahony, 2007; Dziekan & Kottenhoff, 2007; Hine & Scott, 2000;


Lyons, 2006). The success of such information, however, depends on its reliability and

its accessibility by all travellers (Jou, 2001). Regular travellers formally obtain such

information when they first use a particular service, while occasional travellers rely on

informal information from others (Lyons, 2006).

With particular reference to train services, Brons et al. (2009) mentioned a

number of important factors that affect travellers’ perceived satisfaction: station

organization, real-time information, level of comfort, service punctuality, train

frequency, and accessibility. Accessibility includes bicycle parking, connections to the

wider public transport network, and car parking facilities at stations. Furthermore, in

order to increase ridership, Ellaway et al. (2003) and Brons et al. (2009) suggested the

provision of benefits similar to those enjoyed travelling by private transport. Examples

include the integration of cycling and train travel by permitting bicycles on board; the

provision of bike hooks on certain trains during off-peak hours; and the provision of

park-and-ride facilities at major stations to entice motorists to use the train as a part of

their trips (Andersson & Nässén, 2016; Pucher & Buehler, 2009).

In addition, individual characteristics such as age, income, education level,

household composition, travel budget, a drivers’ license, and access to a motor vehicle

affect mobility behaviour and the level of access to train services (Dekker, Hess,

Arentze, & Chorus, 2014; Dieleman, Dijst, & Burghouwt, 2002; Efthymiou et al.,

2017; Geurs & Van Wee, 2004). Specifically, in order to attract potential train riders

and increase ridership, train services should be designed according to the level of

service preferred by current riders (Beirão & Cabral, 2007).

2.1.2 Research gaps

Overall, the majority of the previous studies focussed on users’ overall

satisfaction, and only a few investigated user satisfaction with train fare (Efthymiou &

Antoniou, 2017). Despite the fact that train fares, as the only monetary cost incurred

by train riders, is often used as a powerful tool to influence ridership. Thus, our

understanding of the relationship between train fare and user experience remains

elusive. In addition, most of the previous studies were based on survey data collected

from a single city, thus precluding the possibility of intercity comparison. Hence, this

comparative analysis study, which is complemented by knowledge of the

characteristics of the current fare system in each city, is valuable in revealing the

underlying and causal factors that contribute to travellers’ perceived satisfaction. The


data used in many previous studies only partially capture the potentially significant

factors linked to traveller satisfaction. In particular, characteristics of travellers’ most

recent trips are often not considered (Paramita, Zheng, Haque, Washington, & Hyland,

2018).

2.2 CONSISTENCY BETWEEN PERCEPTIONS AND STATED

PREFERENCES DATA IN A NATIONWIDE MODE CHOICE

EXPERIMENT

2.2.1 Literature review

SP responses are generally obtained from a SP experiment. SP experiment is a

quantitative method for simulating preferences that can be used in the absence of RP

data (Mangham, Hanson, & McPake, 2009). The technique involves asking

respondents to state their preferences in a number of hypothetical scenarios. Each

hypothetical scenario is described by various attributes. With the availability of SP

experiment and data, the demand for new products or services can be estimated. The

SP experiment is also able to stimulate responses about individuals’ behaviour so as

to ultimately identify their inclinations (Cascajo, Garcia-Martinez, & Monzon, 2017).

Earlier studies show that SP data can enrich RP data. The former have the ability

to view the experiment as supplementary to RP data (Hensher, 1994; Wardman, 1988).

SP data incorporates diverse attributes, which are not available in the market, and have

no corresponding RP history (Hensher, Louviere, & Swait, 1998; Mark & Swait,

2004). The explanatory variables of SP are able to introduce variability, and to remove

or reduce collinearity among variables. This enables a more accurate estimate of

contributions to the utility of the goods or services (Mark & Swait, 2004). In general,

in the near future, the analysis of a combination of SP and RP data will be able to

quantify the usage of new and more favourable products and services (Mark & Swait,

2004; Tselentis et al., 2018). The different characteristics of SP and RP data suggest

that the joint utilisation of both data types are able to enhance the modelling and

understanding of choice behaviours (Hensher, 1994). The combination of the two data

sources also provides considerable benefits for those who can focus on each source's

strengths, and has the potential to provide policy makers with more accurate

knowledge and information.


The number of attributes of each choice within SP experiment reflects concerns

about task complexity and non-obligatory decision rules. However, the increasing

number of attributes and choices per respondent could have increased the cognitive

burden of the SP experiment respondents (Bech, Kjaer, & Lauridsen, 2011; de Bekker‐

Grob, Ryan, & Gerard, 2012; DeShazo & Fermo, 2002; J. J. Louviere, Islam, Wasi,

Street, & Burgess, 2008). When respondents are fatigued, bored, and less engaged;

and when they have the opt-out availability for all alternatives, they are more likely to

choose the latter as they proceed through the questionnaires (Miguel et al., 2005).

Some respondents, who were not familiar with SP surveys, simply perceived the

completion of the SP experiment as a complex task, and adopted various strategies to

simplify the process and reduce their effort (Brazil, Caulfield, & Bhat, 2017; Hoang-

Tung & Kubota, 2017). For instance, they failed to consider some information, or only

partially attended to the SP experiment attributes. This strategy is known as ‘non-

attendance’ (Brazil et al., 2017; Hensher, Rose, & Greene, 2012; Hess & Hensher,

2010; Hoang-Tung & Kubota, 2017). Attribute non-attendance can be influenced by

existing knowledge, or lack thereof, of a particular topic of interest (Brazil et al., 2017).

The predictive value of SP experiment can be limited since it measures stated

preferences that can differ from actual travel behaviours. That is, in line with Azevedo

et al.’s (2003) findings, respondents’ stated choices might be inconsistent with their

actual decisions in similar real-life situations (Lambooij et al., 2015; Severin, 2001).

Azevedo et al. highlight a diverse range of potential biases, where respondents might

overlook and underestimate some of the attributes and constraints when answering

hypothetical travel scenarios (Arrow et al., 1993; Kemp & Maxwell, 1993; Loomis,

Gonzalez-Caban, & Gregory, 1994; Wardman, 1988). Additional critique includes the

possibility of SP-based WTP estimates also failing to diverge according to the scope

of the resource being valued; this is known as the “embedding effect” (Azevedo et al.,

2003; Desvousges et al., 1992; Kahneman & Knetsch, 1992).

Despite these criticisms, there is substantial evidence that responses to carefully

designed SP experiment can produce valuable information. Mitchell and Carson

(1989) mention that the valuation estimates of SP methods are typically correlated in

the expected direction of the predicted independent variables. To further examine the

validity of SP information, a group of researchers have compared RP data-based

valuation estimates to estimates based on actual market prices or simulated market


transactions (Adamowicz, Louviere, & Williams, 1994; Azevedo et al., 2003;

Cameron, 1992; Cummings, Brookshire, Bishop, & Arrow, 1986; Cummings, Elliott,

Harrison, & Murphy, 1997; Cummings & Taylor, 1999; J. Louviere, 1996). Carson et

al. (1996; 2007) found that, generally, the ratio of SP to RP valuations lies close to

one, and that the RP and SP estimates are highly correlated. These results suggest that

reliable information can be collected from carefully designed SP experiment (Azevedo

et al., 2003).

Over the past four decades, travel behaviour researchers have conducted

numerous external validation studies based on travellers’ revealed behaviours and

future preferences; however, these studies are limited to values of time (Bates, 1984;

Hensher & Truong, 1985; Louviere et al., 1981); predicted market shares (Benjamin

& Sen, 1982; Kocur et al., 1982; Lerman & Louviere, 1978; J. J. Louviere & Hensher,

1982; J. J. Louviere & Kocur, 1983); and the WTP test (Blumenschein et al., 2001;

Clarke, 2002; Ding, 2007; Hensher, 2010; Lambooij et al., 2015; Ryan, 2004).

Interestingly, almost all studies consistently suggest the need to establish a reliable

validation method to cope with the ever-growing SP literature.

With specific reference to the travel behaviour context, Wardman (1988) tested

the validity of SP data by relating RP and SP values of travel time to respondents’

socio-economic status. This method is consequently described as “segmentation

analysis” (Wardman, 1988). Wardman (1988) provides evidence that respondents’ SP,

as responses to hypothetical scenarios, are a reasonably accurate guide to their true

preferences. This segmentation analysis was able to avoid the unnecessary increases

in the standard errors, and provide detailed assessment of the SP data.

In the agricultural context , Azevedo et al. (2003) suggested that the validity of

SP and RP data can be identified by testing for their consistency. They developed a

combined model containing both SP and RP responses, and constructed hypotheses

concerning potential biases. On completion of the study, after analysing the survey

data of actual wetland usage patterns against the anticipated changes to those

hypothetical patterns with increasing trip costs, they reported conflicting conclusions;

that is, the hypothetical patterns were found to be inconsistent with actual patterns.

Lambooji et al. (2015) performed the consistency test for actual “before” and the

“after” scenarios in the healthcare context. The actual scenarios related to parents’

attitudes to vaccinating their new born child against hepatitis B. Parents were first


asked to respond to a SP experiment related to their inclination to vaccinate their child;

then, after a specific timeframe, their actual vaccination behaviours were monitored.

Lambooji et al. (2015) concluded that the predictive value of SP experiment was

satisfactory for predicting the option to vaccinate, but unsatisfactory for predicting the

option of not vaccinating.

2.2.2 Research gaps

Overall, the fundamental question – of whether individuals’ responses to a

cautiously designed SP experiment are able to characterize their actual preference –

persists. In the on-going efforts to develop a reliable approach to testing the

consistency of SP data in relation to RP data, it is also essential to note that no

published study compares the perception of a particular mode choice with the SP

responses. These two research gaps motivate this study. It qualitatively and

quantitatively assesses the perception of various service factors that influence more

frequent train travel, and the mode choices indicated in response to hypothetical travel

scenarios. Both types of responses are obtained from the same set of respondents at

the same time.

2.3 POLICY INTERVENTIONS STUDY TO ENCOURAGE

BEHAVIOURAL SHIFT FROM CAR TO PUBLIC TRANSPORT

The following literature review starts by identifying and reviewing a number of

psychological theories that underlie mode choice and mode shift behaviours. It is

followed by the discussion of policy intervention methods that can influence the mode

choice process. A number of ways to overcome mode shift barriers and constraints and

to encourage car users to take public transport are then presented. This section ends by

recognizing the research gaps in the travel behaviour context.

2.3.1 Psychological theories that underlie change in travel behaviour

According to the theory, human behaviour is guided by behavioural beliefs,

normative beliefs, and control beliefs (as shown in Figure 2.2). Behavioural beliefs

result in a favourable or unfavourable attitude to a behaviour; normative beliefs relate

to perceived social pressure or subjective norms; and control beliefs result in perceived

behavioural control (Ajzen, 1985). The more helpful the attitude and the subjective

norm, and the larger the perceived control, the stronger should be the person’s


intention to perform the behaviour in question. A few years after stating the theory of

human behaviour, Ajzen (1991) posits the theory of planned behaviour (TPB), which

links a person’s beliefs and their behaviour.

Figure 2.2 : Three guides underlying the theory of human behaviour (Ajzen, 1985)

TPB identifies intention as the most immediate cognitive antecedent of

behaviour (Ajzen, 1991; Mann & Abraham, 2006). It is the most used of all

psychological theories, and was developed to account for goal-directed behaviour that

is beyond complete volitional control (Ajzen, 1985, 1991; Thøgersen, 2006). The TPB

has been applied to studies of the relationships among beliefs, attitudes, perceptions,

intentions, and behaviours in various fields, such as transport choice, marketing,

advertising, public relations, advertising, and healthcare (Ajzen, 1985, 1991). Within

the framework of TPB, the intention to perform certain behaviours – such as the use

of public transport – can be accurately predicted by attitudes toward the behaviour,

subjective norms, and perceived behavioural control (Ajzen, 1991; Bamberg &

Schmidt, 2001). The intentions and perceptions of behavioural control account for

considerable variance in performed behaviour. Behavioural interventions are expected

to change the principal beliefs that ultimately guide the performed behaviour (Ajzen,

1985).

An unresolved issue within the TPB framework is whether the theory is

sufficient to accommodate the prediction of later behaviours based on past behaviours

(Ajzen, 1991). Bamberg, Ajzen, and Schmidt (2003) closely examined this issue in a

transport case study. They concluded that past behaviour is not always a good predictor

of future behaviour, and that prior behaviour contributes significantly to the prediction

of later behaviour only when circumstances remain unchanged (Bamberg et al., 2003;

Thøgersen, 2006). Interventions, such as new information and persuasive acts, were

found to affect and change attitudes toward subject matter, subjective norms, and


perceptions of behavioural control. They also influenced intentions and behaviour in

the desired direction (Bamberg et al., 2003). The consequence of the intervention on

behaviour is mediated by the causal chain hypothesized by the TPB (Bamberg &

Schmidt, 2001). Human social behaviour is controlled by a certain level of cognitive

effort (Bamberg et al., 2003). Even a relatively minor event can disrupt the automatic

execution of behaviour and initiate reasoned action. Indeed, human social behaviours

consist of both automatic and reasoned elements. Consistent with TPB, travellers who

are faced with an unfamiliar choice will deliberate, and then choose the most attractive

goal-directed option (Gardner, 2009; Verplanken, Aarts, Knippenberg, & Moonen,

1998).

Thøgersen (2006), however, claims that travel mode choices are repetitive

behaviours, and are made in a stable situation (Ouellette & Wood, 1998; Ronis, Yates,

& Kirscht, 1989; Wood, Quinn, & Kashy, 2002). It is argued that travel mode choices

are usually performed in a habitual, rather than a reasoned or planned way (Thøgersen,

2006; Verplanken & Aarts, 1999; Verplanken, Aarts, Knippenberg, & Knippenberg,

1994; Verplanken et al., 1998). This perspective of habitual action suggests that when

travellers have strong travel choice habits, motivation has no effect on their behaviour.

This is especially the case in the unusual situation where habits are in conflict with

intentions (Gardner, 2009; Mann & Abraham, 2006). By contrast, the motivational

models neglect the often repetitive travel choice decisions because frequently repeated

behaviours can become habituated and, thus, automated (Gardner, 2009; Verplanken

et al., 1994; Verplanken et al., 1998). In the case where habit is weak, intention predicts

behaviour; however, where it is strong, intention neglects the effect on behaviour

(Gardner, 2009). Strong habit also reduces the likelihood that travellers will consider

alternatives to their regular modes (Aarts, Verplanken, & Knippenberg, 1998; Gärling

& Axhausen, 2003; Thøgersen, 2006; Wood et al., 2002).

Using TPB as a foundation, Thøgersen (2006) traced back the use of public

transport issue to attitudes to it and its reliability, and to attitudes to car ownership

(Hoyer & MacInnis, 1997; Ȫlander & ThØgersen, 1995; van Raaij, Bartels, &

Nelissen, 2002). He assumes that travel mode choices are partly individual (e.g.,

transport habits, car ownership), partly contextual (e.g., the availability of public

transportation), and partly volitional (i.e., influenced by the traveller’s evaluations and

motives) (Thøgersen, 2006). The influence of these variables is weakened when past

behaviour is considered (Fishbein, 1967; Van Raaij & Verhallen, 1983). The


behavioural changes of those who do not own a car are more in line with current

attitudes and perceptions; these attitudes and perceptions, on the other hand, are

inconsequential to car owners. For example, when car owners have a more positive

attitude to driving than to public transport, their attitude to the latter is inconsequential

and rather irrelevant. Thøgersen (2006) also found that the temporal stability of

transport behaviour is higher for car owners than non-owners.

When habitual and intentional tendencies diverge, habits are revealed (Gardner,

2009). Other studies on the formulations of the motivation–behaviour relationship

explore habit-attenuating relations between personal norms for non-car and car travel

(Eriksson, Garvill, & Nordlund, 2008) (Eriksson et al., 2008). In a stable choice setting

such as commuting, behaviour is governed by habit, and is in line with motivation.

Commuting mode choice could be modelled as a planned behaviour (Gardner, 2009).

Socially acceptable economic incentives can restructure the decision making context

to disrupt habits and motivate a change in behaviour (Fujii & Kitamura, 2003; Gardner,

2009; Thøgersen & Møller, 2008). For instance, Thøgersen and Møller (2008) show

that a free one-month travel card is sufficient to break driving habits and to enhance

public transport ridership. Eriksson et al. (2008) claim that personal resources (i.e., the

time and knowledge to change travel behaviour) and contextual factors (i.e., the

availability of alternative travel modes and supportive social norms and policy

strategies) are crucial to reducing car usage.

The hypothesis of habit discontinuity states that a context change disrupts an

individual’s habit: a window opens to encourage an individual to deliberately consider

their behaviour (Verplanken, Walker, Davis, & Jurasek, 2008). The hypothesis of self-

activation states that when values incorporated in the self-concept are activated, they

are more likely to guide behaviour. The combination of these two hypotheses predicts

the context change in order to emphasize the important values that will guide

subsequent sustainable behaviours. For example, travellers who had recently moved

and were environmentally concerned, used the car less frequently for commuting to

work (Hamer, 2010; Verplanken et al., 2008), and the provision of park and ride

facilities near residential areas generated mode shift from private cars to more

sustainable transport modes, especially for commuting to workplaces (Hamer, 2010).

Cognitive work analysis (CWA) is an ever-growing technic in the Human

Factors and Ergonomics field, and is suited to the analysis of complex socio-technical

systems (Rasmussen, Pejtersen, & Goodstein, 1994). It has been implemented in


various fields, such as in military activity allocation (Jenkins, Stanton, Salmon,

Walker, & Young, 2008); cognitive artefact design (Jenkins, Salmon, Stanton, &

Walker, 2010a); accident analysis (Jenkins, Salmon, Stanton, & Walker, 2010b);

large-scale system design process assessment (Bisantz et al., 2003); design concept

evaluation (Naikar & Sanderson, 2001); and train driver constraint comprehension

(Jansson, Olsson, & Erlandsson, 2006). Specifically, Stanton et al. (2013) utilize the

CWA perspective to explore the constraints in shifting to rail transport mode, and to

determine which groups are most impacted by each constraint. Constraints could be

linked to the relationships among travel purposes and functions, journey contexts and

types, and groups in the analysis. The CWA offers a contextual interpretation of these

constraints, rather than simply identifying them. Thus, it offers a new perspective

about mode shift issues, and generates insights into potential solutions.

2.3.2 The policy intervention methods

The habitual nature of mode choice behaviour has been recognized as an

important factor, and is included in models of mode choice (Bamberg et al., 2003;

Gärling & Axhausen, 2003; Verplanken et al., 1994; Verplanken et al., 1998;

Verplanken et al., 2008). Habituation limits the collection of information. (Verplanken

& Holland, 2002) suggest that values influence choices and behaviours only when

two conditions are met. These conditions are: a value should be part of the traveller’s

self-concept, and should be cognitively activated.

Klöckner, Matthies, & Hunecke (2003) propose normative decision making for

mode choice to include a dual-process account that consists of a norm-based route and

an habitual route (Klöckner et al., 2003). The norm-based route follows the principles

of normative decision making (Nordlund & Garvill, 2002; Schwartz & Howard, 1981),

while the habitual route consists of direct responses to situational signals and, thus,

bypasses normative considerations (Aarts et al., 1998; Ouellette & Wood, 1998).

Utilization of the intervention method requires habit measurement because

motivation shift would be sufficient to modify established behaviour when habit is not

considered. Mode choice could be modelled as the reasoned action of travellers with

weak or no habit (Gardner, 2009). Among habitual travellers, behaviour is dominated

by habitual tendencies rather than intentions. Initiatives that target attitude and belief

change, only partially influence these habituated behaviours. Therefore, intervention


methods need to be employed to acknowledge the limited cognitive engagement in

habitual decision-making (Gardner, 2009).

Bamberg et al. (2003) investigate the effects of an intervention, and the logic of

the proposition that past behaviour is the best predictor of later behaviour. The

introduction of a free semester ticket for students caused a drastic decrease in students’

car usage and an increase in their bus usage (Bamberg & Schmidt, 2001). Both

interventions influenced attitudes to bus use, subjective norms, perceptions of

behavioural control, and intentions and behaviours in the desired direction(Bamberg

et al., 2003). The TPB is demonstrated as a conceptual framework for predicting mode

choice, and for understanding the effects of an intervention on behaviours (Bamberg

et al., 2003).

The micro and macro processes of the intervention method are shown in Figure

2.3. While the rationale for the policy intervention method is rather theoretical

(Chaminade & Edquist, 2010), it is important to acknowledge that policy formulation

is not always a rational process: The rationale can emerge as an ex-post analysis, and

not as an a priori exercise.

Figure 2.3 : The micro and macro process of the intervention method (Bamberg & Schmidt,

1998)

Habitual car use was interrupted by prompting the deliberate consideration to

reduce personal car use, and to form the intention to plan changes to travel behaviour


(Eriksson et al., 2008). This intervention resulted in a weakened association between

car use and car habit, and strengthened the relationship between car use and personal

norm. Specifically, car users with a strong car use habit and a strong personal norm

were more likely to reduce car use than those with a weak car use habit and a weak

personal norm (Eriksson et al., 2008).

Frank et al. (2008) and Hu & Schneider (2017) found that urban housing and

workplace forms, and travel time and cost, were significant predictors of travel choice.

Travel time was the strongest predictor of mode choice, while urban form was the

strongest predictor of the location of public transport stops. The dominance of travel

time shows that a reduction in traffic congestion and motorized modes travel times

would lead to lower travel among the non-motorized modes. Better street connectivity,

greater retail density, and mixed land use were associated with increased walking,

public transport services, and cycling (Frank et al., 2008). The overall findings

demonstrate that transportation investment and more effective land use could

collectively and uniquely impact mode choice for both work and non-work purposes.

Thus, comprehensive land-use planning should be considered when formulating

transport and housing policies (Frank et al., 2008; Hamer, 2010; Hu & Schneider,

2017; Limtanakool, Dijst, & Schwanen, 2006).

2.3.3 Overcoming barriers to mode shift

Scheiner (2010) found that regardless of distance travelled, car usage had

increased enormously. This increase was mainly at the expense of walking and public

transport travel, and was more common in small towns than large cities. Due to cities’

mixed land-use structure and the close proximity of various facilities, car owners were

more inclined to walk in central urban areas than they were in small towns. In other

words, spatial and built environments had a strong impact on car usage, especially for

car owners (Limtanakool et al., 2006; Scheiner, 2010; Scheiner & Holz-Rau, 2007;

Simma & Axhausen, 2001). Other important factors affecting the propensity to walk

were personal motivation; available transport modes; financial resources; health; the

attractiveness of the route; and social roles and needs (Scheiner, 2010). Specifically,

Batty et al. (2015) identified 'Pull' and 'Push' mechanisms to encourage the mode shift

in urban areas. The 'Pull' mechanism includes the provision of an attractive, accessible,

and affordable public transport system that meets travellers' needs, while the 'Push'


mechanism focuses on breaking the habit of private car usage (Batty et al., 2015;

Stanton et al., 2013).

Constraints to modal shift can be defined as the temporal, financial, physical,

cognitive, and affective efforts required to change to another mode (Accent, 2009;

Blainey, Hickford, & Preston, 2009; Stanton et al., 2013). In order to change their

travel behaviour, a traveller would need the motivation to change, the means to

facilitate a change, and the ability to overcome the existing constraints (Stanton et al.,

2013). The theory of human factor explains the human role in the transport system,

and offers perspectives on mode shift constraints and the disadvantages of certain

modes. The comprehensive functioning of rail transport depends on connectivity with

other modes that offer transport between origin or destination and station, and on

travellers’ ability to access the connecting modes (Napper, Coxon, & Allen, 2007;

Stanton et al., 2013). Stanton et al. (2013) identified ten key constraints for mode shift

to train. They are: cost; punctuality and reliability; frequency; comfort and cleanliness;

travel time; interchange and station facilities; safety and security; station access;

journey planning and information provision; and ticketing. These constraints highlight

the links between the purposes, functions, and processes within the rail system from a

traveller’s perspective.

The reliability of train journeys is known to be far from perfect (J. Bates et al.,

2001; Stanton et al., 2013). Reliability is described as the degree of difference between

the advertised and the actual departure and arrival times (Derek Halden Consultancy,

2003; Stanton et al., 2013). Reliability also affects the individual travel decision. As

opposed to rail journey, car travel tends to be associated with flexibility and control

over departure and arrival times (J. Bates et al., 2001; Stanton et al., 2013). Given the

positive elasticity in travel time for commuting trips, the promotion of mode shift from

private cars to trains is challenging (Limtanakool et al., 2006). Thus, to outweigh the

unpredictability and unreliability of rail travel, complementary measures are needed

to promote mode shift. Public transport systems need to be restructured to suit both

visitors and occasional users, and back-up transport options (such as taxis) need to be

provided in the event of breakdowns and interrupted networks (Stanton et al., 2013).

Taking public transport is known to be less convenient than driving, especially

for car owners. Mann and Abraham (2006) recognize that taking public transport is

physically demanding because it includes effort, and additional time to travel to/from

the station, and also requires specific cognitive effort in planning and remembering.


Affective demand is also involved because of the reduced comfort and enjoyment, and

increased stress. Mann and Abraham (2006) suggest that psychological and affective

considerations are an important motivation; however, they are not adequately

represented in rational transport choice models.

Thøgersen (2006) argues that our habitual nature, a dependence on both

motivational and personal factors, and external constraints, should all be considered in

an effort to fully understand mode choice. The structure of the public transport system

should be changed to increase its attractiveness, and to reduce the attractiveness of car

usage. Public transport could be marketed in the same way that various consumer

products are. One possibility is to adopt a ‘hierarchy of effects’ approach that informs

travellers that public transport is a sufficiently attractive alternative to private cars

(Paul, Olson, & Gruner, 1999). First, car drivers need to be made aware of the available

public transport services; second, short-term incentives should be offered to encourage

them to give public transport a trial run; and, finally, their choice of public transport

should be reinforced and supported so that they continue to use it, and it eventually

becomes their new and habitual mode choice. The specific elements in such a targeted

approach include entertainment system on-board, sales promotion, and feedback

(Thøgersen, 2006). A number of past studies relating to the promotion of a mode shift

to sustainable transport are explored below.

2.3.4 Encouragements to shift to public transport

‘Travel Plan’ documents in the United Kingdom were found to be effective ways

of increasing awareness that contributed to significant mode shift among employees

(Rye, 2002). Key policies that could be introduced are: setting up a car sharing scheme;

providing cycle facilities; improving bus services, introducing flexi-work practices;

restricting car parking; and telecommuting (Kingham, Dickinson, & Copsey, 2001;

Limtanakool et al., 2006). Employees could also be encouraged to live closer to their

workplace by providing financial incentives for them to do so. ‘Travel plan’

documents are successful because they are site-specific plans that address issues such

as road congestion, limited parking space, and on-board crowding of public transport

services within the vicinity of the workplace. The collaborative work of companies

and organizations in this regard could build morale among employees and lobby

transport authorities and providers. The barrier of poor perceptions of alternative


modes could be weakened by comprehensive marketing and the full support of

managements (Kingham et al., 2001; Rye, 2002).

Nurdden et al. (2007) aimed to study policies that discouraged private cars usage,

and to identify factors that prevented the use of public transport in Malaysia. Their

study found that travel time, cost, and the distance from home/to public transport/to

workplace are the key reasons why cars are chosen over public transport. Therefore,

an efficient public transport system, including higher capacity transit systems, bus

lanes, and intelligent transport systems should be available to promote less reliance on

cars. Finally, they contend that real commuter incentives should be provided to

encourage mode switch to sustainable modes (Nurdden et al., 2007).

To promote walking and cycling to rail access, Rastogi (2010) analysed the

results of a study of willingness to switch mode within suburban rail stations in

Mumbai, India. Given the study’s context, the analysis served as a method to identify

the population segments that could increase the rate of switch to non-motorized modes.

The study concludes that respondent characteristics, an area’s distinctiveness,

available information about alternative transport modes, and the distance travelled, all

contributed to a switch in mode. Specifically, employed respondents, those with a high

income, and those residing at shorter distances from their workplace, were less likely

to shift transport mode (Rastogi, 2010). Following the change in their travel scenarios,

the respondents' behaviour was dynamic. This knowledge is important to the success

of the implementation of improved travel plans.

2.3.5 Research gaps

Having reviewed numerous relevant studies, research gaps were identified to

determine the direction of further study in relation to urban travellers’ mode shift from

car to public transport. It is clear that no study has yet used the nested logit model on

Australian capital city traveller datasets to identify target policy interventions, and to

determine travellers who are most affected by changes in transport policies. When

determining methods to understand revealed travel behaviours and to encourage a shift

in these behaviours, it is also essential to consider travellers’ habitual tendencies,

utility, their affective concerns, and the spatial and built environment. To date, no

mode shift and travel behaviour study has been found to address all of these

considerations in one study. Thus, these research gaps motivate this current study.

Chapter 3: Dataset 31

Chapter 3: Dataset

All the three interconnected sub-studies utilised the same dataset gathered as part

of Project R1.130 Understanding Urban Rail Travel for Improved Patronage

Forecasting funded by the CRC for Rail Innovation (established and supported under

the Australian Government's Cooperative Research Centres program). This chapter

begins with the explanation of data collection process (Section 3.1) and the structure

of mode choice experiment (Section 3.2). It is the followed by the description of the

notable responses of the survey questionnaire (Section 3.3). The data utilisation of the

collected datasets is elaborated in Section 3.4. Section 3.5 describes the data

preparation process for the first sub-study. Section 3.6 presents the factor mapping of

perceptions and their corresponding attributes of mode choice experiment for the

second sub-study. The last section (3.7) specifically illustrates the utilisation of each

city dataset for the third sub-study.

3.1 DATA COLLECTION

The data used in this study are the online survey data collected by the project

team of the “CRC for Rail Innovation Project 1.130: Urban Rail Travel Behaviour”

(Zheng et al., 2013). The team’s survey questionnaire was designed with four primary

objectives: to identify what factors drive the patronage of urban trains; to quantify the

relative importance of these patronage drivers; to allow national comparisons of

satisfaction with train usage in five Australian cities; and to develop a national

database as a resource for further research. In order to efficiently achieve these

objectives, the project team synthesized the earlier published best practice surveys

from Australia and overseas, identified the gaps in these works, and then addressed

these gaps (Zheng et al., 2013).

The final survey questionnaire contained five different sections, i.e. Section A

to E. Each section had its own objectives. Section A to D of the questionnaire

represented the Revealed Preference (RP) type of questions and Section E represented

Stated Preference (SP) type of questions. Correspondingly, the responses obtained

from Section A to D signified the respondents’ revealed travel behaviours, perceptions,

and socio-economic profiles. While the responses obtained from Section E signified

32 Chapter 3: Dataset

the respondents’ stated preferences in mode choice experiments. The five sections of

survey questionnaire were

1. Section A aimed to exclude respondents who were 16 years and younger and

the ones had a conflict of interests to avoid bias responses, as well as to identify

whether they were belong to train rider or non-rider category. The train rider

category in this study is defined as respondents who made more than one trip

by train (excluding light rail and tram) in the last month prior to the survey,

and the others belong to the non-rider category (Zheng et al., 2016; Zheng et

al., 2013);

2. Section B aimed to gather data about the trip characteristics experienced by the

train riders;

3. Section C aimed to acquire information about the trip characteristics

experienced by the non-riders;

4. Section D aimed to collect all respondents’ perceptions of service factors and

their socio-economic profiles; and

5. Section E, the mode choice experiment, aimed to gain all respondents’ stated

mode preferences for six hypothetical travel scenarios (Zheng et al., 2013). The

structure of mode choice experiment is detailed in Section 0.

The detailed responses collected from the survey questionnaire are demonstrated

in Figure 3.1 below. A copy of the survey questionnaire is available on Appendix A.

Figure 3.1 : The diagram of the collected dataset


The survey was conducted via a web-based survey platform in five Australian

cities from 9 April 2013 to 16 May 2013. A consent form was provided to each

respondent at the first page of the questionnaire. The survey was hosted by ORU (The

Online Research Unit: http://www.theoru.com/) utilising their ‘research only’ online

consumer panels. The resources provided access to more than 300 000 profiled

persons. This also allowed structured data collection method that provided

representative sampling from each city (Zheng et al., 2013).

Seven thousand respondents were targeted, 2000 in both Sydney and Melbourne,

and 1000 in each of the remaining cities (i.e., Brisbane, Adelaide, and Perth). The

target number for each city was proportional to its population. Half of the respondents

from each city were expected to be train riders, and the other, non-riders. Eventually,

a total of 6731 respondents participated in the survey. With the exception of the

Adelaide train rider group, all targeted quotas were achieved (Zheng et al., 2013). Age

and gender groups obtained were close to those in the ABS data, which confirms the

representativeness of the sample (Zheng et al., 2013). Out of 6731 respondents, 3231

are train riders and 3500 are non-riders as shown in Table 3.1.

Table 3.1 The breakdown of train riders and non-riders across five cities

Rider

Category

Sydney Melbourne Brisbane Adelaide Perth Total

N=2000 N=2000 N=989 N=742 N=1000 N=6731

Train rider 1000(50%) 1000(50%) 489(49%) 242(33%) 500(50%) 3231(48%)

Non-rider 1000(50%) 1000(50%) 500(51%) 500(67%) 500(50%) 3500(52%)

3.2 MODE CHOICE EXPERIMENT STRUCTURE

The DCE section (Section E) of the survey is useful in gaining insights into the

significance and relative significance of various factors in the mode decision making

process. Ngene (ChoiceMetrics, 2012), a specialised software for designing DCE, was

employed to generate the choice scenarios. An orthogonal design was adopted for the

pilot surveys to achieve balance and independence among the attribute levels.

However, the main survey was upgraded via an efficient design by using parameters

of the factors that were estimated based on the data collected from the pilot surveys.

The efficient design generated parameter estimates with the smallest standard errors

(Bliemer & Rose, 2006; Rose, Bliemer, Hensher, & Collins, 2008). In order to generate

the most realistic combinations of different attribute levels, the mode choice sets

presented to each individual were personalised according to the available modes

during their most recent trip (Hensher, Rose, & Collins, 2011). The realistic attribute

http://www.theoru.com/


levels for access time to/from station/bus stop, waiting time, on-board time, and bus/

train ticket were determined based on the data collected from the pilot surveys. (Details

of how the travel attribute levels were defined and calibrated can be found in Zheng et

al. (2016; 2013). Table 3.2 summarises the final attributes and their levels considered

in the mode choice scenarios in the main survey.

Table 3.2 Attributes and their levels employed in the mode choice experiments

Mode Access time

to train

station/ bus

stop

Waiting

time

On-board

time

(minutes)

Ticket

($)

On-board

crowdedn

ess level

Access time

from train

station/ bus

stop/ car park

Bus Short (2.5),

Medium (5),

Long period

(7.5 minutes)

Short (3),

Medium (6),

Long period

(9 minutes)

16, 32, 48 1.2,

2.4, 3.6

Not

crowdeda,

Crowdedb

Short (2.5),

Medium (5),

Long period

(7.5 minutes)

Train Short (5),

Medium (10),

Long period

(15 minutes)

Short (4),

Medium (8),

Long period

(12 minutes)

15, 30, 45 1.5, 3,

4.5

Not

crowdeda,

Crowdedb

Short (5),

Medium (10),

Long period

(15 minutes)

Car NAc NAc 10, 20, 30 NAc NAc Short (1),

Medium (5),

Long period

(10 minutes)

Mode Availability

of wireless

connection

Availability

of laptop

station

Fuel Cost

($)

Parking

Cost ($)

Toll Cost ($)

Bus Yes, No Yes, No NAc NAc NAc

Train Yes, No Yes, No NAc NAc NAc

Car NAc NAc 0.98,1.31,

1.63,1.96,

2.62,2.95,

3.27,

3.93,

4.91

0, 10, 20,

30

0, 5, 10, 15

a Not crowded: a traveller found a seat for the entire journey; b Crowded experience: a traveller has to stand up prior to finding a vacant seat on-board and a

traveller has to stand up for the entire journey and no vacant seat available (Cantwell et al.,

2009) c NA : This attribute is not applicable to this mode

In total, Ngene generated 216 mode-choice scenarios (ChoiceMetrics, 2012): 72

for train and bus; 72 for train, bus, and car; 36 for train and car; and 36 for bus and car.

Specifically, to minimize respondents’ workload and to optimize the accuracy of their

responses, only six scenarios were randomly presented to each respondent. To avoid

any confusion, each hypothetical travel scenario was presented in the form of a

pictogram.


3.3 DATA DESCRIPTION

3.3.1 Socio-economic profiles of all respondents

Each train rider’s socio-economic profile of all respondents were extracted from

Section D of the questionnaire. Socio-economic background includes information

about their gender, age, employment status, pre-tax household weekly income level,

highest education level, whether they hold a driver’s license, whether they have access

to a vehicle, whether a motor vehicle is required for their work, whether train services

influenced the choice of their current home location, and the composition of their

household. The socio-economic profiles of all respondents from each city are shown

in the Table 3.3.

Table 3.3 Socio-economic profiles of respondents from each city

Socio-

economic

factor

Classification Sydney Melbourne Brisbane Adelaide Perth

N=2000 N=2000 N=989 N=742 N=1000

Gender Female 46% 45% 42% 48% 42%

Male 54% 55% 59% 52% 58%

Age group 16-17 years old 0% 0% 1% 1% 1%

18-30 years old 17% 17% 17% 18% 18%

31-40 years old 21% 22% 24% 18% 19%

41-50 years old 19% 22% 15% 23% 19%

51-60 years old 23% 22% 25% 20% 21%

60+ years old 19% 18% 20% 21% 23%

Employment

status Full time 42% 39% 36% 32% 34%

Part time 17% 18% 16% 18% 18%

Self-employed 6% 8% 7% 6% 6%

Outside

workforce 35% 35% 41% 43% 42%

Pre-tax

household

weekly

income

Low income 28% 27% 31% 38% 34%

Middle income 32% 33% 33% 31% 28%

High income 20% 18% 19% 13% 19%

Not reported 20% 23% 18% 19% 20%

Highest

educational

qualifications

attained

Pre-bachelor

degree

23% 23% 22% 16% 19%

Bachelor degree

level

23% 23% 19% 18% 23%

Graduate level

and above

54% 53% 59% 65% 59%

Driving

licence

ownership

Yes 90% 91% 91% 90% 92%

No 10% 9% 9% 10% 8%

Vehicle

access

No access and

solely

dependent on

public transport 14% 12% 11% 11% 11%


Socio-

economic

factor

Classification Sydney Melbourne Brisbane Adelaide Perth

N=2000 N=2000 N=989 N=742 N=1000

Access to either

privately

owned,

company owned

or shared motor

vehicles 87% 89% 89% 89% 89%

Whether a

motor vehicle

is required for

working

Yes 27% 30% 28% 29% 31%

No 74% 70% 73% 71% 69%

Train

services’

influence on

the current

home location

decision

Significant 47% 26% 13% 14% 11%

Moderate 23% 19% 18% 13% 12%

Insignificant 30% 56% 68% 73% 78%

Household

composition Adults and kids 36% 38% 34% 37% 33%

Adults only 51% 49% 52% 53% 56%

Multiple family

household 13% 13% 14% 10% 11%

Trip purpose Work/School 58% 56% 53% 51% 53%

Other trip

purposes 42% 44% 47% 49% 47%

Departure

time Off-peak 54% 55% 55% 58% 57%

AM peak 37% 32% 33% 31% 31%

PM peak 10% 13% 11% 11% 12% a ”Outside work force” employment status includes being student, retired, on maternity or on other

official working leaves, and being unemployed.

3.3.2 Perception rates towards seven service factors

Each respondent was asked to rate their perceptions of seven service factors in

influencing their decision to take a train more often (Section D). The service factors

included: better access to train station; trains running on schedule; the probability of

getting a seat; the ability to access up-to-date train information; availability of an on-

board entertainment system; increased road congestion; and congestion charge or toll

for private vehicles entering the city centre during peak hours. Their responses are

tabulated in Table 3.4 below.

Table 3.4 Respondents’ perceptions of various service factors in each city

Perception of … to

take a train more

often

Levels of

influence

Sydney

(%)

Melbourne

(%)

Brisbane

(%)

Adelaide

(%)

Perth

(%)

Better access to train

station

Strong/Not

strong influence 23/ 77 23/77 22/ 78 21/ 79 31/ 69

Train running on

schedule

Strong/Not

strong influence 29/ 71 29/ 71 30/ 70 28/ 72 33/ 67


Perception of … to

take a train more

often

Levels of

influence

Sydney

(%)

Melbourne

(%)

Brisbane

(%)

Adelaide

(%)

Perth

(%)

The probability of

getting a seat

Strong/Not


The ability to access

up-to-date train

information

Strong/Not


Availability of an on-

board entertainment

system

Strong/Not


Increased road

congestion

Strong/Not


Congestion charge or

toll for private vehicles

entering city centre

during peak hours

Strong/Not


3.3.3 Trip characteristics of non-riders

The trip characteristics of non-riders’ most recent home-based trip by using bus

or any other motor vehicles were collected in the survey via Section C (the trip

characteristics experienced by the non-riders) of the questionnaire. The description

analysis of the trip characteristics are presented in the Table 3.5 below.

Table 3.5 Trip characteristics of train riders from each city (Non-riders)

Trip

characteristic

Classification Sydney

(n=1000)

Melbourne

(n=1000)

Brisbane

(n=500)

Adelaide

(n=500)

Perth

(n=500)

Main

transport

mode used

Motor vehicle 80% 86% 81% 76% 87%

Passenger in a

motor vehicle 5% 6% 6% 4% 5%

Company motor

vehicle 3% 2% 3% 2% 3%

Taxi 1% 1% 1% 0% 1%

Bus 11% 5% 10% 16% 5%

Departure

time Off Peak 57% 59% 59% 57% 56%

AM Peak 31% 30% 29% 31% 31%

PM Peak 12% 11% 11% 12% 13%

Trip purpose Work/ School 39% 41% 39% 39% 40%

Social/

Recreational 13% 10% 9% 13% 10%

Shopping/

Personal Business 48% 49% 51% 48% 50%

On-board

time

Short period

(<15 minutes) 39% 42% 43% 40% 39%

Medium period

(15-30 minutes) 29% 29% 31% 30% 32%

Long period

(>30 minutes) 32% 30% 26% 30% 29%

Total out-

vehicle timea

Short period

(<15 minutes) 86% 91% 87% 84% 90%

Medium period

(15-30 minutes) 8% 5% 7% 10% 6%


Trip

characteristic


(n=1000)

Melbourne

(n=1000)

Brisbane

(n=500)

Adelaide

(n=500)

Perth

(n=500)

Long period

(>30 minutes) 6% 5% 6% 7% 5%

Total travel

time

Short period

(<15 minutes) 23% 26% 26% 24% 25%

Medium period

(15-30 minutes) 29% 32% 31% 30% 30%

Long period

(>30 minutes) 48% 42% 43% 45% 44%

Total one-way

cost Low fare (<$2) 79% 86% 82% 85% 87%

Medium fare

($2-$6) 13% 6% 11% 8% 7%

High fare (>$6) 9% 8% 8% 7% 6%

Main reasons

for not taking

train

I like driving 10% 11% 6% 7% 11%

Driving was faster 16% 20% 17% 12% 16%

Work-related

vehicle 3% 2% 2% 3% 3%

No available train

service 41% 29% 38% 52% 35%

Personal mobility

constraints 2% 2% 3% 2% 2%

Needed to

transport bulky

items 2% 2% 2% 1% 2%

I was offered a

free ride 2% 2% 1% 1% 1%

I prefer bus/ ferry/

tram 1% 1% 0% 2% 0%

The nearest train

station was too far 9% 9% 15% 8% 12%

The trip was too

short for taking

train 6% 9% 5% 2% 7%

The train is not

sufficiently safe 1% 1% 0% 0% 1%

The train does not

run frequently

enough 1% 1% 1% 1% 0%

The train is not

convenient for

visiting multiple

destinations 2% 2% 2% 2% 3%

The train is

generally for

people who don’t

have access to

motor vehicles 1% 1% 1% 0% 1%

The train is too

crowded 1% 1% 0% 0% 1%

The train fare is

too expensive 1% 2% 2% 0% 1%

Others 5% 6% 4% 6% 4%

Reasons for

choosing

transport

mode used

No other mode

was available 23% 19% 24% 18% 22%

Cheapest mode

available 19% 14% 23% 22% 17%


Trip

characteristic


(n=1000)

Melbourne

(n=1000)

Brisbane

(n=500)

Adelaide

(n=500)

Perth

(n=500)

(multiple

selections

allowed)

Fastest mode

available 46% 47% 46% 41% 45%

Most convenience

mode available 64% 65% 63% 64% 65%

Safest mode

available 11% 12% 11% 9% 10%

Most comfortable

mode available 31% 30% 25% 26% 27%

Work-related

vehicle 4% 3% 4% 4% 4%

Personal mobility

constraints 6% 4% 4% 4% 4%

Needed to

transport bulky

items 5% 6% 8% 7% 7%

Weather condition 3% 5% 3% 2% 4%

Bus on-board

crowding

(only for Bus

users)b

Sydney

(n=114)

Melbourne

(n=48)

Brisbane

(n=50)

Adelaide

(n=82)

Perth

(n=27)

Not crowded 93% 94% 100% 95% 96%

Crowded 2% 2% 0% 1% 0%

Overcrowded 5% 4% 0% 4% 4% aTotal out-vehicle time is defined as the total of the time a respondent spent travelling between their

point of origin and the bus stop/ car park, the waiting time for bus services, when applicable, and the

time they spent travelling between the bus stop/ car park and their destination.

bThe on-board crowding is self-reported with three pre-defined categories: not crowded (a traveller

found a seat for the entire journey); crowded (a traveller had to stand up prior to finding a vacant seat

on-board); and overcrowded (a traveller had to stand up for the entire journey and no vacant seat

available) (Cantwell et al., 2009).

3.4 DATA UTILISATION

According to the structure of the survey questionnaire describe above, at least

four types of responses were obtained from each respondents, such as responses from

Section A, either Section B or C (depending on whether the respondent belongs to

Train rider or Non-rider category), Section D, and Section E. Nonetheless, not all of

those responses were analysed in each of the sub-study. Based on the objectives of

each study, a different set of responses were analysed. The detailed utilisation of the

responses collected from the survey questionnaire are demonstrated in Figure 3.2.

The first sub-study (Urban travellers’ satisfaction with train fares in five

Australian cities) utilized only the train riders dataset consisting of 3231 respondents

from all five Australian capital cities as well as specifically used the data from Section

B and Section D, only the socio-economic profiles, of the questionnaire.


The second sub-study (Consistency between perceptions and stated preferences

data in a nationwide mode choice experiment) utilized both the train rider and non-

rider dataset consisting of 6731 respondents from all five Australian capital cities.

Specifically, it analysed two the data types, the RP data type was obtained from Section

D, both the perception rates and socio-economic profiles, and the SP data type was

acquired from Section E, the mode choice experiment responses.

In order to achieve the objective of investigating the travel behavioural shifts in

Sydney, Melbourne, and Brisbane, the third sub-study (Policy interventions study to

encourage behavioural shift from car to public transport) only utilised both the train

rider and non-rider dataset collected from these three cities. This gave a total of 4989

respondents. Specifically, it analysed responses from Section B, Section C, Section D

(only the socio-economic profiles), and Section E of the questionnaire. The Section B,

C, and socio-economic profiles represented the respondents’ revealed travel

behaviours. On the other hand, the Section E represented the respondents’ future travel

behaviours.

Figure 3.2 : The diagram of data utilisation


3.5 TRAIN RIDERS DATASET

Initially, the sub-study of “User satisfaction with train fares: A comparative

analysis in five Australian cities” utilized the whole train riders dataset consisting of

3231 respondents from all five Australian capital cities. Prior to the analysis process,

the data were then prepared and its outliers were removed. Therefore, only 2927 train

riders data were included in the model estimation and further analysed (Paramita et al.,

2018).

3.5.1 Data preparation

To ensure the quality of the survey data, several factors were checked for

outliers, including one-way train fare, travel time, waiting time, and access time.

Respondents were asked to provide the amount of fare they paid for their most recent

train trip. As a commonly used outlier detection method, train fares outside the range

of Q1 – 3 ∗ (Q3 − Q1) ($ 0) to Q3 + 3 ∗ (Q3 − Q1) ($ 17.6) were regarded as

potential outliers, where Q3 and Q1 were the third and first quantile, respectively

(DiLalla & Dollinger, 2006; Hawkins, 1980). The same procedure was then applied to

on-board time, and any on-board time outside the range of 0 to 135 minutes was

regarded as a potential outlier. Similarly, waiting time outside the range of 0 to 25

minutes was regarded as a potential outlier, and any total access time outside the range

of 0 to 75 minutes was also regarded as a potential outlier (Paramita et al., 2018).

Prior to removing any outliers, the precaution of seeking additional information

to confirm their outlier status was taken. More specifically, by checking the actual train

fare structures in the five cities, this study found that the maximum train fare for a one-

way trip of around 135 minutes is $28. Thus, $28 was used as the upper fare limit

(Paramita et al., 2018). After this data cleansing, 2927 respondents (out of 3231) were

included in the detailed analysis.

Correlation analysis was performed prior to modelling in order to obtain valuable

knowledge about potential relationships among explanatory variables. More

specifically, Pearson correlation analysis was performed for the continuous variables,

while Spearman correlation analysis was performed for the categorical and ordinal

variables (Lund & Lund, 2013a, 2013b). It was found that i) total travel time has a

significant positive correlation with on-board time, waiting time, and total access time;

and that ii) age group has significant correlations with the type of concession fare.

Therefore, total travel time and age group were not included in the model. In addition,


paid fare status and departure time were strongly correlated because paid fare status

was derived from departure time. Thus, departure time was also not included in the

model (Paramita et al., 2018).

3.5.2 Data description

Each train rider’s socio-economic profile and the characteristics of their most

recent train trip were collected in the survey via Section D and B of the questionnaire,

respectively. Socio-economic background includes information about their gender,

age, employment status, pre-tax household weekly income level, highest education

level, whether they hold a driver’s license, whether they have access to a vehicle,

whether a motor vehicle is required for their work, whether train services influenced

the choice of their current home location, and the composition of their household.

Meanwhile, characteristics of their most recent train trip include departure time, the

purpose of their trip, pre-departure information check, home-to-station transport mode,

whether there was any on-board crowding or on-board activities, transport mode from

the station to their destination, the one-way trip cost, the time spent on-board, waiting

time, total access time, and total travel time.

In the five cities chosen for this study, at least two types of concessions fare are

available: A senior concession fare, and a student concession fare (Paramita et al.,

2018). However, respondents were not directly questioned about these fares in the

survey. Given the eligibility requirements for concession fares set by transport

authorities, the following assumptions were made in order to accurately understand

train riders’ satisfaction with their fare: (a) A respondent who is at least 60 years old

and is retired is eligible for a senior concession fare; and (b) A respondent who is

between 16 and 30 years old and is a student is eligible for a student concession fare

(Dell’Olio, Ibeas, & Cecin, 2011; Prideaux, Wei, & Ruys, 2001). The respondents who

did not belong to either of the concession groups were assumed to pay the full adult

fare (Paramita et al., 2018). Meanwhile, train fares can differ depending on the

travelling period. Respondents who departed from home between 7:00 PM and 7:00

AM and between 9:00 AM and 3:00PM were assumed to pay a discounted fare, and

those who departed between 7:00 and 9:00 AM and between 3:00 and 7:00 PM were

assumed to pay full fare (Paramita et al., 2018).

The respondents were also asked to rate their perceived satisfaction with the train

fare for their most recent home-based train trip, using a 5-point Likert scale; i.e.,


extremely satisfied, satisfied, neutral, dissatisfied, and extremely dissatisfied. Figure

3.3 shows the perceived satisfaction in each city. About 46% to 50% of respondents

from Sydney, Melbourne, and Brisbane were satisfied or extremely satisfied with the

fare paid for their most recent train trip, while less than 28% of respondents from each

of those cities were not satisfied or extremely dissatisfied. Over 59% of respondents

of Adelaide and Perth were satisfied or extremely satisfied with the fare paid for their

most recent train trip, and less than 12% of respondents from each of those two cities

were not satisfied or extremely dissatisfied (Paramita et al., 2018).

Figure 3.3 : Satisfaction with train fare in each city

The train riders’ socio-economic background and their most recent train trip

characteristics are summarised in Table 3.6 and Table 3.7, respectively.

Table 3.6 Socio-economic profiles of respondents from each city (Train riders)

Socio-

economic

factor


(n=889)

Melbourne

(n=912)

Brisbane

(n=425)

Adelaide

(n=222)

Perth

(n=479)

Gender Male 48% 49% 40% 51% 41%

Female 52% 51% 60% 49% 59%

Age group

Under 31

years old

22% 25% 20% 23% 21%

Between 31

and 60 years

old

62% 62% 63% 63% 55%

Above 60

years old

16% 14% 17% 14% 24%

Employment

status

Full time 50% 48% 44% 37% 35%

Part time 18% 17% 16% 17% 19%

Self-employed 6% 6% 6% 3% 5%


Socio-

economic

factor


(n=889)

Melbourne

(n=912)

Brisbane

(n=425)

Adelaide

(n=222)

Perth

(n=479)

Outside work

forcea

27% 29% 34% 43% 41%

Pre-tax

household

weekly

income level

Low income 23% 25% 29% 39% 35%

Middle

income

35% 35% 36% 34% 27%

High income 23% 21% 21% 8% 18%

Not reported 19% 20% 14% 19% 20%

Highest

education

level

Pre-bachelor

degree

46% 44% 52% 60% 56%

Bachelor

degree level

27% 29% 24% 24% 26%

Graduate level

and above

27% 27% 24% 16% 19%

Driver

license

Yes 88% 88% 89% 83% 86%

No 12% 12% 11% 17% 14%

Vehicle

access

No access and

solely

dependent on

public

transport

20% 17% 17% 19% 18%

Access to

either

privately

owned,

company

owned, or

shared motor

vehicles

80% 83% 83% 81% 82%

Whether a

motor

vehicle is

required for

work

Yes 21% 25% 25% 23% 25%

No 79% 75% 75% 77% 75%

Train

services’

influence on

the current

home

location

decision

Significant 49% 44% 39% 24% 36%

Moderate 21% 24% 24% 26% 18%

Insignificant 29% 32% 37% 50% 46%

Household

composition

Adults and

children

36% 37% 35% 40% 33%

Adults only 48% 46% 48% 47% 50%

Multiple-

family

16% 17% 17% 13% 17%

a ”Outside work force” employment status includes being student, retired, on maternity or on other

official working leaves, and being unemployed.

Table 3.7 Trip characteristics of train riders from each city (Train riders)


Trip

characteristic


(n=889)

Melbourne

(n=912)

Brisbane

(n=425)

Adelaide

(n=222)

Perth

(n=479)

Departure time

Off-peak 50% 52% 51% 58% 57%

AM peak 42% 34% 37% 33% 31%

PM peak 8% 14% 12% 9% 12%

Trip purpose

Work/School 69% 66% 62% 58% 57%

Social/

Recreational

7% 10% 7% 12% 14%

Shopping/

Personal

Business

25% 24% 31% 30% 29%

Pre-departure

information

check

Yes 47% 49% 51% 54% 37%

No 53% 51% 49% 46% 63%

Transport mode

from home to

the train station

Bus 19% 16% 15% 18% 31%

Walking 52% 43% 33% 45% 23%

Driving 20% 29% 40% 25% 32%

Cycling 0% 1% 0% 1% 2%

Dropped-off 8% 11% 13% 12% 12%

On-board

crowdinga

Not crowded 82% 82% 88% 92% 74%

Crowded 11% 8% 6% 4% 10%

Overcrowded 7% 11% 6% 4% 16%

On-board

activities

Work/study 8% 5% 3% 4% 2%

Leisure 92% 95% 97% 96% 98%

Transport mode

from the train

station to the

destination

Bus/tram 93% 83% 92% 87% 91%

Walking 2% 2% 5% 4% 4%

Picked up 1% 14% 1% 8% 2%

Cycling 3% 2% 3% 2% 4%

Concession fare

Student

concession

fare

5% 5% 4% 10% 6%

Senior

concession

fare

9% 8% 13% 11% 16%

No concession

(Adult) fare

86% 87% 84% 79% 78%

Paid fare status

Full Fare 50% 48% 49% 42% 43%

Discounted

Fare

50% 52% 51% 58% 57%

One-way cost

($)

Minimum 0 0 0 0 0

Median 2.5 1.9 2.95 1.5 0.97

Maximum 25.8 20 26 25 25.5

Mean 3.511 2.460 3.319 2.022 1.891

Standard

deviation

3.535 2.881 4.038 2.627 2.667

On-board time

(Minutes)

Minimum 1 1 3 1 1

Median 30 30 30 25.5 20

Maximum 135 130 135 120 90

Mean 32.285 31.912 33.904 31.059 24.282


Trip

characteristic


(n=889)

Melbourne

(n=912)

Brisbane

(n=425)

Adelaide

(n=222)

Perth

(n=479)

Standard

deviation

20.287 18.316 21.040 17.358 16.055

Waiting time

(Minutes)

Minimum 1 1 1 1 1

Median 8 7 10 10 7

Maximum 25 25 25 25 25

Mean 8.351 8.231 9.299 8.824 7.816

Standard

deviation

4.861 4.911 5.008 4.966 4.435

Total access

timeb (Minutes)

Minimum 1 3 1 1 2

Median 20 20 20 25 20

Maximum 75 75 75 75 67

Mean 23.172 23.113 22.758 26.752 22.132

Standard

deviation

13.570 13.945 13.452 15.399 13.171

Total travel time

(Minutes)

Minimum 8 8 8 3 14

Median 60 59 60 63.5 50

Maximum 180 194 185 165 160

Mean 63.808 63.257 65.960 66.635 54.230

Standard

deviation

25.962 24.876 28.284 26.053 23.189

a The on-board crowding is self-reported with three pre-defined categories: not crowded (a traveller

found a seat for the entire journey); crowded (a traveller had to stand up prior to finding a vacant seat

on-board); and overcrowded (a traveller had to stand up for the entire journey and no vacant seat

available) (Cantwell et al., 2009).

b Total access time is defined as the total of the time a respondent spent travelling between their point

of origin and the train station and the time they spent travelling between the train station and their

destination.

3.6 FACTOR MAPPING OF PERCEPTIONS AND ATTRIBUTES OF

MODE CHOICE EXPERIMENT

In order to examine the consistency in respondents’ perceptions relating to

service factors, and their choices in travel scenarios, the perception data were closely

mapped to the attributes of the mode choice experiments. For six perception data, this

study identified the most suitable attributes that closely represents the similar

information, as presented in Table 3.8.

Table 3.8 Factor mapping of perceptions and attributes of mode choice experiment

Strongly influenced perceptions of… to take a train

more often

Attributes in mode choice experiments

Better access to train station Access time to train station

Train running on schedule Train waiting time

The probability of getting a seat Train on-board crowding level


Strongly influenced perceptions of… to take a train

more often

Attributes in mode choice experiments

The ability to access up-to-date information on train

services (such as current train status)

Not Available

Availability of an on-board entertainment system Availability of wireless connection

Availability of laptop station

Increased road congestion Car on-board time

Congestion charge or toll for private vehicles entering

city centre during peak hours

Car toll cost

3.7 DATASET FOR TRANSIT POLICY INTERVENTIONS STUDY

The third sub-study only utilised the data collected from Sydney, Melbourne,

and Brisbane. This gave a total of 4989 respondents; 2000 respondents from Sydney,

2000 respondents from Melbourne, and 989 respondents from Brisbane. The

knowledge of travellers’ travel behaviours and documented information about each

city were utilised to identify an average traveller’s profile for each transport mode in

Sydney, Brisbane, and Melbourne (Queensland Government, 2016; Transport for

NSW, 2016; Victoria State Government, 2016). While, the responses of the mode

choice experiment section were utilised to estimate a nested logit model for each city.

The detailed utilisation of the dataset in this sub-study is illustrated in Figure 3.4.

Figure 3.4 : The diagram of data utilisation in the transit policy interventions study


To ensure the quality of the RP data for travellers’ profiling, the socio-economic

factors and the revealed travel behaviours were checked for outliers. For example, to

determined realistic one-way train fare for the train riders, the data were extracted from

the average of respondents’ responses on the amount of fare they paid for their most

recent train trip. As a commonly used outlier detection method, train fares outside the

range of Q1 – 3 ∗ (Q3 − Q1) to Q3 + 3 ∗ (Q3 − Q1) were regarded as potential

outliers, where Q3 and Q1 were the third and first quantile, respectively (DiLalla &

Dollinger, 2006; Hawkins, 1980). Prior to removing any outliers, the precaution of

seeking additional documented information to confirm their outlier status was also

taken.

The same procedure was then applied to other socio-economic factors and trip

characteristics. The additional documented information was obtained by checking the

actual train fare structures in the five cities as well as the reported access time to/from

bus stop or train station, waiting time, and on-board time for bus and train from various

related published articles, websites and journal papers (Queensland Government,

2016; Transport for NSW, 2016; Victoria State Government, 2016).

Chapter 4: Methodology 49

Chapter 4: Methodology

In correspond to the defined three interconnected sub-studies (Section 1.2), this

chapter contains three sections, each of them outlines and describes the modelling

methodology of each sub-study. A random parameter ordered logit model was

developed to identify significant factors associated with train riders’ satisfaction with

train fare as part of the first sub-study (Section 4.1). A random parameters binomial

logit model that accounts for heterogeneity in the population was estimated for the

perceptions, and a mixed logit model was employed to model the stated choice

responses as part of the second sub-study (Section Error! Reference source not

found.). Section 4.3 describes the utilisation of a nested logit model to estimate the

utility functions of mode choice responses.

4.1 URBAN TRAVELLERS’ SATISFACTION WITH TRAIN FARES IN

FIVE AUSTRALIAN CITIES

The satisfaction with train fare for the most recent train trip from home was

ordered and categorical. In this analysis, satisfaction was treated as ordinal rather than

nominal to provide simpler interpretations, greater flexibility, greater detection power,

and more similarity to ordinary regression analysis (Efthymiou et al., 2017; Zheng,

Liu, Liu, & Shiwakoti, 2014). In this study, travellers’ satisfaction with train fare was

estimated as a function of a number of socio-economic factors and trip characteristics,

using an ordered logit model. The logit model used the cumulative distribution

function of the logistic distribution (Albright, 2016). The logit model can be

generalized to account for non-constant error variances in more advanced econometric

settings, such as heteroskedastic or random-parameter logit model. Constraining all

parameters to be fixed while they are actually heterogeneous could lead to biased,

inefficient, and inconsistent parameter estimates (Bordagaray, dell'Olio, Ibeas, &

Cecín, 2014; Washington, Karlaftis, & Mannering, 2011).

In the context of this study, heterogeneity could arise in many factors such as

on-board crowding, one-way cost, on-board time, waiting time, total access time, and

total travel time (Beirão & Cabral, 2007; Bordagaray et al., 2014; Brons et al., 2009;

J. de Oña & de Oña, 2014; Hine & Scott, 2000; Zheng et al., 2016). To capture such

50 Chapter 4: Methodology

heterogeneity, these factors were considered as random parameters in the model

development because different respondents might perceive them differently. For

instance, a $2 one-way cost and 15 minutes waiting time can be perceived by one user

as a cheap fare, and a short waiting time; however, another user could consider the fare

to be expensive and the waiting time to be long. Such heterogeneity is influenced by

individuals’ various socio-economic factors and trip experience (Paramita et al., 2018).

The formulation of the ordinal data modelling problem, which is motivated by

the latent regression perspective, is defined as below.

𝑌 = 𝑗 𝑖𝑓 𝛼𝑗−1 < 𝑌∗ ≤ 𝛼𝑗 [4-1]

𝑌∗ is a continuous latent variable and is assumed to underlie the observed ordinal data.

Particularly, 𝑌∗ = 𝛽′𝑋 + 𝜀 and X is a vector of explanatory variables, 𝛽 is a vector of

coefficients, and 𝜀 is an error term. While, 𝑗 is an ordinal responses and 𝛼 is a set of

cut points of the continuous scale for 𝑌∗ (Agresti, 2010; Zheng et al., 2014). 𝑌 is

observed to be in category j when the latent variable falls in the 𝑗𝑡ℎ interval.

In order to maintain the order of ordinal dependent variables, the logit

transformation is applied to the cumulative probabilities, as below.

l𝑜𝑔𝑖𝑡[𝑃(𝑌𝑖 ≤ 𝑗)] = 𝑙𝑜𝑔 (𝑃(𝑌𝑖≤𝑗)

1−𝑃(𝑌𝑖≤𝑗) ) [4-2]

A general model for the cumulative logits is shown below

𝑙𝑜𝑔𝑖𝑡 [𝑃(𝑌𝑖 ≤ 𝑗)] = 𝛽1𝑋1 + 𝛽2𝑋2 + ⋯ + 𝛽𝑛𝑋𝑛 + 𝜀𝑖= 𝛽′𝑋 +𝜀𝑖 [4-3]

,where 𝑗 = 1, … , 𝑐 − 1; 𝑐 is the total number of categories. 𝑋1, 𝑋2, … , 𝑋𝑛 are the 𝑛

explanatory variables; 𝛽1, 𝛽2, … , 𝛽𝑛 are the corresponding coefficients (Zheng et al.,

2014). In this study, 𝑌𝑖 denotes the perceived satisfaction with the train fare of 𝑖𝑡ℎ

respondent.

In the fixed parameter ordered logit model above, the vector of parameters 𝛽 is

the same for all observations. On the other hand, a random-parameter ordered logit

model explicitly accounts for heterogeneities by allowing the regression coefficients

to vary across observations (Agresti, 2010; Long & Freese, 2006), as shown in below.

𝛽𝑖 = 𝛽 + 𝑢𝑖 [4-4]

,where 𝛽𝑖 is a vector of random regression coefficients and 𝑢𝑖 is a vector of randomly

distributed terms for each regression coefficient. The additional error term 𝑢𝑖 is

correlated with the error term 𝜀𝑖 of the perceived satisfaction function, and thus

translates individual heterogeneities into parameter heterogeneities. From Equations

4-3 and 4-4, the function for the perceived satisfaction level becomes


𝑙𝑜𝑔𝑖𝑡 [𝑃(𝑌𝑖 ≤ 𝑗)] = 𝛽𝑖′𝑋 +𝜀𝑖 = 𝛽′ 𝑋𝑖 + (𝑢𝑖′ 𝑋𝑖 + 𝜀𝑖). [4-5]

In this study, coefficients of on-board crowding, one-way cost, on-board time,

waiting time, total access time, and total travel time variables are considered as

candidate random parameters. More specifically, each of the random parameters is

assumed to follow a lognormal distribution restricted to the negative side because the

expected sign of the estimates is known to be negative (Paramita et al., 2018).

A simulation-based maximum likelihood method is employed to estimate the

random-parameter ordered logit model. Halton sequence (that provides more accurate

approximations for numerical integrations than purely random draws) is used to obtain

the simulation-based estimation (Bhat, 2003; Halton, 1960; Train, 2009). Following

the recommendation in the literature, 500 random Halton draws are used in estimating

the random parameters (Bhat, 2003; Hensher & Greene, 2002; Train, 1999). Following

the standard model development process, the best fixed- and the random-parameter

models are obtained based on the statistical significance independent variables and the

Akaike’s Information Criteria (AIC) (Akaike, 1992).

In the case of availability of two points from the explanatory variables, 𝑋𝑎 and

𝑋𝑏, then the cumulative logit is defined as

𝑙𝑜𝑔𝑖𝑡 [𝑃(𝑌𝑖 ≤ 𝑗|𝑋𝑎)] − 𝑙𝑜𝑔𝑖𝑡 [𝑃(𝑌𝑖 ≤ 𝑗|𝑋𝑏)] = 𝛽′( 𝑋𝑎 − 𝑋𝑏). [4-6]

The equation above specifies that the odds of making response 𝑌 ≤ 𝑗 at 𝑋𝑎 are

𝑒𝑥𝑝(𝛽′ ∗ ( 𝑋𝑎 − 𝑋𝑏)) times the odds of 𝑋𝑏. The log odds ratio is proportional to the

distance between these two points, and the proportionality remains constant across

different categories. Hence, Equation 4-3 is referred to as a “proportional odds” model.

Due to its easy interpretation, this model has been widely used in the literature

(Agresti, 2010; Agresti & Kateri, 2011; Greene & Hensher, 2010; Zheng et al., 2014).

4.2 CONSISTENCY BETWEEN PERCEPTIONS AND STATED

PREFERENCES DATA IN A NATIONWIDE MODE CHOICE

EXPERIMENT

As this study employed two data types (perceptions and mode choice experiment

responses) from the same survey, two different statistical models were estimated. A

statistical model for each data type. The probability of taking train service more than

once a month was estimated as a function of a number of perceptions of service factors

and socio-economic factors using random-parameter binomial logit model (Albright,


2016). While, the mixed logit model is employed to fit the mode choice experiment

responses (Hensher et al., 2005; Train, 2009).

4.2.1 Mixed logit model

Based on the mode choice experiment responses, this study develops three utility

functions and three probability functions for each of the three modes: bus, train, and

car. Each utility function has its own set of attributes and socio-economic factors.

Specifically, six mode choice experiment responses are collected from each

respondent. As the six responses come from the same decision maker, they are likely

to be correlated. Hence, the mixed multinomial logit (MMNL), or mixed logit model,

is employed to fit the mode choice experiment responses using the NLOGIT (Greene,

2000). This model is able to capture random taste variation, unrestricted substitution

patterns, and correlation in unobserved factors over repeated measures (Hensher et al.,

2005; Train, 2009).

A mixed logit model accounts for heterogeneities among respondents. A part, or

all, of the parameter estimates in this model are randomly distributed among

respondents (Hensher et al., 2005). The regression coefficients vary across

respondents.

𝛽𝑖𝑗𝑘 = 𝛽𝑗𝑘 + 𝑤𝑖𝑘 [4-7]

where 𝑤𝑖𝑘 is a random variable from some underlying distributions. If 𝐸[𝑤𝑖𝑘] = 0

and ŋ𝑘 is the standard deviation of 𝑤𝑖𝑘 , then

𝛽𝑖𝑗𝑘 = 𝛽𝑗𝑘 + (ŋ𝑘 ∗ 𝑧𝑖𝑘) [4-8]

where 𝛽𝑗𝑘 is the mean marginal utility in the sampled population, and ŋ is the standard

deviation of the marginal utilities held by respondents for attribute 𝑘 belonging to

alternative 𝑗 in choice set 𝑠. 𝑧𝑖𝑘 represents some underlying distributions, such as a

standard normal distribution [𝑧𝑖𝑘~ 𝑁(0,1)] (Hensher et al., 2005). Therefore, the

utility function of the mixed logit model is defined as follows:

𝑈𝑖𝑠𝑗 = ∑(𝛽𝑖𝑗𝑘 ∗

𝑘

𝐾=1

𝑥𝑖𝑠𝑗𝑘) + 𝜀𝑖𝑠𝑗 = ∑(( 𝛽𝑗𝑘 + (ŋ𝑘 ∗ 𝑧𝑖𝑘)) ∗

𝐾

𝑘=1

𝑥𝑖𝑠𝑗𝑘) + 𝜀𝑖𝑠𝑗

= 𝑉𝑖𝑠𝑗 + 𝜀𝑖𝑠𝑗 [4-9]

where 𝑖 signifies the respondent, 𝑠 is choice task, 𝑗 denotes alternative, and

𝑘 represents attribute. 𝑉𝑖𝑠𝑗 is a function of travel attributes, and 𝛽𝑗𝑘 is a function of


socio-economic factors influencing random parameter attributes and

𝜀𝑖𝑗~ 𝐼𝐼𝐷 𝐸𝑥𝑡𝑟𝑒𝑚𝑒 𝑉𝑎𝑙𝑢𝑒 𝑡𝑦𝑝𝑒 (𝐸𝑉)1(Caussade, de Dios Ortúzar, Rizzi, & Hensher,

2005). Congruently, the choice probability is defined as:

𝑃𝑖𝑠𝑗 =exp(∑ ( 𝛽𝑗𝑘+ŋ𝑘𝑧𝑖𝑘)∗𝑥𝑖𝑠𝑗𝑘

𝐾𝑘=1 )

∑ (exp(∑ ( 𝛽𝑗𝑘+ŋ𝑘𝑧𝑖𝑘)∗𝑥𝑖𝑠𝑗𝑘𝐾𝑘=1 ))

𝐽𝑗=1

[4-10]

Heterogeneity can arise in a number of attributes within mode choice

experiments. These attributes could be perceived differently across respondents, as

they are influenced by their socio-economic profiles. The candidates for random

parameters are: access time to/from train station/bus stop, bus/train waiting time, on-

board time, bus/train ticket, car toll cost, car fuel cost, availability of wireless

connection, and availability of laptop station. For instance, the $4.5 one-way train

ticket could be perceived as affordable by a respondent who is employed full time, but

might be perceived differently by an unemployed respondent. Specifically, the

availability of a wireless connection and a laptop station are assumed to follow a

standard normal distribution. The rest of the random parameter candidates are assumed

to follow a lognormal distribution because the sign of the estimates is expected to be

negative.

The marginal utility for a random parameter with an underlying normal

distribution is defined as:

𝛽𝑖𝑗𝑘 = 𝛽𝑗𝑘 + ∑ (𝛽𝑙 ∗ 𝑦𝑙)𝐿𝑙=1 + (𝜎𝑗𝑘 ∗ 𝑛𝑖) , 𝑛𝑖~ 𝑁 (0,1)

𝑎𝑛𝑑 𝑢𝑛𝑖𝑞𝑢𝑒 𝑓𝑜𝑟 𝑒𝑎𝑐ℎ 𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑒𝑛𝑡. [4-11]

while the marginal utility for a random parameter with lognormal as the underlying

distribution is defined as

𝛽𝑖𝑗𝑘 = 𝑒𝑥𝑝 (𝛽𝑗𝑘 + ∑ (𝛽𝑙 ∗ 𝑦𝑙)𝐿𝑙=1 + (𝜎𝑗𝑘 ∗ 𝑛𝑖)) , 𝑛𝑖~ 𝑁 (0,1)

𝑎𝑛𝑑 𝑢𝑛𝑖𝑞𝑢𝑒 𝑓𝑜𝑟 𝑒𝑎𝑐ℎ 𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑒𝑛𝑡 [4-12]

where 𝛽𝑗𝑘 and 𝜎𝑗𝑘 are the coefficient and the standard deviation of the corresponding

random parameter, respectively. 𝛽1 𝑡𝑜 𝛽𝐿 are the coefficients of influential socio-

economic factors, and 𝑦1 𝑡𝑜 𝑦𝐿 signify the corresponding socio-economic factors

(Greene, 2012; Hensher et al., 2005).

4.2.2 Random-parameters binomial logit model

Meanwhile, for the train rider category, the probability of taking a train more

than once a month is estimated as a function of a number of perceptions of service


factors and socio-economic factors by using the binomial logit model. The response

variable, 𝜋𝑖, can take the values between one and zero. 𝜋𝑖 is the respondents’

probability of making more than one train trip in the past month prior to the survey,

while (1 − 𝜋𝑖) is the respondents’ probability of making one trip or less. The various

perceptions of service factors are the candidates for the explanatory variables. In

particular, the choice probability of binomial logit is defined as:

𝜋𝑖 =exp( ∑ (𝛽𝑘∗𝑥𝑖𝑘)𝐾

𝑘=1 )

1+exp( ∑ (𝛽𝑘∗𝑥𝑖𝑘)𝐾𝑘=1 )

[4-13]

where 𝑖 signifies respondent, 𝑘 denotes the number of explanatory variables, and

𝑥𝑖1 to 𝑥𝑖𝐾 represent a series of explanatory variables (Rodrıguez, 2007).

Heterogeneities might arise within a number of perceptions of service factors

(Borins, 1988; Brons et al., 2009; Cox, Houdmont, & Griffiths, 2006; Evans & Wener,

2007; Fan, Guthrie, & Levinson, 2016; Farag & Lyons, 2012; Givoni & Rietveld,

2007; Santos & Rojey, 2004; Schwieterman, Fischer, Field, Pizzano, & Urbanczyk,

2009; Stanton et al., 2013; Stopher, 2004; Van Exel & Rietveld, 2009). Therefore, a

random-parameter binomial logit model is considered. This binomial logit model is a

special case of a mixed logit model for dichotomous response variables. It can be

generalized to account for non-constant error variances in more advanced econometric

settings, such as heteroskedastic or random-parameter binomial logit model (Albright,

2016).

Different respondents might have a different understanding of each of the

perceptions of service factors. For instance, better access to train station could strongly

influence one respondent’s decision to take a train more often, but might not influence

other respondents. Such heterogeneity can be influenced by respondents’ socio-

economic profiles. At the start of the model estimation, all perceptions are considered

as prospective random-parameters. Specifically, the perception of the availability of

an on-board entertainment system is assumed to follow a standard normal distribution

because of the uncertainty of which sign to expect. The rest of the prospective random

parameters are assumed to follow a gamma distribution because the sign of the

estimates is expected to be positive.


The marginal utility for a random parameter with gamma as the underlying

distribution is defined as:

𝛽𝑖𝑚 = (𝛽𝑚 + ∑ (𝛽𝑛 ∗ 𝑦𝑛))𝑁𝑛=1 ∗ 𝑣𝑖 ,

𝑣𝑖~ 𝑔𝑎𝑚𝑚𝑎 (1,4) 𝑎𝑛𝑑 𝑢𝑛𝑖𝑞𝑢𝑒 𝑓𝑜𝑟 𝑒𝑎𝑐ℎ 𝑟𝑒𝑠𝑝𝑜𝑛𝑑𝑒𝑛𝑡 [4-14]

𝛽𝑚 and 𝜎𝑚 are the coefficient and the standard deviation of the corresponding

random parameter, respectively. 𝛽1 𝑡𝑜 𝛽𝑁 are the coefficients of influential socio-

economic factors, and 𝑦1 𝑡𝑜 𝑦𝑁 signify the corresponding socio-economic factors

(Greene, 2012; Hensher et al., 2005).

A simulation-based maximum likelihood method is employed to separately

estimate the mixed logit and random-parameters binomial logit model. Five hundred

draws of Halton sequence each is used to obtain the simulation-based estimation (Bhat,

2003; Halton, 1960; Train, 2009). Various mixed logit and random-parameters logit

models are separately developed, and their performances are assessed using the

significance of the independent variables, the Akaike’s Information Criteria (AIC),

and logical soundness (Akaike, 1998).

4.2.3 Consistency assessment of the perceptions and mode choice experiment

responses

By reference to Table 3.8, the qualitative and quantitative assessments are then

performed to the best-fitted binomial logit model and mixed logit model results. Both

assessments are important approaches to determine whether respondents’ perceptions

are consistently aligned to their responses on mode choice experiments, and also to

estimate the extent of their consistencies.

Qualitative assessment is performed by identifying and mapping the significant

independent variables in the random-parameter binomial logit model, and their

comparable significant attributes in each of three utility functions estimated within the

mixed logit model. If both a perception response and its corresponding attribute are

found to be significant, the former is considered to be consistently aligned to the latter

when respondents decide their mode preferences in the mode choice experiment. If

this is not the case, their consistency cannot be properly assessed.

Subsequently, the quantitative assessment – via probability functions – is

performed to further understand the magnitude of respondents’ preference changes in

the shift of one particular explanatory factor, while having all other factors constant.


The probability is one way of estimating the rate of change in one variable relative to

the rate of change in a second variable, and is expressed as a unit change. More

specifically, the change in probability value as the result of a unit change in one

variable, when other conditions remain the same, or that other things are equal

(Boumans & Morgan, 2001; Hensher et al., 2005).

4.3 POLICY INTERVENTIONS STUDY TO ENCOURAGE

BEHAVIOURAL SHIFT FROM CAR TO PUBLIC TRANSPORT

A restrictive property of a multinomial logit (MNL) model is its independence

from irrelevant alternatives (IIA). IIA assumes independence of disturbance

terms (𝜀𝑖𝑛), among alternative outcomes. In order to overcome the IIA limitation,

McFadden (1981) developed a class of models known as ‘generalized extreme value

(GEV) models’, which addressed the IIA problem (Washington et al., 2011). The

nested logit model is one of the commonly used models in the GEV category.

The nested logit model aims to group into the same nest alternative outcomes

that are suspected of sharing unobserved effects (Washington et al., 2011). Providing

that all alternatives in the nest share unobserved effects, these effects cancel out in

each nest. This cancelling out does not occur unless all alternatives in the same nest

share the unobserved effects. This is an IIA violation in the nest (Washington et al.,

2011). The nested structure is purely an empirical method for eliminating IIA

violations, and does not convey information about a hierarchical decision-making

process. It often contains more than two levels, depending on the number of

alternatives and the hypothesized disturbance correlations (Washington et al., 2011).

Within this study, the hypothesized nested structure accounts for the differences

in the availability, rules of usage, fare, and route status of the transport mode. The

mode choice is divided into two: public transport and private transport. ‘Public

transport’ is defined as the mode of transport that is available for public use; is based

on certain rules of usage; charges set fares; and runs on fixed routes with a certain

schedule (Stevenson, 2010). ‘Private transport’, on the other hand, consists of transport

modes that are not available for the general public; it does not have specific rules of

usage, set fares, or fixed routes and schedules. Bus and train are classified as public

transport, and cars are classified as private transport. Private transport is a degenerate

branch, with only one alternative within the nested logit structure (Hensher et al.,

2005). This mode choice framework is represented as a nested structure in Figure 4.1,


and the feasibility of this hypothesized structure is statistically tested in Section 7.2.

Accordingly, the estimated nested logit model contains a number of utility functions,

each with its own set of key travel attributes and socio-economic factors.

Figure 4.1 : The nested structure of the mode choice experiment

The proposed nested structure Figure 4.1 underlie the nested logit model utilised

in the third sub-study. If only one level would be utilised, then the modelling

methodology would have transformed to mixed logit model instead of nested logit

model. The mixed logit model would not be the preferred statistical model for the third

sub-study because it would not allow the identification of the shift behaviours between

public transport riders and car drivers.

McFadden (1981) shows that the GEV disturbances assumption leads to the

following model structure for observation 𝑛 choosing outcome 𝑖 :

𝑃𝑛(𝑖) = 𝐸𝑋𝑃 [ 𝛽𝑖𝑋𝑖𝑛+∅𝑖𝐿𝑖𝑛]

∑ 𝐸𝑋𝑃 [ 𝛽𝐼𝑋𝐼𝑛+∅𝐼𝐿𝑆𝐼𝑛]∀𝐼 [4-15]

𝑃𝑛(𝑗|𝑖) = 𝐸𝑋𝑃 [ 𝛽𝑗|𝑖𝑋𝑛]

∑ 𝐸𝑋𝑃 [ 𝛽𝐽|𝑖𝑋𝐽𝑛]∀𝐼 [4-16]

𝐿𝑆𝑖𝑛 = 𝐿𝑁 [∑ 𝐸𝑋𝑃 ( 𝛽𝐽|𝑖𝑋𝐽𝑛)]∀𝐼 [4-17]

where 𝑃𝑛(𝑖) is the unconditional probability of observation 𝑛 having discrete outcome

𝑖, 𝑋 are vectors of measurable characteristics that determine the probability of discrete

outcomes, and 𝛽 are vectors of estimable parameters. 𝑃𝑛(𝑗|𝑖) is the probability of

observation 𝑛 having discrete outcome 𝑗, conditioned on the outcome being in outcome

category 𝑖. 𝐿𝑆𝑖𝑛 is the inclusive value (IV), also known as IV, and ∅𝑖 is an estimable

parameter. Meanwhile, the unconditional probability of observation 𝑛 of having

outcome 𝑗 is defined as (McFadden, 1981),

𝑃𝑛(𝑗) = 𝑃𝑛(𝑖)𝑥𝑃𝑛(𝑗|𝑖) [4-18]


The estimation of a nested model was typically done in a sequential fashion

(McFadden, 1981). This method first estimated the conditional model (Equation 2),

using only the observations in the sample that were observed to have discrete

outcomes. Once these estimation results were obtained, the IV was calculated for all

observations, both those selecting 𝑗, and those not. Subsequently, these computed IVs

were used as independent variables in the functions, as shown in Equation 1. It was

not necessary for all unconditional outcomes to have an IV in their respective

functions. The downside of the sequential estimation method was that the variance-

covariance matrices were too small. Thus, t-statistics were inflated by about 10%-15%.

In order to address the disadvantage, the entire model was estimated at once, using

full-information maximum likelihood (FIML) and Nlogit, a modern software package

that provides for a simultaneous estimation of all nests (Greene, 2000). Details of this

simultaneous approach are available at Hensher, Rose, & Greene (2008)

The IV has two roles: to provide a basis for identifying behavioural relationships

among choices at each level of the nest, and to indicate the validity of the nested

structure (Hansen, 1987; Washington et al., 2011). The interpretation of the estimated

parameter associated with the IV (∅𝑖) has the following important elements

(Washington et al., 2011): ∅𝑖 must be between 0 to 1 in magnitude to be consistent

with the nested logit derivation (McFadden, 1981); if ∅𝑖 is equal to 1, the assumed

shared unobserved effects in the nest are not significant, and the nested model reduces

to a simple MNL; if ∅𝑖 is less than zero, then the factors that increase the likelihood of

an outcome being chosen in the lower nest will decrease the likelihood of the nest

being chosen; if ∅𝑖 is equal to zero, then changes in nest outcome probability will not

affect the probability of nest selection, and the correct model is separated (McFadden,

1981; Washington et al., 2011). The t-test is employed to test whether the parameter

estimate is significantly different from 1. The t-test is chosen over other tests because

of the non-normality in the logit model (Washington et al., 2011). It is defined as,

𝑡∗ =𝛽−1

𝑆.𝐸.(𝛽) [4-19]

Chapter 5: Urban Travellers’ Satisfaction with Train Fares in Five Australian Cities 59

Chapter 5: Urban Travellers’ Satisfaction

with Train Fares in Five

Australian Cities

This chapter discusses the findings of the first sub-study (Urban travellers’

satisfaction with train fares in five Australian cities) (Paramita et al., 2018). This study

uses two main data sources: objective information on the characteristics of each city’s

existing train system, and a nationwide survey. Hence, this chapter begins with the

summary of the former dataset in order to provide the study context and information

on the train fare structure in each of the targeted cities, which is the base knowledge

for analysing the nationwide survey data (Section 5.1). Subsequently, this chapter

reports and analyses the modelling results of the first sub-study (Section 5.2 to 5.5). It

also includes the heterogeneity discussions (Section 5.6) and detailed intercity

comparison results of train fare structures in the five Australian cities (Section 5.7).

Section 5.8 summarises the main conclusions of the study.

5.1 TRAIN FARE STRUCTURES IN THE FIVE AUSTRALIAN CITIES

A city’s geographical spread, land-use planning, and overall public transport

network also influence travellers’ perceptions of the overall transportation system, and

especially their perceived satisfaction with public transport services (Glaeser, Kahn,

& Rappaport, 2008). In the data analysis of this study, objective information on the

characteristics of the existing train fare structure in each of five Australian capital cities

was used to explain the diverse perceived satisfaction with train fare. Intuitively,

information on the current train fare structure in each of the targeted cities is the base

knowledge for analysing the nationwide survey data. Particularly, such information

can assist readers in better understanding underlying reasons behind the differences in

user satisfaction across the five cities revealed by our modelling analysis. Thus, the

actual train fare structure in each city based on the practice in 2016 is explained in this

sub section. Although train fare airport surcharge exists in some of the capital cities,

it is not included in the study due to its irrelevancy. The train riders, who have to access

60 Chapter 5: Urban Travellers’ Satisfaction with Train Fares in Five Australian Cities

the airports within their regular commutes, own the airport staff pass and hence, they

do not have to pay any airport surcharge.

5.1.1 Sydney

“Opal” is a smartcard system that has been implemented by Transport for New

South Wales for commuters using various public transport modes across Sydney, the

Blue Mountains, Central Coast, Hunter or the Illawarra region (Transport for NSW,

2016). The correct amount of fare is automatically deducted, and is based on the

distance travelled. Four different fares are available: Adult, Child/Youth, Concession

and Senior/Pensioner. With respect to train services, it costs $3.38 for an adult, $ 1.69

for a child, $ 1.69 for concession-eligible travellers, and $ 1.69 for seniors/pensioners

to travel between 0 to 10 km; and up to $8.30 for an adult, $ 4.15 for a child, $ 4.15

for concession-eligible travellers, and $ 2.50 for seniors/pensioners to travel above 65

km one-way during peak hours.

The off-peak fare is 70% of the peak fare. The off-peak periods are outside the

peak periods of 7:00 to 9:00 AM and 4:00 to 6:30 PM for Sydney train services, and

outside the peak periods of 6:00 to 8:00 AM and 4:00 to 6:30 PM for intercity train

services. There is a $ 2.50 cap for all Opal trips taken on Sundays, and a weekly travel

reward for travellers who make eight paid trips in a week (Transport for NSW, 2016).

Travellers are entitled to save around 20% by using their Opal card rather than

purchasing single trip tickets.

5.1.2 Melbourne

Public Transport Victoria (Victoria State Government, 2016) issues “Myki”, a

smart card, as its method of payment across public transport in Melbourne. This card

automatically deducts the lowest fare possible, based on the travellers’ departing and

alighting locations. The card is divided into two categories, Myki money and Myki

pass. Full and concession fare are available for each category. There is a cap of $6 for

a day fare during the weekend and public holiday when using Myki across Zones 1

and 2 (Victoria State Government, 2016). Travellers who have touched off before 7:15

AM are eligible for a free early bird train travel. Two fare bands, Zones 1+2 and Zone

2 are available for Myki money for both 2-hour usage and daily fare. Zone 1 of

Melbourne’s train, tram, and bus networks is the CBD and inner city suburbs, an

approximately 12 km radius. Zone 2 covers the middle and outer suburbs.


5.1.3 Brisbane

A smartcard system, “Go Card”, has been implemented by Translink in the South

East Queensland Region, including Brisbane. Go Card allows travellers to travel

seamlessly on all public transport; i.e., on Translink’s bus, ferry, train, and tram

services (Queensland Government, 2016). Travellers are entitled to savings when they

use a Go Card rather than paper tickets. The Go Card automatically calculates and

deducts the overall fare at either an adult or concession rate, based on the number of

zones travelled through the trip (Queensland Government, 2016).

The off-peak Go Card fares are 80% of the peak fares. The off-peak periods are

between 8:30 AM and 3:30 PM and 7:00 PM and 6:00 AM on weekdays, and all

weekend. Paper tickets cost 130% of the Go Card fare, while the concession fares are

50% of adult fares (Queensland Government, 2016).

5.1.4 Perth

Perth public transport has two types of ticket: A SmartRider card and cash

tickets. The SmartRider card is a refillable smart card with seven categories: standard,

concession, senior, pensioner, veteran, student, and tertiary (Government of Western

Australia, 2016).

With the exception of two-section fares (valid for a single one-way trip), cash

tickets have an expiry time; however, within the allowable time travellers are free to

ride on any number of buses, trains, and ferry services to complete their trip

(Government of Western Australia, 2016). Travellers who use SmartRider should

remember to touch on and off upon boarding and alighting from public transport

services. The expiry times are two hours for a trip of one to four zones, and three hours

for a trip of five or more zones. Overall, Transperth (public transport system serving

the city and suburban areas of Perth) covers services across nine different zones. Each

zone has an approximately 8 km radius. Within the city centre, there is a free zone

area, where all public transport is completely free.


5.1.5 Adelaide

There are two public transport tickets available in Adelaide, “Metrotickets” and

“Metrocard” (Government of South Australia, 2016). Metrotickets are paper tickets

for both single and day trips across Adelaide Metro trains, trams, and buses, and are

the best option for infrequent public transport users. Metrocard, on the other hand, is

an electronic smart card designed for multiple public transport trips on Adelaide Metro

trains, trams, and buses.

There is no public transport zone division in Adelaide; therefore, fares are not

calculated according to the distance travelled. Overall, there are four different fares

offered to Adelaide public transport users: regular fares, concession and tertiary

student fares, primary and secondary student fares, and senior Metrocard (Government

of South Australia, 2016). The interpeak, regular Metrocard fares are 55% - 75% of

the regular peak fares. The peak periods are before 9:01 AM and after 3:00 PM on

weekdays, and all day Saturday. Inter peak periods are Monday to Friday 9:01 AM to

3:00 PM, and all day Sundays and public holidays. Concession and tertiary student as

well as senior Metrocard fares are 50% of regular Metrocard fares, while primary and

secondary student Metrocard fares are 30% of regular Metroard fares.

5.2 MODELLING RESULTS

Table 5.1 shows the summary of the best fixed-parameter and random-parameter

ordered logit models of perceived satisfaction with train fare. A likelihood ratio test

was used to compare the performance of these models; the result shows that in terms

of explaining travellers’ perceived satisfaction with their train fare, the random-

parameter ordered logit model performs statistically better than the fixed-parameter

model, at a 95% significance level.

Table 5.1 Summary of the best fixed-parameter and random-parameter logit models

Explanatory

Variables

Fixed-parameter model Random-parameter model

Coefficient z-statistics p-value Coefficient z-statistics p-value

Constant 3.684 25.18 <0.05 4.349 22.14 <0.05

Female -0.255 -3.69 <0.05 -0.293 -4.15 <0.05

Sydney -0.328 -2.29 <0.05 -0.357 -2.14 <0.05

Melbourne -0.371 -2.75 <0.05 -0.419 -2.63 <0.05

Brisbane -0.478 -3.12 <0.05 -0.552 -3.16 <0.05

Perth 0.406 2.55 <0.05 0.398 2.22 <0.05

Employment

status: Outside

work force

0.205

1.91 0.057 0.118 1.07 0.284

Transport mode

from home to

<0.05 0.404 3.83 <0.05


Explanatory

Variables

Fixed-parameter model Random-parameter model

Coefficient z-statistics p-value Coefficient z-statistics p-value

the train station:

Bus

0.233

2.32

Student

concession fare

-0.488

-2.56 <0.05 -0.435 -2.16 <0.05

Senior

concession fare

2.349

11.4 <0.05 2.448 11.70 <0.05

Sydney *

Employment

status: Outside

work force

0.460

2.35

<0.05 0.608 3.11 <0.05

Perth *

Transport mode

from home to

the train station:

Bus

-0.438

-2.06

<0.05 -0.491 -2.23 <0.05

Sydney *

Student

concession fare

-0.856

-2.35

<0.05 -1.024 -2.74 <0.05

Melbourne *

Senior

concession fare

-1.288

-4.36

<0.05 -1.350 -4.31 <0.05

Brisbane *

Senior

concession fare

-1.155

-3.48

<0.05 -1.077 -2.95 <0.05

One-way cost

($)a 0.082 7.37 <0.05 -2.482 -16.49 <0.05

Waiting time

(Minutes)a 0.038 5.29 <0.05 -2.988 -16.51 <0.05

µ1b 1.514 35.80 <0.05 1.848 20.51 <0.05

µ2b 2.917 72.84 <0.05 3.404 33.90 <0.05

µ3b 4.781 84.09 <0.05 5.354 47.48 <0.05

Log-likelihood

at convergence

-4024.145

-3976.970

AIC 2.763 2.739 a random parameters b µ1 , µ2 and µ3 are the thresholds of perceived satisfaction with train fare estimated by the model.

The random-parameter ordered logit model includes two random parameters in

the final model: one-way cost and waiting time. For these two parameters, the standard

deviations of the lognormal distribution are significantly different from zero, as shown

in Table 5.2

Table 5.2 Distribution of the random parameters

Random parameter Underlying distribution

and constrained

Standard

deviation

z-statistics p-value

One-way cost ($) Lognormal - negative 1.268 14.56 <0.05

Waiting time

(Minutes)

Lognormal - negative 0.549 7.65 <0.05

As shown in Table 5.3, the random-parameter ordered logit model identifies a

number of socio-economic factors and trip characteristics that contribute to the


heterogeneities in the influence of one-way cost and waiting time on perceived

satisfaction with train fare.

Table 5.3 Heterogeneities in the random parameters

Random

Parameter

Attribute Coefficient

Estimate

z-statistics p-value

One-way cost ($) Household composition:

Adults and children -0.464 -4.13 <0.05

Transport mode from home

to the train station: Dropped-

off -0.727 -3.69 <0.05


to the train station: Driving -0.485 -3.63 <0.05

Waiting time

(Minutes)

Employment status: Self-

employed

0.619 3.59

<0.05


to the train station: Driving

-0.345

-2.17 <0.05

Trip purpose:

Shopping/Personal/ Business -0.364 -2.35 <0.05

Paid fare status: Full fare 0.615 4.66 <0.05

Train services’ influence on

the current home location

decision: Significant -0.604 -4.27 <0.05



decision: Moderate -0.747 -3.53 <0.05

5.3 KEY SOCIO-ECONOMIC FACTORS

The best fitted random-parameter ordered logit model is estimated by the socio-

economic factors discussed below.

5.3.1 Gender

Female respondents are significantly (p-value < 0.05) less satisfied with their

train fare compared with their male counterparts. When other factors are controlled, a

female respondent’s estimated odds of responding “extremely satisfied”, rather than

“satisfied”, “extremely dissatisfied”, “dissatisfied”, or “neutral”, decrease by 25% in

comparison to a male respondent.

This finding is in line with gender differences in general. Women tend to be more

sensitive to monetary costs, more likely to shop for higher quality products or services

than men, and more effective in distributing their income (Dwyer, 1983; Furnham,

2016) . Dwyer (1983) also found that women allocate most of their income to buying

goods rather than procuring services, and allocate a smaller amount to their travel

budgets. Consequently, women are more likely to have a higher expectation of train

services for the fare paid compared with their male counterparts (Paramita et al., 2018).


In addition, Ellaway et al. (2003) found that male travellers prefer to use private

vehicles, and are most likely to have alternatives to public transport (Andersson &

Nässén, 2016; Paulley et al., 2006). This implies that they tend to ride trains less often

and have less familiarity with train services when compared with female travellers.

Thus, male travellers tend to be more accepting of the asking fare, and do not have the

same high expectation of train services as female travellers. (Paramita et al., 2018)

5.3.2 City of origin

Table 5.1 shows that respondents’ perceived satisfaction with train fare differs

significantly across cities. When other factors are controlled, Sydney, Melbourne, and

Brisbane respondents feel less satisfied with train fare compared with respondents in

Adelaide. Specifically, by holding other factors constant, when compared with a

respondent from Adelaide, the estimated odds of a Sydney respondent responding that

they were “extremely satisfied”—rather than “satisfied”, “extremely dissatisfied”,

“dissatisfied”, or “neutral”—decrease by 30% (i.e., (1 − exp(−0.357)) ∗ 100%)

(Agresti & Kateri, 2011). Similarly, the estimated odds of a Melbourne respondent

responding that they were “extremely satisfied”—rather than “satisfied”, “extremely

dissatisfied”, “dissatisfied”, or “neutral”—decrease by 34%. Also, for a respondent

from Brisbane, these estimated odds decrease by 42%.

However, when other factors are controlled, compared with respondents from

Adelaide, Perth respondents feel more satisfied with their train fare. Specifically, when

controlling for other factors, when compared with a respondent from Adelaide, a Perth

respondent’s estimated odds of responding that they are “extremely satisfied”, rather

than “satisfied”, “extremely dissatisfied”, “dissatisfied”, or “neutral”, increase by 49%

(i.e., (exp(0.398) − 1) ∗ 100%) (Agresti & Kateri, 2011).

These findings are not surprising, given the different characteristics of the public

transport network and the diverse train fare structures across the five cities as

elaborated in the Context: train fare structures in the five Australian cities section. This

issue is further discussed in the Intercity comparison section (Paramita et al., 2018).

5.4 KEY TRIP CHARACTERISTICS

Our modelling results also clearly show that characteristics of their most recent

train trip significantly influence respondents’ satisfaction with the fare for that trip, as

elaborated below.


5.4.1 Transport mode from home to the train station

Respondents who take the bus to the train station are significantly (p-value <

0.05) more satisfied with their paid train fare than other respondents. Specifically,

when controlling for other factors, the estimated odds of an “extremely satisfied”

response for someone who takes the bus rather than other transport modes to the

station—rather than a “satisfied”, “extremely dissatisfied”, “dissatisfied”, or “neutral”

response—increase by 50%.

Again, this finding is not surprising, as the importance of train station access is

widely acknowledged in the literature (Brons et al., 2009; Caulfield & O'Mahony,

2007; Daniels & Mulley, 2013; Rissel, Curac, Greenaway, & Bauman, 2012). First,

the transport mode from home to the station influences how much effort travellers need

to exert and, also, the travel cost. This latter factor is especially vital in this study

because of the availability of fare transfer between public transport trips within a

certain time window in Sydney, Melbourne, Brisbane, Adelaide, and Perth. This

transferability often leads to a reduced train fare for respondents who take the bus to

the station (Government of South Australia, 2016; Government of Western Australia,

2016; Queensland Government, 2016; Transport for NSW, 2016; Victoria State

Government, 2016). A respondent’s experience with their mode of station access is

able to influence their satisfaction with the train trip in general, and the paid train fare

in particular (Brons et al., 2009). The existence of advance transit systems in major

train stations in Sydney, Melbourne, Brisbane, Adelaide, and Perth enables an easy

transfer between bus and train. Therefore, as revealed in our analysis, bus access to the

train station appears to positively contribute to a respondent’s satisfaction with their

train fare (Paramita et al., 2018).

5.4.2 Concession fare

Eligibility for a concession fare significantly (p-value < 0.05) affects a

respondent’s satisfaction with their train fare. Respondents who are eligible for a

student concession fare feel less satisfied with train fare than respondents who are

ineligible. When other factors are controlled, the estimated odds of an “extremely

satisfied” response for a respondent who is eligible for student concession fare —rather

than a “satisfied”, “extremely dissatisfied”, “dissatisfied”, or “neutral” response—

decrease by 35% in comparison to a respondent who is ineligible. However, compared


with respondents that are not eligible for any concession fare, respondents who are

eligible for a senior concession fare feel more satisfied with their fare. Specifically,

when controlling for other factors, compared with a respondent who is not eligible for

any concessional fare, the estimated odds of an “extremely satisfied” response from a

respondent who is eligible for a senior concession fare —rather than a “satisfied”,

“extremely dissatisfied”, “dissatisfied”, or “neutral” response—increase by 10.6

times.

Student and senior concession fare are designed to provide subsidies for students

and senior citizens respectively. The availability of subsidies influences seniors’ and

student travellers’ perceived satisfaction with train fare. Respondents who are eligible

for a senior concession fare pay less than respondents who are ineligible for any

concession fare. Having received a subsidy through a reduced train fare, senior

travellers feel more satisfied with their paid train fare. Subsidies are also directed to

the disadvantaged transport group (Starrs & Perrins, 1989) who have decreased

mobility, earn a low income, or both.

Interestingly, this study found that respondents who are eligible for student

concession fares feel less satisfied with their train fare, despite having received a

subsidy and only paying part of the adult non- concessional fare. This might simply

imply that the student concession fare in the five targeted Australian cities is still

perceived as too expensive by most students, who have no regular income and a limited

travel budget (Paramita et al., 2018).

5.4.3 One-way cost

Not surprisingly, the one-way cost paid by each respondent significantly (p-

value < 0.05) influenced his/her perceived satisfaction with train fare. Specifically, this

study identified heterogeneity in the influence of one-way cost on the perceived

satisfaction with train fare over the sampled population. The parameter of one-way

cost varies significantly across respondents, as indicated by the significant standard

deviation (p-value < 0.05) of its parameter. This finding is in line with earlier studies

(Hine & Scott, 2000; Wardman, 2004; Weisbrod & Reno, 2009). The marginal utility

of one-way cost is linked with both socio-economic factors (household composition)

and trip characteristics (i.e., transport mode from home to the train station) (Paramita

et al., 2018). Further discussion on the heterogeneity of one-way cost is provided in

the Heterogeneity sub-section.


5.4.4 Waiting time

This study found that the impact of waiting time varies across respondents, as

indicated by the significant (p-value < 0.05) standard deviation of its parameter. Table

5.3 shows how the perception of waiting time differs according to employment status,

transport mode from home to the train station, trip purpose, paid fare status, and the

influence of train services on the current home location. This finding is not surprising

because the waiting time is perceived as an unproductive period, and travellers (who

often experience long waiting times) feel less satisfied with public transport service

and, consequently, more stressed than their counterparts (Beirão & Cabral, 2007;

Cantwell et al., 2009; Dell’Olio et al., 2010; Dziekan & Kottenhoff, 2007). Long

waiting times are caused by services not running to schedule. This lack of reliability

of public transport results in travellers feeling a diminished sense of control (Cantwell

et al., 2009). Over time, long waiting times also lead to intense and prolonged feelings

of stress. On the other hand, an increase on-board accessibility leads to both an increase

in the train ridership increment, and a perceived satisfaction with the paid train fare

(Brons et al., 2009; Delmelle, Haslauer, & Prinz, 2013).

The heterogeneity of waiting time is also not surprising because the perception

of waiting time can be influenced by numerous factors (Paramita et al., 2018). For

example, the availability of at-stop, real-time information and a comfortable waiting

area can improve the perceived quality of public transport services by reducing the

perceived waiting time (Dziekan & Kottenhoff, 2007; Litman, 2008). The availability

of real-time information also reduces the unit costs of waiting time because travellers

experience a more organized trip and reduced stress (Litman, 2008). The marginal

utility of one-way cost and further discussion on heterogeneity of waiting time is

provided in the Heterogeneity sub-section.

5.5 INTERACTION VARIABLES

The significant (p-value < 0.05) key interaction variables identified in the best-

fitted ordered logit model and how they affect the respondents’ level of satisfaction

with train fare are discussed below.


5.5.1 Sydney and Employment status: Outside work force

There is a positive relationship between interaction of Sydney and outside work

force against the level of satisfaction with train fare. By controlling for other factors,

the estimated odds of an outside work force Sydney respondent responding that they

were “extremely satisfied”—rather than “satisfied”, “extremely dissatisfied”,

“dissatisfied”, or “neutral”—increase by 1.07 times (i.e., (exp(0.608+0.118)-1)

*100%), compared with a respondent who is from Sydney but is not outside work

force. In addition, regardless of the city, the outside work force employment status is

positively related to a respondent’s satisfaction level with the paid train fare.

The aforementioned finding might be related to the impact of employment status

on trip frequency and travel budget allocation (Taylor & Morris, 2015). Travellers’

perceptions of their trip is an important determinant of their satisfaction level (Beirão

& Cabral, 2007), and the frequent and regular train rides of employed commuters

enable them to establish a sense of familiarity with, and expectations of services. Not

surprisingly, regular riders tend to have high expectations of train services for the fare

they paid. However, outside workforce respondents’ train travel is generally less

frequent and more irregular (Daniels & Mulley, 2013). Thus, they are more likely to

be less familiar with, and have lower expectations of train services. In addition, they

are more likely to be satisfied with the fare paid for their most recent trip (Paramita et

al., 2018).

5.5.2 Perth and Transport mode from home to the train station: Bus

Interaction of Perth and taking bus to train station is negatively related with the

level of satisfaction with train fare. This is quite interesting because such interaction

makes taking bus from home to train station impact differently on the level of

satisfaction with the paid train fare for a respondent from Perth, compared with that

for a respondent from the other cities. More specifically, for a Perth respondent who

takes bus from home to the train station, by controlling for other factors, the estimated

odds of being “extremely satisfied”—rather than “satisfied”, “extremely dissatisfied”,

“dissatisfied”, or “neutral”—decrease approximately by 8%, compared with a

respondent who is from the same city but uses other modes for travelling from home

to the train station. In contrast, for a respondent from any other city who takes bus

from home to the train station, by controlling for other factors, the estimated odds of

being “extremely satisfied”—rather than “satisfied”, “extremely dissatisfied”,


“dissatisfied”, or “neutral”—increase by about 50%, compared with a respondent who

is from the same city but uses other modes for travelling from home to the train station.

The finding above highlights the importance of the supporting role of bus

service. The poor quality of the bus service that is part of a train trip can significantly

and negatively impact a train user’s satisfaction with the train fare paid. It appears that

Perth respondents are less pleased with the reliability of bus services to connect their

homes to the nearest train stations than other cities’ respondents. It is possible that the

bus stops are not evenly located across the surrounding neighbourhood or the

schedules’ are not properly aligned with train services to cater the train riders’

demands (Paramita et al., 2018).

5.5.3 Sydney and Student concession fare

A negative relationship is found between the interaction of Sydney and Student

concession fare and the level of satisfaction with train fare. This demonstrates that

students in Sydney have less appreciation towards student concession fare offered by

Transport for New South Wales (NSW). The Transport for NSW official website

mentions that student concession fare is only provided for primary and secondary

students in NSW, full time Australian tertiary students and limited full time

international students who are fully funded by specified Australian Government

scholarships (Transport for NSW, 2016). The best-fitted model perhaps reflects the

scenario that many Sydney respondents who are 16-30 years old and are student do

not actually meet the eligibility requirements for receiving the student concession fare

(Paramita et al., 2018).

5.5.4 City of origin and Senior concession fare

Our modelling analysis also reveals some complex interactions between city of

origin and senior concession fare in terms of respondents’ satisfaction level with the

paid train fare. Although receiving senior concession fare generally increases a

respondent’s level of satisfaction with the paid train fare, this trend can be significantly

influenced by respondents’ city of origin. For example, for a Melbourne respondent

who receives the senior concession fare, by controlling for other factors, the estimated

odds of being “extremely satisfied”—rather than “satisfied”, “extremely dissatisfied”,

“dissatisfied”, or “neutral”— increase approximately by 2 times, compared with a

Melbourne respondent who is not eligible for receiving the senior concession fare. A


similar impact of receiving the senior concession fare is found for respondents from

Brisbane as the estimated odds for a Brisbane respondent of being “extremely

satisfied”—rather than “satisfied”, “extremely dissatisfied”, “dissatisfied”, or

“neutral”— increase approximately by 2.9 times , by controlling for other factors. The

positive impact of receiving the senior concession fare in the other cities are even

stronger.

These findings establish that Melbourne and Brisbane respondents who are 60

or more years old and are retired appear to have less appreciation towards the senior

concession fare offered by Victoria State Government (Victoria State Government,

2016) and by Queensland Government (Queensland Government, 2016), respectively.

This could be due to different eligibility criteria, such as minimum age limit, minimum

paid working hours, and residency status set by transport authorities in each state

(Government of South Australia, 2016; Government of Western Australia, 2016;

Queensland Government, 2016; Transport for NSW, 2016; Victoria State

Government, 2016). Our model may have reflected the scenario that many Melbourne

and Brisbane respondents, who are at least 60 years old and are retired, are not actually

eligible for the senior concession fare in their respective city of origin (Paramita et al.,

2018).

5.6 HETEROGENEITY

As discussed in the previous sub-sections, notable heterogeneity is detected

across respondents in their perceived satisfaction with train fare. Specifically,

significant heterogeneity is observed for one-way cost and waiting time. The analysis

shows that this heterogeneity can be explained, to a certain extent, by socio-economic

factors and trip characteristics, as elaborated below.

5.6.1 Household composition

Respondents who belong to the household of adults and children have a

significant influence on the heterogeneity in respondents’ sensitivity to one-way costs.

The marginal utility of one-way cost is: − 𝑒𝑥𝑝 [−2.482 −

0.464 𝑥 (ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑜𝑓 𝑎𝑑𝑢𝑙𝑡𝑠 𝑎𝑛𝑑 𝑐ℎ𝑖𝑙𝑑𝑟𝑒𝑛) −

0.727 𝑥 (𝑏𝑒𝑖𝑛𝑔 𝑑𝑟𝑜𝑝𝑝𝑒𝑑 𝑜𝑓𝑓 𝑎𝑡 𝑡ℎ𝑒 𝑡𝑟𝑎𝑖𝑛 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) − 0.485 𝑥 (𝑑𝑟𝑖𝑣𝑒 𝑡𝑜 𝑡ℎ𝑒 𝑡𝑟𝑎𝑖𝑛 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) +

1.268 𝑥 𝑛], where 𝑛 is a random number generated from a standard normal distribution.

To gain insights into the influence of household composition on respondents’


sensitivity to one-way cost, this marginal utility has been simulated for 100 randomly

selected individuals by holding both the transport modes from home to the train station

factors constant, as shown in the top-left subfigure of Figure 5.1. Specifically, 50% of

the randomly selected individuals belong to a household of adults and children. This

figure clearly shows strong heterogeneity across individuals, regardless of their

household composition.

By controlling randomness and other factors, individuals who belong to a

household of adults and children are less sensitive to one-way cost than individuals

from other type of households. Although individuals who belong to household of

adults and children are likely to pay a large amount of train fare when travelling as a

family, a heavily discounted fare is available for the children. On weekdays, the

children fare is a small portion of the adult fare and, under certain circumstances, is

free during weekends (Government of South Australia, 2016; Government of Western

Australia, 2016; Queensland Government, 2016; Transport for NSW, 2016; Victoria

State Government, 2016). As a result, this group of individuals is less sensitive to the

variation in one-way cost than their counterparts (Paramita et al., 2018).


Figure 5.1 : The simulated marginal utility of one-way cost for 100 randomly selected

individuals

5.6.2 Transport mode from home to the train station

Similarly, two other simulation exercises of marginal utility of one-way cost are

performed for the mode to the train station variables, as shown in the top right and the

bottom subfigures of Figure 5.1. Each simulation run aims to attain a deeper

understanding on the influence of a particular key variable on the marginal utility of

one-way cost, by controlling for the rest of variables. Each simulation specifies that

half of the randomly selected individuals belong to a key variable in focus and the

other half do not. In both simulations, clear heterogeneities across individuals were


detected, regardless of their mode of transport from home to the station. Specifically,

travellers who were dropped-off at, and who drove to the train station appear to be less

sensitive to one-way cost than other travellers. This is probably because they are able

to save money on fuel, parking, and toll costs by taking the train for the rest of their

trip. They are also able to reduce their access time by driving to or being dropped off

at the station instead of taking other modes of transport. (Paramita et al., 2018)

The marginal utility for waiting time is: − 𝑒𝑥𝑝 [−2.988 + 0.619 𝑥 (𝑠𝑒𝑙𝑓 −

𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑) – 0.345 𝑥 (𝑑𝑟𝑖𝑣𝑒 𝑡𝑜 𝑡ℎ𝑒 𝑡𝑟𝑎𝑖𝑛 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) − 0.364 𝑥 (𝑠ℎ𝑜𝑝𝑝𝑖𝑛𝑔/

𝑝𝑒𝑟𝑠𝑜𝑛𝑎𝑙 𝑏𝑢𝑠𝑖𝑛𝑒𝑠𝑠 𝑡𝑟𝑖𝑝) +

0.615 𝑥 (𝑓𝑢𝑙𝑙 𝑓𝑎𝑟𝑒) – 0.604 𝑥 (𝑠𝑖𝑔𝑛𝑖𝑓𝑖𝑐𝑎𝑛𝑡 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 𝑜𝑓 𝑡𝑟𝑎𝑖𝑛 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 𝑜𝑛 𝑡ℎ𝑒 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 ℎ𝑜𝑚𝑒

𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛) – 0.747 𝑥 (𝑚𝑜𝑑𝑒𝑟𝑎𝑡𝑒 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 𝑜𝑓 𝑡𝑟𝑎𝑖𝑛 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 𝑜𝑛 𝑡ℎ𝑒 𝑐𝑢𝑟𝑟𝑒𝑛𝑡

ℎ𝑜𝑚𝑒 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛) + 0.549 𝑥 𝑛], where 𝑛 is a random number generated from

a standard normal distribution. Like the previous one, this marginal utility has been

simulated for 100 randomly selected individuals as shown in the top-left subfigure of

Figure 5.2. Specifically, half of the randomly selected individuals are driving to the

station. This study finds travellers who drove to the station are less sensitive to waiting

time compared to other travellers. This is consistent with our everyday experience

because driving to the station often gives a traveller more control over their departure

time and they are more likely to arrive at the station as scheduled to endure less waiting

time (Paramita et al., 2018).


Figure 5.2 : The simulated marginal utility of waiting time for 100 randomly selected

individuals

Similarly, another five simulations of the marginal utility of waiting time are

performed for the rest key variables (three of them are presented in Figure 5.2 for

illustration purpose), and notable heterogeneities of marginal utility values are also

found in the simulation results, as elaborated below.

5.6.3 Employment status

By controlling randomness and other factors, the marginal disutility values of

waiting time of self-employed individuals are much larger than respondents with other

employment status. This finding is not surprising because it is in self-employed


respondents’ best interest to be time-conscious so as not to disadvantage business;

hence, they tend to be very sensitive to waiting time. In addition, as shown in the top-

right subfigure of Figure 5.2, regardless of the employment status, significant

heterogeneities exist across individuals (Paramita et al., 2018).

5.6.4 Trip purpose

When randomness and other factors are controlled, the marginal disutility values

of waiting time of individuals whose trips are for shopping or personal business

purposes are smaller than those of individuals whose trips are for other purposes.

Generally, the group of individuals who are travelling for shopping or personal

business purposes have irregular travel patterns. Hence, they tend not to be sensitive

to fluctuations in waiting time (Paramita et al., 2018).

5.6.5 Paid fare status

The marginal utility function for waiting time also indicates that by controlling

randomness and other factors, individuals who pay full fare and travel during peak

hours are generally more sensitive to waiting time than individuals who pay a

discounted fare and travel during off-peak hours. This finding is also consistent with

our daily experience. As a trade-off for paying full fare and experiencing on-board

crowding, peak hour travellers are more likely to expect the advantage of higher

frequency train services compared with travelling off-peak. Consequently, they tend

to be more sensitive to waiting for their targeted train services. In addition, as shown

in the bottom-left subfigure of Figure 5.2, significant heterogeneities exist across

individuals, regardless of the amount of fare they pay (Paramita et al., 2018).

5.6.6 Influence of train services on the current home location decision

As shown in the bottom-right subfigure of Figure 5.2 the moderate and

significant influences of train services on the current home location decision also

contribute to the explanation of preference heterogeneity detected in the marginal

utility for waiting time. This marginal utility function reveals that when randomness

and other factors are controlled, the marginal disutility values of individuals who are

moderately influenced by train services on the current home location decision are

smaller than those of their counterparts. Similar trend has been found on the marginal

disutility values of individuals who are significantly influenced by train services on


the current home location decision are smaller in comparison to those of their

counterparts.

These findings are both interesting and somewhat surprising. Respondents who

are moderately and significantly influenced by train services in making their current

home location decision would be expected to be sensitive to waiting time (Noland,

Weiner, DiPetrillo, & Kay, 2017). Nevertheless, this study finds quite the opposite. It

seems that once this group of respondents have considered train services in making

their current home location decision, they tend to accept variations in waiting time


5.7 INTERCITY COMPARISON

“Context: train fare structures in the five Australian cities” section has described

the actual fare structure in each city based on the practice in 2016, which reveals

notable differences in how train fares are structured across these five cities. More

specifically, train fares in Sydney, Melbourne, and Brisbane rise as the number of

zones or distance travelled increase. Our data show that a number of respondents from

Sydney, Melbourne, and Brisbane travelled about 15 km during their most recent

home-based train trip. In Sydney, Melbourne, and Brisbane, public transport travellers

respectively pay $4.20, $3.90, and $5.96 for a one-way, 15 km trip during peak hours

on any weekday (Queensland Government, 2016; Transport for NSW, 2016; Victoria

State Government, 2016). The regular public transport fare in Adelaide for a single

trip using MetroCard is $ 3.54, regardless of the distance travelled (Government of

South Australia, 2016). According to our data, the structure of Adelaide’s public

transport fares benefits most Adelaide respondents whose trip origins are within 15 km

of their destinations. For the same travelled distance, Adelaide respondents pay the

least one-way fare. Consequently, Adelaide respondents feel more satisfied with their

train fares than Sydney, Melbourne, and Brisbane respondents (Paramita et al., 2018).

Meanwhile, when comparing Perth and Adelaide, the fact that many Adelaide

respondents are required to pay a fixed fare regardless of the distance travelled can

negatively impact their satisfaction with their train fare. In Perth, on the other hand,

public transport fares are determined by the number of zones travelled. According to

the survey responses, Perth respondents’ origins for their most recent train trips were

often located around 16 km from their destinations. Since one public transport zone in

Perth is approximately 8 km in radius, the respondents commute daily for at least two


zones one-way, and pay at least $ 3.91 for a regular SmartRider fare during peak

periods (Government of Western Australia, 2016). In contrast, Adelaide respondents

who travel a short distance from home on a train can pay the same fare as those who

travel a long distance. Therefore, Adelaide respondents tend to be less contented with

their train fares than Perth respondents (Paramita et al., 2018).

5.8 CONCLUSIONS

Based on a nationwide survey, this study focuses on train riders’ satisfaction

with the fare they paid for their most recent trip. The influence of their socio-economic

profiles and their specific trip characteristics on these perceived satisfactions are

assessed and quantified using an ordered logit model.

This study identifies that train riders’ socio-economic profiles can significantly

influence their satisfaction with their train fare(Paramita et al., 2018). Specifically,

female respondents tend to be less satisfied with the fare than their male counterparts.

In addition, respondents’ perceived satisfaction with train fare significantly differs

across cities. When other factors are controlled, Sydney, Melbourne and Brisbane

respondents feel less satisfied with train fare than Adelaide respondents . To attain a

deeper knowledge, modelling results are discussed in the context of the different train

fare structures in the five cities. Specifically, for the same travelled distance, Adelaide

respondents pay the least for a one-way trip compared with Sydney, Melbourne, and

Brisbane respondents. On the contrary, Perth respondents are likely to be more

contented with their train fares than Adelaide respondents (Paramita et al., 2018).

Around one fifth of respondents in Sydney, Melbourne, Adelaide, and Perth did

not report their income level, while about 15% of respondents in Brisbane chose not

to report their income level (Paramita et al., 2018). Respondents’ reluctance of

reporting their income is a widely acknowledged challenge in the survey literature

(Moore, 1988; Tourangeau & Yan, 2007). Such reluctance’s impact on the data

analysis can be very complex. The literature suggests that people who are not willing

to disclose their income tend to feel more insecure, and are consequently more

sensitive to monetary costs (Alm, Bahl, & Murray, 1993; Hurst, Li, & Pugsley, 2014;

Johansson, 2005). It would be interesting to test and differentiate high and low non-

reported income groups’ effect. Unfortunately, by the very nature of being not

reported, it would be very difficult to categorize a non-reported income into a high or

low income group, which makes it almost impossible to detect and scrutinize any


phenomenon caused by different non-reported incomes (Paramita et al., 2018). In this

study, the ‘Not reported income’ variable has been found not to be statistically

significant in the best-fitted ordered logit model.

Meanwhile, characteristics of the train trip also significantly influence a rider’s

satisfaction with their train fare (Paramita et al., 2018). Respondents’ perceived

satisfaction is significantly impacted by station access and their eligibility for a

concession fare. Respondents who take the bus to the train station appeared to be more

contented with their paid train fares in comparison to other respondents. Nonetheless,

it is not the case for Perth respondents. A Perth respondent who takes bus from home

to the train station tend to be less satisfied compared with a respondent who is from

the same city but uses other modes to reach the train station. It is possible that the bus

services in Perth are not seamlessly connected with the train stations.

Respondents who are eligible for a student concession fare feel less satisfied with

train fares than ineligible respondents. Specifically, a negative relationship is found

between the interaction of Sydney and Student concession fare and the level of

satisfaction with train fare. Sydney respondents who are 16-30 years old and are

students tend to have a low appreciation towards the concession fare offered by

Transport for New South Wales (NSW) (Transport for NSW, 2016). It may be due to

strict eligibility rules imposed by Transport for New South Wales (NSW). Conversely,

respondents who are eligible for a senior concession fare feel more pleased with their

train fares compared with respondents that are not eligible for any concession fare


Having taken into account findings in previous literature (Hensher et al., 2011;

Zheng et al., 2016), on-board crowding is considered as a random-parameter at the

start of model estimation. However, this variable turns out to be not statistically

significant in the best-fitted ordered logit model. This result may be caused by the

potential confounding effect of the self-reported crowdedness because the self-

reported crowdedness can be inconsistent with the actual crowdedness (Paramita et al.,

2018). Future studies could use some objective measure crowdedness (e.g., number of

passengers) into the train fare satisfaction model developed in this study.

Moreover, notable heterogeneity in their perceived satisfaction with train fare is

detected across respondents. Specifically, one-way cost and waiting time are found to

be the significant random parameters (Paramita et al., 2018). The marginal utility of

one-way cost is linked with both socio-economic factors and trip characteristics.


Household composition and station access are found to have a significant influence on

heterogeneity in respondents’ sensitivity to one-way cost. Meanwhile, the disutility

value of waiting time varies significantly across respondents, and is influenced by their

employment status, transport mode from home to the station, trip purpose, paid fare

status, and the influence of train services on their current home location.

A number of earlier studies mentioned the important role of pre-departure

information. This information allows travellers to plan their trip in advance, and

provides a basis for their loyalty to a particular transport mode (Cantwell et al., 2009;

Caulfield & O'Mahony, 2007; Dziekan & Kottenhoff, 2007; Hine & Scott, 2000;

Lyons, 2006). Having obtained their trip information, travellers can anticipate a

straightforward trip, as well as the ability to make route changes should the unexpected

occur (Lyons, 2006). However, this study finds that once they had chosen the train as

their transport mode, respondents’ perceived satisfaction with their train fare is not

significantly influenced by whether they had checked pre-departure information or not

(Paramita et al., 2018). Despite the apparent demand for the provision of complete trip

information at train stations, there is still a need to scientifically examine the impact

of the availability, reliability, and usefulness of such information on train ridership and

traveller satisfaction (Jou, 2001). This is especially the case, given that the provision

of unreliable trip information can be a major source of decline in train ridership

(Cantwell et al., 2009).

The importance and urgency of determining factors that influence travellers’

perceived satisfaction with train fares is frequently recognized in the literature (Beirão

& Cabral, 2007; Brons et al., 2009; Dieleman et al., 2002; Ellaway et al., 2003; Geurs

& Van Wee, 2004), as such knowledge is critical for policy makers and transport

operators in developing effective future transit policies. In this regard, the findings of

this study are significant and useful. On-going efforts in this study area need to

determine the influence of specific socio-economic factors and trip characteristics on

travellers’ choice of transport mode (Paramita et al., 2018). It will be useful, for

example, to determine whether the key variables found in this study affect

respondents’ preference for a particular transport mode, and their loyalty to that mode.

The stated-preference data collected in this survey can assist in this further

investigation.

Chapter 6: Consistency between Perceptions and Stated Preferences Data in A Nationwide Mode Choice

Experiment 81

Chapter 6: Consistency between Perceptions

and Stated Preferences Data in

A Nationwide Mode Choice

Experiment

The first two sections of this chapter reports and analyses the modelling results

from each model separately, the random-parameters binomial logit model (Section 6.1)

and the mixed logit mode (Section 6.2). It is then followed by the elaboration of the

qualitative consistency assessment (Section 6.3) and quantitative consistency

assessment (Section 6.4) between the perceptions and the corresponding attributes of

SP experiment from a nationwide survey. Section 6.5 synthesises the study’s findings,

and discusses its limitations.

6.1 THE RANDOM-PARAMETERS BINOMIAL LOGIT MODEL FOR

PERCEPTIONS

Table 6.1 shows the five significant (p-values < 0.05) service factors that

respondents claimed to be strong influences on the frequency of their train usage in the

best-fitted random-parameter binomial logit model. These factors are: whether a train

is running on schedule; the probability of getting a seat; the ability to access up-to-

date information on train services; the availability of an on-board entertainment

system; and increased road congestion. Notable heterogeneity was detected across

respondents, who stated that trains running on schedule and the probability of getting

a seat significantly influence the frequency of their train usage.

Table 6.1 Summary of the best-fitted random-parameter binomial logit model

Strongly influenced perception of… to take

a train more often

Coefficient Estimate Z-statistics

Train running on schedulea 0.418 8.54

The corresponding standard deviation 0.672 14.48

The probability of getting a seata 0.234 4.64

The corresponding standard deviation 0.157 3.7

The ability to access up-to-date information

on train services (such as current train status) 0.120 2.33

82 Chapter 6: Consistency between Perceptions and Stated Preferences Data in A Nationwide Mode Choice

Experiment

Strongly influenced perception of… to take

a train more often

Coefficient Estimate Z-statistics

Availability of an on-board entertainment

system 0.582 7.32

Increased road congestion 0.167 3.61

Socio-economic factors Coefficient Estimate Z-statistics

Age group 0.092 7.23

Influence of train services on the current home

location decision -0.284 -21.56

Vehicle access: Access to either privately

owned, company owned, or shared motor

vehicles -0.568 -7.37

Gender: Female 0.078 2.0

Own driving licence 0.244 3.06

Employment status: Full time 0.494 9.54

Employment status: Part time 0.286 4.76

Employment status: Self-employed 0.352 4.06

Trip purpose: Work/School 0.522 12.36

Require a motor vehicle for working -0.335 -6.93

Number of observations 6,731

Log-likelihood -3,888.453

AIC 1.16 a random parameters with gamma as the underlying distribution

A respondent’s probability of taking more than one train trip in a month is

mathematically given below:

𝜋𝑖 =exp( ∑ 𝛽𝑘𝑥𝑖𝑘

15𝑘=1 )

1+exp( ∑ 𝛽𝑘𝑥𝑖𝑘15𝑘=1 )

, where [6-1]

∑ (𝛽𝑘 ∗ 𝑥𝑖𝑘

15

𝑘=1) =

[ (0.419 ∗ 𝑣_𝑖) ∗ (𝑡𝑟𝑎𝑖𝑛 𝑟𝑢𝑛𝑛𝑖𝑛𝑔 𝑜𝑛 𝑠𝑐ℎ𝑒𝑑𝑢𝑙𝑒) + (0.234 ∗ 𝑣_𝑖) ∗

(𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑔𝑒𝑡𝑡𝑖𝑛𝑔 𝑎 𝑠𝑒𝑎𝑡) + 0.120 ∗ (𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑡𝑜 𝑎𝑐𝑐𝑒𝑠𝑠 𝑢𝑝 − 𝑡𝑜 −

𝑑𝑎𝑡𝑒 𝑖𝑛𝑓𝑜 𝑜𝑛 𝑡𝑟𝑎𝑖𝑛 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠) + 0.582 ∗ (𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑎𝑛 𝑜𝑛 −

𝑏𝑜𝑎𝑟𝑑 𝑒𝑛𝑡𝑒𝑟𝑡𝑎𝑖𝑛𝑚𝑒𝑛𝑡 𝑠𝑦𝑠𝑡𝑒𝑚) + 0.167 ∗ (𝑖𝑛𝑐𝑟𝑒𝑎𝑠𝑒𝑑 𝑟𝑜𝑎𝑑 𝑐𝑜𝑛𝑔𝑒𝑠𝑡𝑖𝑜𝑛) + 0.092 ∗

(𝑎𝑔𝑒 𝑔𝑟𝑜𝑢𝑝) + (−0.284) ∗

(𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 𝑜𝑓 𝑡𝑟𝑎𝑖𝑛 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 𝑜𝑛 𝑡ℎ𝑒 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 ℎ𝑜𝑚𝑒 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛) + (−0.568) ∗

(𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑎𝑐𝑐𝑒𝑠𝑠) + 0.078 ∗ (𝑏𝑒𝑖𝑛𝑔 𝑓𝑒𝑚𝑎𝑙𝑒) + 0.244 ∗ (𝑑𝑟𝑖𝑣𝑖𝑛𝑔 𝑙𝑖𝑐𝑒𝑛𝑐𝑒 𝑜𝑤𝑛𝑒𝑟𝑠ℎ𝑖𝑝) + 0.494 ∗

(𝑓𝑢𝑙𝑙 − 𝑡𝑖𝑚𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡) + 0.286 ∗ (𝑝𝑎𝑟𝑡 − 𝑡𝑖𝑚𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡) + 0.352 ∗ (𝑠𝑒𝑙𝑓 −

𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑 𝑠𝑡𝑎𝑡𝑢𝑠) + 0.522 ∗ (𝑤𝑜𝑟𝑘 𝑜𝑟 𝑠𝑐ℎ𝑜𝑜𝑙 𝑡𝑟𝑖𝑝 𝑝𝑢𝑟𝑝𝑜𝑠𝑒) + (−0.336) ∗

(𝑟𝑒𝑞𝑢𝑖𝑟𝑒 𝑎 𝑚𝑜𝑡𝑜𝑟 𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑓𝑜𝑟 𝑤𝑜𝑟𝑘𝑖𝑛𝑔)

where 𝑣𝑖 is from a gamma distribution where the shape parameter is equal to 1, and

the scale parameter is equal to 4 (Greene, 2012).


Experiment 83

6.2 THE MIXED LOGIT MODEL FOR THE SP DATA

Table 6.2 shows the result of the best-fitted mixed logit model for the SP data.

Waiting time, on-board time, ticket, and on-board crowding level have been identified

as significant attributes within the bus utility function (p-value < 0.05). The significant

attributes of the train utility function are: waiting time, on-board time, ticket, on-board

crowding level, access time from the train station, and availability of a laptop station.

Meanwhile, the utility function of car consists of on-board time, fuel cost, parking

cost, and toll cost.

Table 6.2 Summary of the best-fitted mixed logit model

Mode choice Attributes Coefficient Estimate Z-statistics

Bus Bus waiting time -0.147 -9.91 Bus on-board time -0.040 -40.92 Bus ticket -0.275 -22.77

Bus on-board crowding level -0.373 -12.72

Train Train waiting timea -1.820 -14.46

The corresponding standard

deviation 2.120 16.34 Train on-board time -0.035 -34.17 Train ticket -0.213 -21.78 Train on-board crowding level -0.301 -9.85 Access time from train station -0.167 -10.97

Availability of laptop station -0.101 -3.39

Car Car on-board timea -0.615 -6.62

The corresponding standard

deviation 0.932 29.41 Car fuel cost 0.956 35.01

Car parking cost -0.094 -58.63

Car toll cost -0.110 -39.34

Log likelihood -30519.187 AIC 1.513

Replication

for simulated

probabilities

500 Halton sequences used for

simulations

Number of group for

RPL model with

panel

6,731

Fixed number

of group 6

Number of

observations 40,386

a random parameters with lognormal as the underlying distribution

Two significant (p-value < 0.05) random parameters are found in the model:

train waiting time and car on-board time. Their standard deviations of the assumed

lognormal distribution are significantly different from zero, as shown in Table 6.2. The

heterogeneity of train waiting time cannot be explained by the observed socio-

economic factors. On the other hand, the heterogeneity of car on-board time is

explained by a series of observed socio-economic factors, as shown in Table 6.3.


Experiment

Table 6.3 Heterogeneities in car on-board time of car utility function within best-fitted

mixed logit model

Random parameter Socio-economic factors Coefficient Estimate Z-statistics

Car on-board time Age group 0.0402 3.49

Train service influence on the

current home location decision -0.257 -12.86

Vehicle access: Access to either

privately owned, company

owned, or shared motor vehicles -0.514 -8.07

Require a motor vehicle for

working -0.215 -6.51

Full time employeda 0.158 4.51

Part time employeda 0.147 3.45

Weekly income level -0.051 -3.95

Being female -0.121 -4.03 a Both full time and part time variables are two significant dummy variables derived from the final

mixed logit model. It is important to note that an individual can only have one employment status and

cannot be both part time and full time.

The three utility functions are mathematically given below,

𝑉_𝐵𝑢𝑠 = (−0.147) ∗ (𝑤𝑎𝑖𝑡𝑖𝑛𝑔 𝑡𝑖𝑚𝑒) + (−0.040) ∗ (𝑜𝑛 − 𝑏𝑜𝑎𝑟𝑑 𝑡𝑖𝑚𝑒) + (−0.275) ∗

(𝑏𝑢𝑠 𝑡𝑖𝑐𝑘𝑒𝑡) + (−0.373) ∗ (𝑏𝑢𝑠 𝑜𝑛 − 𝑏𝑜𝑎𝑟𝑑 𝑐𝑟𝑜𝑤𝑑𝑖𝑛𝑔 𝑙𝑒𝑣𝑒𝑙) [6-2]

𝑉_𝑇𝑟𝑎𝑖𝑛 = [(−1) ∗ 𝑒𝑥𝑝 (−1.820 + (2.120 ∗ 𝑛_𝑖 ))] ∗ (𝑤𝑎𝑖𝑡𝑖𝑛𝑔 𝑡𝑖𝑚𝑒) + (−1.820) ∗ (𝑜𝑛 −

𝑏𝑜𝑎𝑟𝑑 𝑡𝑖𝑚𝑒) + (−0.035) ∗ (𝑡𝑟𝑎𝑖𝑛 𝑡𝑖𝑐𝑘𝑒𝑡) + (−0.301) ∗ (𝑡𝑟𝑎𝑖𝑛 𝑜𝑛 − 𝑏𝑜𝑎𝑟𝑑 𝑐𝑟𝑜𝑤𝑑𝑖𝑛𝑔 𝑙𝑒𝑣𝑒𝑙) +

(−0.167) ∗ (𝑎𝑐𝑐𝑒𝑠𝑠 𝑡𝑖𝑚𝑒 𝑓𝑟𝑜𝑚 𝑡𝑟𝑎𝑖𝑛 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) + (−0.101) ∗ (𝑎𝑣𝑎𝑖𝑙𝑎𝑏𝑖𝑙𝑖𝑡𝑦 𝑜𝑓 𝑙𝑎𝑝𝑡𝑜𝑝 𝑠𝑡𝑎𝑡𝑖𝑜𝑛)

where 𝑛𝑖 is a random number generated from a standard normal distribution.

[6-3]

𝑉𝐶𝑎𝑟 = [(−1) ∗ exp〖(−0.615 + (0.0402) ∗ (𝑎𝑔𝑒 𝑔𝑟𝑜𝑢𝑝) + (−0.257)

∗ (𝑡𝑟𝑎𝑖𝑛 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 𝑜𝑛 𝑡ℎ𝑒 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 ℎ𝑜𝑚𝑒 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛)

+ (−0.514) ∗ (𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑎𝑐𝑐𝑒𝑠𝑠) + (−0.215)

∗ (𝑟𝑒𝑞𝑢𝑖𝑟𝑒 𝑎 𝑚𝑜𝑡𝑜𝑟 𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑓𝑜𝑟 𝑤𝑜𝑟𝑘𝑖𝑛𝑔) + (0.158) ∗ (𝑓𝑢𝑙𝑙 − 𝑡𝑖𝑚𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑)

+ (0.147) ∗ (𝑝𝑎𝑟𝑡 − 𝑡𝑖𝑚𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑) + (−0.0512) ∗ (𝑤𝑒𝑒𝑘𝑙𝑦 𝑖𝑛𝑐𝑜𝑚𝑒 𝑙𝑒𝑣𝑒𝑙)

+ (−0.121) ∗ (𝑏𝑒𝑖𝑛𝑔 𝑓𝑒𝑚𝑎𝑙𝑒) + (0.9318 ∗ 𝑛𝑖))]〗 ∗ (𝑜𝑛 − 𝑏𝑜𝑎𝑟𝑑 𝑡𝑖𝑚𝑒)

+ (0.956) ∗ (𝑓𝑢𝑒𝑙 𝑐𝑜𝑠𝑡) + (−0.094) ∗ (𝑝𝑎𝑟𝑘𝑖𝑛𝑔 𝑐𝑜𝑠𝑡) + (−0.110) ∗ (𝑡𝑜𝑙𝑙 𝑐𝑜𝑠𝑡)

where 𝑛𝑖 is a random number generated from a standard normal distribution.

[6-4]

To illustrate how the equation [6-4] works, the following details are offered. The

“weekly income level” is chosen as an example independent variable. The income

variable is part of the equation of coefficient of random parameter “car on-board time”.

This random parameter follows log-normal distribution. The equation [6-4] is


Experiment 85

essential to explain the heterogeneity of “car on-board time”. When all other factors

are constant, the higher the “weekly income level”, the overall coefficient of “car on-

board time” is also increasing. Eventually, it improves the utility of car.

Correspondingly, the probability function is mathematically given below:

𝑃𝑖 =exp(𝑉𝑖)

[exp(𝑉𝑖)+exp(𝑉𝑗)+exp (𝑉𝑘)] , where 𝑖, 𝑗, 𝑘 represents bus, train, and car,

respectively. [6-5]

6.3 QUALITATIVE ASSESSMENT

The best-fitted random-parameter binomial logit (RP) model and mixed logit

(SP) model are assessed qualitatively and quantitatively against one-another. With

reference to Table 3.8, the qualitative assessment starts by the mapping of the

significant explanatory variables from the RP model against the SP model, as shown

in Table 6.4.

Table 6.4 Mapping of significant variables of the random-parameter binomial logit (RP)

model against the mixed logit (SP) model

Significant variables of best-fitted RP model:

Strongly influenced perception of … to take a

train more often

Significant variables of best-fitted SP

model

Trains running on schedulea Train waiting timea

The probability of getting a seata Train on-board crowding

Availability of an on-board entertainment system Availability of a laptop station

Increased road congestion

Car on-board timea (part of utility Car)

Its heterogeneity is explained by the

following socio-economic factors: Age

group; Train service influence on current

home location decision; Vehicle access;

Whether car is required for work; Employed

full time; Employed part time; Weekly

income level; and Being female.

The ability to access up-to-date information on train

services (such as current train status) Not Applicableb

Not Applicableb Car toll cost (part of utility Car) a random parameters

b Not Applicable: the particular variable is not found to be significant

The qualitative analysis reveals that four out of the seven perceptions of service

factors that strongly influenced respondents’ frequency of train usage, are significant

variables in the RP model. The attributes that are either identical or similar to these,

also appear to be significant variables in the model for the mode choice experiment

data. More specifically, the perception of trains running on schedule is a significant

variable in the RP model. A similar attribute, train waiting time, appears to be


Experiment

significant in the SP model. In addition, significant heterogeneities have been detected

in each of these two variables. Both the RP model and the SP model reveal that

crowding level is a significant factor that influences respondents’ train usage: It is

reflected in the significance of perception of the probability of getting a seat in the RP

model, and in the significance of on-board crowding in the SP model. The perception

of the probability of getting a seat in the RP model shows significant heterogeneity,

while on-board crowding in the SP model is treated as a fixed parameter. Although

the perception of the availability of on-board entertainment system, which is a

significant variable in the RP model, does not have an exact counterpart in the SP

model, the availability of a laptop station, which has a close association with this

variable, appears to be also significant in the SP model. Similarly, even though the

perception of increased road congestion – a significant variable in the RP model –

does not have an exact counterpart in the SP model, car on-board time, which is

assumed to capture similar information, also appears to be significant in the SP model.

The car on-board time is treated as a random parameter, and its heterogeneity is

explained by a series of socio-economic factors, as shown in Table 6.3. In addition,

neither the perception of better access to train station nor similar attributes turn out to

be significant in either the RP model or the SP model.

Meanwhile, some inconsistences in the results of these two models are also

observed. The perception of the availability of up-to-date information on train services

is a significant variable in the RP model; however, no corresponding or similar

attribute is included in the mode choice experiments. Therefore, their consistency

cannot be assessed in this study. On the other hand, toll cost is a significant variable in

the SP model, yet the corresponding variable in the RP model, congestion charge and

toll cost, appears to be insignificant.

In summary, the result of the qualitative assessment shows that a number of

variables, which are gathered as respondents’ perceptions or as responses to the mode

choice experiment, consistently appear in both the RP model and the SP model.

Particularly, the respondents’ views on train waiting time, train on-board crowding

level, availability of laptop station, and increased road congestion within the SP

experiment are regarded as being aligned to respondents’ perceptions of the same or

similar service factors.


Experiment 87

6.4 QUANTITATIVE ASSESSMENT

To further appraise the consistency in respondents’ perceptions of service factors

on the frequency of train usage and their responses to the same or similar attributes in

the SP experiment, a quantitative assessment was subsequently performed. This was

based on the comparison of probability values estimated from both best-fitted models,

as elaborated below.

6.4.1 ‘Perception of train running on schedule’ versus ‘Train waiting time’

Perception of train running on schedule in the best-fitted RP model is found to

be a random parameter that follows the gamma distribution restricted to the positive

side. However, its heterogeneity cannot be explained by the observed socio-economic

factors of the respondents. Other factors, for example, weather condition and security

level on-board and on the station – factors that are not contained in the dataset used by

this study – could have influenced the respondents’ perceptions for train running on-

schedule (Fan et al., 2016). The marginal utility of this variable is[0.419 ∗ 𝑣𝑖], where

𝑣𝑖 is a random number generated from a gamma distribution where the shape

parameter equals 1, and the scale parameter equals 4 (Greene, 2012).

Train waiting time in the best-fitted SP model is found to be a random parameter,

which follows the lognormal distribution restricted to the negative side. Similarly, its

heterogeneity cannot be explained by the observed socio-economic factors. The

marginal utility of this variable is [(−1) ∗ exp(−1.820 + (2.120 ∗ 𝑛𝑖))], where 𝑛𝑖is a

random number generated from a standard normal distribution (Greene, 2012).

To better understand the taste variation across individuals captured through these

two random parameters, their impacts on the probability of respondents taking the

train have been separately simulated for 200 randomly selected individuals by

controlling for all other factors, as shown in Figure 6.1 and Figure 6.2. Specifically, in

the first simulation, half of the randomly selected individuals have a strongly

influenced perception of train running on schedule, and in the second simulation, half

of the randomly selected individuals experience a medium waiting time (that is, 8

minutes), while the other half experience a short wait time (that is, 4 minutes).


Experiment

Figure 6.1 : The simulated probability of individuals’ perception of train running on schedule

from the RP model

Figure 6.2 : The simulated probability of train waiting time from the SP model


Experiment 89

Figure 6.1 and Figure 6.2 reveal that by holding all other factors constant, there

are variations across different individuals in the impact of perception of train running

on schedule and train waiting time. To gain more insight, the difference in the

probability of taking the train more than once a month between each pair of selected

individuals in the first simulation is depicted in Figure 6.3. This figure shows that for

the randomly selected 200 individuals, their different perception of train running on

schedule can cause the difference in the probability of taking the train more than once

per month, up to approximately 28.9%. More specifically, for 30% of the cases, the

difference in the probability of taking the train more than once per month is in the

range of [0.1%, 4.9%]; for almost 25% of the cases, the difference in the probability

of taking the train more than once per month is in the range of (4.9%, 9.7%]; for about

15% of the cases, the difference in the probability of taking the train more than once

per month is in the range of (9.7%, 14.5%]; for 20% of the cases, the difference in the

probability of taking the train more than once per month is in the range of (14.5%,

19.3%]; and for about 10% of the cases, the difference in the probability of taking train

more than once per month is more than 19.3%.


Experiment

Figure 6.3 : The difference in probability value of the impact of a one unit increase in

individuals’ perception level of train running on schedule

Figure 6.4 : The difference in probability value of the impact of one unit increase in train

waiting time in the SP experiment

Similarly, the difference in the probability of choosing the train because of the

impact of train waiting time between each pair of the selected individuals in the second

simulation is depicted in Figure 6.4. This figure shows that, for the randomly selected

200 individuals, the increment of train waiting time can cause a difference of up to

27.1% in the probability of their taking the train or not . Particularly, for almost 43%

of cases, the difference in the probability of taking the train due to the impact of train

waiting time is less than 4.6%; for about 23% of the cases, the difference in the

probability of the taking train is in the range of (4.6%, 9.1%]; for about 18% of the

cases, the difference in the probability of taking the train is in the range of (9.1%,

13.6%]; and for about 16% of the cases, the difference in the probability is more than

13.6%.


Experiment 91

6.4.2 Perception of the ‘probability of getting a seat’ compared to ‘Train on-board

crowding’

The perception of the probability of getting a seat is found to be a random

parameter that follows the gamma distribution restricted to the positive side.

Nonetheless, its heterogeneity cannot be explained by the observed socio-economic

factors. Other factors that are not contained in the dataset used in this study (for

example, mental and physical health conditions) could have influenced the

respondents’ perception of the probability of getting a seat (Cox et al., 2006; Evans &

Wener, 2007; Singleton, 2018). The marginal utility of this variable is: 0.234 ∗ 𝑣𝑖 ,

where 𝑣𝑖 is a random number generated from a gamma distribution, where the shape

parameter equals 1, and the scale parameter equals 4 (Greene, 2012).

To gain more knowledge of the taste variation across individuals captured

through perception of the probability of getting a seat, its impact on the probability of

taking the train more than once per month has been simulated for 200 randomly

selected individuals by controlling for all other factors, as shown in Figure 6.5. In this

simulation, half of the randomly selected individuals have a strongly influenced

perception of the probability of getting a seat.

Figure 6.5 : The simulated probability of individuals’ perception of the probability of getting

a seat from the RP model


Experiment

Observation of Figure 6.5 exposes that by holding all other factors constant,

there are variations across different individuals due to the diversely perception of the

probability of getting a seat. Figure 6.6 presents the difference in the probability of

taking the train more than once a month for each pair of selected individuals in the

simulation. For the randomly selected 200 individuals, their different perception of

train running on schedule can cause an approximately 25% difference in the

probability of taking the train more than once per month. In particular, for the majority

(43%) of cases, the difference in the probability of taking the train more than once per

month is in the range of (0%, 5%]; for 38% of the cases, the difference is in the range

of (5%, 15%]; and for almost 13% and 6% of the cases, the difference is in the range

of (15%, 20%] and (20%, 25%], respectively.

Figure 6.6 : The difference in probability value of the impact of a one unit increase in

individuals’ perception level of the probability of getting a seat

Train on-board crowding in the best-fitted SP model is found to be a fixed

parameter. When controlling for all factors, respondents who feel a one unit increase

in the train on-board crowding experience are less likely to take the train than their

counterparts. In particular, the estimated odds of their taking the train decrease by 26%

(i.e. (1-exp(-0.301))*100%) compared with other respondents (Agresti & Kateri,

2011). As they experience increased on-board crowding, they feel less comfortable,

and their personal space is diminished (Cantwell et al., 2009; Dziekan & Kottenhoff,

2007). These travellers would take train services less frequently in the future.


Experiment 93

6.4.3 Perception of ‘availability of an on-board entertainment system’ versus

‘Availability of laptop station’

The perception of the availability of an on-board entertainment system in the

best-fitted RP model is found to be a fixed parameter. All other things being equal,

respondents who have a strongly influenced perception of the availability of on-board

entertainment system are more likely to take a train more than once a month compared

with the other respondents. Particularly, the estimated odds of their doing so increase

by 79% (i.e. (exp(0.582)-1)*100%) for respondents who have a strongly influenced

perception of the availability of on-board entertainment system compared with their

counterparts (Agresti & Kateri, 2011).

An increasing number of travellers have consciously selected bus, train, and

flight services that provide on-board support for portable technology, such as wireless

connection, LCD screens, digital music players, and electrical plugs (Schwieterman et

al., 2009). Travellers who are influenced by the availability of on-board entertainment

systems would enjoy videos, music, or wireless connection as well as have the chance

to work or study rather than being idle. Despite the fact that public transport services

that are equipped with advanced technology cost more than conventional services,

travellers are not deterred from choosing such services (Delclòs-Alió, Marquet, &

Miralles-Guasch, 2017; Naudts et al., 2013). The provision of advanced technology on

train services appears to be positively correlated with travellers’ perceptions of a

greater level of comfort.

Availability of laptop station in the best-fitted SP model is found to be a fixed

parameter. The negative sign of its coefficient belies the findings of previous

researches, where it was found that the availability of broadband internet and other

portable on-board technologies, including laptop stations, strongly motivated

travellers to take trains and buses more often (Delclòs-Alió et al., 2017; Naudts et al.,

2013; Schwieterman et al., 2009; Stanton et al., 2013). In contrast, in this study, the

estimated odds of taking the train decrease by 10% for respondents who are on trains

with laptop stations. It appears that trains with laptop stations correspond with the

increased dissatisfaction of train travellers. Having examined the details of mode

choice experiments, there are two possible explanations: first, the installation of laptop

stations requires additional space and, consequently, can reduce passengers’ personal

space and create cramped conditions; and, second, the longest train on-board time


Experiment

shown in mode choice experiments was 45 minutes, and it is unlikely that train riders

would need charging stations during this time.

6.4.4 Perception of ‘increased road congestion’ versus ‘car on-board time’

The perception of increased road congestion in the best-fitted RP model was

found to be a fixed parameter. While controlling all other factors, respondents who

have a strongly influenced perception of increased road congestion are more likely to

take a train more than once a month than the other respondents. In particular, the

estimated odds of taking a train more than once a month increase by 18% for

respondents who have a strongly influenced perception of increased road congestion.

Travellers who are sensitive to road congestion are more likely to care more about the

reliability of all road-based transport modes. As an alternative, these travellers would

opt for train services, as trains have their own ‘right of way’, and contribute to the

reduction of carbon emissions (Stopher, 2004). In the long run, these travellers would

take the train more frequently to avoid road congestion (Nguyen, Soltani, & Allan,

2018; Van Exel & Rietveld, 2009).

Car on-board time in the best-fitted SP model is found to be a random parameter

that follows the lognormal distribution restricted to the negative side. Its heterogeneity

is explained by a series of socio-economic factors to a certain extent, and its marginal

utility is: [(−1) ∗ exp(−0.615 + (0.0402) ∗ (𝑎𝑔𝑒 𝑔𝑟𝑜𝑢𝑝) + (−0.257) ∗

(𝑡𝑟𝑎𝑖𝑛 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 𝑜𝑛 𝑡ℎ𝑒 𝑐𝑢𝑟𝑟𝑒𝑛𝑡 ℎ𝑜𝑚𝑒 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛 𝑑𝑒𝑐𝑖𝑠𝑖𝑜𝑛) + (−0.514) ∗

(𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑎𝑐𝑐𝑒𝑠𝑠) + (−0.215) ∗ (𝑟𝑒𝑞𝑢𝑖𝑟𝑒 𝑎 𝑚𝑜𝑡𝑜𝑟 𝑣𝑒ℎ𝑖𝑐𝑙𝑒 𝑓𝑜𝑟 𝑤𝑜𝑟𝑘𝑖𝑛𝑔) + (0.158) ∗

(𝑓𝑢𝑙𝑙 − 𝑡𝑖𝑚𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑) + (0.147) ∗ (𝑝𝑎𝑟𝑡 − 𝑡𝑖𝑚𝑒 𝑒𝑚𝑝𝑙𝑜𝑦𝑒𝑑) + (−0.0512) ∗

(𝑤𝑒𝑒𝑘𝑙𝑦 𝑖𝑛𝑐𝑜𝑚𝑒 𝑙𝑒𝑣𝑒𝑙) + (−0.121) + (−0.121) ∗ (𝑏𝑒𝑖𝑛𝑔 𝑓𝑒𝑚𝑎𝑙𝑒) + (0.9318 ∗ 𝑛𝑖))],

where 𝑛𝑖 is a random number generated from a standard normal distribution (Greene,

2012).

To better understand the taste variation across individuals captured through car

on-board time, this parameter’s impact on the probability of respondents using their

car was simulated for 200 randomly selected individuals by controlling for all other

factors, as shown in Figure 6.7. The corresponding probabilities of taking the train for

the same 200 randomly selected individuals are illustrated in Figure 6.8. Specifically,

half of the randomly selected individuals experience a 20-minutes car on-board time,

while the other half experience a 10-minute car on-board time.


Experiment 95

Figure 6.7 : The simulated probability of using a car for individuals experiencing two different

periods of car on-board time from the SP model

Figure 6.8 : The corresponding simulated probability of taking the train for individuals who

experience two different periods of car on-board time from the SP model


Experiment

A detailed examination of Figure 6.7 and Figure 6.8 reveal that by holding all

other factors constant, there are variations in the probability of different individuals

using a car, and in the corresponding probability of their taking the train. To further

observe the implication of the probability variations of the two groups, the difference

in the probability of each pair of the selected individuals using a car is illustrated in

Figure 6.9. The change in car on-board time can affect the difference in the probability

of using a car up to approximately 60.2% for the randomly selected 200 individuals.

For the majority (50%) of cases, the difference in the probability of using car is in the

range of [0%, 8.6%]; for approximately 25% of the cases, it is in the range of (8.6%,

17.2%]; for 10% of the cases, it is in the range of (17.2%, 25.8%]; and for 15% of the

cases, it is in the range of (25.8%, 60.2%].

Figure 6.9 : The difference in the probability value of using a car as the impact of a one unit

increase in car on-board time in the SP experiment


Experiment 97

Figure 6.10 : The difference in the probability value of taking the train as the impact of a one

unit increase in car on-board time in the SP experiment

Similarly, the difference in the corresponding probability of taking the train due

to the change of car on-board time between each pair of the selected individuals is

shown in Figure 6.10. For the same randomly selected 200 individuals, the figure

reveals that the difference in the probability of their decision to take the train can be

up to 39.2%. Specifically, for almost 45% of the cases, the difference in the probability

of taking the train is in the range of [0%, 5.6%]; for about 28% of the cases, it is in the

range of (5.6%, 11.2%]; for about 15% of the cases, it is in the range of (11.2%,

16.8%]; and for about 12% of the cases, the difference is more than 16.8%.

6.4.5 The summary of quantitative assessment

The quantitative assessment evaluates and compares the magnitude of the

changes in probability values. Specifically, when holding all other factors constant,

our simulation analysis clearly shows that there are heterogeneities in the probability

changes across randomly selected individuals as the impact of a one unit increase in a

random parameter – either the perception of a service factor, or the corresponding

attribute in SP experiment. Despite the range of variations, the magnitude of

differences as the result of a one unit increase in the perception of a service factor is

comparable to those differences that are the result of a one unit increase in the

corresponding attribute in SP experiment.


Experiment

6.5 CONCLUSIONS

By utilizing a nationwide survey that targeted urban travellers from five

Australian capital cities, this study investigated the consistency of travellers’

perceptions of service factors and their stated travel mode choices. More specifically,

this study utilized nationwide survey data related to the most recent home-based trip

of urban travellers in Sydney, Melbourne, Brisbane, Adelaide, and Perth. In so doing,

it addressed two limitations of many previous studies: their small sample size, and the

use of data from respondents from a particular user group in a single city.

This study offers a notable theoretical contribution to the ongoing validation

issue with regard to RP and SP data. Previous studies often used RP data to assess the

reliability of SP responses. In this study, however, another valuable data source is used;

that is, respondents’ perceptions of the influence of various service factors. A random-

parameter binomial logit (RP) model and a mixed logit (SP) model are estimated for

the perceptions of service factors, and the responses to mode choice experiments,

respectively. In terms of explaining respondents’ mode choice behaviours, the

consistency between the model based on their perceptions, and the model based on the

SP responses has been assessed from both a qualitative and a quantitative perspective.

At the qualitative level, the significant factor mapping from the two models

reveals that perception of the four service factors and their corresponding attributes

including in SP experiment are well aligned; more specifically, the respondents’ views

on train waiting time, on-board crowding level, availability of laptop station, and

increased road congestion within the SP experiment are regarded as being aligned to

their perceptions of the same or similar service factors.

At the quantitative level, the marginal utilities of choosing the train mode in

these two models are tested through rigorous numerical simulations for the same four

service factors, and the estimated probabilities of choosing the train mode from the SP

model are found to be similar to the probabilities of choosing the train mode estimated

from the RP model.

The RP model identifies two significant random parameters: the perception of

train running on schedule and of the probability of getting a seat. Correspondingly, the

SP model also identifies two random parameters: train waiting time and car on-board

time. To gain a further understanding of the underlying reasons for the heterogeneities,

the significant random parameters are traced back to respondents’ diverse socio-


Experiment 99

economic backgrounds, and a series of numerical simulations was implemented to

better detect and interpret the heterogeneities in these random parameters. While the

heterogeneity of car on-board time can be explained by the observed socio-economic

factors, this is not the case for the heterogeneities of perception of train running on

schedule, of the probability of getting a seat, and of train waiting time.

Our simulation analysis clearly shows significant heterogeneities in the

probability changes across randomly selected individuals as the impact of a one unit

increase in a random parameter – either the perception of a service factor, or the

corresponding attribute in SP experiment. Nevertheless, the magnitudes of differences

as a result of a one unit increase in the perception of a service factor are comparable

to the results for a one unit increase in the corresponding attribute in SP experiment.

Thus, the quantitative assessment confirms that while choosing their mode preference

in SP experiment, respondents’ perception of the four service factors are consistent

with their reasoning with regard to their corresponding attributes.

However, the overall consistency between the respondents’ perceptions and their

responses to the corresponding attribute in SP experiment, does not indicate that one

dataset can replace the other. Rather, our analysis confirms that these two types of data

sources are complementary in helping us to better understand travellers’ complex

mode choice behaviour. Our qualitative and quantitative assessments demonstrate that

respondents’ perception of four service factors are constructively consistent with their

reasoning surrounding their corresponding attributes in SP experiment.

Finally, this study has at least two limitations. First, access to real-time train

service information is not considered in the SP experiment, and this makes it

impossible to assess the consistency of respondents’ perceptions and their

corresponding responses in the SP experiment on this important attribute. In addition,

previous studies mention the importance of the weather, security, safety, and the health

and psychological condition of the respondents in understanding the underlying

heterogeneity of perception of train running on schedule, train waiting time, and the

probability of getting a seat (Cox et al., 2006; Evans & Wener, 2007; Fan et al., 2016;

Singleton, 2018). However, such information is not collected in the survey used in this

study.

Chapter 7: Policy Interventions Study to Encourage Behavioural Shift from Car to Public Transport 101

Chapter 7: Policy Interventions Study to

Encourage Behavioural Shift

from Car to Public Transport

The findings of the third sub-study (Policy interventions study to encourage

behavioural shift from car to public transport) are reported and analysed in Section 7.1

to 7.4. Section 7.5 presents the corresponding probability function of bus, train, car,

and public transport. Section 7.6 identifies the average traveller’s profile for each mode

in each city. Section 7.7 simulates and presents the impacts of over a hundred scenarios

of policy intervention. Section 7.8 discusses the most efficient combined policy

interventions based on the simulations. The last section (7.9) concludes the study,

highlights its limitations, and recommends further related studies.

7.1 MODELLING RESULTS

With reference to Figure 4.1, three nested logit models are estimated for each of

the three cities (Sydney, Melbourne, and Brisbane). Each nested logit estimation (using

Nlogit) produces a simple MNL model, and a FIML of nested logit (Greene, 2000).

The three best-fitted FIML of nested logit models produced by Nlogit are summarized

in Table 7.1.

Table 7.1 Summary of the best-fitted FIML of nested logit model

Mode Choice Explanatory Variables

Sydney

dataset:

Estimated

Coefficient

Melbourne

dataset:

Estimated

Coefficient

Brisbane

dataset:

Estimated

Coefficient

Lower Nestb

Bus ASC_Busa 1.573 2.161 2.062 Bus waiting time -0.046 -0.035 -0.028 Bus on-board time -0.033 -0.034 -0.032 Bus ticket -0.168 -0.233 -0.263 Bus on-board crowding -0.220 -0.191 -0.269 Employment status -0.129 -0.171 -0.219

Age group -0.111

Train ASC_Traina 2.221 2.250 2.761

Access time to train

station -0.032 -0.104 -0.250

Train waiting time -0.017 -0.005 -0.058

Train on-board time -0.031 -0.033 -0.033

Train ticket -0.200 -0.174 -0.260

102 Chapter 7: Policy Interventions Study to Encourage Behavioural Shift from Car to Public Transport

Mode Choice Explanatory Variables

Sydney

dataset:

Estimated

Coefficient

Melbourne

dataset:

Estimated

Coefficient

Brisbane

dataset:

Estimated

Coefficient

Train on-board crowding -0.241 -0.328

Access time from train

station -0.027 -0.118 -0.209

Train services’ influence

on the current home

location decision

-0.346 -0.227 -0.195

Age group 0.037 -0.111 0.070

Employment status -0.129 -0.171 -0.219

Car Car on-board time 0.017 0.012 0.018 Car parking cost -0.055 -0.055 -0.062 Car toll cost -0.054 -0.058 -0.073

Car fuel cost -0.060 Employment status -0.129 -0.171 -0.219

Whether car is required

for working 0.339 0.242 0.446

Driving licence 0.452 0.618 Income level 0.122 0.135 Age group -0.111 0.070

Gender 0.559

Upper Nest

Public

Transport ASC_Publica 0.503 0.424 0.273

Highest educational

qualification -0.253 -0.119 -0.307

Employment status -0.129 -0.171 -0.219

Gender -0.180 -0.123 0.559

Age group -0.111 0.070

IV Parameters Public Transport 0.424 0.536 0.410

Public Transport :

Standard Error 0.052 0.051 0.069

Private Transportc 1 1 1

Log likelihood function -10,178.073 -10,223.316 -4,939.586

P-value <0.01 <0.01 <0.01

Degrees of freedom 24 25 24

McFadden Pseudo R-

squared

0.287

0.278 0.306

Number of group for

RPL model with panel

2,000

2,000 989

Number of observations 12,000 12,000 5,934

Fixed number of

observations/

group

6

Replication

for simulated

probability

500 Halton

sequences

used for

simulations a

ASC: alternative specific constant b All three nested logit models are being normalized at the lower level (i.e. RU1 command in Nlogit). c Due to its degenerative nature, the IV parameter of the private transport branch is being normalized

at 1 for all three nested logit models.


7.2 THE GOODNESS-OF-FIT TEST OF THE NESTED STRUCTURE

The p-value from each city dataset (presented in Table 7.1) is calculated based

on the likelihood ratio test between the nested logit model and the simple MNL model

from each city dataset. It shows that the nested logit model is statistically better than

the simple MNL model in explaining the mode choice responses in the Sydney,

Melbourne, and Brisbane dataset at a 99% significance level. The “McFadden Pseudo

R-squared” (𝜌2) value represents the second indicator for the goodness-of-fit of the

nested logit model (Hensher & Johnson, 1981; McFadden, 1973). McFadden (1973)

constructed the goodness-of-fit measures for logit models from the values of log of the

model’s likelihood function when the coefficients assume various values (Sobel,

1980). The likelihood ratio index is equivalent to “McFadden Pseudo R-squared” (𝜌2)

(Hensher & Johnson, 1981; McFadden, 1973). The ideal value of 𝜌2 is between 0.2

and 0.4. As the nested logit model from Sydney, Melbourne, and Brisbane has 𝜌2 of

0.287, 0.278, and 0.306, respectively, the three nested logit models are considered

extremely-good-fit models.

An additional method to validate the nested structure is to assess whether the

estimated coefficient of the IV parameter for public transport from each best-fitted

model is significantly above zero, and significantly less than one (Hansen, 1987;

Washington et al., 2011). The significant (p-value < 0.05) estimated coefficients of the

IV parameter for public transport for Sydney, Melbourne, and Brisbane are 0.424,

0.536, and 0.410, respectively. The associated standard errors for Sydney, Melbourne,

and Brisbane datasets are 0.052, 0.051, and 0.069, respectively. In order to test if each

of the estimated coefficients of the IV is significantly less than one, Equation 4-19 is

applied for each dataset (Washington et al., 2011).

𝑡 − 𝑆𝑦𝑑𝑛𝑒𝑦∗ =𝛽 − 1

𝑆. 𝐸. (𝛽)=

0.424 − 1

0.052= −11.108

𝑡 − 𝑀𝑒𝑙𝑏𝑜𝑢𝑟𝑛𝑒∗ =𝛽 − 1

𝑆. 𝐸. (𝛽)=

0.536 − 1

0.051= −9.098

𝑡 − 𝐵𝑟𝑖𝑠𝑏𝑎𝑛𝑒∗ =𝛽 − 1

𝑆. 𝐸. (𝛽)=

0.410 − 1

0.069= −8.489

A one-tailed t test gives a confidence level of over 99% for each city dataset.

This provides convincing evidence that the estimated coefficient of the IV parameter

for public transport is significantly less than 1.0 for each dataset. Thus, the nested

logit model structure (as shown in Figure 4.1) is preferred to the simple MNL structure


for all three datasets. There is a correlation between the bus and train riders’

disturbance terms in each dataset.

7.3 THE LOWER NEST OF THE NESTED STRUCTURE

7.3.1 The lower nest of the Sydney dataset

The lower nest of each best-fitted nested model consists of bus and train

alternatives under the public transport branch, and the car alternative under the private

transport branch. Based on the best-fitted model presented in Table 7.1, the utility

functions of bus, train, and car of the Sydney dataset are defined as below.

7.3.1.1 The bus branch

The bus utility function consists of six independent variables; an alternative

specific constant, ‘employment status’, and four travel attributes. The four travel

attributes that characterize the probability of taking the bus are waiting time, on-board

time, ticket, and on-board crowding. As expected, the increments of waiting time, on-

board time, fare, and on-board crowding correspond to a reduction in the probability

of taking the bus. The bus utility function is mathematically given below as,

𝑉𝐵𝑢𝑠 = 1.573 + (−0.046) ∗ (𝑏𝑢𝑠 − 𝑤𝑎𝑖𝑡𝑖𝑛𝑔 − 𝑡𝑖𝑚𝑒) + (−0.033) ∗ (𝑏𝑢𝑠 − 𝑜𝑛𝑏𝑜𝑎𝑟𝑑 −

𝑡𝑖𝑚𝑒) + (−0.168) ∗ (𝑏𝑢𝑠 − 𝑡𝑖𝑐𝑘𝑒𝑡) + (−0.220) ∗ (𝑏𝑢𝑠 − 𝑐𝑟𝑜𝑤𝑑𝑒𝑑𝑛𝑒𝑠𝑠 − 𝑜𝑛𝑏𝑜𝑎𝑟𝑑) +

(−0.129) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) [7-1]

7.3.1.2 The train branch

The train utility function is estimated by an alternative specific constant, access

time to the train station, waiting time, on-board time, ticket, on-board crowding, access

time to train station, ‘level of influence of train service on the choice of current home

location’, ‘employment status’, and ‘age group’. By having the last nine variables at

zero, the utility value of taking the train is positive at 2.221, as shown by the sign and

magnitude of its estimated alternative specific constant. The sign of each estimated

coefficient of key factors is as expected. The train utility function is mathematically

given below as,


𝑉𝑇𝑟𝑎𝑖𝑛 = 2.221 + (−0.032) ∗ (𝑎𝑐𝑐𝑒𝑠𝑠 − 𝑡𝑖𝑚𝑒 − 𝑡𝑜 − 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) + (−0.017) ∗ (𝑡𝑟𝑎𝑖𝑛 −

𝑤𝑎𝑖𝑡𝑖𝑛𝑔 − 𝑡𝑖𝑚𝑒) + (−0.031) ∗ (𝑡𝑟𝑎𝑖𝑛 − 𝑜𝑛𝑏𝑜𝑎𝑟𝑑 − 𝑡𝑖𝑚𝑒) + (−0.200) ∗ (𝑡𝑟𝑎𝑖𝑛 −

𝑡𝑖𝑐𝑘𝑒𝑡) + (−0.241) ∗ (𝑡𝑟𝑎𝑖𝑛 − 𝑐𝑟𝑜𝑤𝑑𝑒𝑑𝑛𝑒𝑠𝑠 − 𝑜𝑛𝑏𝑜𝑎𝑟𝑑) + (−0.027) ∗ (𝑎𝑐𝑐𝑒𝑠𝑠 −

𝑡𝑖𝑚𝑒 − 𝑓𝑟𝑜𝑚 − 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) + (−0.346) ∗ (𝑡𝑟𝑎𝑖𝑛 − 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 − 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 − 𝑜𝑛 − ℎ𝑜𝑚𝑒 −

𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛) + ∗ 0.037 (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝) + (−0.129) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) [7-2]

7.3.1.3 The car branch

The car utility function is estimated by six independent variables; three travel

attributes, and three socio-economic factors. The key travel attributes are on-board

time, parking cost, and toll cost. The vital socio-economic factors are ‘employment

status’, ‘whether a car is required for work’, and ‘whether a driving licence is held’.

The car utility function is mathematically given below as,

𝑉𝐶𝑎𝑟 = (0.017) ∗ (𝑐𝑎𝑟 − 𝑜𝑛𝑏𝑜𝑎𝑟𝑑 − 𝑡𝑖𝑚𝑒) + (−0.055) ∗ (𝑐𝑎𝑟 − 𝑝𝑎𝑟𝑘𝑖𝑛𝑔 − 𝑐𝑜𝑠𝑡) +

(−0.054) ∗ (𝑐𝑎𝑟 − 𝑡𝑜𝑙𝑙 − 𝑐𝑜𝑠𝑡) + (−0.129) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + 0.339 ∗ (𝑐𝑎𝑟 −

𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 − 𝑓𝑜𝑟 − 𝑤𝑜𝑟𝑘𝑖𝑛𝑔) + 0.452 ∗ (𝑑𝑟𝑖𝑣𝑖𝑛𝑔 − 𝑙𝑖𝑐𝑒𝑛𝑐𝑒 ) [7-3]

7.3.2 The lower nest of the Melbourne dataset

Based on the best-fitted model presented in Table 7.1, the utility functions of

bus, train, and car of the Melbourne dataset are defined as below.


The bus utility function consists of seven independent variables; an alternative

specific constant, ‘employment status’, ‘age group’, and four travel attributes. The four

travel attributes that characterize the probability of taking the bus are waiting time, on-

board time, ticket, and on-board crowding. As expected, the increments of waiting

time, on-board time, fare, and on-board crowding correspond to a reduction in the

probability of taking the bus. The bus utility function is mathematically given below

as,



(−0.171) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + (−0.111) ∗ (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝) [7-4]



time to train station, waiting time, on-board time, ticket, access time to train station,


‘level of influence of train services on the current choice of home location’,

‘employment status’, and ‘age group’. By having the last eight variables at zero, the

utility value of taking the train is positive at 2.250, as shown by the sign and magnitude

of its estimated alternative specific constant. The sign of each estimated coefficient of

key factors is as expected. The train utility function is mathematically given below as,



𝑡𝑖𝑐𝑘𝑒𝑡) + (−0.118) ∗ (𝑎𝑐𝑐𝑒𝑠𝑠 − 𝑡𝑖𝑚𝑒 − 𝑓𝑟𝑜𝑚 − 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) + (−0.227) ∗ (𝑡𝑟𝑎𝑖𝑛 −

𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 − 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 − 𝑜𝑛 − ℎ𝑜𝑚𝑒 − 𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛) + ∗ 0.111 (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝) +



The car utility function is estimated by nine independent variables; four travel

attributes and five socio-economic factors. The key travel attributes are on-board time,

parking cost, toll cost, and fuel cost. The vital socio-economic factors are ‘employment

status’, ‘whether a car is required for work’, ‘whether a driving licence is held’,

‘income level’, and ‘age group’. The car utility function is mathematically given below

as,


(−0.058) ∗ (𝑐𝑎𝑟 − 𝑡𝑜𝑙𝑙 − 𝑐𝑜𝑠𝑡) + (−0.060) ∗ (𝑐𝑎𝑟 − 𝑓𝑢𝑒𝑙 − 𝑐𝑜𝑠𝑡) + (−0.171) ∗

(𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + 0.242 ∗ (𝑐𝑎𝑟 − 𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 − 𝑓𝑜𝑟 − 𝑤𝑜𝑟𝑘𝑖𝑛𝑔) + 0.618 ∗

(𝑑𝑟𝑖𝑣𝑖𝑛𝑔 − 𝑙𝑖𝑐𝑒𝑛𝑐𝑒 ) + 0.122 ∗ (𝑖𝑛𝑐𝑜𝑚𝑒 − 𝑙𝑒𝑣𝑒𝑙 ) + 0.111 ∗ (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝 ) [7-6]

7.3.3 The lower nest of the Brisbane dataset

Based on the best-fitted model presented in Table 7.1, the utility functions of

bus, train, and car of the Brisbane dataset are defined as below.


The bus utility function consists of six independent variables; an alternative

specific constant, ‘employment status’, and four travel attributes. The four travel

attributes that characterize the probability of taking the bus are waiting time, on-board

time, ticket, and on-board crowding. As expected, the increments of waiting time, on-

board time, fare, and on-board crowding correspond to a reduction in the probability

of taking the bus. The bus utility function is mathematically given below as,







time to the train station, waiting time, on-board time, ticket, on-board crowding, ‘level

of influence of train services on the current choice of home location’, ‘employment

status’, and ‘age group’. By having the last nine variables at zero, the utility value of

taking the train is positive at 2.761, as shown by the sign and magnitude of its estimated

alternative specific constant. The sign of each estimated coefficient of key factors is

as expected. The train utility function is mathematically given below as,



𝑡𝑖𝑐𝑘𝑒𝑡) + (−0.328) ∗ (𝑡𝑟𝑎𝑖𝑛 − 𝑐𝑟𝑜𝑤𝑑𝑒𝑑𝑛𝑒𝑠𝑠 − 𝑜𝑛𝑏𝑜𝑎𝑟𝑑) + (−0.209) ∗ (𝑎𝑐𝑐𝑒𝑠𝑠 −

𝑡𝑖𝑚𝑒 − 𝑓𝑟𝑜𝑚 − 𝑠𝑡𝑎𝑡𝑖𝑜𝑛) + (−0.195) ∗ (𝑡𝑟𝑎𝑖𝑛 − 𝑠𝑒𝑟𝑣𝑖𝑐𝑒𝑠 − 𝑖𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 − 𝑜𝑛 − ℎ𝑜𝑚𝑒 −

𝑙𝑜𝑐𝑎𝑡𝑖𝑜𝑛) + ∗ 0.070 (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝) + (−0.219) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) [7-8]


The car utility function is estimated by eight independent variables; three travel

attributes, and five socio-economic factors. The key travel attributes are on-board time,

parking cost, and toll cost. The vital socio-economic factors are ‘employment status’,

‘whether a car is required for work’, ‘income level’, ‘age group’, and ‘gender’. The

car utility function is mathematically given below as,


(−0.073) ∗ (𝑐𝑎𝑟 − 𝑡𝑜𝑙𝑙 − 𝑐𝑜𝑠𝑡) + (−0.219) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + 0.446 ∗ (𝑐𝑎𝑟 −

𝑟𝑒𝑞𝑢𝑖𝑟𝑒𝑑 − 𝑓𝑜𝑟 − 𝑤𝑜𝑟𝑘𝑖𝑛𝑔) + 0.135 ∗ (𝑖𝑛𝑐𝑜𝑚𝑒 − 𝑙𝑒𝑣𝑒𝑙 ) + 0.070 ∗ (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝 ) +

0.559 ∗ (𝑔𝑒𝑛𝑑𝑒𝑟 ) [7-9]

7.4 THE UPPER NEST OF THE NESTED STRUCTURE

The upper nest of each best-fitted nested model consists of a public transport and

a private transport branch.


7.4.1 The private transport branch for all datasets

Because of the degenerative nature of the private transport branch (refer to

Figure 4.1), the utility function of private transport for each city is equal to the utility

function of car for each city. The IV parameters of private transport have been

normalized to 1 during model estimation (𝜆(𝐶𝑎𝑟|𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) =

1; 𝜆𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡 = 1; 𝑎𝑛𝑑 𝐼𝑉𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡 = 1). There is no more variance at

the top (Private transport) of a degenerative branch than there is at the bottom of the

branch (Car) (Hensher et al., 2005; Hunt, 2000). Hence, the private transport utility

function is exactly the same as the car utility function: 𝑉𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡 = 𝑉𝐶𝑎𝑟.

7.4.2 The public transport branch of the Sydney dataset

The best-fitted model presented in Table 7.1 shows that public transport users in

Sydney are characterized by an alternative specific constant, ‘highest educational level

attained’, ‘employment status’, and ‘gender’. The utility function of public transport

is also influenced by the estimated coefficient of IV parameter of public transport and

the utility function of bus and train at the lower nest. The public transport utility

function is mathematically given below as,

𝑉𝑃𝑢𝑏𝑙𝑖𝑐 𝑡𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡 = 0.424 ∗ [0.503 + (−0.253) ∗ (ℎ𝑖𝑔ℎ𝑒𝑠𝑡 − 𝑒𝑑𝑢𝑐𝑎𝑡𝑖𝑜𝑛𝑎𝑙 −

𝑞𝑢𝑎𝑙𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠) + (−0.129) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + (−0.180) ∗ (𝑔𝑒𝑛𝑑𝑒𝑟) + 1 ∗

ln (𝐸𝑥𝑝(𝑉𝐵𝑢𝑠) + 𝐸𝑥𝑝(𝑉𝑇𝑟𝑎𝑖𝑛))], where 𝑉𝐵𝑢𝑠 and 𝑉𝑇𝑟𝑎𝑖𝑛 are calculated using Equation 7-

1 and 7-2, respectively. [7-10]

7.4.3 The public transport branch of the Melbourne dataset


Melbourne are characterized by an alternative specific constant, ‘highest educational

level attained’, ‘employment status’, ‘gender’, and ‘age group’. The utility function of

public transport is also influenced by the estimated coefficient of IV parameter of

public transport and the utility function of bus and train at the lower nest. The public

transport utility function is mathematically given below as,


𝑞𝑢𝑎𝑙𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠) + (−0.171) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + (−0.123) ∗ (𝑔𝑒𝑛𝑑𝑒𝑟) +

(−0.111) ∗ (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝) + 1 ∗ ln (𝐸𝑥𝑝(𝑉𝐵𝑢𝑠) + 𝐸𝑥𝑝(𝑉𝑇𝑟𝑎𝑖𝑛))], where 𝑉𝐵𝑢𝑠 and 𝑉𝑇𝑟𝑎𝑖𝑛

are calculated using Equation 7-4 and 7-5, respectively. [7-11]


7.4.4 The public transport branch of the Brisbane dataset


Brisbane are characterized by an alternative specific constant, ‘highest educational

level attained’, ‘employment status’, ‘gender’, and ‘age group’. The utility function of

public transport is also influenced by the estimated coefficient of IV parameter of

public transport, and the utility function of bus and train at the lower nest. The public

transport utility function is mathematically given below as,


𝑞𝑢𝑎𝑙𝑖𝑓𝑖𝑐𝑎𝑡𝑖𝑜𝑛𝑠) + (−0.219) ∗ (𝑒𝑚𝑝𝑙𝑜𝑦𝑚𝑒𝑛𝑡 − 𝑠𝑡𝑎𝑡𝑢𝑠) + (−0.559) ∗ (𝑔𝑒𝑛𝑑𝑒𝑟) +

(−0.070) ∗ (𝑎𝑔𝑒 − 𝑔𝑟𝑜𝑢𝑝) + 1 ∗ ln (𝐸𝑥𝑝(𝑉𝐵𝑢𝑠) + 𝐸𝑥𝑝(𝑉𝑇𝑟𝑎𝑖𝑛))], where 𝑉𝐵𝑢𝑠 and 𝑉𝑇𝑟𝑎𝑖𝑛

are calculated using Equation 7-7 and 7-8, respectively. [7-12]

7.5 THE PROBABILITY FUNCTIONS

With reference to Figure 4.1, 𝑃(𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) is the unconditional

probability of a respondent taking public transport, and 𝑃(𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) is the

unconditional probability of a respondent taking private transport. As previously

discussed in Section 7.4.1, the utility of car is equal to the utility of private transport.

Consequently, the unconditional probability of a respondent driving a car – 𝑃(𝐶𝑎𝑟) –

is equal to the unconditional probability of a respondent taking private transport,

𝑃(𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡). Utilising the utility functions derived in the previous section,

the probability functions at the upper nest are mathematically defined below (Hensher

et al., 2005; Hunt, 2000).

𝑃(𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) =𝐸𝑋𝑃[ 𝑉𝑃𝑢𝑏𝑙𝑖𝑐 T𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡]

𝐸𝑋𝑃[ 𝑉𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡]+𝐸𝑋𝑃[ 𝑉𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡] [7-13]

𝑃(𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) =𝐸𝑋𝑃[ 𝑉𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡]

𝐸𝑋𝑃[ 𝑉𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡]+𝐸𝑋𝑃[ 𝑉𝑃𝑟𝑖𝑣𝑎𝑡𝑒 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡]= 𝑃(𝐶𝑎𝑟) [7-14]

,where 𝑃(𝐵𝑢𝑠|𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) is the probability of a respondent taking the bus

(conditional on their choosing public transport), and 𝑃(𝑇𝑟𝑎𝑖𝑛|𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) is

the probability of a respondent taking the train (conditional on their choosing public

transport). These probabilities are mathematically defined below (Hensher et al., 2005;

Hunt, 2000).

𝑃(𝐵𝑢𝑠|𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) =𝐸𝑋𝑃[ 𝑉𝐵𝑢𝑠]

𝐸𝑋𝑃[ 𝑉𝐵𝑢𝑠]+𝐸𝑋𝑃[ 𝑉𝑇𝑟𝑎𝑖𝑛] [7-15]


𝑃(𝑇𝑟𝑎𝑖𝑛|𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡)𝐸𝑋𝑃[ 𝑉𝑇𝑟𝑎𝑖𝑛]

𝐸𝑋𝑃[ 𝑉𝐵𝑢𝑠]+𝐸𝑋𝑃[ 𝑉𝑇𝑟𝑎𝑖𝑛] [7-16]

Hence, at the lower nest, 𝑃(𝐵𝑢𝑠) is the unconditional probability of a

respondent taking the bus, and 𝑃(𝑇𝑟𝑎𝑖𝑛) is the unconditional probability of a

respondent taking the train. These probabilities are estimated by the following

equations. They are mathematically defined below (Hensher et al., 2005; Hunt, 2000).

𝑃(𝐵𝑢𝑠) = 𝑃(𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) ∗ 𝑃(𝐵𝑢𝑠|𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) [7-17]

𝑃(𝑇𝑟𝑎𝑖𝑛) = 𝑃(𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) ∗ 𝑃(𝑇𝑟𝑎𝑖𝑛|𝑃𝑢𝑏𝑙𝑖𝑐 𝑇𝑟𝑎𝑛𝑠𝑝𝑜𝑟𝑡) [7-18]

7.6 THE AVERAGE PROFILE OF TRAVELLERS FOR THE BASELINE

SCENARIO

Responses related to train riders’ experience, non-riders’ experience, and socio-

economic factors, reveal the respondents’ travel behaviours (Section 3.4). Using this

knowledge of travel behaviours, documented information about each city, and the key

variables from the best-fitted utility functions (Section 7.3 and 7.4), this section

identifies an average traveller’s profile for each transport mode in Sydney, Brisbane,

and Melbourne (Queensland Government, 2016; Transport for NSW, 2016; Victoria

State Government, 2016). In the next section (7.7), the average traveller’s profile and

experience are used to estimate the utility and probability value of taking the bus, train,

car, and public transport for each city at the baseline scenario.

7.6.1 The average profile of Sydney travellers

Based on the Sydney respondents’ travel experience and socio-economic

profiles at the time of their most recent home-based journey, their profile and trip

characteristics have been identified as follows.

Table 7.2 The average profile of Sydney travellers for a baseline scenario

Type of

traveller

Key variables from the

corresponding utility

function

Gathered from revealed travel behaviours

dataset and has been checked against

documented information

A bus rider Bus waiting time 8 minutes

Bus on-board time 32 minutes

Bus ticket $ 2.37

Bus on-board crowding Experience on-board crowding

Employment Status Full time employed

A train rider Access time to train station 16 minutes

Train waiting time 8 minutes

Train on-board time 33 minutes

Train ticket $ 3.40

Train on-board crowding Experience on-board crowding

Access time from train station 9 minutes


Type of

traveller



function






decision Significant

Age group 31-40 years old


A public Highest educational

qualification Bachelor degree

transport Employment status Full time employed

user Gender Female

A car driver Car on-board time 28 minutes

Car parking costa $ 5.00

Car toll costa $ 2.50

Employment status Full time employed

Whether car is required for

working No

Driving licence Yes a The baseline car parking and toll costs may not be a true representation of current parking and toll

costs. Rather, both costs represent the average parking and toll cost paid by all Sydney car driver

respondents. Based on their responses to the ‘Most recent experience’ questions, there are two very

different groups of car drivers: those who manage to pay nothing for parking and tolls, and those who

pay enormous amounts for them.

7.6.2 The average profile of Melbourne travellers

Based on the Melbourne respondents’ travel experience and socio-economic



Table 7.3 The average profile of Melbourne travellers for a baseline scenario

Type of

traveller



function






Bus ticket $ 2.18







Train ticket $ 2.62




decision Significant






user Gender Female



Type of

traveller



function







Car fuel costb $ 2.62



working No

Driving licence Yes

Income level

Pre-tax household weekly income $1,600 and

above a The baseline car parking and toll costs may not be a true representation of current costs. Rather, both

costs represent the average parking and toll costs paid by all Melbourne car driver respondents. Based

on their responses on the ‘Most recent experience’ questions, there are two very different groups of

Melbourne car driver respondents: those who manage to pay nothing for parking and tolls, and those

who pay enormous amounts for them. b The baseline car fuel toll cost represents the average fuel cost spent by Melbourne car-driving

respondents on a one-way, home-based trip to a single destination.

7.6.3 The average profile of Brisbane travellers

Based on the Brisbane respondents’ travel experience and socio-economic



Table 7.4 The average profile of Brisbane travellers for a baseline scenario

Type of

traveller



function






Bus ticket $ 3.23


Employment Status Outside workforce




Train ticket $ 2.92

Train on-board crowding Experience on-board crowding




decision Insignificant






user Gender Female







Type of

traveller



function





working No

Income level Unreported income


Gender Female a The baseline car parking and toll costs may not be a true representation of current parking and toll

costs. Rather, both costs represent the average parking and toll costs paid by all Brisbane car driving

respondents. Based on their responses to the ‘Most recent experience’ questions, there are two very

different groups of Brisbane car driving respondents: those who manage to pay nothing for parking and

tolls, and those who pay enormous amounts for them.

7.7 THE POLICY INTERVENTION SCENARIO ANALYSIS

With reference to the defined utility function of bus, train, car, and public

transport in Section 7.3 and 7.4, two policy interventions have been determined, each

with a different purpose. The two different intervention purposes are: 1) to encourage

public transport (i.e., both bus and train) ridership; and 2) to discourage regular car

usage. Increased public transport ridership can be achieved by improving passengers’

level of service (LOS); specifically, by reducing public transport waiting times. On the

other hand, a reduction in regular car usage can be achieved by increasing its cost,

particularly by increasing parking and toll costs.

By holding all other factors constant, a sequence of scenarios were simulated by

changing a particular area of policy intervention for each city dataset, and for each

policy intervention purpose. Take, for example, a simulation to encourage public

transport ridership in Sydney: By holding all other factors constant and employing the

average traveller’s profile and experience (as defined in Section 7.6.1), bus waiting

time was reduced from the baseline scenario (i.e., 100%) to a 50% scenario with a 5%

interval. The calculated increment of public transport mode shares (public transport

shares) and the corresponding decrement of car mode shares (car shares) in Sydney

were observed.

Meanwhile, consider an example of a simulation to discourage regular car usage

in Sydney: By holding all other factors constant and employing the average traveller’s

profile and experience (as defined in Section 7.6.1), the parking cost was increased

from the baseline scenario (i.e., 100%) to a 200% scenario with a 10% interval. The

calculated decrement of car shares and the corresponding increment of public transport

shares for Sydney were detected. The single policy intervention scenario analysis was


replicated for bus waiting time and toll cost factors for each city. The resulting rapid

change in car and public transport shares are acknowledged in each simulation.

The single policy intervention scenario analyses indicate that the change in mode

shares – that is, either the increment of public transport share or the decrement of car

share – is not optimized, and is not an accurate or realistic representation of the way

policy interventions operate. Hence, in order to replicate actual policy intervention,

and to optimise the change in mode share, subsequent analyses incorporated both

policy intervention purposes, and held all other factors constant. In particular, to gain

insights into the direct impact of policy interventions on car and public transport mode

shares, 176 different travel scenarios were simulated for each city. By controlling all

other factors, the bus and train waiting time were reduced from the baseline scenario

(i.e., 100%) to a 50% scenario with a 5% interval. Furthermore, the combined parking

and toll cost was increased from a baseline scenario (i.e., 100%) to a 250% scenario

with a 10% interval. The calculated decrement of car shares and the corresponding

increment of public transport shares for each city were tabulated, and illustrated with

surface graphs. The findings from each city are presented below.

7.7.1 The impact of policy interventions in Sydney

With reference to the utility functions of bus, train, car, and public transport for

the Sydney dataset (Equations 7-1, 7-2, 7-3, and 7-10, respectively), and to the profile

of Sydney travellers (Table 7.2), the following probability of driving a car and the

probability of taking public transport are calculated for each cell of travel scenarios,

by controlling for all other factors. The probability values of the results of each travel

scenario are tabulated in Table 7.5.

Table 7.5 The probability values of driving a car and taking public transport in Sydney: 176

different policy intervention scenarios

Parking and Toll cost

100% 110% 120% 130% 140% 150%

$7.5 $8.25 $9 $9.75 $10.5 $11.25

P(Car)

P(Public Transport)

Bus and

Train

waiting

time

(Minutes)

100% 8 63.83% 62.88% 61.93% 60.96% 59.99% 59.01%

36.17% 37.12% 38.07% 39.04% 40.01% 40.99%

95% 7.6 63.70% 62.75% 61.79% 60.82% 59.85% 58.87%

36.30% 37.25% 38.21% 39.18% 40.15% 41.13%

90% 7.2 63.56% 62.61% 61.65% 60.69% 59.71% 58.73%

36.44% 37.39% 38.35% 39.31% 40.29% 41.27%


85% 6.8 63.43% 62.48% 61.52% 60.55% 59.57% 58.59%

36.57% 37.52% 38.48% 39.45% 40.43% 41.41%

80% 6.4 63.29% 62.34% 61.38% 60.41% 59.43% 58.45%

36.71% 37.66% 38.62% 39.59% 40.57% 41.55%

75% 6 63.16% 62.21% 61.24% 60.27% 59.29% 58.31%

36.84% 37.79% 38.76% 39.73% 40.71% 41.69%

70% 5.6 63.02% 62.07% 61.10% 60.13% 59.15% 58.16%

36.98% 37.93% 38.90% 39.87% 40.85% 41.84%

65% 5.2 62.89% 61.93% 60.97% 59.99% 59.01% 58.02%

37.11% 38.07% 39.03% 40.01% 40.99% 41.98%

60% 4.8 62.75% 61.79% 60.83% 59.85% 58.87% 57.88%

37.25% 38.21% 39.17% 40.15% 41.13% 42.12%

55% 4.4 62.61% 61.65% 60.69% 59.71% 58.73% 57.74%

37.39% 38.35% 39.31% 40.29% 41.27% 42.26%

50% 4 62.48% 61.52% 60.55% 59.57% 58.58% 57.59%

37.52% 38.48% 39.45% 40.43% 41.42% 42.41%


160% 170% 180% 190% 200%

$12 $12.75 $13.5 $14.25 $15

P(Car)

P(Public Transport)

Bus and

Train

waiting

time

(Minutes)

100% 8 58.02% 57.02% 56.021% 55.015% 54.005%

41.98% 42.98% 43.979% 44.985% 45.995%

95% 7.6 57.88% 56.88% 55.879% 54.872% 53.862%

42.12% 43.12% 44.121% 45.128% 46.138%

90% 7.2 57.74% 56.74% 55.737% 54.730% 53.718%

42.26% 43.26% 44.263% 45.270% 46.282%

85% 6.8 57.60% 56.60% 55.594% 54.586% 53.575%

42.40% 43.40% 44.406% 45.414% 46.425%

80% 6.4 57.45% 56.46% 55.451% 54.443% 53.431%

42.55% 43.54% 44.549% 45.557% 46.569%

75% 6 57.31% 56.31% 55.308% 54.299% 53.286%

42.69% 43.69% 44.692% 45.701% 46.714%

70% 5.6 57.17% 56.17% 55.164% 54.154% 53.141%

42.83% 43.83% 44.836% 45.846% 46.859%

65% 5.2 57.03% 56.03% 55.020% 54.009% 52.996%

42.97% 43.97% 44.980% 45.991% 47.004%

60% 4.8 56.88% 55.88% 54.875% 53.864% 52.850%

43.12% 44.12% 45.125% 46.136% 47.150%

55% 4.4 56.74% 55.74% 54.730% 53.719% 52.704%

43.26% 44.26% 45.270% 46.281% 47.296%

50% 4 56.60% 55.59% 54.584% 53.573% 52.558%

43.40% 44.41% 45.416% 46.427% 47.442%

Parking and Toll cost 210% 220% 230% 240% 250%


$15.75 $16.5 $17.25 $18 $18.75

P(Car)

P(Public Transport)

Bus and

Train

waiting

time

(Minutes)

100% 8 52.991% 51.975% 50.958% 49.939% 48.921%

47.009% 48.025% 49.042% 50.061% 51.079%

95% 7.6 52.848% 51.832% 50.814% 49.796% 48.778%

47.152% 48.168% 49.186% 50.204% 51.222%

90% 7.2 52.704% 51.688% 50.670% 49.652% 48.633%

47.296% 48.312% 49.330% 50.348% 51.367%

85% 6.8 52.560% 51.544% 50.526% 49.507% 48.489%

47.440% 48.456% 49.474% 50.493% 51.511%

80% 6.4 52.416% 51.399% 50.381% 49.362% 48.344%

47.584% 48.601% 49.619% 50.638% 51.656%

75% 6 52.271% 51.254% 50.236% 49.217% 48.200%

47.729% 48.746% 49.764% 50.783% 51.800%

70% 5.6 52.126% 51.108% 50.090% 49.072% 48.054%

47.874% 48.892% 49.910% 50.928% 51.946%

65% 5.2 51.980% 50.963% 49.944% 48.926% 47.909%

48.020% 49.037% 50.056% 51.074% 52.091%

60% 4.8 51.834% 50.817% 49.798% 48.780% 47.763%

48.166% 49.183% 50.202% 51.220% 52.237%

55% 4.4 51.688% 50.670% 49.652% 48.634% 47.617%

48.312% 49.330% 50.348% 51.366% 52.383%

50% 4 51.541% 50.523% 49.505% 48.487% 47.470%

48.459% 49.477% 50.495% 51.513% 52.530%

In order to identify the best combination of policy interventions, a certain target

of probability of driving the car and taking public transport has to be set at the outset.

This target was set at a 10% decrease in the probability of driving car in Sydney. Table

7.5 demonstrates that there are two possible travel scenarios that can reduce the

probability of driving a car in Sydney by at least 10% (i.e., from 64% to 54%). The

first scenario is where the bus and train waiting times are reduced to 80% of the

baseline scenario (i.e., from 8 minutes to 6.4 minutes), and the total parking and toll

cost is increased to 1.9 times the baseline scenario (i.e., from $7.5 to $14.25). The

second scenario is where the bus and train waiting times are the same as for the

baseline scenario (i.e., 8 minutes), and the total parking and toll costs are increased to

twice the baseline scenario (i.e., from $7.5 to $15).

Of these two scenarios, the second seems to be more realistic than the first. This

is because the implementation of plans to reduce bus and train waiting times requires


a higher initial cost than the cost of imposing higher parking and toll costs (Boardman,

Greenberg, Vining, & Weimer, 2017; Guo & Wilson, 2011; Wardman, 2004). For

instance, the additional funds obtained from the increment of parking and toll costs

can be invested to improve other aspects of passenger LOS, such as the provision of

real time public transport service information at major bus stops and train stations.

While holding the bus and train waiting times at the baseline scenario, and all other

factors constant, a further increment of total parking and toll costs up to 2.5 times the

baseline scenario reduces the car mode share from 64% to 49%. When the most

extreme travel scenario is applied (i.e., a 50% reduction in bus and train waiting times,

and a 250% increase in total parking and toll costs), the probability of driving a car in

Sydney decreases from 64% to 47%.

The change in the probability of driving a car and taking public transport are

separately illustrated in Figure 7.1 and Figure 7.2, respectively. Figure 7.1 shows that

the probability of driving a car in Sydney decreases with both the increment of parking

and toll costs (i.e., from $ 7.50 to $18.75), and the decrement of bus and train waiting

times (i.e., from 8 minutes to 4 minutes). Meanwhile, Figure 7.2 demonstrates that the

probability of taking public transport in Sydney increases both with the increment of

parking and toll costs (i.e., from $ 7.50 to $18.75), and the decrement of bus and train

waiting times (i.e., from 8 minutes to 4 minutes).


Figure 7.1 : The probability of driving a car in Sydney as the result of policy intervention

related to bus and train waiting times and parking and toll costs, while holding all other factors

constant

Figure 7.2 : The probability of taking public transport in Sydney as the result of policy

interventions in both bus and train waiting times and parking and toll costs, holding all other

factors constant

8

7.2

6.4

5.6

4.84

40.00%

45.00%

50.00%

55.00%

60.00%

65.00%

Bus a

nd T

rain

waitin

g t

ime (

min

ute

s)

Pro

babili

ty o

f drivin

g c

ar

Combined parking and toll cost ($)

40.00%-45.00% 45.00%-50.00% 50.00%-55.00% 55.00%-60.00% 60.00%-65.00%

8

7.2

6.4

5.6

4.84

30.00%

35.00%

40.00%

45.00%

50.00%

55.00%

Bus a

nd T

rain

waitin

g t

ime (

min

ute

s)

Pro

babili

ty o

f ta

kin

g p

ublic

tra

nsport


30.00%-35.00% 35.00%-40.00% 40.00%-45.00% 45.00%-50.00% 50.00%-55.00%


7.7.2 The impact of policy interventions in Melbourne

With reference to the utility function of bus, train, car, and public transport in

the Melbourne dataset (Equations 7-4, 7-5, 7-6, and 7-11, respectively) and the

Melbourne traveller’s profile (Table 7.3), the probability of driving a car and the

probability of taking public transport are calculated for each cell of travel scenarios,

by controlling for all other factors. The probability values of the results of each travel

scenario are tabulated in Table 7.6.

Table 7.6 The probability values of driving a car and taking public transport in Melbourne:

176 different policy intervention scenarios


100% 110% 120% 130% 140% 150%

$7.5 $8.25 $9 $9.75 $10.5 $11.25

P(Car)

P(Public Transport)

Bus and

Train

waiting

time

(Minutes)

100% 8 58.16% 57.129% 56.097% 55.059% 54.017% 52.971%

41.84% 42.87% 43.90% 44.94% 45.98% 47.03%

95% 7.6 57.99% 56.96% 55.93% 54.89% 53.84% 52.80%

42.01% 43.04% 44.07% 45.11% 46.16% 47.20%

90% 7.2 57.82% 56.79% 55.75% 54.71% 53.67% 52.62%

42.18% 43.21% 44.25% 45.29% 46.33% 47.38%

85% 6.8 57.65% 56.62% 55.58% 54.54% 53.50% 52.45%

42.35% 43.38% 44.42% 45.46% 46.50% 47.55%

80% 6.4 57.48% 56.45% 55.41% 54.37% 53.32% 52.28%

42.52% 43.55% 44.59% 45.63% 46.68% 47.72%

75% 6 57.31% 56.27% 55.24% 54.20% 53.15% 52.10%

42.69% 43.73% 44.76% 45.80% 46.85% 47.90%

70% 5.6 57.14% 56.10% 55.06% 54.02% 52.98% 51.93%

42.86% 43.90% 44.94% 45.98% 47.02% 48.07%

65% 5.2 56.96% 55.93% 54.89% 53.85% 52.80% 51.75%

43.04% 44.07% 45.11% 46.15% 47.20% 48.25%

60% 4.8 56.79% 55.76% 54.72% 53.67% 52.63% 51.58%

43.21% 44.24% 45.28% 46.33% 47.37% 48.42%

55% 4.4 56.62% 55.59% 54.54% 53.50% 52.45% 51.40%

43.38% 44.41% 45.46% 46.50% 47.55% 48.60%

50% 4 56.45% 55.41% 54.37% 53.33% 52.28% 51.23%

43.55% 44.59% 45.63% 46.67% 47.72% 48.77%


160% 170% 180% 190% 200%

$12 $12.75 $13.5 $14.25 $15

P(Car)

P(Public Transport)

Bus and

Train

100% 8 51.922% 50.872% 49.821% 48.770% 47.720%

48.08% 49.13% 50.179% 51.230% 52.280%


waiting

time

(Minutes)

95% 7.6 51.75% 50.70% 49.647% 48.596% 47.547%

48.25% 49.30% 50.353% 51.404% 52.453%

90% 7.2 51.57% 50.52% 49.473% 48.422% 47.373%

48.43% 49.48% 50.527% 51.578% 52.627%

85% 6.8 51.40% 50.35% 49.299% 48.249% 47.200%

48.60% 49.65% 50.701% 51.751% 52.800%

80% 6.4 51.23% 50.18% 49.125% 48.074% 47.026%

48.77% 49.82% 50.875% 51.926% 52.974%

75% 6 51.05% 50.00% 48.950% 47.900% 46.852%

48.95% 50.00% 51.050% 52.100% 53.148%

70% 5.6 50.88% 49.83% 48.776% 47.726% 46.678%

49.12% 50.17% 51.224% 52.274% 53.322%

65% 5.2 50.70% 49.65% 48.601% 47.552% 46.504%

49.30% 50.35% 51.399% 52.448% 53.496%

60% 4.8 50.53% 49.48% 48.427% 47.377% 46.330%

49.47% 50.52% 51.573% 52.623% 53.670%

55% 4.4 50.35% 49.30% 48.252% 47.203% 46.157%

49.65% 50.70% 51.748% 52.797% 53.843%

50% 4 50.18% 49.13% 48.077% 47.028% 45.982%

49.82% 50.87% 51.923% 52.972% 54.018%


210% 220% 230% 240% 250%

$15.75 $16.5 $17.25 $18 $18.75

P(Car)

P(Public Transport)

Bus and

Train

waiting

time

(Minutes)

100% 8 46.673% 45.628% 44.587% 43.551% 42.520%

53.327% 54.372% 55.413% 56.449% 57.480%

95% 7.6 46.500% 45.455% 44.415% 43.380% 42.350%

53.500% 54.545% 55.585% 56.620% 57.650%

90% 7.2 46.326% 45.283% 44.243% 43.209% 42.180%

53.674% 54.717% 55.757% 56.791% 57.820%

85% 6.8 46.153% 45.110% 44.072% 43.038% 42.010%

53.847% 54.890% 55.928% 56.962% 57.990%

80% 6.4 45.980% 44.938% 43.900% 42.867% 41.841%

54.020% 55.062% 56.100% 57.133% 58.159%

75% 6 45.807% 44.765% 43.728% 42.696% 41.671%

54.193% 55.235% 56.272% 57.304% 58.329%

70% 5.6 45.634% 44.593% 43.556% 42.526% 41.501%

54.366% 55.407% 56.444% 57.474% 58.499%

65% 5.2 45.460% 44.420% 43.385% 42.355% 41.332%

54.540% 55.580% 56.615% 57.645% 58.668%

60% 4.8 45.287% 44.247% 43.213% 42.184% 41.162%

54.713% 55.753% 56.787% 57.816% 58.838%

55% 4.4 45.113% 44.075% 43.041% 42.014% 40.993%

54.887% 55.925% 56.959% 57.986% 59.007%


50% 4 44.940% 43.902% 42.870% 41.843% 40.824%

55.060% 56.098% 57.130% 58.157% 59.176%


of probability of driving a car and taking public transport has to be set at the outset. In

this case, the target was a 10% decrease (i.e., from 58% to 48%) in the probability of

driving a car in Melbourne. Table 7.6 demonstrates that, to reach this target, there are

two possible travel scenarios. The first is the reduction of bus and train waiting time

to 60% of the baseline scenario (i.e., from 8 minutes to 4.8 minutes), and an increase

in the total parking and toll cost to 1.8 times the baseline scenario (i.e., from $7.5 to

$13.5). The second travel scenario is where the bus and train waiting time is the same

as the baseline scenario (i.e., 8 minutes), and the total parking and toll cost is increased

to twice the baseline scenario (i.e., from $7.5 to $15).

Of these two scenarios, the second seems more realistic. This is because the

implementation of plans to reduce bus and train waiting time requires a higher initial

cost than the imposition of a higher parking and toll cost (Boardman et al., 2017; Guo

& Wilson, 2011; Wardman, 2004). In this instance, the additional funds obtained from

the increment of parking and toll cost could be invested to improve other aspects of

passenger LOS; for example, the provision of charging stations on board public

transport services. While holding the bus and train waiting time at the baseline

scenario, and all other factors constant, a further increment in total parking and toll

cost of up to 2.5 times the baseline scenario decreases the car mode share from 58%

to 43%. When the most extreme scenario is applied (i.e., a 50% reduction in bus and

train waiting time and a 250% increase in total parking and toll costs), the probability

of driving a car in Melbourne decreases from 58% to 41%.

The change in the probability of driving a car and the probability of taking public

transport are separately illustrated in Figure 7.3 and Figure 7.4, respectively. Figure

7.3 shows that the probability of driving a car in Melbourne decreases with both the

increment of parking and toll costs (i.e., from $ 7.50 to $18.75) and the decrement of

bus and train waiting time (i.e., from 8 minutes to 4 minutes). Meanwhile, Figure 7.4

demonstrates that the probability of taking public transport in Melbourne increases

with both the increment of parking and toll cost (i.e., from $ 7.50 to $18.75) and the

decrement of bus and train waiting time (i.e., from 8 minutes to 4 minutes).


Figure 7.3 : The probability of driving a car in Melbourne as the result of policy interventions

in bus and train waiting time and parking and toll costs, holding all other factors constant

Figure 7.4 : The probability of taking public transport in Melbourne as the result of policy

interventions in bus and train waiting time and parking and toll costs, holding all other factors

constant

8

7.2

6.4

5.6

4.84

40.00%

45.00%

50.00%

55.00%

60.00%

7.5910.5

1213.5

15

16.5

18

Bus a

nd T

rain

waitin

g t

ime (

min

ute

s)

Pro

babili

ty o

f drivin

g c

ar

Combined parking and toll

cost ($)

40.00%-45.00% 45.00%-50.00% 50.00%-55.00% 55.00%-60.00%

8

7.2

6.4

5.6

4.8

4

40.00%

45.00%

50.00%

55.00%

60.00%B

us a

nd T

rain

waitin

g t

ime (

min

ute

s)

Pro

babili

ty o

f ta

kin

g p

ublic

tra

nsport


40.00%-45.00% 45.00%-50.00% 50.00%-55.00% 55.00%-60.00%


7.7.3 The impact of policy interventions in Brisbane

With reference to the utility function of bus, train, car, and public transport for

the Brisbane dataset (Equations 7-7, 7-8, 7-9, and 7-12, respectively) and the Brisbane

traveller’s profile (Table 7.4), the probability of driving a car and the probability of

taking public transport is calculated for each cell of travel scenarios, by controlling for

all other factors. The probability values of the result of each travel scenario are

tabulated in Table 7.7.

Table 7.7 The probability values of driving a car and taking public transport in Brisbane:

176 different policy intervention scenarios


100% 110% 120% 130% 140% 150%

$7.5 $8.25 $9 $9.75 $10.5 $11.25

P(Car)

P(Public Transport)

Waiting

time

(Minutes)

100% 9 for Bus 82.96% 82.25% 81.52% 80.77% 79.99% 79.19%

11 for Train 17.04% 17.75% 18.48% 19.23% 20.01% 20.81%

95% 8.55 for Bus 82.88% 82.18% 81.44% 80.69% 79.91% 79.11%

10.45 for Train 17.12% 17.82% 18.56% 19.31% 20.09% 20.89%

90% 8.1 for Bus 82.81% 82.10% 81.36% 80.61% 79.82% 79.02%

9.9 for Train 17.19% 17.90% 18.64% 19.39% 20.18% 20.98%

85% 7.65 for Bus 82.74% 82.02% 81.28% 80.52% 79.74% 78.93%

9.35 for Train 17.26% 17.98% 18.72% 19.48% 20.26% 21.07%

80% 7.2 for Bus 82.66% 81.94% 81.20% 80.44% 79.65% 78.85%

8.8 for Train 17.34% 18.06% 18.80% 19.56% 20.35% 21.15%

75% 6.75 for Bus 82.58% 81.87% 81.12% 80.36% 79.57% 78.76%

8.25 for Train 17.42% 18.13% 18.88% 19.64% 20.43% 21.24%

70% 6.3 for Bus 82.51% 81.79% 81.04% 80.27% 79.48% 78.67%

7.7 for Train 17.49% 18.21% 18.96% 19.73% 20.52% 21.33%

65% 5.85 for Bus 82.43% 81.71% 80.96% 80.19% 79.40% 78.58%

7.15 for Train 17.57% 18.29% 19.04% 19.81% 20.60% 21.42%

60% 5.4 for Bus 82.36% 81.63% 80.88% 80.11% 79.31% 78.49%

6.6 for Train 17.64% 18.37% 19.12% 19.89% 20.69% 21.51%

55% 4.95 for Bus 82.28% 81.55% 80.80% 80.02% 79.23% 78.40%

6.05 for Train 17.72% 18.45% 19.20% 19.98% 20.77% 21.60%

50% 4.5 for Bus 82.20% 81.47% 80.72% 79.94% 79.14% 78.31%

5.5 for Train 17.80% 18.53% 19.28% 20.06% 20.86% 21.69%


160% 170% 180% 190% 200%

$12 $12.75 $13.5 $14.25 $15

P(Car)

P(Public Transport) Waiting

time

(Minutes)

100% 9 for Bus 78.37% 77.53% 76.66% 75.76% 74.85%

11 for Train 21.63% 22.47% 23.34% 24.24% 25.15%

95% 8.55 for Bus 78.28% 77.43% 76.56% 75.67% 74.75%


10.45 for

Train 21.72% 22.57% 23.44% 24.33% 25.25%

90% 8.1 for Bus 78.19% 77.34% 76.47% 75.57% 74.65%

9.9 for Train 21.81% 22.66% 23.53% 24.43% 25.35%

85% 7.65 for Bus 78.10% 77.25% 76.37% 75.47% 74.55%

9.35 for Train 21.90% 22.75% 23.63% 24.53% 25.45%

80% 7.2 for Bus 78.01% 77.16% 76.28% 75.38% 74.45%

8.8 for Train 21.99% 22.84% 23.72% 24.62% 25.55%

75% 6.75 for Bus 77.92% 77.06% 76.18% 75.28% 74.35%

8.25 for Train 22.08% 22.94% 23.82% 24.72% 25.65%

70% 6.3 for Bus 77.83% 76.97% 76.09% 75.18% 74.25%

7.7 for Train 22.17% 23.03% 23.91% 24.82% 25.75%

65% 5.85 for Bus 77.74% 76.88% 75.99% 75.08% 74.15%

7.15 for Train 22.26% 23.12% 24.01% 24.92% 25.85%

60% 5.4 for Bus 77.65% 76.78% 75.90% 74.98% 74.05%

6.6 for Train 22.35% 23.22% 24.10% 25.02% 25.95%

55% 4.95 for Bus 77.56% 76.69% 75.80% 74.88% 73.95%

6.05 for Train 22.44% 23.31% 24.20% 25.12% 26.05%

50% 4.5 for Bus 77.47% 76.60% 75.70% 74.79% 73.85%

5.5 for Train 22.53% 23.40% 24.30% 25.21% 26.15%


210% 220% 230% 240% 250%

$15.75 $16.5 $17.25 $18 $18.75

P(Car)

P(Public Transport)

Waiting

time

(Minutes)

100% 9 for Bus 73.91% 72.95% 71.97% 70.97% 69.94%

11 for Train 26.09% 27.05% 28.03% 29.03% 30.06%

95% 8.55 for Bus 73.81% 72.85% 71.86% 70.86% 69.83%

10.45 for

Train 26.19% 27.15% 28.14% 29.14% 30.17%

90% 8.1 for Bus 73.71% 72.74% 71.76% 70.75% 69.72%

9.9 for Train 26.29% 27.26% 28.24% 29.25% 30.28%

85% 7.65 for Bus 73.61% 72.64% 71.65% 70.64% 69.61%

9.35 for Train 26.39% 27.36% 28.35% 29.36% 30.39%

80% 7.2 for Bus 73.50% 72.53% 71.54% 70.53% 69.50%

8.8 for Train 26.50% 27.47% 28.46% 29.47% 30.50%

75% 6.75 for Bus 73.40% 72.43% 71.44% 70.42% 69.39%

8.25 for Train 26.60% 27.57% 28.56% 29.58% 30.61%

70% 6.3 for Bus 73.30% 72.32% 71.33% 70.31% 69.27%

7.7 for Train 26.70% 27.68% 28.67% 29.69% 30.73%

65% 5.85 for Bus 73.20% 72.22% 71.22% 70.20% 69.16%

7.15 for Train 26.80% 27.78% 28.78% 29.80% 30.84%

60% 5.4 for Bus 73.09% 72.11% 71.11% 70.09% 69.05%

6.6 for Train 26.91% 27.89% 28.89% 29.91% 30.95%

55% 4.95 for Bus 72.99% 72.01% 71.00% 69.98% 68.94%

6.05 for Train 27.01% 27.99% 29.00% 30.02% 31.06%


50% 4.5 for Bus 72.88% 71.90% 70.90% 69.87% 68.82%

5.5 for Train 27.12% 28.10% 29.10% 30.13% 31.18%


for the probability of driving a car and taking public transport has to be determined at

the outset; in this case, the target of achieving a 10% decrease (i.e., from 83% to 73%)

in the probability of driving a car in Brisbane was set. Table 7.7 demonstrates that

there are two possible travel scenarios that would achieve this target. The first is a

reduction in bus and train waiting time to 75% of the baseline scenario (i.e., from 9

minutes to 6.75 minutes for the bus waiting time, and 11 minutes to 8.25 minutes for

train waiting time), and an increase in the total parking and toll cost to 2.1 times the

baseline scenario (i.e., from $7.5 to $15.75). The second is where the bus and train

waiting time are the same as the baseline scenario (i.e., 9 minutes for bus waiting time,

and 11 minutes for train waiting time) and the total parking and toll cost is increased

to 2.2 times the baseline scenario (i.e., from $7.5 to $16.5).

Of these two travel scenarios, the second seems more realistic. This is because

implementation plans to reduce bus and train waiting times require a higher initial cost

than the cost of imposing higher parking and toll cost (Boardman et al., 2017; Guo &

Wilson, 2011; Wardman, 2004). In this case, the additional funds obtained from the

increment of parking and toll cost could be invested to improve other aspects of

passenger LOS, such as the provision of new buses or trains with better seats and

reduced carbon emissions. While holding the bus and train waiting times at the

baseline scenario and all other factors constant, a further increment of total parking

and toll cost up to 2.5 times the baseline scenario reduces the car mode share from

83% to 70%. When the most extreme travel scenarios are applied (i.e., a 50% reduction

in bus and train waiting time, and a 2.5 times increase in total parking and toll costs),

the probability of driving a car in Brisbane decreases from 83% to 69%.

The change in the probability of driving a car and the probability of taking public

transport are separately illustrated in Figure 7.5 and Figure 7.6, respectively. Figure

7.5 shows that the probability of driving a car decreases with an increase in parking

and toll cost (i.e., from $ 7.50 to $18.75) and a decrease in bus and train waiting times

(i.e., from 9 minutes to 4.5 minutes for bus waiting time, and from 11 minutes to 5.5

minutes for train waiting time). Meanwhile, Figure 7.6 demonstrates that the

probability of taking public transport increases with an increase in parking and toll cost


(i.e., from $ 7.50 to $18.75) and a decrease in bus and train waiting times (i.e., from 9

minutes to 4.5 minutes for bus waiting time, and from 11 minutes to 5.5 minutes for

train waiting time).

Figure 7.5 : The probability of driving a car in Brisbane as the result of policy interventions in

bus and train waiting time and parking and toll cost, holding all other factors constant

Figure 7.6 : The probability of taking public transport in Brisbane as the result of policy

interventions in bus and train waiting times and parking and toll costs, holding all other factors

constant

9 for bus and 11 for train

8.1 for bus and 9.9 for train



5.4 for bus and 6.6 for train4.5 for bus and 5.5 for train

60.00%

65.00%

70.00%

75.00%

80.00%

85.00%

7.59.75

12

14.25

16.5

18.75

Bus a

nd T

rain

waitin

g t

ime (

min

ute

s)

Pro

babili

ty o

f drivin

g c

ar

Combined parking and toll

cost ($)

60.00%-65.00% 65.00%-70.00% 70.00%-75.00% 75.00%-80.00% 80.00%-85.00%

9 for bus and 11 for train






15.00%

20.00%

25.00%

30.00%

35.00%

Bus a

nd T

rain

waitin

g t

ime (

min

ute

s)

Pro

babili

ty o

f ta

kin

g p

ublic

tra

nsport


15.00%-20.00% 20.00%-25.00% 25.00%-30.00% 30.00%-35.00%


7.8 THE DISCUSSIONS OF COMBINED POLICY INTERVENTION

SCENARIOS

Holding all other factors constant, the impacts of 176 different travel scenarios,

involving a wide range of bus and train waiting times and various parking and toll

costs in Sydney, Melbourne, and Brisbane, have been analysed. Results show that

policy interventions to improve public transport services can work effectively to

discourage regular car use and vice versa. More specifically, results of this analysis

show that the mode shares of car and public transport rapidly reverse as the result of

combined policy interventions; that is, car use rapidly decreases, while public transport

use rapidly increases.

In particular, as suggested by the literature (Axhausen & Polak, 1991; Feeney,

1989; Marsden, 2006; Young, Thompson, & Taylor, 1991), the increment in parking

cost is in the form of an increment in the hourly rate for on-street parking spots, and

the decrement in on-street parking spots in higher density areas. Another possibility

is to tighten the regulations around, and to increase the permit and licensing costs of

building multi-storey car park lots in high density traffic areas. As the total cost of

building is high, developers would be forced to charge exorbitant parking fees for their

car parks. The implementation of at least one of these possibilities would certainly

discourage car driving and foster the use of public transport services, especially if trips

are on a regular basis. If car drivers experience difficulties in finding parking spots,

while at the same time, paying an exorbitant parking cost for a single trip, they will

carefully reconsider their mode choices.

The issue of tolls, a form of a congestion tax for drivers, was raised in the survey

questionnaires. The charging of tolls for expressways connecting various suburbs with

the city centre or sites of major attraction can prevent drivers from using them. In turn,

this lack of expressway patronage could eventually reduce traffic pressure when

driving to, or within such areas. In the wider context, another form of congestion tax

for drivers is a congestion charge, which has been implemented successfully in London

and Singapore (Santos, 2005). The implementation of a congestion charge for entering

areas of high traffic density during peak hours would have an impact similar to the

impact of tolls; that is, it would discourage drivers from regularly using their cars in

such areas at such times. The roads and areas most suited to tolls and congestion taxes

can be determined through the analysis of past and forecasted traffics data in each city.

Such charges are structured based on operational costs and traffic demands (Santos,


2005). The increment of both costs has led car drivers to re-consider their mode choice

and driving habits (Beirão & Cabral, 2007; Feeney, 1989; Gardner & Abraham, 2010;

Santos & Rojey, 2004).

At the same time, policy interventions to encourage public transport ridership

should be introduced. The co-timing of these interventions is vital to maintaining and

improving the attractiveness of public transport services for both existing users and

mode shifting car users. The supply of public transport services needs to be adjusted

in anticipation of an increased demand. Ideally, the additional funds obtained from

increasing parking and toll costs or other congestion taxes can be allocated to

improving the passenger LOS across all public transport services.

Increasing the frequency of high demand bus and train services during AM or

PM peak hours is one way to improve LOS. For example, there is the need for more

frequent services from residential areas to working areas during the morning peak

period, and more frequent services in the opposite direction during PM peak hours.

Another approach is to identify and synchronise bus and train services which, based

on passenger data related to their past trips, are believed to be connected. Hence,

travellers would have a minimum waiting time between the two connecting services;

for example, between a feeder bus service from a residential area to the nearest train

station, and a feeder bus service from a train station to a working area during morning

peak hours.

From a different perspective, the improvement of perceived waiting time would

also improve passenger LOS. This perception can be achieved by improving the

condition of waiting areas in major train stations, and major bus stops and terminals.

Improvements could include the provision of entertainment systems, shelter, real-time

arrival/departure information, and printed service schedules.

7.9 CONCLUSIONS

Having undertaken a comprehensive literature review of policy intervention

theories that influence travel behaviours, this study found that no earlier study had

conducted a DCE or used a nested logit model to identify policy intervention targets

for influencing transport mode shift behaviour. To address these research gaps, this

study used DCE and travel behaviour data provided by urban travellers in the three

largest Australian capital cities (i.e., Sydney, Melbourne, and Brisbane). Specifically,

this study focused on determining the socio-economic factors and travel attributes that


significantly influence travellers’ mode choice in each city. Analyses of a sequence of

combined policy intervention scenarios then identified the most efficient combination

of policy interventions for urban travel.

The nested structure presented in Figure 4.1 provides a reasonable econometric

specification of the key factors influencing mode choice, and a means to avoid biased

estimates. It divides mode choice into two upper nest branches, public transport and

private transport. The public transport branch is further divided into two lower nest

branches, bus and train. Both bus and train are assumed to contain unobserved

attributes of public transport modes. Meanwhile, private transport is a degenerative

branch that contains only a single lower nest branch, car. This nested structure served

as a foundation for nested logit model estimation. Three best-fitted nested models were

estimated for the Sydney, Melbourne, and Brisbane dataset. Each of the utility

functions estimated within the nested logit model contains key travel attributes and

socio-economic factors. The travel attribute elements were useful in identifying the

specific area of policy intervention, while the socio-economic factor elements were

important to an understanding of the underlying profile of travellers who are affected

by a change in transport policies.

Based on the defined utility functions and the main objective of this study, two

policy interventions – each with a different purpose – were determined. These

intervention purposes are: to encourage public transport ridership; and to discourage

regular car usage. To obtain the optimal change in car mode share and public mode

share, both types of policy interventions were implemented at the same time. In order

to gain insights into the direct impacts of policy interventions on mode share (i.e., of

car and public transport), 176 different travel scenarios were simulated for Sydney,

and then replicated for the other two capital cities. By controlling all other factors, the

bus and train waiting time was reduced from the baseline scenario (i.e., 100%) to a

50% scenario, with a 5% interval; at the same time, the total parking and toll cost was

increased up to 2.5 times the baseline scenario, with a 10% interval.

While controlling all other factors, the overall scenario analysis suggests that the

total parking and toll cost needs to increase by 1.9, 1.8, and 2.1 times the baseline

scenario as well as the bus and train waiting time needs to decrease by 80%, 60%, and

75% of the baseline scenario to achieve a 10% reduction in car mode share in Sydney,

Melbourne, and Brisbane, respectively. When the most extreme travel scenarios are

applied (i.e., a 50% reduction in bus and train waiting time and a 2.5 times increase in


total parking and toll cost), the probability of driving a car in Sydney, Melbourne, and

Brisbane decreases from 64% to 47%, from 58% to 41%, and from 83% to 69%,

respectively. (While it is acknowledged that there is mode share competition between

public transport services [i.e., between bus and train services] in each city, this issue

is not the focus of this study.)

Overall, this study has recognised the continuing practical value of nested logits.

It has also highlighted an appealing empirical strategy, at the centre of which is a

carefully crafted DCE that contributes to the mode choice modelling environment data

that is capable of satisfying many of the statistical properties. This study has also

attained its objective to successfully address the existing research gaps. The

establishment of a novel method for the analysis of travellers’ behavioural changes

and policy interventions is a theoretical contribution to the field of SP experiment. The

successfully replicated nested model estimation and policy intervention scenario

analysis for three different urban traveller datasets also provide evidence that there are

many opportunities to replicate the overall framework of the policy intervention study

to influence mode shift behaviours in other urban traveller datasets.

The findings of this study provide useful knowledge for policy makers and

transport authorities. This knowledge will contribute to the formulation of future

transit policies that focus on mode shift from car to public transport for urban travellers

in any Australian or international capital city. The analysis of the simulated travel

scenarios demonstrates that combined policy interventions dedicated to each mode

choice can go a long way in influencing long-term behavioural change.

Although the nested logit mode is widely recognised as a way of building a more

parsimonious predictive method, where the real focus of the analysis is on the policy

value of the explanatory variables, it is not the most advanced choice of modelling

methods, and this is a limitation of this study. Another limitation is that the nested logit

model employed does not account for unobserved heterogeneity in the population.

The random-parameter nested logit model is considered as a new topic in the

field of statistics and econometrics, and will be useful in identifying random

parameters and their heterogeneities within the utility functions estimated in a nested

logit model. Another meaningful extension of this study will be a comprehensive cost-

benefit analysis of each possible travel scenario to determine which combined policy

interventions would be the most beneficial and economical in optimising behavioural

change and mode share shift from car to public transport.

Chapter 8: Conclusions 131

Chapter 8: Conclusions

This chapter consists of three sections: the first (Section 8.1) synthesises the

overall thesis, elaborates the linkages among the three interconnected sub-studies, and

highlights the findings; the second (Section 8.2) discusses its contributions to wider

theoretical and practical contexts as well as its policy implications; the third (Section

8.3) acknowledges the study’s limitations; and the last section (Section 8.4) suggests

directions for future study.

8.1 OVERARCHING CONCLUSIONS

The combined findings of the three interconnected sub-studies have successfully

addressed all the defined research questions, and achieved the target objectives. These

major findings and innovations are elaborated below.

The comprehensive investigation of train riders’ satisfaction with train fares and

a comparative analysis of train fares in five Australian cities (Sydney, Melbourne,

Brisbane, Adelaide, and Perth) indicate that the train fare structure in each city

significantly influences the train riders’ satisfaction levels. In particular, when other

factors are controlled, this study finds that Sydney, Melbourne, and Brisbane train

riders feel less satisfied with train fare than those in Adelaide. This difference is most

likely caused by the different train fare structures imposed by the transport authorities

in each city. Specifically, Sydney, Melbourne, and Brisbane apply zone-based fares,

whereas Adelaide applies fixed fares regardless of the distance travelled. The majority

of train rider respondents from all cities travelled for 15 KM one-way. Therefore,

Sydney, Melbourne, and Brisbane train riders paid $4.20, $3.90, and $5.96

(respectively) for this journey, while Adelaide respondents only paid $3.54; that is, for

the same travelled distance, Adelaide train riders pay the least amount and, hence, have

the highest satisfactions level of the four cities.

Satisfaction levels with train fares are also influenced by gender; eligibility for

a concession fare; transport mode from home to train station; waiting time; one-way

cost; and a number of interaction variables between city of origin and socio-economic

factors. Waiting time and one-way cost are also found to be significant random

parameters in the best-fitted random parameters ordered logit model. Urban travellers’

132 Chapter 8: Conclusions

views on both variables are significantly affected by their socio-economic profiles,

such as their employment status, household composition, and trip purpose.

Having successfully completed the traveller satisfaction studies utilising the

revealed travel behaviour dataset, a subsequent consistency study developed a novel

test to scrutinize the consistency between perceptions of various quality of service

related factors and their SP responses. The latter data were collected concurrently from

the same group of travellers in the five Australian capital cities. This study was a

crucial prior step to utilising and analysing the stated travel behaviours dataset to

understand travellers’ future mode choice behaviours.

In this novel consistency test, a statistical model was estimated for each type of

dataset: the random parameters binomial logit model (RP model) for the perceptions

dataset; and the mixed logit model (SP model) for the responses to the SP experiment.

The factor mapping from these two models shows that the perception rates of four

service factors – such as train waiting time, on-board crowding, the availability of

laptop stations, and increased road congestion - and their corresponding attributes in

the SP experiment were constructively aligned. Furthermore, the estimated

probabilities of choosing the train mode from the SP model are found to be similar to

the probabilities of choosing the train mode estimated from the RP model.

Having confirmed the consistency between the collected mode choice responses,

and understanding the influence of socio-economic profiles and trip characteristics on

urban travellers’ mode choice behaviours, the mode shift study then analysed the

travellers’ SP responses, their past trip experiences, and their socio-economic profiles.

This study aimed to determine the significant socio-economic factors and travel

attributes that influence travellers’ future mode choice, and the interventions that could

be used to encourage mode shift behaviour, especially from car to public transport.

Unlike the previous two studies, this study separately analysed the datasets from

Sydney, Melbourne, and Brisbane.

To suit the behaviour-shift objective of this study, a nested logit model based on

a particular nested structure, was estimated for each city’s dataset. The nested structure

divided the mode choice into two upper nest branches, public transport and private

transport. The public transport branch was further divided into two lower nest

branches, bus and train. Both bus and train were assumed to contain unobserved

elements of the public transport mode. Private transport was a degenerative branch that

contained a single lower nest branch only, namely, car.


According to the modelling results, two policy interventions – each with a

different purpose – were determined. The two purposes were: to encourage public

transport ridership; and to discourage regular car usage. To obtain the optimal change

from car mode share to public mode share, both types of policy intervention were

implemented at the same time.

In order to gain insights into the direct impacts of these interventions on car and

public transport mode shares, 176 different travel scenarios were simulated for

Sydney, and replicated for the other two capital cities. For instance, while controlling

all other factors, the bus and train waiting times were reduced from the baseline

scenario (i.e., 100%) to a 50% scenario with a 5% interval, and the total parking and

toll cost was increased up to 2.5 times the baseline scenario with a 10% interval.

The overall scenario analysis suggests that by increasing twice the total parking

and toll cost, controlling the public transport waiting time at a baseline scenario, and

keeping other factors constant, the car mode share decreases from 64% to 54% for the

Sydney dataset. A further increase to 250% in total parking and toll cost, while holding

all other factors constant, further decreases the car mode share to 49%.

Overall, it can be inferred that combined policy interventions to increase the cost

of travelling by car would certainly encourage car drivers to leave their cars at home

and take public transport services, especially if their trips are on a regular basis. The

increment in parking cost can be implemented together with a decrement in on-street

parking spots in higher density areas. Tolls for expressways connecting various

suburbs with the city centre or areas of major attractions, or a congestion tax for private

vehicles entering areas of high traffic density during peak hours can reduce traffic

pressure in such areas.

Ideally, the additional funding obtained from increasing car travel costs can be

concurrently utilised to encourage public transport ridership. The concurrent timing of

both interventions is vital to maintaining and improving the attractiveness of public

transport services for both existing, and mode-shift users. For example, in anticipation

of an increment in demand for public transport services, the supply of public transport

services needs to be adjusted accordingly (for example, increasing the frequency of

public transport services during peak hours). Hence, the passengers’ level of service

across all public transport services can be maintained and improved over time.


8.2 CONTRIBUTIONS AND POLICY IMPLICATIONS

Each of the sub-studies has successfully achieved its target objectives and

provides significant contributions. These achievements and policy implications are

detailed below.

8.2.1 Urban travellers’ satisfaction with train fares in five Australian cities

In a wider context, the overall satisfaction modelling framework can be

replicated to understand and to quantify the diverse factors influencing satisfaction

with the whole journey experience. The findings of the satisfaction study constitute

significant knowledge for both policy makers and transport operators. Specifically,

this knowledge provides them with a comprehensive understanding of traveller

behaviours. This understanding, in turn, can guide their formulation of effective transit

policies to increase train rider satisfaction with the paid fare and, eventually, with the

overall journey experience.

8.2.2 Consistency between perceptions and stated preferences data in a

nationwide mode choice experiment

Specifically, the consistency assessment study demonstrated that the views of

diverse travellers on train waiting time, on-board crowding, the availability of laptop

stations, and increased road congestion were reasonably consistent across the

perceptions data and the SP experiment data. Our analysis confirms that these two

types of data source (i.e., the perceptions and the SP responses) are complementary in

helping us to better understand travellers’ complex mode choice behaviour.

In order to amplify the benefits of the consistency test, the overall modelling

framework could be implemented in other types of behavioural research, where both

revealed preference and stated preference data are concurrently collected from groups

of diverse respondents. For example, in the health field, research into people’s attitudes

to vaccination would be a prime candidate for such a framework.


8.2.3 Policy interventions study to encourage behavioural shift from car to

public transport

This policy intervention mode shift study has recognised the continuing practical

value of nested logits, and has established a novel method for the analysis of travellers’

behavioural changes. The successfully replicated nested model estimation and policy

interventions scenario analysis in three different urban traveller datasets provide

evidence that the overall framework of the policy intervention study could be used to

influence mode shift behaviours in any other urban traveller dataset.

The findings of this study also provide useful knowledge for policy makers and

transport authorities to contribute to the formulation of future transit policies that focus

on encouraging mode shift from car to public transport services for urban travellers in

any urban setting. The analysis of the simulated travel scenarios demonstrates that the

combined policy interventions can significantly influence behavioural change. At the

same time, these interventions should comply with current social norms, regulations,

and laws, and the existing policies.

8.3 LIMITATIONS OF THIS STUDY

The following limitations of the overall study are acknowledged:

The complex impact on the data analysis caused by the fact that 15-20% of

respondents (from each city dataset) did not report on their income.

The reported on-board crowding could be inconsistent with the actual crowding,

and this could cause a potential confounding effect. Future studies could

incorporate some objective measure of crowding (e.g., the number of passengers on

a particular service) in the urban travellers’ behavioural study.

The access to real-time train service information was not considered in the SP

Experiment. This made it impossible to assess the consistency between

respondents’ perceptions and their corresponding SP responses on this important

attribute.

Previous studies noted the importance of weather, security, safety, and the health

and psychological condition of respondents to an understanding of the underlying

heterogeneity of key travel attributes in the SP Experiment (such as waiting time

and on-board crowding). Unfortunately, such information was not collected in the

survey dataset used in this study.


8.4 RECOMMENDATIONS FOR FUTURE STUDY

The following recommendations are offered for future studies:

The random-parameter nested logit model is considered as a new topic in the field

of statistics and econometrics. It will be useful in identifying random parameters

and their heterogeneities within the utility functions estimated in a nested logit

model.

A comprehensive cost benefit analysis of each possible travel scenario to strongly

quantify which combined policy interventions would be the most beneficial and

economical to implement (that is, in order to optimise the behavioural and mode

share shift from car to public transport services).

[Extra page

[Extra page inserted to ensure correct even-page footer for this section. Delete

this when chapter is at least 2 pages long.]

Appendices 137

Appendices

Appendix A

Urban Rail Travel Behaviour, Web-based Survey

This appendix is a soft-copy of the revised questionnaire, dated 19th November 2012,

that was distributed to the respondents. This questionnaire was developed by a project

team as part of Project R1.130 Understanding Urban Rail Travel for Improved

Patronage Forecasting funded by the CRC for Rail Innovation (established and

supported under the Australian Government's Cooperative Research Centres program).

This questionnaire has been approved by QUT Office of Research Ethics and Integrity

and I have been officially granted permission to utilise the dataset for my PhD study.


this when bibliography is at least 2 pages long.]

Urban Rail Travel Behaviour Web-based Survey

Revised Questionnaire – November 19, 2012

Note:

• The survey can be customized for particular cities, with a core of comparable questions; • The estimated workload for train riders: 15 to 18 minutes; • The estimated workload for non train rider: 13 to 16 minutes;

SCREENING & QUOTA SELECTION (Estimated Workload: 1 minute)

A research team from Southern Cross University, Queensland University of Technology and The University of South Australia are conducting a study for the Commonwealth Cooperative Research Centre for Rail Innovation to more fully understand the travel patterns and travel choices of people like you. Your experiences, opinions, and travel behaviour are extremely important in order for us to deliver improved future train services. The survey will take you approximately 15 - 20 minutes to complete. The study is being conducted for research purposes only, and no attempt will be made to sell you anything at any time. Your participation is entirely voluntary. All comments and responses are anonymous and will be treated confidentially. Data from the survey will be saved on secure servers and de-identified by ORU before transmission to the research team. The data held by ORU will be destroyed at the conclusion of the research project. The de-identified data will be retained for use in further research. Research findings flowing from the survey will be incorporated in a report to the Commonwealth Cooperative Research Centre for Rail Innovation, which will normally make the report or core outcomes available on its publications website page http://www.railcrc.net.au/publications. If you have any questions or require any further information about this survey please contact Adjunct Associate Professor Keith Sloan Associate or Professor Michael Charles of Southern Cross University at 02 662 000 or by email via the link given here …………. This survey has Southern Cross University ethics approval number ECN-12-307. If you have concerns about the ethical conduct of this research or the researchers you may contact: The Ethics Complaints Officer Southern Cross University PO Box 157 Lismore NSW 2480 Email: [email protected]

SQ1 How old are you? (1) Younger than 16 years (Survey Close) (2) 16-17 years (3) 18-30 years (4) 31-40 years (5) 41-50 years (6) 51-60 years (7) 60+ years

SQ2 Do you or does anyone in your household work for a motor vehicle manufacturer, a public transport

provider, city rail company, or the city transport department? (1) Yes (Survey Close) (2) No

SQ3 How often did you travel by train LAST MONTH?

(1) Eight times or more; (Go to TR1) (2) Four times or more, but less than eight times; (Go to TR1) (3) Two times or more, but less than four times; (Go to TR1) (4) Once; (Go to NR1) (5) None; (Go to NR1)

Assign respondent to REGULAR RIDER if the answer to SQ3 is 1 or2, and assign respondent to INFREQUENT RIDER if the answer to SQ3 is 3; otherwise assign respondent to NONRIDER. For REGULAR RIDER and INFREQUENT RIDER, go to TR1; For NONRIDER, go to NR1. IF RELEVANT QUOTAS ARE FULL, CLOSE THE SURVEY.

TRAIN RIDER EXPERIENCE

(Estimated Workload: 6-7 minutes) The questions in this section are going to ask you to recall the details of your most recent TRAIN trip from home.

TR1 Besides train, which modes of transport were available to you when you made your most recent train trip

from home (even if you never use these methods)? (Please check all that apply) (1) Motor vehicle (financed, leased, or owned) (2) Company/work vehicle (3) Taxi (4) Motor Bike or Scooter (5) Bus (6) Tram (7) Cycling (8) Walking

TR2 Please state the nearest intersection to your home (please write down the postcode and two street names): Postcode: _______ Street 1: _______ Street 2: _______

TR3 Please state the nearest intersection to your destination of your most recent TRAIN trip from home (please write down postcode and two street names): Postcode: _______ Street 1: _______ Street 2: _______

TR4 What time of day did you commence your most recent TRAIN trip from home?

(1) Before 7am (2) Between 7:01 am and 9 am (3) Between 9:01 am and 3 pm (4) Between 3:01 pm and 7 pm (5) Between 7:01 pm and 2 am (6) Not sure

TR5 What was the main purpose of this most recent TRAIN trip from home?

(1) Employment (2) Business (3) Education (4) Leisure (holiday)/Recreation/Social/Volunteer (5) Shopping (6) Personal activity (e.g., picking up kids, medical appointment, errand, banking, etc.) (7) Other (Please indicate) __________________

TR6 Before you departed on your most recent TRAIN trip from home, did you check the most current information on the train services, such as arrival time of the next train or service updates on any delays or cancellations, etc.? (1) Yes (2) No

TR7 How did you get to the station on your most recent TRAIN trip from home?

(1) Bus (Go to TR7a) (2) Walked (Go to TR8) (3) Parked (free) at the nearest station (Go to TR7b) (4) Paid for parking at the nearest station (Go to TR7b) (5) Dropped off at the station, including taxi (Go to TR8) (6) Bicycle (Go to TR8) (7) Other (please indicate)_______________________ (Go to TR8)

TR7a: How long did you have to wait for the bus?

_________hours and _________ minutes Go to TR8

TR7b: If the parking facility was not available when you took this trip, you would have:

(1) driven to the destination (2) driven to another station (3) taken bus to the station (4) taken bus to the destination (5) walked to the station (6) cycled to the station (7) asked somebody for a ride to the station (8) parked elsewhere and then taken the train (9) parked elsewhere and then taken the bus (10) walked to the destination (11) cycled to the destination (12) asked somebody for a ride to the destination (13) cancelled the trip (14) Other (please indicate) _____________________

TR8 About how long does it take you to get from your home to the nearest train station?

_________hours and _________ minutes

TR9 On your most recent train trip about how long did you wait for the train? _________hours and _________ minutes

TR10 About how long were you on the train (time between getting on and off from the train including switching trains if applicable)? _________hours and _________ minutes

TR11 Which statement best describes the crowding condition during your most recent TRAIN trip from home? (1) It was easy to find a seat for the entire journey (2) I found a seat for the entire journey but most seats were occupied (3) I was standing up for 5 minutes prior to finding a vacant seat (4) I was standing up for 6 - 15 minutes prior to finding a vacant seat (5) I was standing up for 16 - 25 minutes or more prior to finding a vacant seat (6) I stood for the entire journey, as no seats were available (7) I stood for the entire journey, as I didn’t want to sit (8) Other (please indicate) ___________________

TR12 How did you spend your time on the train during your most recent TRAIN trip from home?

(1) Work or study related activities (e.g., writing) (2) Reading (e.g., news, book, magazine, etc) (3) Entertaining myself using personal digital devices (e.g., smart phone, tablet, iPod, iPad, etc) (4) Chatting with other travelling companion (e.g., chatting with friends, minding kids) (5) Relaxing or doing nothing in particular (e.g., sitting, looking out the window at local scenery) (6) Other (please indicate) ___________________

TR13 Once you got off the train on your most recent TRAIN trip from home, how did you get to your final

destination? (1) Bus (Go to TR13a) (2) Tram (Go to TR14) (3) Walked (Go to TR14) (4) Passenger pick up, including taxi (Go to TR14) (5) Bicycle (Go to TR14) (6) Other (please indicate) ____________________ (Go to TR14)

TR13a: How long did you wait for bus?

_________hours and _________ minutes

TR14 About how long did it take you to get to the destination? _________hours and _________ minutes

TR15 How much did you pay for tickets for this ONE-WAY trip?

(1) _________ dollars and ______cents (2) I don’t know.

TR16 Thinking about the services you received, on a 5-point rating scale how satisfied were you with the PRICE

you paid for your most recent TRAIN trip from home? 1 = “Extremely dissatisfied”; 5 = “Extremely satisfied”.

¨ ¨ ¨ ¨ ¨

Extremely dissatisfied Dissatisfied Neutral Satisfied Extremely satisfied

1 2 3 4 5

TR17 All things considered, on a 5-point rating scale how satisfied were you with the experience of your most recent TRAIN trip from home? 1 = “Extremely dissatisfied”; 5 = “Extremely satisfied”.

¨ ¨ ¨ ¨ ¨

Extremely dissatisfied Dissatisfied Neutral Satisfied Extremely satisfied

1 2 3 4 5

BUS/MOTOR VEHICLE EXPERIENCE (Estimated Workload: 4-5 minutes)

The questions in this section are going to ask you to recall the details of your most recent trip from home that was taken by motor vehicle or by bus.

NR1 For your most recent trip from home that was taken by motor vehicle or by bus, which methods of transport listed below were available to you when you were planning this trip (even if you never use these methods)? (Please check all that apply) (1) Motor vehicle (financed, leased, or owned) (2) Company/work vehicle (3) Taxi (4) Motor Bike or Scooter (5) Bus (6) Tram (7) Train (8) Cycling (9) Walking

NR2 Please state the nearest intersection to your home (please write down the postcode and two street names): Postcode: _______ Street 1: _______ Street 2: _______

NR3 For your most recent trip from home that was taken by motor vehicle or by bus , please state the nearest intersection to your destination (please write down the postcode and two street names): Postcode: _______ Street 1: _______ Street 2: _______

NR4 For your most recent trip from home that was taken by motor vehicle or by bus, what time of day did you depart? (1) Before 7am (2) Between 7:01 am and 9 am (3) Between 9:01 am and 3 pm (4) Between 3:01 pm and 7 pm (5) Between 7:01 pm and 2 am (6) Not sure

NR5 For your most recent trip from home that was taken by motor vehicle or by bus , what was its main purpose? (1) Employment (2) Business (3) Education (4) Leisure (holiday)/Recreation/Social/Volunteer (5) Shopping (6) Personal activity (e.g., picking up kids, medical appointment, errand, banking, etc.) (7) Other (Please indicate) __________________

NR6 Approximately what was your in-vehicle travel time for this trip? _________hours and _________ minutes

NR7 For your most recent trip from home that was taken by motor vehicle or by bus, what transport mode did you

use? (1) Motor vehicle (financed, leased, or owned) (Go to NR7a-b) (2) Passenger in a motor vehicle driven by someone else (Go to NR7a-b) (3) Company/work vehicle (Go to NR7a-b) (4) Taxi(Go to NR7c) (5) Bus (Go to NR7d-g)

NR7a Approximately how much did you spend on parking for this trip? If you didn’t pay for parking, enter ‘0’.

_________ dollars and ______cents

NR7b Approximately how much did you spend on tolls for this trip? If you didn’t pay for toll, enter ‘0’.

Go to NR8

NR7c How much did you pay for this ONE-WAY trip?

_________ dollars and ______cents

NR7d How much did you pay for tickets for this ONE-WAY trip? _________ dollars and ______cents

NR7e About how long did it take you to walk to the bus stop from your home? If you did NOT walk to the bus stop, please estimate the would-be walking time.

_________ hours and _________ minutes

NR7f About how long did you wait for the bus? _________ hours and _________ minutes

NR7g Which statement best describes the crowding condition on the bus during your most recent trip? (1) It was easy to find a seat for the entire journey (2) I found a seat for the entire journey but most seats were occupied (3) I was standing up for 5 minutes prior to finding a vacant seat (4) I was standing up for 6 - 15 minutes prior to finding a vacant seat (5) I was standing up for 16 - 25 minutes or more prior to finding a vacant seat (6) I stood for the entire journey, as no seats were available (7) I stood for the entire journey, as I didn’t want to sit (8) Other (please indicate) ___________________

NR8 Why did you choose this transport mode? (1) No other mode was available (2) Cheapest mode available (3) Fastest mode available (4) Most convenient mode available (5) Safest mode available (6) Most comfortable mode available (7) Work-related vehicle (8) Personal mobility constraints (e.g., disabled) (9) Needed to transport bulky items (10) Weather conditions (11) Other (please indicate)________________________

NR9 What are the main reasons that you did not take the train for this trip? (1) I like driving (2) Driving was faster (3) Work-related vehicle (4) No available train service (5) Personal mobility constraints (e.g., disabled) (6) Needed to transport bulky items (7) I was offered a free ride (e.g., sitting in a car driven by another person) (8) I prefer bus/ferry/tram (9) The nearest train station was too far (10) The trip was too short for taking train (11) The train is not sufficiently safe (12) The train does not run frequently enough (13) The train is not convenient for visiting multiple destinations (14) The train is generally for people who don’t have access to motor vehicles (15) The train is too crowded (16) The train fare is too expensive (17) Other (please indicate) ________________________

NR10 After getting out of the motor vehicle or getting off the bus, about how long did it take you to reach your final destination? _________ hours and _________ minutes

GENERAL QUESTIONS

(Estimated Workload: 4-5 minutes) The questions in this section are designed to collect some general information regarding your travel behaviour and your socio demographic background. These questions have NOTHING to do with your most recent trip.

GQ1 On a 5-point rating scale, to what extent would the following factors influence you to take a train more often? 1 = “no influence at all”; 5 = “very strong influence”. (1) Better access (e.g., more bus services, more comfortable walking environment) to the station by bus

(2) Train runs on schedule

(3) Higher likelihood of getting a seat / less crowding

(4) The ability to access up to date information on train services such as current train status

(5) Availability of an entertainment system (e.g., video, audio, TV etc. )

(6) Increased road congestion

(7) Private vehicle drivers are charged a congestion tax or toll every time they enter the city in peak hours

GQ2 To what extent did access to rail services influence your decision to live in your current location? (1) Very significant (2) Significant (3) Moderate (4) Low (5) Was not a consideration

GQ3 What is your motor vehicle access status? (1) I don’t have access to a motor vehicle (2) I have access to my own motor vehicle (3) I have access to company/work vehicles (4) I have access to a shared motor vehicle

GQ4 Do you have a driver’s license?

(1) Yes (2) No

GQ5 Do you need a motor vehicle for your work or business?

(1) Yes (2) No (3) I do not work

GQ6 Your current employment status

(1) Full-time (paid employment) (2) Part-time (paid employment) (3) Self-employed (4) Not in the work force (e.g., maternity leave, unemployed) (5) Retired (6) Student (1) Other (please indicate)__________________________

GQ7 Which of the following best describes your current household? (1) Couple family with dependent children (2) One parent family with dependent children (3) Couple only (4) Multiple family household (5) Lone person (6) Group household (7) Other one family household (8) Other (please indicate)__________________________

GQ8 In which country were you born? (1) Australia (2) England (3) New Zealand (4) Italy (5) Vietnam (6) Scotland (7) Greece (8) China (9) India (10) Philippines (11) South Africa (12) Germany (13) Malaysia (14) Other (please indicate) __________________________

GQ9 What is your gender?

(15) Male (16) Female

GQ10 Into which of these groups does your pre-tax household weekly income fall?

(1) Under $200 (2) $200 to less than $300 (3) $300 to less than $400 (4) $400 to less than $500 (5) $500 to less than $600 (6) $600 to less than $700 (7) $700 to less than $800 (8) $800 to less than $900 (9) $900 to less than $1,000 (10) $1000 to less than $1,100 (11) $1,100 to less than $1,200 (12) $1,200 to less than $1,300 (13) $1,300 to less than $1,400 (14) $1,400 to less than $1,500 (15) $1,500 to less than $1,600 (16) $1,600 to less than $1,700 (17) $1,700 to less than $1800 (18) $1800 or more (19) Did not draw a wage or salary

GQ11 What is your highest educational qualification?

(1) Postgraduate Degree Level (2) Graduate Diploma and Graduate Certificate Level (3) Bachelor Degree Level (4) Advanced Diploma and Diploma Level (5) Certificate Level (6) School Education Level

HOW WOULD YOU TRAVEL? (Estimated Workload: 4-5 minutes)

Based on the information you have provided, we understand that your most recent trip from home by [train/bus/motor vehicle] started from [TR2/NR2] and ended at [TR3/NR3]. For this trip, the available transport modes were [TR1/NR1]. As you are probably aware, changes that are out of your control can take place, including availabilities of different transport options. In this section, you will be asked to participate in 6 HYPOTHETICAL experiments. In each of these experiments, a number of transport choices with different features are described. If these options were available to you with the features listed when you were planning your most recent trip described above, please indicate which travel option you would choose. Due to the large amount, it is impossible to attach all the experiments here. Instead, for illustration purpose, 6 experiments from the same block for {train, bus} are provided below.

Mode choice experiment 1:

Bibliography 139

Bibliography

Aarts, H., Verplanken, B., & Knippenberg, A. (1998). Predicting behavior from

actions in the past: Repeated decision making or a matter of habit? Journal of

Applied Social Psychology, 28(15), 1355-1374.

Accent. (2009). Perception towards integrated transport: literature review .London:

Passenger Focus.

Adamowicz, W., Louviere, J., & Williams, M. (1994). Combining revealed and stated

preference methods for valuing environmental amenities. Journal of

Environmental Economics and Management, 26(3), 271-292.

Agresti, A. (2010). Analysis of ordinal categorical data (2nd ed.). Florida: John Wiley

& Sons.

Agresti, A., & Kateri, M. (2011). Categorical data analysis. International

Encyclopedia of Statistical Science: Berlin: Springer.

Ajzen, I. (1985). From intentions to actions: A theory of planned behavior Action

control (pp. 11-39). Berlin Heidelberg: Springer.

Ajzen, I. (1991). The theory of planned behavior. Organizational behavior and human

decision processes, 50(2), 179-211.

Akaike, H. (1992). Information theory and an extension of the maximum likelihood

principle Breakthroughs in statistics (pp. 610-624). New York: Springer.

Akaike, H. (1998). Information theory and an extension of the maximum likelihood

principle. In Selected Papers of Hirotugu Akaike (pp. 199-213). New York,

NY: Springer.

Albright, J. (2016). What is the difference between logit and probit models. Retrieved

from https://www.methodsconsultants.com/tutorial/what-is-the-difference-

between-logit-and-probit-models/

Alm, J., Bahl, R., & Murray, M. N. (1993). Audit selection and income tax

underreporting in the tax compliance game. Journal of development

Economics, 42(1), 1-33.

Andersson, D., & Nässén, J. (2016). The Gothenburg congestion charge scheme: A

pre–post analysis of commuting behavior and travel satisfaction. Journal of

Transport Geography, 52, 82-89.

Arrow, K., Solow, R., Portney, P. R., Leamer, E. E., Radner, R., & Schuman, H.

(1993). Report of the NOAA panel on contingent valuation. Federal register,

58(10), 4601-4614.

Azevedo, C. D., Herriges, J. A., & Kling, C. L. (2003). Combining revealed and stated

preferences: consistency tests and their interpretations. American Journal of

Agricultural Economics, 85(3), 525-537.

Bamberg, S., Ajzen, I., & Schmidt, P. (2003). Choice of travel mode in the theory of

planned behavior: The roles of past behavior, habit, and reasoned action. Basic

and applied social psychology, 25(3), 175-187.

Bamberg, S., & Schmidt, P. (1998). Changing travel-mode choice as rational choice:

Results from a longitudinal intervention study. Rationality and Society, 10(2),

223-252.

Bamberg, S., & Schmidt, P. (2001). Theory‐driven subgroup‐specific evaluation of an

intervention to reduce private car use. Journal of Applied Social Psychology,

31(6), 1300-1329.

https://www.methodsconsultants.com/tutorial/what-is-the-difference-between-logit-and-probit-models/

https://www.methodsconsultants.com/tutorial/what-is-the-difference-between-logit-and-probit-models/

140 Bibliography

Bates, J., Polak, J., Jones, P., & Cook, A. (2001). The valuation of reliability for

personal travel. Transportation Research Part E: Logistics and Transportation

Review, 37(2), 191-229.

Bates, J. J. (1983). Stated preference techniques for the analysis of transport behavior.

Paper presented at the World Conference on Transport Research: Research for

Transport Policies in a Changing World.

Bates, J. J. (1984). Values of time from stated preference data. Planning and Transport

Resesarch and Computation.

Bates, J. J., & Roberts, M. (1986). Value of time research: summary of methodology

and findings. Paper presented at the 14th PTRC Summer Annual Meeting.

Batty, P., Palacin, R., & González-Gil, A. (2015). Challenges and opportunities in

developing urban modal shift. Travel Behaviour and Society, 2(2), 109-123.

doi:http://dx.doi.org/10.1016/j.tbs.2014.12.001 .

Bech, M., Kjaer, T., & Lauridsen, J. (2011). Does the number of choice sets matter?

Results from a web survey applying a discrete choice experiment. Health

economics, 20(3), 273-286.

Beirão, G., & Cabral, J. S. (2007). Understanding attitudes towards public transport

and private car: A qualitative study. Transport Policy, 14(6), 478-489.

Benjamin, J., & Sen, L. (1982). Comparison of the predictive ability of four

multiattribute approaches to attitudinal measurement. Transportation

Research Record(890).

Bhat, C. R. (2003). Simulation estimation of mixed discrete choice models using

randomized and scrambled Halton sequences. Transportation Research Part

B: Methodological, 37(9), 837-855.

Bhat, C. R., & Sardesai, R. (2006). The impact of stop-making and travel time

reliability on commute mode choice. Transportation Research Part B:

Methodological, 40(9), 709-730.

Bisantz, A. M., Roth, E., Brickman, B., Gosbee, L. L., Hettinger, L., & McKinney, J.

(2003). Integrating cognitive analyses in a large-scale system design process.

International Journal of Human-Computer Studies, 58(2), 177-206.

Blainey, S., Hickford, A., & Preston, J. M. (2009). Barriers to modal shift: literature

review. . Retrieved from Transportation Research Group, University of

Southampton, UK:

Bliemer, M. C., & Rose, J. M. (2006). Designing stated choice experiments: state of

the art. ETH. Zurich.

Blumenschein, K., Johannesson, M., Yokoyama, K., & Freeman, P. (2001).

Hypothetical versus real willingness to pay in the health care sector: results

from a field experiment. Value in Health, 4(2), 79-79.

Boardman, A. E., Greenberg, D. H., Vining, A. R., & Weimer, D. L. (2017). Cost-

benefit analysis: concepts and practice: Cambridge University Press.

Bordagaray, M., dell'Olio, L., Ibeas, A., & Cecín, P. (2014). Modelling user perception

of bus transit quality considering user and service heterogeneity.

Transportmetrica A: Transport Science, 10(8), 705-721.

Borins, S. F. (1988). Electronic road pricing: an idea whose time may never come.

Transportation Research Part A: Policy and Practice, 22(1), 37-44.

Boumans, M., & Morgan, M. S. (2001). Ceteris paribus conditions: materiality and the

application of economic theories. Journal of Economic Methodology, 8(1), 11-

26.

Brazil, W., Caulfield, B., & Bhat, C. R. (2017). The potential role of eye tracking in

stated preference survey design and piloting. Trinity College. Dublin, Ireland.

http://dx.doi.org/10.1016/j.tbs.2014.12.001

Bibliography 141

Brons, M., Givoni, M., & Rietveld, P. (2009). Access to railway stations and its

potential in increasing rail use. Transportation Research Part A: Policy and

Practice, 43(2), 136-149.

Buys, L., & Miller, E. (2011). Conceptualising convenience: Transportation practices

and perceptions of inner-urban high density residents in Brisbane, Australia.

Transport Policy, 18(1), 289-297.

Cameron, T. A. (1992). Combining contingent valuation and travel cost data for the

valuation of nonmarket goods. Land Economics, 302-317.

Cantwell, M., Caulfield, B., & O’Mahony, M. (2009). Examining the factors that

impact public transport commuting satisfaction. Journal of Public

Transportation, 12(2), 1.

Carson, R. T., Flores, N. E., Martin, K. M., & Wright, J. L. (1996). Contingent

valuation and revealed preference methodologies: comparing the estimates for

quasi-public goods. Land Economics, 80-99.

Carson, R. T., & Groves, T. (2007). Incentive and informational properties of

preference questions. Environmental and Resource Economics, 37(1), 181-

210.

Cascajo, R., Garcia-Martinez, A., & Monzon, A. (2017). Stated preference survey for

estimating passenger transfer penalties: design and application to Madrid.

European Transport Research Review, 9(3), 42.

Caulfield, B., & O'Mahony, M. (2007). An examination of the public transport

information requirements of users. IEEE transactions on intelligent

transportation systems, 8(1), 21-30.

Caussade, S., de Dios Ortúzar, J., Rizzi, L. I., & Hensher, D. A. (2005). Assessing the

influence of design dimensions on stated choice experiment estimates.

Transportation Research Part B: Methodological, 39(7), 621-640.

Chaminade, C., & Edquist, C. (2010). Rationales for public policy intervention in the

innovation process: A systems of innovation approach. The theory and practice

of innovation policy. An international research handbook, 95-114.

Chatterjee, A., Wegmann, F. J., & McAdams, M. A. (1983). Non-commitment bias in

public opinion on transit usage. Transportation, 11(4), 347-360.

Chen, C. F. (2008). Investigating structural relationships between service quality,

perceived value, satisfaction, and behavioral intentions for air passengers:

Evidence from Taiwan. Transportation Research Part A: Policy and Practice,

42(4), 709-717.

ChoiceMetrics. (2012). Ngene 1.1. 1 User Manual & Reference Guide. Retrieved

from http://www.choice-metrics.com/index.html

Clarke, P. M. (2002). Testing the convergent validity of the contingent valuation and

travel cost methods in valuing the benefits of health care. Health Economics,

11(2), 117-127.

Corpus, G. (2008). Public transport or private vehicle: Factors that impact on mode

choice. Retrieved from

Couture, M. R., & Dooley, T. (1981). Analyzing traveler attitudes to resolve intended

and actual use of a new transit service. Transportation Research Record(794).

Cox, T., Houdmont, J., & Griffiths, A. (2006). Rail passenger crowding, stress, health

and safety in Britain. Transportation Research Part A: Policy and Practice,

40(3), 244-258.

Cummings, R. G., Brookshire, D. S., Bishop, R. C., & Arrow, K. J. (1986). Valuing

environmental goods: an assessment of the contingent valuation method.

Totowa,NJ: Rowman Allanheld.

http://www.choice-metrics.com/index.html

142 Bibliography

Cummings, R. G., Elliott, S., Harrison, G. W., & Murphy, J. (1997). Are hypothetical

referenda incentive compatible?. Journal of Political Economy, 105(3), 609-

621.

Cummings, R. G., & Taylor, L. O. (1999). Unbiased value estimates for environmental

goods: a cheap talk design for the contingent valuation method. The American

Economic Review, 89(3), 649-665.

Daniels, R., & Mulley, C. (2013). Explaining walking distance to public transport: The

dominance of public transport supply. Journal of Transport and Land Use,

6(2), 5-20.

de Bekker‐Grob, E. W., Ryan, M., & Gerard, K. (2012). Discrete choice experiments

in health economics: a review of the literature. Health Economics, 21(2), 145-

172.

de Oña, J., & de Oña, R. (2014). Quality of service in public transport based on

customer satisfaction surveys: A review and assessment of methodological

approaches. Transportation Science, 49(3), 605-622.

de Oña, J., de Oña, R., Eboli, L., & Mazzulla, G. (2013). Perceived service quality in

bus transit service: a structural equation approach. Transport Policy, 29, 219-

226.

de Oña, R., Machado, J. L., & de Oña, J. (2015). Perceived service quality, customer

satisfaction, and behavioral intentions: structural equation model for the Metro

of Seville, Spain. Transportation Research Record: Journal of the

Transportation Research Board(2538), 76-85.

Dekker, T., Hess, S., Arentze, T., & Chorus, C. (2014). Incorporating needs-

satisfaction in a discrete choice model of leisure activities. Journal of


Delclòs-Alió, X., Marquet, O., & Miralles-Guasch, C. (2017). Keeping track of time:

A Smartphone-based analysis of travel time perception in a suburban

environment. Travel Behaviour and Society, 9, 1-9.

Dell’Olio, L., Ibeas, A., & Cecin, P. (2011). The quality of service desired by public

transport users. Transport Policy, 18(1), 217-227.

Dell’Olio, L., Ibeas, A., & Cecín, P. (2010). Modelling user perception of bus transit

quality. Transport Policy, 17(6), 388-397.

Delmelle, E. C., Haslauer, E., & Prinz, T. (2013). Social satisfaction, commuting and

neighborhoods. Journal of Transport Geography, 30, 110-116.

Derek Halden Consultancy. (2003). Barriers to modal shift. Edinburgh: Scottish

Executive, Social Research.

DeShazo, J., & Fermo, G. (2002). Designing choice sets for stated preference methods:

the effects of complexity on choice consistency. Journal of Environmental

Economics and Management, 44(1), 123-143.

Desvousges, W. H., Johnson, F. R., Dunford, R. W., Boyle, K. J., Hudson, S. P., &

Wilson, K. N. (1992). Measuring nonuse damages using contingent valuation:

An experimental evaluation of accuracy (2nd Ed.). NC,USA: Research

Triangle Institute Research Triangle Park.

Dieleman, F. M., Dijst, M., & Burghouwt, G. (2002). Urban form and travel behaviour:

micro-level household attributes and residential context. Urban studies, 39(3),

507-527.

DiLalla, D. L., & Dollinger, S. J. (2006). Cleaning up data and running preliminary

analyses. The psychology research handbook: A guide for graduate students

and research assistants, 2, 241-254.

Bibliography 143

Ding, M. (2007). An incentive-aligned mechanism for conjoint analysis. Journal of

Marketing Research, 44(2), 214-223.

Dwyer, D. H. (1983). Women and income in the Third World: implications for policy.

International Programs Working Paper No. 18. Population Council. New York.

Dziekan, K., & Kottenhoff, K. (2007). Dynamic at-stop real-time information displays

for public transport: effects on customers. Transportation Research Part A:

Policy and Practice, 41(6), 489-501.

Eboli, L., & Mazzulla, G. (2012). Performance indicators for an objective measure of

public transport service quality.

Efthymiou, D., & Antoniou, C. (2017). Understanding the effects of economic crisis

on public transport users’ satisfaction and demand. Transport Policy, 53, 89-

97.

Efthymiou, D., Antoniou, C., Tyrinopoulos, Y., & Skaltsogianni, E. (2017). Factors

affecting bus users’ satisfaction in times of economic crisis. Transportation

Research Part A: Policy and Practice.

Ellaway, A., Macintyre, S., Hiscock, R., & Kearns, A. (2003). In the driving seat:

psychosocial benefits from private motor vehicle transport compared to public

transport. Transportation Research Part F: Traffic Psychology and Behaviour,

6(3), 217-231.

Eriksson, L., Garvill, J., & Nordlund, A. M. (2008). Interrupting habitual car use: The

importance of car habit strength and moral motivation for personal car use

reduction. Transportation Research Part F: Traffic Psychology and

Behaviour, 11(1), 10-23.

Ettema, D., Friman, M., Gärling, T., Olsson, L. E., & Fujii, S. (2012). How in-vehicle

activities affect work commuters’ satisfaction with public transport. Journal of


Evans, G. W., & Wener, R. E. (2007). Crowding and personal space invasion on the

train: Please don’t make me sit in the middle. Journal of Environmental

Psychology, 27(1), 90-94.

Fan, Y., Guthrie, A., & Levinson, D. (2016). Waiting time perceptions at transit stops

and stations: Effects of basic amenities, gender, and security. Transportation

Research Part A: Policy and Practice, 88, 251-264.

Farag, S., & Lyons, G. (2012). To use or not to use? An empirical study of pre-trip

public transport information for business and leisure trips and comparison with

car travel. Transport Policy, 20, 82-92.

Feeney, B. P. (1989). A review of the impact of parking policy measures on travel

demand. Transportation Planning and Technology, 13(4), 229-244.

Fellesson, M., & Friman, M. (2012). Perceived satisfaction with public transport

service in nine European cities. Journal of the Transportation Research

Forum, 47(3), 99-103.

Fishbein, M. (1967). Attitude and the prediction of behavior. Readings in Attitude

Theory and Measurement, 477-492.

Fornell, C. (1992). A national customer satisfaction barometer: The Swedish

experience. Journal of Marketing, 56(1), 6-21.

Frank, L., Bradley, M., Kavage, S., Chapman, J., & Lawton, T. K. (2008). Urban form,

travel time, and cost relationships with tour complexity and mode choice.

Transportation, 35(1), 37-54.

Friman, M., Edvardsson, B., & Gärling, T. (2001). Frequency of negative critical

incidents and satisfaction with public transport services. I. Journal of Retailing

and Consumer Services, 8(2), 95-104.

144 Bibliography

Friman, M., & Gärling, T. (2001). Frequency of negative critical incidents and

satisfaction with public transport services. II. Journal of Retailing and

Consumer Services, 8(2), 105-114.

Fujii, S., & Kitamura, R. (2003). What does a one-month free bus ticket do to habitual

drivers? An experimental analysis of habit and attitude change. Transportation,

30(1), 81-95.

Furnham, A. (2016). Gender and Money: Are There Sex Differences in Money

Usage?. Are men and women very different on how they think about and use

money?. Retrieved from https://www.psychologytoday.com/blog/sideways-

view/201511/gender-and-money-are-there-sex-differences-in-money-usage

Gadziński, J., & Radzimski, A. (2016). The first rapid tram line in Poland: How has it

affected travel behaviours, housing choices and satisfaction, and apartment

prices?. Journal of Transport Geography, 54, 451-463.

Gardner, B. (2009). Modelling motivation and habit in stable travel mode contexts.

Transportation Research Part F: Traffic Psychology and Behaviour, 12(1), 68-

76.

Gardner, B., & Abraham, C. (2010). Going green? Modeling the impact of

environmental concerns and perceptions of transportation alternatives on

decisions to drive. Journal of Applied Social Psychology, 40(4), 831-849.

Gärling, T., & Axhausen, K. W. (2003). Introduction: Habitual travel choice.


Geurs, K. T., & Van Wee, B. (2004). Accessibility evaluation of land-use and transport

strategies: review and research directions. Journal of Transport Geography,

12(2), 127-140.

Givoni, M., & Rietveld, P. (2007). The access journey to the railway station and its

role in passengers’ satisfaction with rail travel. Transport Policy, 14(5), 357-

365.

Glaeser, E. L., Kahn, M. E., & Rappaport, J. (2008). Why do the poor live in cities?

The role of public transportation. Journal of urban Economics, 63(1), 1-24.

Government of South Australia. (2016). Adelaide Metro. Retrieved from

https://www.adelaidemetro.com.au/

Government of Western Australia. (2016). Transperth - Let's connect. Retrieved from

http://www.transperth.wa.gov.au/

Green, P. E., & Srinivasan, V. (1978). Conjoint analysis in consumer research: issues

and outlook. Journal of Consumer Research, 5(2), 103-123.

Greene, W. H. (2000). Econometric analysis 4th edition. International edition, New

Jersey: Prentice Hall.

Greene, W. H. (2012). Nlogit Version 5 Reference Guide. .

Greene, W. H., & Hensher, D. A. (2010). Modeling ordered choices: A primer. New

York: Cambridge University Press.

Grönroos, C. (1984). A service quality model and its marketing implications.

European Journal of marketing, 18(4), 36-44.

Grönroos, C. (1990). Service management and marketing: managing the moments of

truth in service competition. Massachusetts: Lexington Books.

Guo, Z., & Wilson, N. H. (2011). Assessing the cost of transfer inconvenience in

public transport systems: A case study of the London Underground.

Transportation Research Part A: Policy and Practice, 45(2), 91-104.

Halton, J. H. (1960). On the efficiency of certain quasi-random sequences of points in

evaluating multi-dimensional integrals. Numerische Mathematik, 2(1), 84-90.

https://www.psychologytoday.com/blog/sideways-view/201511/gender-and-money-are-there-sex-differences-in-money-usage

https://www.psychologytoday.com/blog/sideways-view/201511/gender-and-money-are-there-sex-differences-in-money-usage

https://www.adelaidemetro.com.au/

http://www.transperth.wa.gov.au/

Bibliography 145

Hamer, P. (2010). Analysing the effectiveness of park and ride as a generator of public

transport mode shift. Road & Transport Research, 19(1), 51-61.

Hansen, E. R. (1987). Industrial location choice in Sao Paulo, Brazil: a nested logit

model. Regional Science and Urban Economics, 17(1), 89-108.

Hawkins, D. M. (1980). Identification of outliers (Vol. 11): Springer.

Hensher, D. (1994). Stated preference analysis of travel choices: the state of practice.


Hensher, D., Louviere, J., & Swait, J. (1998). Combining sources of preference data.

Journal of Econometrics, 89(1), 197-221.

Hensher, D. A. (2007). Bus transport: Economics, policy and planning. Amsterdam:

Elsevier.

Hensher, D. A. (2010). Hypothetical bias, choice experiments and willingness to pay.


Hensher, D. A., & Greene, W. H. (2002). The mixed logit model: The state of practice

and warnings for the unwary. Paper presented at the Proceedings of Institute

of Transportation Studies of University of Sydney, Sydney.

Hensher, D. A., & Johnson, L. W. (1981). Applied discrete choice modelling.

Hensher, D. A., Li, Z., & Ho, C. (2015). The role of source preference and subjective

probability in valuing expected travel time savings. Travel Behaviour and

Society, 2(1), 42-54.

Hensher, D. A., Rose, J. M., & Collins, A. T. (2011). Identifying commuter

preferences for existing modes and a proposed Metro in Sydney, Australia with

special reference to crowding. Public Transport, 3(2), 109-147.

Hensher, D. A., Rose, J. M., & Greene, W. H. (2005). Applied choice analysis: a

primer. Cambridge, UK: Cambridge University Press.

Hensher, D. A., Rose, J. M., & Greene, W. H. (2008). Combining RP and SP data:

biases in using the nested logit ‘trick’ – contrasts with flexible mixed logit

incorporating panel and scale effects. Journal of Transport Geography, 16(2),

126-133. doi:http://dx.doi.org/10.1016/j.jtrangeo.2007.07.001

Hensher, D. A., Rose, J. M., & Greene, W. H. (2012). Inferring attribute non-

attendance from stated choice data: implications for willingness to pay

estimates and a warning for stated choice experiment design. Transportation,

39(2), 235-245.

Hensher, D. A., & Truong, T. P. (1985). Valuation of travel time savings: a direct

experimental approach. Journal of Transport Economics and Policy, 237-261.

Hess, S., & Hensher, D. A. (2010). Using conditioning on observed choices to retrieve

individual-specific attribute processing strategies. Transportation Research

Part B: Methodological, 44(6), 781-790.

Hine, J., & Scott, J. (2000). Seamless, accessible travel: users’ views of the public

transport journey and interchange. Transport Policy, 7(3), 217-226.

Hitayezu, P., Wale, E., & Ortmann, G. F. (2016). Assessing agricultural land-use

change in the Midlands region of KwaZulu-Natal, South Africa: application of

mixed multinomial logit. Environment, Development and Sustainability, 18(4),

985-1003.

Hoang-Tung, N., & Kubota, H. (2017). Application of attitude theory for identifying

the effects of non-attendance attributes in stated choice surveys. Travel

Behaviour and Society.

Horowitz, J. L. (1985). Travel and location behavior: State of the art and research

opportunities. Transportation Research Part A: Policy and Practice, 19(5-6),

441-453.

http://dx.doi.org/10.1016/j.jtrangeo.2007.07.001

146 Bibliography

Hoyer, W., & MacInnis, D. (1997). Consumer Behavior. Boston and NY: Houghton

Mifflin Company.

Hu, L., & Schneider, R. J. (2017). Different ways to get to the same workplace: How

does workplace location relate to commuting by different income groups?

Transport Policy, 59, 106-115.

Hunt, G. L. (2000). Alternative nested logit model structures and the special case of

partial degeneracy. Journal of Regional science, 40(1), 89-113.

Hurst, E., Li, G., & Pugsley, B. (2014). Are household surveys like tax forms?

Evidence from income underreporting of the self-employed. Review of

economics and statistics, 96(1), 19-33.

Jansson, A., Olsson, E., & Erlandsson, M. (2006). Bridging the gap between analysis

and design: improving existing driver interfaces with tools from the framework

of cognitive work analysis. Cognition, Technology & Work, 8(1), 41-49.

Jen, W., & Hu, K. C. (2003). Application of perceived value model to identify factors

affecting passengers' repurchase intentions on city bus: A case of the Taipei

metropolitan area. Transportation, 30(3), 307-327.

Jenkins, D. P., Salmon, P. M., Stanton, N. A., & Walker, G. H. (2010a). A new

approach for designing cognitive artefacts to support disaster management.

Ergonomics, 53(5), 617-635.

Jenkins, D. P., Salmon, P. M., Stanton, N. A., & Walker, G. H. (2010b). A systemic

approach to accident analysis: a case study of the Stockwell shooting.

Ergonomics, 53(1), 1-17.

Jenkins, D. P., Stanton, N. A., Salmon, P. M., Walker, G. H., & Young, M. S. (2008).

Using cognitive work analysis to explore activity allocation within military

domains. Ergonomics, 51(6), 798-815.

Johansson, E. (2005). An estimate of self-employment income underreporting in

Finland. Nordic Journal of Political Economy, 31(1), 99-109.

Johnson, M., & Gustafsson, A. (2006). Improving customer satisfaction, loyalty and

profit: An integrated measurement and management system. New Delhi: Wiley

India Pvt. Limited.

Jou, R. C. (2001). Modeling the impact of pre-trip information on commuter departure

time and route choice. Transportation Research Part B: Methodological,

35(10), 887-902.

Kahneman, D., & Knetsch, J. L. (1992). Valuing public goods: the purchase of moral

satisfaction. Journal of Environmental Economics and Management, 22(1),

57-70.

Kamruzzaman, M., Baker, D., Washington, S., & Turrell, G. (2014). Advance transit

oriented development typology: case study in Brisbane, Australia. Journal of


Kemp, M. A., & Maxwell, C. (1993). Exploring a budget context for contingent

valuation estimates Contingent valuation: A critical assessment (pp. 217-269):

Emerald Group Publishing Limited.

Kingham, S., Dickinson, J., & Copsey, S. (2001). Travelling to work: will people move

out of their cars. Transport Policy, 8(2), 151-160.

Klöckner, C. A., Matthies, E., & Hunecke, M. (2003). Problems of operationalizing

habits and integrating habits in normative decision‐making models. Journal of

Applied Social Psychology, 33(2), 396-417.

Kocur, G., Hyman, W. A., & Aunet, B. (1982). Wisconsin work mode-choice models

based on functional measurement and disaggregate behavioral data

(0309035074). Retrieved from

Bibliography 147

Lambooij, M. S., Harmsen, I. A., Veldwijk, J., de Melker, H., Mollema, L., van Weert,

Y. W., & de Wit, G. A. (2015). Consistency between stated and revealed

preferences: a discrete choice experiment and a behavioural experiment on

vaccination behaviour compared. BMC medical research methodology, 15(1),

19.

Leigh, T. W., MacKay, D. B., & Summers, J. O. (1984). Reliability and validity of

conjoint analysis and self-explicated weights: A comparison. Journal of

Marketing Research, 456-462.

Lerman, S. R., & Louviere, J. J. (1978). Using functional measurement to identify the

form of utility functions in travel demand models. Transportation Research

Record(673).

Li, Z., & Hensher, D. A. (2011). Crowding and public transport: A review of

willingness to pay evidence and its relevance in project appraisal. Transport

Policy, 18(6), 880-887.

Limtanakool, N., Dijst, M., & Schwanen, T. (2006). The influence of socioeconomic

characteristics, land use and travel time considerations on mode choice for

medium- and longer-distance trips. Journal of Transport Geography, 14(5),

327-341. doi:http://dx.doi.org/10.1016/j.jtrangeo.2005.06.004

Litman, T. (2008). Valuing transit service quality improvements. Journal of Public


Long, J. S., & Freese, J. (2006). Regression Models for Categorical Dependent

Variables using Stata (2nd ed.). Texas: Stata Press.

Loomis, J., Gonzalez-Caban, A., & Gregory, R. (1994). Do reminders of substitutes

and budget constraints influence contingent valuation estimates? Land

Economics, 499-506.

Louviere, J. (1996). Combining revealed and stated preference data: the rescaling

revolution. Paper presented at the AERE conference.

Louviere, J. J., Henley, D. H., Woodworth, G., Meyer, R. J., Levin, I. P., Stoner, J. W.,

. . . Anderson, D. A. (1981). Laboratory-simulation versus revealed-preference

methods for estimating travel demand models. Transportation Research

Record(794).

Louviere, J. J., & Hensher, D. A. (1982). Design and analysis of simulated choice or

allocation experiments in travel choice modeling.

Louviere, J. J., Hensher, D. A., & Swait, J. D. (2000). Stated choice methods: analysis

and applications: Cambridge University Press.

Louviere, J. J., Islam, T., Wasi, N., Street, D., & Burgess, L. (2008). Designing discrete

choice experiments: Do optimal designs come at a price?. Journal of Consumer

Research, 35(2), 360-375.

Louviere, J. J., & Kocur, G. (1983). The magnitude of individual-level variations in

demand coefficients: a Xenia, Ohio case example. Transportation Research

Part A: Policy and Practice, 17(5), 363-373.

Lucas, J. L., & Heady, R. B. (2002). Flextime commuters and their driver stress,

feelings of time urgency, and commute satisfaction. Journal of Business and

Psychology, 16(4), 565-571.

Lund, A., & Lund, M. (2013a). Pearson Product-Moment Correlation. Retrieved from

https://statistics.laerd.com/statistical-guides/pearson-correlation-coefficient-

statistical-guide.php .

Lund, A., & Lund, M. (2013b). Spearman's Rank-Order Correlation. Retrieved from

https://statistics.laerd.com/statistical-guides/spearmans-rank-order-

correlation-statistical-guide.php .


https://statistics.laerd.com/statistical-guides/pearson-correlation-coefficient-statistical-guide.php

https://statistics.laerd.com/statistical-guides/pearson-correlation-coefficient-statistical-guide.php

https://statistics.laerd.com/statistical-guides/spearmans-rank-order-correlation-statistical-guide.php

https://statistics.laerd.com/statistical-guides/spearmans-rank-order-correlation-statistical-guide.php

148 Bibliography

Lyons, G. (2006). The role of information in decision-making with regard to travel.

Intelligent Transport Systems, 153(2), 199-212.

Mangham, L. J., Hanson, K., & McPake, B. (2009). How to do (or not to do)…

Designing a discrete choice experiment for application in a low-income

country. Health Policy and Planning, 24(2), 151-158.

Mann, E., & Abraham, C. (2006). The role of affect in UK commuters' travel mode

choices: An interpretative phenomenological analysis. British Journal of

Psychology, 97(2), 155-176.

Mark, T. L., & Swait, J. (2004). Using stated preference and revealed preference

modeling to evaluate prescribing decisions. Health Economics, 13(6), 563-

573.

McDougall, G. H., & Levesque, T. (2000). Customer satisfaction with services: putting

perceived value into the equation. Journal of services marketing, 14(5), 392-

410.

McFadden, D. (1973). Conditional logit analysis of qualitative choice behavior.

McFadden, D. (1981). Econometric models of probabilistic choice Structural analysis

of discrete data with econometric applications.

McIntosh, J., Newman, P., & Glazebrook, G. (2013). Why Fast Trains Work: An

assessment of a fast regional rail system in Perth, Australia. Journal of

Transportation Technologies, 3, 37-47.

McMillan, T. E. (2007). The relative influence of urban form on a child’s travel mode

to school. Transportation Research Part A: Policy and Practice, 41(1), 69-79.

Miguel, F. S., Ryan, M., & Amaya‐Amaya, M. (2005). ‘Irrational’stated preferences:

a quantitative and qualitative investigation. Health Economics, 14(3), 307-

322.

Mitchell, R. C., & Carson, R. T. (1989). Using surveys to value public goods: the

contingent valuation method. Washington DC: Resources for the Future Press.

Montgomery, D. B., & Wittink, D. R. (1979). Predictive Validity of Trade-Off Analysis

for Alternative Segmentation Schemes.

Moore, J. C. (1988). Miscellanea, self/proxy response status and survey response

quality, a review of the literature. Journal of Official Statistics, 4(2), 155.

Naikar, N., & Sanderson, P. M. (2001). Evaluating design proposals for complex

systems with work domain analysis. Human Factors, 43(4), 529-542.

Napper, R., Coxon, S., & Allen, J. (2007, 25-27 September 2007). Bridging the divide:

Design’s role in improving multi-modal transport. Paper presented at the 30th

Australasian Transport Research Forum: Managing Transport in a Climate of

Change and Uncertainty, Melbourne, Victoria.

Naudts, B., Van Ooteghem, J., Lannoo, B., Verbrugge, S., Colle, D., & Pickavet, M.

(2013). On the right tracks? continuous broadband internet on trains. Journal

of the Institute of Telecommunications Professionals, 7(1), 31-36.

Nguyen, H. A., Soltani, A., & Allan, A. (2018). Adelaide’s East End tramline: Effects

on modal shift and carbon reduction. Travel Behaviour and Society, 11, 21-30.

Noland, R. B., Weiner, M. D., DiPetrillo, S., & Kay, A. I. (2017). Attitudes towards

transit-oriented development: Resident experiences and professional

perspectives. Journal of Transport Geography, 60, 130-140.

Nordlund, A. M., & Garvill, J. (2002). Value structures behind proenvironmental

behavior. Environment and Behavior, 34(6), 740-756.

Nurdden, A., Rahmat, R., & Ismail, A. (2007). Effect of transportation policies on

modal shift from private car to public transport in Malaysia. Journal of applied

Sciences, 7(7), 1013-1018.

Bibliography 149

Ȫlander, F., & ThØgersen, J. (1995). Understanding of consumer behaviour as a

prerequisite for environmental protection. Journal of Consumer Policy, 18(4),

345-385.

Oliver, R. L. (2014). Satisfaction: A behavioral perspective on the consumer (2nd ed.).

New York: Routledge.

Olsen, S. O. (2007). Repurchase loyalty: The role of involvement and satisfaction.

Psychology & Marketing, 24(4), 315-341.

Ouellette, J. A., & Wood, W. (1998). Habit and intention in everyday life: The multiple

processes by which past behavior predicts future behavior. Psychological

Bulletin, 124(1), 54.

Paramita, P., Zheng, Z., Haque, M. M., Washington, S., & Hyland, P. (2018). User

satisfaction with train fares: A comparative analysis in five Australian cities.

PLoS ONE 13(6): e0199449.

doi:https://doi.org/10.1371/journal.pone.0199449

Parasuraman, A., Zeithaml, V. A., & Berry, L. L. (1985). A conceptual model of

service quality and its implications for future research. Journal of Marketing,

49(4), 41-50.

Parasuraman, A., Zeithaml, V. A., & Berry, L. L. (1994). Reassessment of

expectations as a comparison standard in measuring service quality:

implications for further research. Journal of Marketing, 58(1), 111-124.

Parker, B. R., & Srinivasan, V. (1976). A consumer preference approach to the

planning of rural primary health-care facilities. Operations Research, 24(5),

991-1025.

Passenger Demand Forecasting Council (PDFC). (2013). Passenger Demand

Forecasting Handbook.

Paul, P. J., Olson, J. C., & Gruner, K. (1999). Consumer behaviour and marketing

strategy: European edition. London: McGraw Hill.

Paulley, N., Balcombe, R., Mackett, R., Titheridge, H., Preston, J., Wardman, M., . . .

White, P. (2006). The demand for public transport: The effects of fares, quality

of service, income and car ownership. Transport Policy, 13(4), 295-306.

Prideaux, B., Wei, S., & Ruys, H. (2001). The senior drive tour market in Australia.

Journal of Vacation Marketing, 7(3), 209-219.

Pucher, J., & Buehler, R. (2009). Integrating bicycling and public transport in North

America. Journal of Public Transportation, 12(3), 5.

Queensland Government. (2016). TransLink. Retrieved from http://translink.com.au/

.

Rasmussen, J., Pejtersen, A. M., & Goodstein, L. P. (1994). Cognitive systems

engineering.

Rastogi, R. (2010). Willingness to shift to walking or bicycling to access suburban rail:

case study of Mumbai, India. Journal of urban planning and development.

Rissel, C., Curac, N., Greenaway, M., & Bauman, A. (2012). Physical activity

associated with public transport use—a review and modelling of potential

benefits. International journal of environmental research and public health,

9(7), 2454-2478.

Rodrıguez, G. (2007). Chapter 3 Logit Models for Binary Data. Retrieved from

http://data.princeton.edu/wws509/notes/c3.pdf

Ronis, D. L., Yates, J. F., & Kirscht, J. P. (1989). Attitudes, decisions, and habits as

determinants of repeated behavior. Attitude Structure and Function, 213-239.

https://doi.org/10.1371/journal.pone.0199449

http://translink.com.au/

http://data.princeton.edu/wws509/notes/c3.pdf

150 Bibliography

Rose, J. M., Bliemer, M. C., Hensher, D. A., & Collins, A. T. (2008). Designing

efficient stated choice experiments in the presence of reference alternatives.


Ryan, M. (2004). A comparison of stated preference methods for estimating monetary

values. Health Economics, 13(3), 291-296.

Rye, T. (2002). Travel plans: do they work? Transport Policy, 9(4), 287-298.

Santos, G. (2005). Urban congestion charging: a comparison between London and

Singapore. Transport reviews, 25(5), 511-534.

Santos, G., & Rojey, L. (2004). Distributional impacts of road pricing: The truth

behind the myth. Transportation, 31(1), 21-42.

Scheiner, J. (2010). Interrelations between travel mode choice and trip distance: trends

in Germany 1976–2002. Journal of Transport Geography, 18(1), 75-84.

doi:http://dx.doi.org/10.1016/j.jtrangeo.2009.01.001 .

Scheiner, J., & Holz-Rau, C. (2007). Travel mode choice: affected by objective or

subjective determinants?. Transportation, 34(4), 487-511.

Schwartz, S. H., & Howard, J. A. (1981). A normative decision-making model of

altruism. Altruism and helping behavior, 189-211.

Schwieterman, J. P., Fischer, L. A., Field, S., Pizzano, A., & Urbanczyk, S. (2009). Is

portable technology changing how Americans travel? A survey of the use of

electronic devises on intercity buses, trains, and planes. Depaul University,

Chicago, Illinois. Retrieved from

http://las.depaul.edu/chaddick/docs/Docs/Chaddick_Institute_Survey_of_Tec

hnology_1.pdf .

Severin, V. (2001). Comparing statistical and respondent efficiency in choice

experiments. The University of Sydney, Sydney.

Simma, A., & Axhausen, K. W. (2001). Structures of commitment in mode use: a

comparison of Switzerland, Germany and Great Britain. Transport Policy,

8(4), 279-288.

Singleton, P. A. (2018). Walking (and cycling) to well-being: Modal and other

determinants of subjective well-being during the commute. Travel Behaviour

and Society.

Sobel, K. L. (1980). Travel demand forecasting by using the nested multinomial logit

model (No. 775).

Soltani, A., & Allan, A. (2006). Analyzing the impacts of microscale urban attributes

on travel: Evidence from suburban Adelaide, Australia. Journal of urban

planning and development, 132(3), 132-137.

Solvoll, G., & Hanssen, T. E. S. (2017). User satisfaction with specialised transport

for disabled in Norway. Journal of Transport Geography, 62, 1-7.

Stanton, N. A., McIlroy, R. C., Harvey, C., Blainey, S., Hickford, A., Preston, J. M.,

& Ryan, B. (2013). Following the cognitive work analysis train of thought:

exploring the constraints of modal shift to rail transport. Ergonomics, 56(3),

522-540.

Starrs, M., & Perrins, C. (1989). The markets for public transport: the poor and the

transport disadvantaged. Transport reviews, 9(1), 59-74.

Stevenson, A. (2010). Oxford dictionary of English: Oxford University Press, USA.

Stopher, P. R. (2004). Reducing road congestion: a reality check. Transport Policy,

11(2), 117-131.

Swait, J., Louviere, J. J., & Williams, M. (1994). A sequential approach to exploiting

the combined strengths of SP and RP data: Application to freight shipper

choice. Transportation, 21(2), 135-152.


http://las.depaul.edu/chaddick/docs/Docs/Chaddick_Institute_Survey_of_Technology_1.pdf

http://las.depaul.edu/chaddick/docs/Docs/Chaddick_Institute_Survey_of_Technology_1.pdf

Bibliography 151

Taplin, J. H., & Qiu, M. (1997). Car trip attraction and route choice in Australia.

Annals of Tourism Research, 24(3), 624-637.

Taylor, B. D., & Morris, E. A. (2015). Public transportation objectives and rider

demographics: are transit’s priorities poor public policy? Transportation,

42(2), 347-367.

Thøgersen, J. (2006). Understanding repetitive travel mode choices in a stable context:

A panel study approach. Transportation Research Part A: Policy and Practice,

40(8), 621-638.

Thøgersen, J., & Møller, B. (2008). Breaking car use habits: The effectiveness of a

free one-month travelcard. Transportation, 35(3), 329-345.

Thompson, K., & Schofield, P. (2007). An investigation of the relationship between

public transport performance and destination satisfaction. Journal of Transport

Geography, 15(2), 136-144.

Tourangeau, R., & Yan, T. (2007). Sensitive questions in surveys. Psychological

bulletin, 133(5), 859.

Train, K. E. (1999). Halton sequences for mixed logit. Department of Economics,

University of California Berkeley, California.

Train, K. E. (2009). Discrete choice methods with simulation (2nd ed.). New York:

Cambridge University Press.

Transport for NSW. (2016). Transport. Retrieved from http://www.transportnsw.info/

.

Tselentis, D. I., Theofilatos, A., Yannis, G., & Konstantinopoulos, M. (2018). Public

opinion on usage-based motor insurance schemes: A stated preference

approach. Travel Behaviour and Society, 11, 111-118.

Tyrinopoulos, Y., & Antoniou, C. (2008). Public transit user satisfaction: Variability

and policy implications. Transport Policy, 15(4), 260-272.

Van Exel, N., & Rietveld, P. (2009). Could you also have made this trip by another

mode? An investigation of perceived travel possibilities of car and train

travellers on the main travel corridors to the city of Amsterdam, The

Netherlands. Transportation Research Part A: Policy and Practice, 43(4), 374-

385.

van Raaij, W. F., Bartels, G., & Nelissen, W. (2002). Stages of behavioural change:

motivation, ability and opportunity. Marketing for Sustainability

towardsTransactional Policy-Making, 321-333.

Van Raaij, W. F., & Verhallen, T. M. (1983). A behavioral model of residential energy

use. Journal of Economic Psychology, 3(1), 39-63.

Verplanken, B., & Aarts, H. (1999). Habit, attitude, and planned behaviour: is habit an

empty construct or an interesting case of goal-directed automaticity? European

Review of Social Psychology, 10(1), 101-134.

Verplanken, B., Aarts, H., Knippenberg, A., & Knippenberg, C. (1994). Attitude

versus general habit: Antecedents of travel mode choice. Journal of Applied

Social Psychology, 24(4), 285-300.

Verplanken, B., Aarts, H., Knippenberg, A., & Moonen, A. (1998). Habit versus

planned behaviour: A field experiment. British Journal of Social Psychology,

37(1), 111-128.

Verplanken, B., & Holland, R. W. (2002). Motivated decision making: effects of

activation and self-centrality of values on choices and behavior. Journal of

Personality and Social Psychology, 82(3), 434.

http://www.transportnsw.info/

152 Bibliography

Verplanken, B., Walker, I., Davis, A., & Jurasek, M. (2008). Context change and travel

mode choice: Combining the habit discontinuity and self-activation

hypotheses. Journal of Environmental Psychology, 28(2), 121-127.

Victoria State Government. (2016). Public Transport Victoria. Retrieved from

https://www.ptv.vic.gov.au/ .

Wardman, M. (1988). A comparison of revealed preference and stated preference

models of travel behaviour. Journal of Transport Economics and Policy, 71-

91.

Wardman, M. (2004). Public transport values of time. Transport Policy, 11(4), 363-

377.

Washington, S. P., Karlaftis, M. G., & Mannering, F. L. (2011). Statistical and

econometric methods for transportation data analysis (2nd ed.). Florida: CRC

press.

Weisbrod, G., & Reno, A. (2009). Economic impact of public transportation

investment. Retrieved from Massachusetts:

Wood, W., Quinn, J. M., & Kashy, D. A. (2002). Habits in everyday life: thought,

emotion, and action. Journal of Personality and Social Psychology, 83(6),

1281.

Zeithaml, V. A., Berry, L. L., & Parasuraman, A. (1988). Communication and control

processes in the delivery of service quality. Journal of Marketing, 52(2), 35-

48.

Zheng, Z., Liu, Z., Liu, C., & Shiwakoti, N. (2014). Understanding public response to

a congestion charge: A random-effects ordered logit approach. Transportation

Research Part A: Policy and Practice, 70, 117-134.

Zheng, Z., Washington, S., Hyland, P., Sloan, K., & Liu, Y. (2016). Preference

heterogeneity in mode choice based on a nationwide survey with a focus on

urban rail. Transportation Research Part A: Policy and Practice, 91, 178-194.

Zheng, Z., Wijeweera, A., Sloan, K., Washington, S., Hyland, P., To, H., . . . Holyoak,

N. (2013). Understanding Urban Rail Travel for Improved Patronage

Forecasting – Final Report.

https://www.ptv.vic.gov.au/

Bibliography 153


this when bibliography is at least 2 pages long.]

MODELLING COMMUTERS · 2018. 10. 10. · MODELLING COMMUTERS’ MODE CHOICE: INTEGRATING TRAVEL...

Documents

Transcript of MODELLING COMMUTERS · 2018. 10. 10. · MODELLING COMMUTERS’ MODE CHOICE: INTEGRATING TRAVEL...