i

Demand for Taxi and Hire Car Services in Melbourne, Victoria

The Hensher Group Pty Ltd

i

Report prepared by The Hensher Group Pty Ltd 215 Excelsior Avenue Castle Hill 2154 NSW Australia Hgroup@optusnet.com.au (authors: David Hensher and John Rose) For Victorian Taxi Industry Inquiry http://www.taxiindustryinquiry.vic.gov.au/ Ph: + 61 3 9655 4506 Mob: + 407 909 812 Warwick.Davis@transport.vic.gov.au Version: 15 March 2012(revised final 23 April 2012) The Hensher Group

Professor David A. Hensher, PhD FASSA

David Hensher is a Director of the Hensher Group Pty Ltd (and Econometric Software) and Choice Metric Pty Ltd, Professor of Management, and Founding Director of the Institute of Transport and Logistics Studies (ITLS): The Australian Key Centre of Teaching and Research in Transport Management in The Business School at The University of Sydney. David is a Fellow of the Academy of Social Sciences in Australia (FASSA), Recipient of the 2009 IATBR (International Association of Travel Behaviour Research) Lifetime Achievement Award in recognition for his long-standing and exceptional contribution to IATBR as well as to the wider travel behaviour community, Recipient of the 2006 Engineers Australia Transport Medal for lifelong contribution to transportation, and Recipient of the 2009 Bus NSW (Bus and Coach Association) Outstanding Contribution to Industry Award. Director of Volvo Educational and Research Foundation Centre of Excellence in Bus Rapid Transit (2010 onwards), Member of Singapore Land Transport Authority International Advisory Panel (Chaired by Minister of Transport), Honorary Fellow Singapore Land Transport Authority Academy, Past President of the International Association of Travel Behaviour Research and a Vice-Chair of the International Scientific Committee of the World Conference of Transport Research. David is the Executive Chair and Co-Founder of The International Conference in Competition and Ownership of Land Passenger Transport (the Thredbo Series), now in its 22nd year. David is on the editorial boards of 10 of the leading transport journals and Area Editor of Transport Reviews. David was appointed in 1999 by one of the worlds most prestigious academic publishing houses - Elsevier Science press - as series and volume editor of a handbook series Handbooks in Transport. In 2010 he was appointed by Routledge Publishers (UK) as Editor of a four-volume major works in Transport Economics as well as Edward Elgar Publishers as Series Editor for volumes on Transport and the

ii

Environment. He has published extensively (over 475 papers) in the leading international transport and economics journals as well as 12 books and is Australia’s most cited transport academic and number three academic economist. His books include the Demand for Automobiles, published by North-Holland, the Bus and Coach Business (with Ann Brewer published - Allen and Unwin), Transport: An Economics and Management Perspective (With Ann Brewer – Oxford University Press), Stated Choice Methods (with Jordan Louviere and Joffre Swait – Cambridge University Press), Applied Choice Analysis - a Primer (with John Rose and Bill Greene – Cambridge University Press) and Ordered Choice Models (with Bill Greene –Cambridge University Press). His particular interests are transport economics, transport strategy, sustainable transport, productivity measurement, traveller behaviour analysis, choice analysis, stated choice experiments, and institutional reform (PPPs, privatisation tendering and contracting). David has advised numerous government and private sector organisations on matters related to transportation, especially matters related to forecasting demand for existing and new transportation services; for example the Speedrail project, the Liverpool-Parramatta Transitway, the North-West Rail project, the Sydney Metro, and numerous tollroad projects throughout Australia and internationally. David is regarded as Australia’s most eminent expert on matters relating to travel demand and valuation and transport reform. Appointments over recent years include: a member of the executive committee that reviewed bus transport bids for the Olympic Games, the NSW Government’s Peer Review Committee for the Sydney Strategic Transport Plan, Peer reviewer for Transfund (NZ) of the New Zealand project evaluation program, Peer reviewer of the NZ Land Passenger Transport Procurement Strategy for Land Transport NZ, member of the executive committee of ATEC, a consortium promoting a freight rail system between Melbourne and Darwin; economic adviser to Gilbert+Tobin Lawyers on valuation methods in IP context; panel member of Transport NSW benchmarking program; specialist toll road project adviser to Thiess and member of Infrastructure Australia’s reference panel on public transport.

Professor John M. Rose, PhD

John is a Director of Choice Metrics Pty Ltd and Associate with The Hensher Group, and Professor of Transport and Logistics Modelling and a Deputy Director at Institute of Transport and Logistics Studies (ITLS). John began his academic career in the field of marketing, commencing as an associate lecturer in the Discipline of Marketing at the University of Sydney in 1995. As an associate lecturer, John taught marketing principles, consumer behaviour, introductory and advanced marketing research techniques, and new product development, all at the undergraduate and postgraduate levels. In 1999, John was promoted to the level of Lecturer were he continued in his teaching role. In 2001, John moved over to the ITLS to complete his PhD under Professor David Hensher, which he completed in 2005. Whilst at ITLS, John has been responsible for running the industry program which includes courses taught to the Roads and Traffic Authority of NSW, to NSW bus operators, as well as other professional development courses open to academics and public

iii

companies. John has also taught introductory statistics, transport economics, and discrete choice modelling. John’s research interests are in the areas of discrete choice modelling and efficient stated choice experiments. He has several articles published in the leading journals in the fields of transportation and logistics (including Transportation, Transportation Research A, B and E), marketing (Marketing letters) and environmental economics (including the Australian Journal of Environmental and Resource Economics) and is a co-author of (with Professors David Hensher and William Greene) Applied Choice Analysis: A Primer, (2005) by Cambridge University Press. He is currently writing a book on generating efficient stated choice experimental designs (with Mike Bliemer, Delft). Currently John is active in consulting, working in the areas of Toll Road evaluation and modelling, demand and take up for pharmaceutical and agricultural products.

iv

March 1, 2012 The most recent quarterly telephone surveys of travellers using Melbourne’s trains, trams and buses show that passenger satisfaction is at its highest level in years, according to Victoria’s Public Transport Ministry. The overall satisfaction of Melbourne’s public transport users rose to 62.7 index points, the highest quarterly result since the December quarter 2006. Victoria’s Public Transport Minister Terry Mulder, pictured, says surveyed commuters using Melbourne’s trains expressed the highest satisfaction for five years with an index score of 64, a marked rise of 1.7 points on the September 2011 quarter. “Tram users expressed greater satisfaction with tram drivers with overall satisfaction being consistent with the 70 index point result also recorded in the September 2011 quarter,” he says. “The increased passenger satisfaction for Melbourne’s Metro trains and Yarra Trams is particularly pleasing because these two modes account for 79 per cent of Melbourne’s 523.1 million per annum public transport trips.” But bus commuters also gave a significant tick to the Victorian Government’s introduction of more than 2,000 extra trips a week with the satisfaction level of 71.5 index points. Mulder says the bus rating was the highest recorded since the September quarter 2003. He says that while passenger satisfaction with V/Line trains was higher at 71.5 index points than for metropolitan trains, it had dropped in recent years. While improving Melbourne’s taxi services has been a hotly discussed item in Victoria recently, it still lags behind other modes in terms of customer satisfaction.

“Taxi satisfaction rose in the latest quarter but its result of 59.8 index points is way below any other mode and confirms the importance of the Government’s Taxi Industry Inquiry chaired by Professor Allan Fels,” Mulder says.

The University of Sydney Institute of Transport and Logistics Studies (ITLS) Transport Opinion Survey The Transport Opinion Survey is a quarterly survey of 1,000 adults aged 18 years and over across Australia launched in March 2010. The sample is representative of Australia’s population distribution and demographic characteristics. Interviews are conducted by telephone by Taverner Research using trained interviewers. The Quarter 1, March 2012 survey was conducted over 10-26 February 2012.

v

1. Taxi services in local area In the March 2012 quarter, a new question was asked about taxi services in the local area. Do you think that Taxi services in your local area have improved in the last 12 months? In the March 2012 quarter, only 14 percent of Australians said that taxi services in their local areas have improved in the last 12 months; while 51 percent said there had been no improvement. • Thirty five percent of respondents answered ‘don’t know’. Based on a reasonable assumption that those respondents who answered ‘don’t know’ may not have used a taxi recently or have limited experience of using a taxi and hence they are not regular taxi users, only 21 percent of regular taxi users said that taxi services have improved, compared to 79 percent who said there had been no improvement (worse or unchanged). • Nineteen percent of West Australian residents said that taxi services have improved in the last 12 months, followed by NSW residents (16 percent); while only 11 percent of Queensland residents said that taxi services have improved, lowest among all states. 21 percent of frequent public transport users (at least 3 days a week) said that taxi services have improved compared to 12-15 percent for less-frequent PT user groups.

Note: The % in brackets below X-axis for each category indicates the % of respondents in that category of state, town size, public transport use frequency, sex, age and labour force status. Town size is self-reported by respondent.

vi

Summary This report outlines a study designed to investigate the behavioural influences on traveller choice of mode for specific trips in the Melbourne metropolitan area, with a special focus on understanding the factors that influence the choice of, and hence demand for, taxis and hire car services in Melbourne. Given the importance of positioning preferences for taxi and car hire services within the broader set of modal options (including car, train, tram and bus), we develop a modal choice model capability for all available modes of transport for trips undertaken by individuals or groups of individuals in the broad categories of corporate travellers, international tourists, travellers undertaking general day to day activities (including daily social visitors resident in Melbourne), late night social (including visiting friends and relatives) users, and users that hold multi-purpose trip program (MPTP) cards, which provide for a 50 percent discount on taxi fares. A sample of recent trips in February 2012 defines the context in which the influences on modal preferences can be revealed, emphasising the need to account for trip length, time of day and day of week trip activity. Key outputs include a decision support system (DSS) which can be used to investigate ‘what if’ scenarios in respect of traveller responses towards or out of taxi and hire car services, which produces direct (and cross) elasticities of interest for cost and service level attributes (such as fares and waiting time). The underlying behavioural driver of the DSS are a series of modal choice models corresponding to five travel segments; tourism, corporate travel, general day to day activity travel, late night travel, and trips undertaken by MPTP card holders. Some factors are significant drivers of modal choice, and other factors are sources of reinforced utility, but not of strong significance to result in modal switching.

vii

Table of Contents Summary ..................................................................................................................................................... vi 1. Study Background ............................................................................................................................... 1 2. The Study Research Questions ........................................................................................................... 1 3. The Study Approach ............................................................................................................................ 2 4. The Stated Choice Experiment Approach .......................................................................................... 7 5. The Modelling Approach ................................................................................................................... 11 6. The Pilot Study ................................................................................................................................... 12 7. Main Field Phase Sample Design ...................................................................................................... 14 8. Empirical Data ................................................................................................................................... 16

8.1 Socio-Demographic Descriptive Statistics by Segment ............................................................... 16

8.2 Modal Travel Time and Cost Descriptive Statistics by Segment ................................................. 16

8.3 Descriptive Statistics by Segment of Altitudinal data ................................................................. 18 9. Study Results ..................................................................................................................................... 26

9.1 Tourism Segment.......................................................................................................................... 27

9.2 Business Traveller Segment ......................................................................................................... 32

9.3 General Day to Day Activity Travel Segment .............................................................................. 33

9.4 Night Time Travel Segment ......................................................................................................... 35

9.5 MPTP Card Holder Travel Segment ............................................................................................. 36 10. A Decision Support System (DSS) ................................................................................................ 38 10.1 Constructing a Population-Level Modal Share ....................................................................... 40 10.2 The DSS ...................................................................................................................................... 42 11. Existing Evidence on Taxi Elasticities .......................................................................................... 47 12. Conclusions .................................................................................................................................... 47 Appendix A: Statistical Properties of Stated Choice Designs ................................................................. 50 Appendix B: The CAPI Screens ................................................................................................................. 54 Introduction ....................................................................................................................................... 54 Trip Explanation ................................................................................................................................ 55 Taxi / Hire Car Usage ........................................................................................................................ 56 Recent Trip, Part 1 ............................................................................................................................. 57

viii

Recent Trip, Part 1 (Full Fields) ............................................................................................................... 58 Recent Trip, Part 2 ............................................................................................................................. 59 Your Trip - Bus ................................................................................................................................... 60 Your Trip – Bus (Full Fields) ............................................................................................................ 61 Your Trip - Tram ................................................................................................................................ 62 Your Trip - Train ................................................................................................................................ 63 Your Trip - Car ................................................................................................................................... 64 Your Trip - Walk ................................................................................................................................ 65 Your Trip - Taxi .................................................................................................................................. 66 Your Trip – Hire Car .......................................................................................................................... 67 Game Introduction ............................................................................................................................ 68 Practice Game – Bus, Car, Taxi ......................................................................................................... 69 Practice Game – Car, Taxi .................................................................................................................. 70 Waiting for game start ...................................................................................................................... 71 Game – Bus, Car, Taxi ........................................................................................................................ 72 Game– Car, Taxi ................................................................................................................................. 73 Attitudes to Public Transport ........................................................................................................... 74 Attitudes to Taxis/Hire Cars ............................................................................................................. 75 Socioeconomic Characteristics ......................................................................................................... 76 Thanks ................................................................................................................................................ 77 Appendix C: Recruitment Screener .......................................................................................................... 78 Taxi and Hire Car Study Recruitment Screener – TRC 4303 ................................................................. 78 References .................................................................................................................................................. 82 References on Previous Elasticity Studies Associated with Taxis ......................................................... 83

1

1. Study Background The Victorian Government is currently undertaking an inquiry into the structure of the Victorian taxi industry which commenced in June 2011, headed by Professor Allan Fels. Since its establishment, the enquiry has determined that there currently exists very little information as to the demand for taxi and hire car services for Victoria. Such a lack of information is likely to represent a serious barrier to understanding the impact of any reforms that the Inquiry recommends. This document outlines the results from a study undertaken by The Hensher Group (THG). The report describes the approach taken by THG to gather and analyse data to address the requirements of the Victorian Government Taxi and Hire Car Industry Inquiry in terms of understanding demand and the key drivers of demand for taxis and hire cars. In particular, discrete choice modelling methods applied to stated choice (SC) data, together with a decision support system (DSS) used to investigate ‘what if’ scenarios in respect of traveller responses towards or out of taxi and hire car services, are used to satisfy the needs of the Inquiry. The remainder of this document is set out as follows. Section 2 outlines the research objectives of the Victorian Government inquiry. Section 3 provides an overview of the approach to address these research objectives, whilst Section 4 provides a brief summary of stated choice data and survey techniques, as well as how such a method is applied to this particular research project. Next, Section 5 outlines how THG analyse data collected for the inquiry before Section 6 details of a pilot study undertaken to test the survey instrument and survey logistics. Section 7 discusses the sampling used for the study. Section 8 provides information related to the data collected before the model results are discussed in Section 9. A decision support system is discussed in Section 10, after which a brief discussion about elasticities is given in Section 11. Concluding comments are provided in Section 12.

2. The Study Research Questions A key task for the Inquiry is to understand traveller preferences and consequent demand for taxi and hire car services in Victoria. The findings are a critical input into decisions about various issues before the Inquiry. One part of the Inquiry’s demand analysis relates to the sensitivity of consumer demand to changes in taxi fares (e.g., how many more trips will occur if fares fall), and waiting/response times (e.g., how many more trips will occur from a reduction in current taxi waiting times), as well as users and non-users of taxis attitudes about the softer but important features of service quality such as the general state of the vehicle and the quality of drivers. A second aspect of this study is to consider whether there is scope for taxis to be effective substitutes for forms of public transport, particularly buses, in areas of relatively low demand. Given the above, we summarise the study research questions/objectives as follows. 1. To determine the key drivers of demand for the taxi and hire car markets in Victoria across a number of travel segments;

2

2. To provide (direct) elasticity and other demand related estimates related to the taxi and hire car markets; and 3. To determine the scope for taxis to be effective substitutes for forms of public transport, particularly buses, in areas of relatively low demand. 3. The Study Approach

The approach proposed herein focuses on the development of a modelling approach designed to gain an understanding of the preferences of actual and potential travellers for each of the available modes of transport for specific trip purposes or use segments such as corporate travel, tourism, and late night social outings in the Melbourne Metropolitan Area. Although the primary modal interest is in identifying the role and performance of taxis and hire cars, it is necessary to consider alternative modes as a way of identifying the contribution of taxis and hire cars to the overall passenger transport task in a metropolitan area. To do this requires the development of a modal choice model in which all relevant or feasible modal alternatives are assessed as chosen or non-chosen alternatives for a specific trip. Given that the modal shares of each mode for specific trip purposes and user segments are dominated by the car, and the focus of this study is on service levels of taxis and hire cars, we have to ensure that all modes are studied with a sufficiently large sample in order to identify the key factors that influence the choices made by individual travellers, be they travelling by themselves or in a group. We are as much interested in non-users, as we are in users, of taxis and hire cars, since their non-use may be a consequence of the cost and service levels offered by taxis and hire cars. To understand user preferences for taxis and hire cars in contrast to preferences for car, and public transport modes (i.e., bus, tram, train), and walking, we have to gather data from a sample of individuals who have recently experienced using one or more of the available modes of transport for a specific trip purpose in Melbourne. It is important to note that such trips need not necessarily have been undertaken via a taxi or hire car, but that the sampled trips potentially could have used either a taxi or hire car, for access, main linehaul or egress legs of travel. As such, sampling for the study represents a critical factor to successfully address the research questions identified. The sample must include individuals from the main user segments that taxis and hire cars service such as corporate users, tourists (both international and domestic visiting Melbourne), and locals undertaking social outings. Furthermore, the role that taxis may play in each segment could be influenced by the timing of the trip (e.g., evenings) and the specific destination (e.g., an airport or nightclub). Given the lack of relevant data on travel demand for taxis and hire cars, we make use of a stated choice (SC) experiment embedded within a larger survey questionnaire. The survey instrument has six sections (summarised in Table 1; screen shots of the survey may be located in Appendix B). The SC experiment which represents the centrepiece of the survey is designed specifically to obtain an understanding of the key attributes that influence mode choice. The purpose behind conducting experiments is to determine the independent influence of different variables (attributes or factors depending on the literature cited) on some observed outcome. In SC studies, this translates into the desire to determine the influence of the design attributes upon the choices that are observed to be

3

made by sampled respondents undertaking the experiment. Unlike most survey data where information on both the dependent and independent variables is captured directly from respondents, stated preference data, of which SC data is a subset, is unique in that typically only the dependent variable is provided by the respondent. With the exception of covariate information which is often ignored in most analysis, the primary variables of interest, consisting of attributes and their associated levels, are designed in advance and presented to the respondent in the form of competing alternatives in SC studies. Table 1: An Overview of the Structure of the CAPI Survey Instrument

Section Sections of Survey Instrument1 Introduction screen explaining purpose of survey 2 Selecting one recent trip as reference trip for SC experiment This is a recent trip mode and up to two feasible alternative modes Ensure we have a good spread of all modes and user segments and trip times as chosen Ask questions on trip attributes that are in SC experiment for each candidate mode (see list of attributes for each mode) Who pays for trip? (you, your own business, your employer, a client) Number of people in travelling group? Specific purpose (e.g., Going out on the town, get to an airport on a business trip, etc.) Was taxi booked or hailed from roadside? (online, phone, hail) If booked, did it arrive on time or was it late/early? If late/early, how many minutes? If booked, did you have to call again due to non-arrival? 3 Stated choice scenarios (12 per respondent) (See Table 2 and Appendix B) 4 Taxi/Hire car and Public Transport Attitudinal and other questions (everyone answers) Would you regard yourself as a frequent or infrequent user of taxis/hire cars? (typical use per month) Do you have an account with a taxi company? What do you think are the good features of the taxis services, and hire car services (separate) What do you think are the bad features of the taxis services, and hire car services (separate) 5 Socioeconomic questions

Personal income, age, gender, household size, permanent residence location (postcode or country if not Australia)

6 Thank you For the current study context, we develop a modal choice model in which, for each sampled respondent, we identify the levels of key attributes that are the sources of utility, and which describe the preferences of each sampled respondent. The central feature of the study is a detailed investigation of preferences for alternative modes of travel in the setting of one of the recent trips undertaken. The approach draws on ideas in prospect theory that individuals need familiarity with a real trip experience (a reference point) if they are going to be asked to consider a number of possible future scenarios of modal service level and fares, and to indicate which one would be their preferred modal choice if it were to have been available at the time of the selected recent trip. Specifically, the proposed approach acts to frame the decision context of the modal choice task within some existing memory schema of the individual respondents, and hence make preference-revelation more meaningful at the level of the individual. Theoretically, the role of reference alternatives in SC tasks, the approach proposed (see below), is well supported within the literature. For example, prospect theory (Kahneman and Tversky 1979), which argues that individuals use

4

decision heuristics when making choices, promotes the idea that the very specific context in which a decision is made by each individual is an important determinant of the selection of choice-heuristic, supporting the use of reference alternatives in SC tasks. Framing effects, of which reference dependence is a popular interpretation, provides context support in trading off the desire to make a good choice against the cognitive effort involved in processing the additional information provided in a SC task (Hensher 2010). Case-based decision theory (Gilboa et al. 2002) promotes the role of accumulated experience represented by a reference alternative. Starmer (2000, p353) in particular argues strongly for the use of reference alternatives (e.g., a current trip) in decision theory: “While some economists might be tempted to think that questions about how reference points [alternatives] are determined sound more like psychological than economic issues, recent research is showing that understanding the role of reference points[alternatives] may be an important step in explaining real economic behaviour in the field” The approach that we implement will enable us to evaluate the influence of the set of attributes that represent the items of interest is a SC experiment, which will provide the empirical inputs for a discrete choice model, so that parameter estimates can be obtained to indicate the role specific attributes play in determining mode choices made by the sample of travellers in Melbourne. Once identified, specific and meaningful levels are attached to each attribute, which are then systematically varied in the SC experiment by the researcher Candidate attributes are summarised in Table 2. Some attributes are relevant for all modes and others are relevant to a sub set of modes. In addition to the traditional influences on mode choice such as travel times and costs (or fares), we include a number of the important attributes that describe service levels of specific modes, especially taxis and hire cars. In total, Table 2 reveals 36 potential attribute dimensions that are likely to drive mode choice decisions. Given such a large number of attributes, we limit the number of attribute within the SC experiment to the first 14 listed in Table 1. Data related to the remaining attributes, whilst not part of the SC experiment, are captured as additional variables in the form of attitudinal questions. The SC experiment involves a universal choice set of up to seven modes but recognise that for most trips, only a few of these modes are available or feasible. For example, train may not be available to get to some locations such as the airport at Tullamarine, but bus is (the sky bus service which is very competitive and a genuine alternative to taxis and hire cars, and indeed a private car). To ensure that we obtain enough observations where each mode is chosen and other modes are feasible alternatives, we adopt a choice-based sampling strategy (with weighting of sample to line up with population shares in model estimation). Choice-based sampling is an efficient way of ensuring sufficient observation on each mode so as to be able to obtain reliable estimates of parameters in a mode choice model to construct preference equations for each mode (within each user segment, recognising the need to also accommodate trip length, time of day and day of week and weekend).

5

Table 2: An Overview of the Key Attributes Attributes Taxi Hire car Car Bus Tram Train Walk Time to get to main mode (access time) (mins) × × × ×Time waiting for main mode (mins) × × ×Time in main mode (mins) × × × × × × × Time to get from main mode to destination (egress time ) (mins) × × × × × × Frequency of delay (every X number of trips) Level of crowding (% seats occupied, number of people standing) Cost or fare ($ per one-way trip) × × × × × × Tolls incurred ($) × × × Parking cost ($ per day) × Probability of specific wait times occurring × × Probability of specific trip travel times occurring × × × Probability of specific fare payments occurring × × Time taxi/hire car arrives relative to booked time (mins early, late, on time=0) × × Time to get taxi from when decide to look for one (hail) (mins) ×

Vehicle and Driver Quality: Internal cleanliness of vehicle × × × × ×Air conditioning × × × × ×Mechanical state of vehicle × × Attitude of driver × × External cleanliness of vehicle × × Personal safety × × × × × Exposure to the weather × × × × × Driver knowledge of route × × Seating comfort × × × × ×Seating condition × × Driver presentation × × Vehicle is disable friendly × × English ability of driver × × Driver driving ability × × Ease of booking × × Shops in precinct × × × Getting a seat × × × Crowding at station/stop × × ×Crowding in the vehicle × × ×Travelling in a tunnel × × × Graffiti free × × × Noise level × × ×The centrepiece of the behavioural research is a discrete choice model of choice amongst relevant modes (within the agreed segments). Figure 1 summarises the eligible trip settings which may be an access mode situation or a main mode setting. This choice model provides weights (or parameter estimates) associated with each statistically significant influence on the choice amongst the feasible alternative modes for a specific user segment. To be able to identify the set of weights associated with each influencing variable (or attribute defining each modal alternative), conditioned as appropriate on contextual effects such as trip maker socioeconomics, we need to collect suitable data. The main feature of data definition is a SC experiment in which, for metropolitan-wide travel in the catchment area, we begin by defining a current or recent trip experience made in the past five days (from the set of all travel in Melbourne during this period). Knowing this provides rich data to assist in selecting a specific recent trip that contributes to the quotas that we will select to ensure sufficient sample sizes for each user segment. Importantly, eligible respondents would have been screened in-scope prior to their participation in the choice experiment. The quotas for proposed segments are set out in a later section.

6

Figure 1: Trip Contexts The SC experiment involves a respondent comparing the levels of agreed attributes of various competing alternatives which are related in terms of their levels to the same current or recent trip, thus providing a realistic choice context to the respondent. The alternatives that are shown to individual respondents include the current mode and up to two alternative modes that could be used in the context of the recent trip experience. The respondent is then asked to choose one of these alternatives. The process of choosing amongst the alternatives is repeated an agreed number of times, each repetition is known as a choice scenario in which each choice situation involves varying the levels of each attribute associated with the different alternatives. The data from the SC experiment is then used in estimating choice models for each market segment (e.g., corporate traveller, tourists). Two types of econometric discrete choice models are estimated as part of this study. Given the limited sample size obtained for the MPTP card holder segment, a multinomial logit (MNL) model was estimated on the data captured for this segment. For the remaining travel segments, the choice model form estimated is known as a mixed multinomial logit (MMNL) model (detailed in Section 5). The MMNL, also sometimes referred to as the random parameters logit model (RPL) or mixed logit model, represents the state of art in econometric modelling for discrete choice type data. Several versions of the MMNL model exist which provide various potential benefits over other econometric models of discrete choice. Advantages of the MMNL model over other models include i) the incorporation of ‘unexplained’ preference

7

heterogeneity (see e.g., Hensher and Greene 2003; McFadden and Train 2000), and ii) non-constant error variances across alternatives via a relaxation of the IID assumption (see e.g., Hensher and Greene 2003, or Train 2009). A third possible advantage of the model, depending on how the model is specified, is the accommodation of within-respondent preference correlation across repeated choice observations (see e.g., Revelt and Train 1998). The final set of estimated choice models delivers the required weights attached to each explanatory variable (or attributes of alternatives and characteristics of respondents as separate or interaction effects), which are used in constructing utility equations for use in derivation of the attribute elasticities and other outputs such as attribute-specific willingness to pay values. 4. The Stated Choice Experiment Approach

The method applied in the present study involves the use of SC experiments. The alternative to the revealed preference (RP) method is a stated choice experiment, in which we systematically vary combinations of levels of each attribute to reveal new opportunities relative to the existing circumstance of attribute levels on offer (see Hensher 1994; Louviere et al. 2000; Hensher et al. 2005; Rose et al. 2008). SC data enables us to investigate levels of attributes that do not exist in real markets (as well as new modal alternatives; see Figure 2). RP data is useful when we stay within the attribute range of attributes associated with each modal alternative and that we can be confident that the levels of non-chosen alternatives re reliable (even when they are perceived levels which are the key drivers of choice).

Figure 2: Attribute space of Revealed Preference and Stated Choice data Through the experimental design paradigm we are able to observe a sample of travellers making choices between a recent modal trip attribute level bundle (or a package of alternative time, cost, service level descriptors), and other alternative attribute level bundles describing competing modes. This approach is the most powerful method capable of separating out the independent contributions of each attribute component between a number of modal options for a specific use segment (including time of day and day and trip length). It is the preferred approach, capable of providing disaggregated estimates of direct and cross attribute elasticities of interest. To ensure that the elasticity outputs are meaningful, it is necessary to calibrate the estimates model by

1cost

speed

1cost

speed

FRONTIER OF EXISTING ALTERNATIVES

Walk

Coach

Heavy Rail

Car

Walk

Coach

HeavyRail

CarRP Data SCData

Air Air

Speedrail

8

reweighting the mode-specific constants by a known ratio of the sample modal shares to the population modal shares. Without such calibration, the elasticities might be brought into question (see Hensher et al. 2005 for more details). The data are collected by way of a computer assisted programming instrument (CAPI). The current survey, despite being programmed along similar lines of many such surveys we have developed and programmed in the past, uses state of the art methods to generate the SC experiment so that it is meaningful at the individual level. The SC experiment offers a maximum total of seven possible alternative modes for metropolitan trips. These alternatives are taxi, hire car, car, bus, tram, train, and walk. Any one respondent however is limited to choosing amongst a maximum of three alternatives (with a minimum of two). The variation in the number and types of alternatives shown to individual specific respondents is determined by the responses given by respondents early in the survey in terms of the availability of the various alternatives for the recent trip being examined. For example, if the respondent reports not having a car available for the recent trip, then the car alternative will not be present in the SC experiment. Hence, at the commencement of the survey, respondents are asked about a recent trip that they took in which either a taxi or hire car, or both, were possible means of transport for at least part of the trip (see Figure 1). Respondents could therefore select a recent trip which in addition to having a taxi, hire car, or both, as an available modes of transport, at least five other modes were potentially available (see Figure 3).

Figure 3: Alternatives available for a recent trip The selection of what alternatives are to be shown to each respondent proceeded as follows. A respondent who was presented with a forced choice for their recent trip between either a taxi or

9

hire car (that is, they could only take a taxi or a hire car with no other mode possible), were presented with SC scenarios involving the choice between two hypothetical taxis or two hypothetical hire cars (the later scenario was not empirically observed in any of the data segments however). Where two alternatives were reported as being potential modes for the recent trip (one of which was a taxi and/or hire car and only one of which was actually used for the trip, the other representing a non-chosen alternative), the SC scenarios were generated to reflect these two modes. Likewise, for a trip in which three modes were available to the respondent to choose from, these three modes were selected to form the alternatives present within the SC scenarios shown to that respondent. In cases where respondents had more than three alternative modes of transport available to them for the recent trip used to form the context of the SC experiment, the survey instrument selected as one mode either a taxi or hire car as one alternative in the SC experiment, and two of the remaining alternatives from the set as the last two alternatives in the SC scenario. Once the alternatives to be shown in the SC scenarios have been determined, the CAPI survey seeks respondent information, either real or perceived, related to the levels of the relevant alternatives or a recent trip that they undertook. The SC experiment then ‘pivots’ the attribute levels of the various alternatives, where a pivot from the reference trip makes sense. As well as the number and types of alternatives varying across respondents, several attributes may vary also. For example, access and egress attributes relate to different mode possibilities. As with the main mode alternatives, these attributes only appear if the respondent indicates that they are a valid option for the trip being examined. Exactly how analysts distribute the levels of the design attributes over the course of an experiment (which typically is via the underlying experimental design), may play a big part in whether or not an independent assessment of each attribute’s contribution to the choices observed to have been made by sampled respondents can be determined. Conceptually, an experimental design may be viewed as nothing more than a matrix of values that is used to determine what goes where in a SC survey. The values that populate the matrix represent the attribute levels that will be used in the SC survey, whereas the columns and rows of the matrix represent the choice situations, attributes and alternatives of the experiment. For the current survey, the combinations of levels of each attribute in the SC experiment are designed using the latest experimental techniques developed by Rose and Bliemer and implemented in NGene, the software developed by Rose, Bliemer, Collins and Hensher. A D-efficient design is used to structure the SC experiment (see Rose and Bliemer 2009 and NGene (http://www.choice-metrics.com)). This software establishes the combinations of levels of attributes that are shown to each respondent. Although the design is based on percentage variations around a reference base (typically the levels of each attribute experience by a respondent in their current or recent trip), the levels shown to a respondent are in the units that matter in the real market (i.e., dollars or minutes). See Appendix A for more details. Given a lack of prior knowledge as to the precise alternatives faced by a respondent for their specific trip, the SC experiment needed to cater for up to 44 different sets of potential alternative combinations (e.g., taxi versus taxi, taxi versus bus, hire car versus train, taxi versus bus versus tram). Given 44 different potential combinations of alternatives, no single experimental design is possible. As such, it was necessary to build an interface between the survey instrument and NGene

10

whereby the individual specific attribute levels for the recent trip were downloaded to NGene, which generated an individual specific efficient design, which was subsequently fed back to the SC scenarios shown within the survey. This process represents the state of art in survey design. The SC scenarios consist of a number of different choice tasks. Individual respondents are asked to review 12 of the maximum designed choice tasks and to indicate which alternative they prefer under each of the 12 choice scenarios. The exact levels shown to each respondent vary depending on the attribute levels that each respondent reports for a recently undertaken trip. The pivot-related attributes in the choice scenarios associated with each existing mode are pivoted around current perceived levels; however for some attributes such as level of crowding, the range is fixed and unlinked from the current trip experience. The majority of the attributes have five levels, which were found to deliver a sufficiently wide range as well as levels within the range that would be meaningful to respondents, giving us enough variability in the range that future service changes might be delivered. The attributes and attribute levels for the study have been determined in conjunction with the Victorian Taxi Inquiry Team. The existing literature was also consulted to determine relevant attributes and candidate pivot levels. A pilot study tested the logistical aspects implemented for the main field phase of the project, as well to test the operational capabilities of the CAPI software. Results from the pilot have been used to provide priors for constructing the design in the main field phase. An Example choice scenario screen from the final CAPI is shown in Figure 4. The backend of the instrument captures all data in a format that is ready for immediate input into our choice modelling software (Nlogit 5.0).

Figure 4: Example Mode Choice Scenario Screen

Please examine the following information. Below are three alternative modes that you told us about earlier. Consider the modes as shown below and compare these in terms of the times, costs and crowding levels shown.Once you have compared the modes, select your most preferred and least preferred mode out of the those shown.In answering the questions, imagine everything else being the same as when you took the original trip (i.e., same time of day, same weather conditions, etc.). The only difference is the times, costs and crowding of the modes shown.

My most preferred alternative is Bus Train Taxi

My least preferred alternative is Bus Train Taxi

Delay of X minutes every: 1 in XX trips Delay of X minutes every: 1 in XX trips

XX

XX

Travel time

Probab lity of YY minute travel time XX

XX

XX

Probab lity of XX minute travel time XX

Probab lity of XX minute wait for taxi

Probab lity of 0 minute wait for taxi

Probab lity of X minute wait for taxi

Given the above information

Probab lity X of fare: $XX.XXLevel of Crow ding

XX% of seats are occupied, XX people

are standing

Level of Crow dingXX% of seats are

occupied, XX people are standing

$YY.YY

$ZZ.ZZ

Probability of ZZ minute travel time

Total fare for trip $X.XX Total fare for trip $X.XX

Fare Probab lity Y of fare:

Probab lity Z of fare:

Waiting time for the bus XXX Waiting time for the train XXX

Waiting for the

taxi

Time spent on the bus XXX Time spent on the train XXX

Time spent getting from the bus to your destination

XXXTime spent getting from the train to your destination

XXX

Bus Train Taxi

Time to get to the bus XXX Time to get to the train XXX XXXTime taken to get to where the taxi picked you up

Game 1 of XX

11

5. The Modelling Approach Let nsjU denote the utility of alternative j perceived by respondent n in choice situation s. nsjU may be partitioned into two separate components, an observed component of utility, nsjV and a residual unobserved (and un-modelled) component, ,nsjε such that

.nsj nsj nsjU V ε= + (1)

The observed component of utility is typically assumed to be a linear relationship of observed attribute levels, x, of each alternative j and their corresponding weights (parameters), ,β such that

1

,K

nsj nk nsjk nsjk

U xβ ε=

= + (2)

where nkβ represents the marginal utility or parameter weight associated with attribute k for respondent n and the unobserved component, ,nsjε is assumed to be independently and identically (IID) extreme value type 1 (EV1) distributed with 0.57721nsjE ε = and ( ) ,

6var

2

2

nnsj σ

πε = wherenσ represents a positive scale factor that is typically normalised to one in most applications. As well as containing information on the levels of the attributes, x in Equation (2) may also contain up to J-1 alternative specific constants (ASCs) capturing the residual mean influences of the unobserved effects on choice associated with their respective alternatives; where x takes the value 1 for the alternative under consideration or zero otherwise. The utility specification in Equation (2) is flexible in that it allows for the possibility that different respondents may have different marginal utilities for each attribute being modelled. Unfortunately, in practice it is not generally feasible to estimate individual specific parameter weights. As such, it is typical to estimate parameter weights for the population moments of the sample, such that ignoring subscript j,

,nskknk zηββ += (3) where kβ represents the mean or some other measure of central tendency for the distribution of marginal utilities held by the sampled population and kη represents a deviation or spread of preferences amongst sampled respondents around the mean (or other measure of central tendency) marginal utility. The resulting model structure is based on integrals without a close form solution. In simulation, zns in Equation (3) represents random draws taken from a pre-specified distribution for each respondent n and choice task s. Rather than assuming that the marginal utility has some distribution over both n and s as dictated by zns, an alternative model specification allows

12

for a distribution over only n such that zns becomes zn. In this version of the model, preferences are assumed to vary between individuals but not within given a sequence of observed choices. The assumption that preferences vary between and not within respondents accounts for the pseudo panel nature of SC data (Train 2009). Within the literature, when zns is employed, the resulting model is known as a cross sectional discrete choice model, whilst zn produces what is referred to as a panel discrete choice model as it takes into account the pseudo panel nature of repeated choice observations. The distinction between the cross sectional and panel specifications of the model lie not only in how the draws are taken, but also in how the log-likelihood function of the two models are set-up. In the cross-sectional version of the model, the choices made over choice tasks, S, are assumed to be independent, both within and between individual respondents resulting in the following simulated log-likelihood function ( )

1 1 1

log ( ) log ,N S J

nsj nsjn s j

E L y E P= = =

= (4) where nsjy equals one if alternative j is the chosen alternative in choice situation s shown to respondent n, and zero otherwise, and ( )nsjE P is the expected choice probabilities calculated over draws zns. In the panel version of the model, the choice tasks, S, are no longer assumed to be independent and the simulated log-likelihood function of the model becomes

( )*

1

log ( ) log .N

nn

E L E P=

= (5a) where

( )*

1 1

.nsj

S J y

n nsjs j

P P∈ ∈

= ∏∏ (5b) and where the draws are now taken only over n. See Train (2009) for a more in-depth discussion of the differences between these models.

6. The Pilot Study A pilot study was undertaken to test the survey instrument and identify any issues with the survey process. A pilot briefing took place in Melbourne on Wednesday 1st February, with the pilot undertaken over the period 4th to 9th February. A total of 34 full surveys were undertaken to test the logistics associated with the approved interview locations as well as the effectiveness of iPads as a data collection technology. The data collection process entailed the interviewer accessing the survey instrument from a remote server located at the URL

13

http://surveyitls.econ.usyd.edu.au/victaxi/survey.php. In addition to the survey instrument URL used by interviewers, we also have a URL that the study team can use to monitor the data being collected an URL for survey monitoring, http://surveyitls.econ.usyd.edu.au/victaxi/dashboard/ (which requires a user name and password to access). Of the 35 interviews, only one failed, due to a respondent indicating a travel time of 480 minutes which was outside the Melbourne travel time range. It turned out that this respondent had driven from Mildura. We subsequently issued a revised screener instruction to the interviewers that only trips that begin and finish in the Greater Melbourne Metropolitan Area are candidate trip contexts. The average survey time was 22 minutes, 21 were business trips, six tourists, six day trips for non-business locals, and one other (The one ‘other trip’ was to go to the airport as a non-business trip). Six out of the 34 respondents travelled after 7pm.; of these six, five were for business trips and one was for day to day activities. The model estimated on the pilot data is a multinomial logit (MNL) model with alternative (or mode) specific constants (ASCs) estimated with respect to taxi. Given limited sample size, all parameters have been treated as generic across the alternatives, with a few exceptions. Walking time (as a main mode) was estimated as an alternative specific parameter estimate and two fare parameters were estimated, one generic across the public transport alternatives and the second generic across the taxi and hire car alternatives. The model results are presented in Table 3. Table 3: Pilot Data set Model

Par. (t-rat.) Constants (w.r.t Taxi)Bus -1.153 (-1.20) Tram 1.818 (1.92)Train 1.420 (1.35)Car -1.137 (-2.13) Hire Car 0.094 (0.39)

ParametersAccess Time -0.114 (-3.63) Wait Time -0.034 (-0.80) INV Time -0.032 (-3.06) Egress Time -0.028 (-2.02) Walk (main mode) time -2.563 (-3.55) Number of people sitting -0.887 (-0.98) Number of people standing -0.031 (-2.27) Delay (mins × 1 out of every # trips) 1.070 (2.05)PT fare -0.062 (-2.20) Car costs (Fuel + Toll + Parking) -0.014 (-2.27) Taxi/Hire Car fare -0.042 (-5.92) Model fitLL(ASC) -281.107LL(β) -236.205ρ2 0.160Adj. ρ2 0.141# Respondents 34# choice observations 408The only parameter of concern is the delay parameter. The delay attribute consists of two components; a time component representing how long the delay lasts on average and a frequency of occurrence (a delay occurs once every # trips). The delay variable was constructed as time × occurrence). Hence, a larger value of the delay variable implies either a longer delay, that delays

14

occur more frequently, or both. As such, the delay parameter would be expected to be negative as opposed to what is observed in the model. This is likely however to be a function of the very small sample size. All other parameters are of the expected sign. 7. Main Field Phase Sample Design

Whilst the theory of sampling, as related to discrete choice models, is well developed for revealed preference (RP) data (see Louviere et al. 2000), until recently, little was known about sample size requirements for SC data. Experience suggests however that theory is often discarded in any case for more practical considerations such as issues related to budget and time. Hensher et al. (2005) report that in the experience of many, the minimum sample size requirement for discrete choice experiments is 50 respondents per alternative modelled. Given that each respondent in an SP choice experiment undertakes multiple choices over a variety of choice sets, each sampled respondent in reality provides multiple data observations. For example, assuming an experiment with three alternatives, a total of 150 respondents should be sampled from the population. Assuming each respondent is given 12 choice sets, the total number of observations for the study will be 1,800. More recently however, Bliemer and Rose (2010) have developed specific theory to calculate precise sample size requirements for SC experiments. This sample size theory relates to generating SC experiments that will produce robust estimates for models estimated using data collected in small samples. Although the theory and resultant methods can be applied to large populations, the sample size calculations used become less important. As such, given the scale of the proposed study, it is suggested that a combination of sampling theories be applied. Firstly, it is suggested that more traditional sampling approaches be used for purposes of market segmentation, and that the methods developed by Bliemer and Rose (2010) be applied within each segment. This will ensure not only robust estimates for forecasting, but the ability to generalise the results to the wider population. Specifically, we propose segmenting the market along geographical lines combined with a choice based sampling approach. For the current study, we use a quota based sampling. The use of quota based sampling allows for robust parameter estimates across all segments given that respondents of all types are represented in the data. Whilst random sampling is often preferred to quota sampling, if one wishes to estimate models allowing for say, long distance trips, and such respondents are relatively rare within the population, then no such model can be estimated, or where such a model is capable of being estimated, the parameter estimates will not be robust, and hence the resulting model outputs less useful. The collection of quota based samples not only allows for the estimation of robust parameter estimates, but if required, the data can be re-weighted to match external data on known population distributions. As such, quota based sampling represents the best sampling method for the current study proposal. The original proposal called for sample size of 505 respondents to be collected, spread across four use segment quotas and within each segment to ensure that we have enough observations to obtain a good spread of trip time (short, medium and long trips) and time of day (weekday, weekend and evening) conditions. The breakdown of the original segments is shown in Table 4. Note however

15

that models will be estimated within each use segment, pooling the data by trip length and time of day. Within each use segment, we ensure a good spread of the sampled recent trip across the modal groups (i) car (ii), public transport (bus, tram, train), (iii) taxi, (iv) hire car, and (v) walk. An example of the recruitment script used is provided in Appendix C. Table 4: Quota segments City: Crowne, Hilton, Hyatt, Flinders St Stn and Southern Cross Stn Regional Centres: Chadstone Shopping Centre, Chapel St., Doncaster shopping Centre

135 Corporate Users (Anyone on business-related travel at any time of day (24hrs) 7 days a week)Location Airport 65City 25Regional Centres 45

135 International and Domestic Tourists (including Local non-work related trips) Trips at any time of day (24hrs) 7 days a week of persons visiting (or living in) Melbourne for non

business related activities. Location Airport 45City 45Regional Centres 45

135 Late Night Transport (Anyone who is not in the categories above who live in Melbourne who are out and about seeking transport after 10pm and up to 4am in morning)

Location Airport 25City 65Regional Centres 45100 Holder of multi-purpose trip program (MPTP) cards Location City 50Regional Centres 50Table 5 presents the final sample sizes by segment achieved. Although the previously defined ‘tourist segment’ was to originally include both tourists travelling to/from Melbourne who used a taxi and or/hire car as well as members of the public who travelled to Melbourne for some specific activity such as shopping or education, at the time of modelling, it was found that the preferences for these two groups were markedly different and hence the two were split into two separately defined segments. As can be seen from Table 5, the number of respondents for the business and night time travel segments matched the quotas required, whilst the number of respondents sampled under the previously defined International and Domestic Tourists (including Local non-work related trips) segment exceeded the 135 required sample size. The final MPTP card holder segment however fell well short of the 100 respondents, with only 39 interviews concluded at the time of the writing of this report. The total sample size collected represented 497 respondents. Extensive data cleaning was undertaken of the data to ensure that only representative and plausibly meaningful trips were used in the final model estimation. For example, data for one respondent was deleted due to the fact that their reported trip travel time for one mode was less than five minutes compared to a travel time of 50 minutes for an alternative competing mode, whilst a second respondent was deleted from the analysis due to a hire car fare of $2 compared to a taxi fare of $30 for the same trip. The final sample sizes of effective interviews are shown in Table 5.

16

Table 5: Final sample sizes by segments Tourism Segment Day to Day Activity Segment (Combined Tourist and day to day

activities segments)

Business Segment Night Time Travel Segment MPTP Card Holder Segment Total Sample

Size Collected 71 117 (188) 135 135 39 497 Effective interviews 65 112 (177) 128 119 39 463

8. Empirical Data The main field survey was undertaken between 11th February and 3rd March 2012, sampling travellers in the Melbourne Metropolitan area. A total of 463 effective interviews were undertaken in the five segments (see Table 5). The 463 interviews represent 5,556 choice observations for model estimation (i.e., 463×12 treatments).

8.1 Socio-Demographic Descriptive Statistics by Segment Descriptive statistics for each of the five segments are presented in Table 6. Overall, the average age of respondents for the general day to day activity and night time travel segments were 10 years lower than the tourist and business travel segments and 20 years younger than the average age of the MPTP card holder segment. As is to be expected, the average income of the Business segment was the highest out of all of the segments and the MPTP card holder segment the lowest. With regards to gender, the Business segment was skewed towards males whilst the MPTP card holder segment had a larger proportion of females. The remaining segments were almost evenly split between male and female. Table 6: Descriptive statistics of socio-demographic characteristics of final sample Tourist Segment Day to Day Travel Activity Segment Business Segment Night time travel Segment MPTP card holder segment

General informationAge (years) 42.4 33.95 43.53 34.99 55.59 Income ($000 per annum) 48.48 52.2 114.9 48.02 26.67 Gender (1 = female) 58.46% 42.86% 23.44% 45.38% 74.36%Has a drivers licence 92.31% 90.18% 90.63% 80.67% 56.41%EmploymentFull Time 35.38% 58.04% 93.75% 44.54% 12.82%Part Time 18.46% 9.82% 3.13% 22.69% 10.26%Casual 12.31% 14.29% 3.13% 15.97% 7.69% Not Employed 33.85% 17.86% 0.00% 16.81% 69.23%

8.2 Modal Travel Time and Cost Descriptive Statistics by Segment Table 7 provides descriptive statistics for the mode splits and mode travel times and costs broken down by travel segment. Across all data segments, respondents reported having only one or two modes available to them for the specific recent trip used to generate the SC experiment. Between hire car and taxi, taxi dominated in all travel segments, with hire car being an available alternative in only 0.72 percent of the cases for the tourist segment, 0.85 percent of the cases for the general day to day travel activity segment, and 1.22 and 2.44 percent of the cases for the night time and MPTP card holder segments respectively. For the business travel segment however, the hire car

17

was available in 9.54 percent of the available set of alternatives. Note that for the MPTP card holder segment, walk was never selected as an available alternative. Table 7: Descriptive statistics of socio-demographic characteristics of final sample Tourist Segment Day to Day Travel Activity Segment Business Segment Night time travel Segment MPTP card holder segment

Number of alternatives present in recent trip1 or 2 Alternatives 56 101 122 111 353 + Alternatives 9 11 6 8 4Number of times an alternative was availableBus 26 11 38 6 3Tram 9 12 7 25 2Train 6 13 2 16 6Car 18 54 45 35 9Walk 6 12 5 18 0Taxi 73 131 140 143 60Hire Car 1 2 25 3 2

Total 139 235 262 246 82Model Specific Travel Times and Costs

Access Times (mins)Bus 6.14 6.55 5.94 5.11 2.7Tram 4.45 1.89 4.92 3.87 1.89 Train 9.04 17.52 10.6 6.83 17.52Taxi 2.09 2.05 2.28 3.13 4.84Hire Car 2.03 1.87 2.78 0.91 20.69Waiting Times (mins)Bus 4.72 4.58 4.68 4.58 6.27Tram 4.64 4.64 4.88 6 4.41Train 4.61 4.84 5.15 4.62 9.61Taxi 7.58 7.56 7.53 7.63 7.55Hire Car 7.52 7.54 7.85 7.43 7.3

Main Mode Travel Times (mins)Bus 30.87 35.37 30.46 19.89 86.56Tram 16.47 31.34 28.47 21.91 37.63Train 27.95 31.63 28.09 26.4 27.16Car 34.45 37.51 36.42 28.77 34.28Walk 31.51 14.97 31.06 27.22 -Taxi 27.65 35.82 29.84 22.1 31.19Hire Car 35.64 38.74 40.95 30.93 50.44Egress Travel Times (mins)Bus 4.56 4.75 5.72 8.1 20.93Tram 2.42 7 4.98 6.07 11Train 6.47 5.04 6.91 7.95 4.34Car 2.53 2.86 2.6 1.94 5.88

Trip Specific Costs ($)Bus (fare) $15.65 $14.75 $10.14 $8.80 $11.23Tram (fare) $4.60 $3.44 $3.81 $3.51 $2.39Train (fare) $6.45 $5.88 $4.38 $5.34 $3.29Taxi (fare) $43.62 $51.35 $40.40 $30.13 $23.25Hire Car (fare) $62.56 $69.38 $65.00 $52.83 $70.94Car (petrol costs) $2.74 $2.96 $2.78 $2.28 $1.57Car (toll costs) $3.58 $3.82 $3.03 $2.37 $2.17Car (parking costs) $2.58 $16.02 $12.21 $21.20 $21.20 Table 7 also reports the mean travel times and costs by mode and segment. With the exception of the MPTP card holder segment, the average time taken to access a taxi or hire car was less than for

18

all other modes, however the average reported waiting time (i.e., time spent at a taxi rank, bus stop or station, or after the time a taxi or hire car was booked) for each travel segment was longer for taxi or hire car than for any other mode of transport. For the MPTP card holder segment, the access times for taxi and hire car were reported as being substantially longer than for other segments, however waiting times appear to be similar to those reported for other segments. For all segments, the hire car was reported to have the largest average main mode vehicle time was reported with the exception of the average bus time reported by the MPTP card holder segment. The average taxi mode was reported as being similar to other main mode vehicle times for all segments. This suggests that hire cars are typically considered for longer trips than other modes. Again, as is to be expected, the fares for taxi and hire cars are on average substantially larger than for all other modes, with hire car fares being greater than for taxis. This again, suggests that within the sample, hire cars are mainly seen as an alternative for longer trips. 8.3 Descriptive Statistics by Segment of Altitudinal data At the conclusion of the 12 SC scenarios, respondents were also asked to complete a series of questions related to their attitudes towards the various public transport, taxis and hire car modes (see Figures 5 and 6).

Figure 5: CAPI screen capturing attitudes towards taxis and hire car modes In order to capture these attitudes, we used a derivation of the Fishbein multi-attribute model which states that the attitudes towards an object is a function of both the strength of the belief that some object has a particular attribute and the evaluation of the goodness or badness of the attribute

19

(Fishbien 1972). In the current context, we replace the evaluation component of the attribute within the model with a question related to the importance that attribute has for the respondent and the strength of the belief that some object has a particular attribute with a rating of the respondents satisfaction towards that object.

Figure 6: CAPI screen capturing attitudes towards public transport modes Although respondents answered both importance and satisfaction questions using a traditional 1 to 5 point likert scale, as with the original Fishbein multi-attribute model, we convert the satisfaction scale to range between -2 (formally rated 1) and 2 (formally rated 5). For each attribute or item being measured, the attitude then becomes simply the product of the importance scale ranking by the newly converted satisfaction scale. In this way, the overall satisfaction for an attribute or item will range between -10 and 10 (5×-2 to 5×2) with a mean of zero representing indifference towards that attribute, whilst negative values represent a negative attitude and positive values positive attitudes. Tables 8 and 9 present the results of these attitudinal questions broken down by mode and travel segment. For each item or attribute, mean and standard deviations are presented, alongside the minimum and maximum attitude reported.

20

Table 8: Public Transport attitudinal results by item

Tourism Segment Bus Tram Train Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Cleanliness 1.723 3.731 (3.72) 6.15% -10 10 1.538 4.294 (2.89) 10.77% -10 10 1.800 4.097 (3.54) 7.69% -10 10 Air conditioning 2.538 3.437 (5.95) 0.00% 0 10 2.015 3.629 (4.48) 4.62% -4 10 1.523 3.675 (3.34) 7.69% -8 10 Personal safety 3.923 3.650 (8.67) 0.00% 0 10 1.323 4.701 (2.27) 15.38% -10 10 3.015 4.980 (4.88) 7.69% -10 10 Exposure to weather 1.877 3.693 (4.10) 6.15% -10 10 2.246 4.074 (4.45) 6.15% -6 10 1.954 3.573 (4.41) 6.15% -8 10 Shops in precinct 1.892 2.862 (5.33) 1.54% -1 10 1.031 2.953 (2.81) 6.15% -6 10 1.662 2.879 (4.65) 1.54% -3 10 Seating comfort 1.862 3.933 (3.82) 6.15% -10 10 1.662 4.139 (3.24) 9.23% -8 10 2.462 3.632 (5.46) 4.62% -5 10 Getting a seat 2.523 3.910 (5.20) 4.62% -5 10 0.308 4.454 (0.56) 21.54% -10 10 2.615 4.599 (4.59) 7.69% -10 10 Crowding at station/stop 2.231 2.962 (6.07) 0.00% 0 10 1.185 3.230 (2.96) 6.15% -10 10 2.015 3.044 (5.34) 3.08% -3 10 Crowding in the vehicle 3.554 3.733 (7.67) 4.62% -5 10 2.692 4.058 (5.35) 7.69% -8 10 1.046 3.958 (2.13) 16.92% -10 10 Travel in a tunnel 1.400 2.530 (4.46) 4.62% -5 10 0.800 2.605 (2.48) 10.77% -5 10 0.785 2.446 (2.59) 10.77% -6 10 Graffiti free 1.969 3.298 (4.81) 1.54% -4 10 1.492 2.705 (4.45) 4.62% -5 10 1.154 3.456 (2.69) 7.69% -8 10 Noise level 2.446 3.226 (6.11) 6.15% -3 10 1.831 3.219 (4.59) 9.23% -5 10 1.338 3.285 (3.29) 10.77% -6 10 Business Segment Bus Tram Train Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Cleanliness 0.938 3.250 (3.26) 7.81% -8 10 0.367 3.565 (1.17) 10.94% -10 10 0.914 3.663 (2.82) 7.81% -10 10 Air conditioning 1.156 2.852 (4.59) 2.34% -5 10 0.930 2.591 (4.06) 2.34% -5 10 0.648 3.024 (2.43) 6.25% -10 10 Personal safety 1.727 3.569 (5.47) 3.91% -10 10 0.789 3.651 (2.44) 10.16% -10 10 1.484 3.960 (4.24) 6.25% -10 10 Exposure to weather 1.086 3.043 (4.04) 3.91% -10 10 0.750 4.053 (2.09) 9.38% -10 10 0.914 2.924 (3.54) 5.47% -8 10 Shops in precinct 0.594 1.667 (4.03) 3.13% -3 8 0.484 2.358 (2.32) 3.91% -10 10 0.563 1.561 (4.08) 0.78% -2 10 Seating comfort 0.406 3.168 (1.45) 10.94% -10 10 0.789 3.235 (2.76) 6.25% -8 10 0.930 2.753 (3.82) 7.03% -10 10 Getting a seat 1.031 2.828 (4.13) 6.25% -10 10 -0.141 3.045 (-0.52) 15.63% -10 10 0.797 3.242 (2.78) 9.38% -8 10 Crowding at station/stop 1.195 2.593 (5.21) 0.78% -5 10 0.867 2.533 (3.87) 3.13% -6 10 1.000 2.629 (4.30) 2.34% -6 10 Crowding in the vehicle 1.875 3.275 (6.48) 3.13% -5 10 0.875 3.823 (2.59) 10.94% -10 10 1.141 3.250 (3.97) 7.81% -8 10 Travel in a tunnel 0.719 1.734 (4.69) 3.13% -5 10 0.336 1.523 (2.50) 5.47% -5 10 0.227 1.712 (1.50) 9.38% -8 10 Graffiti free 0.938 2.296 (4.62) 2.34% -3 10 0.664 2.210 (3.40) 1.56% -8 10 0.750 1.920 (4.42) 2.34% -2 10 Noise level 1.203 3.013 (4.52) 5.47% -8 10 0.891 3.025 (3.33) 7.03% -10 10 0.813 3.196 (2.88) 8.59% -10 10

21

Table 8: Public Transport attitudinal results by item (cont’d) Day to Day Activity Segment Bus Tram Train Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Cleanliness 2.098 3.654 (6.08) 10.71% -5 10 1.259 4.512 (2.95) 19.64% -10 10 2.268 4.800 (5.00) 10.71% -10 10 Air conditioning 1.634 3.241 (5.34) 4.46% -8 10 1.286 3.402 (4.00) 8.93% -8 10 1.304 3.631 (3.80) 7.14% -10 10 Personal safety 2.991 3.714 (8.52) 3.57% -5 10 0.946 4.223 (2.37) 19.64% -10 10 2.750 4.946 (5.88) 8.04% -10 10 Exposure to weather 3.179 3.818 (8.81) 4.46% -10 10 1.911 4.940 (4.09) 13.39% -10 10 1.714 3.362 (5.40) 9.82% -6 10 Shops in precinct 1.420 2.713 (5.54) 4.46% -5 10 0.929 2.605 (3.77) 4.46% -8 10 1.161 2.538 (4.84) 5.36% -6 10 Seating comfort 0.750 4.046 (1.96) 20.54% -10 10 0.875 4.045 (2.29) 12.50% -10 10 1.884 3.176 (6.28) 6.25% -10 10 Getting a seat 2.670 4.048 (6.98) 9.82% -10 10 -0.134 4.533 (-0.31) 33.93% -10 10 1.723 4.056 (4.50) 16.96% -6 10 Crowding at station/stop 2.107 3.041 (7.33) 1.79% -4 10 1.232 3.288 (3.97) 6.25% -10 10 1.482 2.806 (5.59) 2.68% -4 10 Crowding in the vehicle 3.295 3.499 (9.96) 2.68% -5 10 1.964 4.036 (5.15) 7.14% -10 10 1.125 3.343 (3.56) 15.18% -5 10 Travel in a tunnel 1.866 2.506 (7.88) 6.25% -6 10 0.563 2.265 (2.63) 17.86% -8 10 0.482 2.185 (2.34) 18.75% -8 10 Graffiti free 1.134 3.284 (3.65) 9.82% -10 10 0.929 1.999 (4.92) 3.57% -3 10 0.813 2.566 (3.35) 6.25% -8 10 Noise level 2.438 3.541 (7.29) 6.25% -6 10 1.375 2.758 (5.28) 7.14% -10 10 1.536 2.819 (5.77) 8.93% -5 10 Night Time Travel Segment Bus Tram Train Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Cleanliness 1.067 3.746 (3.11) 13.45% -10 10 0.840 4.109 (2.23) 14.29% -10 10 1.689 4.610 (4.00) 11.76% -10 10 Air conditioning 2.521 3.405 (8.08) 1.68% -5 10 1.655 3.557 (5.08) 7.56% -10 10 1.361 4.106 (3.62) 14.29% -10 10 Personal safety 2.345 3.758 (6.80) 5.04% -10 10 0.370 4.690 (0.86) 19.33% -10 10 3.185 4.951 (7.02) 7.56% -10 10 Exposure to weather 2.118 3.915 (5.90) 9.24% -10 10 1.521 4.483 (3.70) 10.92% -10 10 1.370 3.505 (4.26) 8.40% -10 10 Shops in precinct 1.681 2.574 (7.12) 4.20% -5 10 1.008 2.854 (3.85) 8.40% -8 10 1.538 2.466 (6.80) 3.36% -4 10 Seating comfort 0.807 3.423 (2.57) 12.61% -10 10 0.933 3.995 (2.55) 10.92% -10 10 1.042 3.171 (3.58) 11.76% -10 10 Getting a seat 1.866 3.513 (5.79) 6.72% -10 10 0.118 3.825 (0.34) 22.69% -10 10 0.899 4.045 (2.42) 17.65% -10 10 Crowding at station/stop 2.395 3.190 (8.19) 1.68% -3 10 1.773 3.222 (6.00) 5.04% -10 10 1.639 3.137 (5.70) 4.20% -8 10 Crowding in the vehicle 2.479 3.647 (7.41) 5.04% -10 10 1.765 3.965 (4.85) 10.08% -10 10 0.076 3.683 (0.22) 17.65% -10 10 Travel in a tunnel 1.261 2.188 (6.28) 2.52% -10 10 0.681 2.470 (3.01) 10.08% -10 10 0.580 2.334 (2.71) 10.92% -10 10 Graffiti free 1.672 2.882 (6.33) 3.36% -6 10 1.529 2.861 (5.83) 2.52% -5 10 1.286 2.832 (4.95) 5.88% -10 10 Noise level 1.899 3.307 (6.26) 4.20% -10 10 1.092 3.141 (3.79) 7.56% -10 10 0.311 3.124 (1.09) 14.29% -10 10

22

Table 8: Public Transport attitudinal results by item (cont’d) MPTP Card Holder Segment Bus Tram Train Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Cleanliness 1.462 4.745 (1.92) 15.38% -10 10 1.487 5.414 (1.72) 23.08% -10 10 1.154 5.076 (1.42) 17.95% -10 10 Air conditioning 2.128 4.137 (3.21) 7.69% -5 10 1.795 4.691 (2.39) 12.82% -8 10 0.590 4.399 (0.84) 12.82% -10 10 Personal safety 1.615 3.544 (2.85) 2.56% -2 10 1.385 4.296 (2.01) 7.69% -10 10 0.846 3.249 (1.63) 5.13% -10 10 Exposure to weather 3.077 4.094 (4.69) 5.13% -4 10 2.974 5.299 (3.51) 10.26% -10 10 3.154 4.976 (3.96) 10.26% -8 10 Shops in precinct 2.051 4.186 (3.06) 10.26% -10 10 1.128 4.432 (1.59) 15.38% -10 10 1.513 4.559 (2.07) 12.82% -10 10 Seating comfort 0.949 3.612 (1.64) 7.69% -8 10 0.308 4.485 (0.43) 12.82% -10 10 1.308 4.181 (1.95) 7.69% -10 10 Getting a seat 3.128 4.219 (4.63) 7.69% -2 10 1.410 4.470 (1.97) 20.51% -6 10 3.000 4.353 (4.30) 5.13% -5 10 Crowding at station/stop 2.487 4.154 (3.74) 7.69% -5 10 1.410 4.387 (2.01) 12.82% -10 10 1.538 4.962 (1.94) 15.38% -10 10 Crowding in the vehicle 1.667 3.723 (2.80) 5.13% -5 10 0.615 4.165 (0.92) 10.26% -10 10 1.179 4.103 (1.80) 7.69% -10 10 Travel in a tunnel 2.154 3.617 (3.72) 7.69% -4 10 1.590 3.393 (2.93) 15.38% -3 10 1.872 3.607 (3.24) 10.26% -5 10 Graffiti free 2.026 4.023 (3.14) 5.13% -6 10 1.026 4.283 (1.50) 15.38% -10 10 1.795 4.066 (2.76) 10.26% -5 10 Noise level 0.692 3.213 (1.35) 7.69% -8 10 1.103 3.251 (2.12) 7.69% -5 10 1.154 3.551 (2.03) 5.13% -8 10

23

Table 9: Taxi and hire Car attitudinal results by item Tourism SegmentTaxi Hire CarMean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Internal cleanliness of vehicle 6.431 2.549 (20.34) 0.00% 0 10 5.277 1.924 (22.11) 0.00% 0 10 Air conditioning 5.400 3.201 (13.60) 0.00% 0 10 4.431 2.250 (15.88) 0.00% 0 10 Mechanical state of vehicle 6.400 3.045 (16.94) 1.54% -5 10 5.246 2.165 (19.53) 1.54% -3 10 Attitude of driver 5.646 2.912 (15.63) 1.54% -3 10 5.277 2.322 (18.32) 1.54% -3 10 External cleanliness of vehicle 3.138 4.930 (5.13) 12.31% -10 10 2.692 3.762 (5.77) 12.31% -6 10 Personal safety 7.308 3.122 (18.87) 1.54% -6 10 5.615 2.177 (20.79) 1.54% -6 10 Exposure to weather 4.538 4.070 (8.99) 4.62% -6 10 3.462 2.932 (9.52) 4.62% -6 8 Driving knowledge of route 6.677 2.676 (20.12) 0.00% 0 10 5.846 1.481 (31.82) 0.00% 0 10 Seat comfort 5.385 3.320 (13.08) 1.54% -5 10 4.354 2.308 (15.21) 1.54% -3 10 Seating condition 5.923 2.922 (16.34) 0.00% 0 10 4.831 2.140 (18.20) 0.00% 0 10 Driver presentation 4.692 3.992 (9.48) 4.62% -10 10 4.077 3.144 (10.45) 4.62% -6 10 Vehicle is disable friendly 2.354 4.587 (4.14) 16.92% -8 10 1.938 4.290 (3.64) 16.92% -6 6 English ability of driver 4.785 3.069 (12.57) 1.54% -4 10 4.354 2.328 (15.08) 1.54% -3 10 Driver driving ability 6.769 2.714 (20.11) 0.00% 0 10 5.785 1.709 (27.28) 0.00% 0 10 Ease of booking 6.138 3.427 (14.44) 1.54% -6 10 4.923 2.489 (15.94) 1.54% -6 10 Business SegmentTaxi Hire CarMean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Internal cleanliness of vehicle 3.477 4.674 (8.42) 9.38% -10 10 1.734 3.619 (5.42) 0.00% 0 10 Air conditioning 4.039 5.158 (8.86) 8.59% -10 10 1.727 3.613 (5.41) 0.00% 0 10 Mechanical state of vehicle 3.938 4.857 (9.17) 7.81% -10 10 1.773 3.661 (5.48) 0.00% 0 10 Attitude of driver 2.438 5.591 (4.93) 17.19% -10 10 1.766 3.581 (5.58) 0.00% 0 10 External cleanliness of vehicle 3.258 4.077 (9.04) 6.25% -10 10 1.547 3.365 (5.20) 0.00% 0 10 Personal safety 5.008 5.423 (10.45) 8.59% -10 10 1.867 3.759 (5.62) 0.00% 0 10 Exposure to weather 4.156 4.657 (10.10) 4.69% -10 10 1.391 3.205 (4.91) 0.78% -2 10 Driving knowledge of route 1.891 6.618 (3.23) 24.22% -10 10 1.695 3.479 (5.51) 0.00% 0 10 Seat comfort 3.633 4.056 (10.13) 3.13% -10 10 1.664 3.412 (5.52) 0.00% 0 10 Seating condition 3.391 4.382 (8.75) 7.03% -10 10 1.781 3.596 (5.60) 0.00% 0 10 Driver presentation 2.148 5.409 (4.49) 15.63% -10 10 1.563 3.343 (5.29) 0.00% 0 10 Vehicle is disable friendly 0.883 4.015 (2.49) 20.31% -10 10 0.805 2.845 (3.20) 3.13% -8 10 English ability of driver 1.344 5.703 (2.67) 21.09% -10 10 1.602 3.349 (5.41) 0.00% 0 10 Driver driving ability 2.336 6.209 (4.26) 19.53% -10 10 1.836 3.758 (5.53) 0.00% 0 10 Ease of booking 4.555 5.386 (9.57) 7.03% -10 10 1.852 3.752 (5.58) 0.00% 0 10

24

Table 9: Taxi and hire Car attitudinal results by item (cont’d) Day to Day Activity SegmentTaxi Hire CarMean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Internal cleanliness of vehicle 3.321 4.497 (7.82) 8.04% -10 10 1.170 3.102 (3.99) 0.00% 0 10 Air conditioning 4.313 3.713 (12.29) 0.89% -4 10 1.170 3.078 (4.02) 0.00% 0 10 Mechanical state of vehicle 3.946 4.069 (10.26) 2.68% -10 10 1.196 3.156 (4.01) 0.00% 0 10 Attitude of driver 2.027 4.728 (4.54) 12.50% -10 10 1.116 2.998 (3.94) 0.00% 0 10 External cleanliness of vehicle 3.080 3.441 (9.47) 3.57% -4 10 1.018 2.728 (3.95) 0.00% 0 10 Personal safety 4.866 4.905 (10.50) 5.36% -10 10 1.295 3.339 (4.10) 0.00% 0 10 Exposure to weather 4.411 4.006 (11.65) 1.79% -3 10 1.063 2.908 (3.87) 0.00% 0 10 Driving knowledge of route 1.750 5.468 (3.39) 21.43% -10 10 1.063 2.948 (3.81) 0.00% 0 10 Seat comfort 4.009 3.745 (11.33) 2.68% -5 10 1.054 2.856 (3.90) 0.00% 0 10 Seating condition 4.018 4.076 (10.43) 4.46% -10 10 1.196 3.125 (4.05) 0.00% 0 10 Driver presentation 2.571 4.347 (6.26) 9.82% -10 10 1.063 2.826 (3.98) 0.00% 0 10 Vehicle is disable friendly 2.098 3.961 (5.61) 10.71% -5 10 0.732 2.515 (3.08) 1.79% -2 10 English ability of driver 1.268 4.554 (2.95) 18.75% -10 10 1.045 2.788 (3.97) 0.00% 0 10 Driver driving ability 2.813 4.964 (6.00) 11.61% -10 10 1.143 3.055 (3.96) 0.00% 0 10 Ease of booking 3.643 5.197 (7.42) 9.82% -10 10 1.080 2.966 (3.86) 0.00% 0 10 Night Time Travel SegmentTaxi Hire Car Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Internal cleanliness of vehicle 3.966 4.141 (10.45) 4.20% -5 10 1.319 3.286 (4.38) 0.00% 0 10 Air conditioning 4.269 3.837 (12.14) 1.68% -3 10 1.269 3.132 (4.42) 0.00% 0 10 Mechanical state of vehicle 4.580 4.494 (11.12) 3.36% -8 10 1.294 3.237 (4.36) 0.00% 0 10 Attitude of driver 2.580 4.501 (6.25) 10.08% -10 10 1.286 3.184 (4.40) 0.00% 0 10 External cleanliness of vehicle 3.277 3.800 (9.41) 4.20% -10 10 1.235 3.118 (4.32) 0.00% 0 10 Personal safety 5.336 4.383 (13.28) 1.68% -5 10 1.294 3.216 (4.39) 0.00% 0 10 Exposure to weather 4.261 3.999 (11.62) 3.36% -4 10 1.118 3.020 (4.04) 0.00% 0 10 Driving knowledge of route 2.210 5.483 (4.40) 16.81% -10 10 1.210 3.033 (4.35) 0.00% 0 10 Seat comfort 3.933 3.642 (11.78) 0.84% -4 10 1.193 3.062 (4.25) 0.00% 0 10 Seating condition 3.681 3.675 (10.92) 1.68% -5 10 1.277 3.154 (4.42) 0.00% 0 10 Driver presentation 2.916 4.331 (7.34) 9.24% -10 10 1.126 2.956 (4.16) 0.00% 0 10 Vehicle is disable friendly 2.328 4.691 (5.41) 8.40% -10 10 0.630 2.350 (2.93) 0.84% -5 10 English ability of driver 1.185 5.030 (2.57) 22.69% -10 10 1.252 3.090 (4.42) 0.00% 0 10 Driver driving ability 3.025 5.115 (6.45) 11.76% -10 10 1.210 3.105 (4.25) 0.00% 0 10 Ease of booking 4.193 5.007 (9.14) 7.56% -10 10 1.193 3.114 (4.18) 0.00% 0 10

25

Table 9: Taxi and hire Car attitudinal results by item (cont’d) MPTP Card Holder SegmentTaxi Hire Car Mean Std dev. (t-ratio) % -ve Min Max Mean Std dev. (t-ratio) % -ve Min Max Internal cleanliness of vehicle 3.958 4.600 (9.39) 20.51% -10 10 1.202 3.150 (4.16) 2.56% -1 10 Air conditioning 4.286 4.078 (11.46) 10.26% -10 10 1.109 3.014 (4.02) 5.13% -2 10 Mechanical state of vehicle 4.479 4.818 (10.14) 20.51% -10 10 1.126 3.035 (4.05) 5.13% -2 10 Attitude of driver 3.151 4.979 (6.90) 35.90% -10 10 1.109 2.965 (4.08) 5.13% -2 10 External cleanliness of vehicle 3.378 4.182 (8.81) 20.51% -10 10 1.092 2.963 (4.02) 5.13% -2 10 Personal safety 5.252 4.831 (11.86) 17.95% -10 10 1.176 3.121 (4.11) 5.13% -2 10 Exposure to weather 4.529 4.376 (11.29) 17.95% -5 10 0.933 2.801 (3.63) 5.13% -2 10 Driving knowledge of route 2.067 5.728 (3.94) 58.97% -10 10 1.008 2.739 (4.02) 5.13% -2 10 Seat comfort 4.185 4.320 (10.57) 12.82% -10 10 0.966 2.768 (3.81) 5.13% -2 10 Seating condition 4.008 4.242 (10.31) 10.26% -10 10 1.134 3.059 (4.05) 5.13% -2 10 Driver presentation 2.966 4.712 (6.87) 35.90% -10 10 0.950 3.019 (3.43) 5.13% -10 10 Vehicle is disable friendly 2.899 5.210 (6.07) 33.33% -10 10 0.706 2.640 (2.92) 7.69% -5 10 English ability of driver 1.714 5.571 (3.36) 69.23% -10 10 1.109 3.204 (3.78) 2.56% -10 10 Driver driving ability 3.193 5.567 (6.26) 48.72% -10 10 0.950 2.991 (3.46) 5.13% -10 10 Ease of booking 4.101 5.364 (8.34) 33.33% -10 10 1.067 3.311 (3.52) 5.13% -10 10

26

Given a distribution of attitudes, it is possible to perform t-tests to determine whether each attitude is statistically different from zero. Within each table, values shown in bold are not statistically significant at the 95 percent confidence level. Where statistically significant, the values are all positive suggesting on average, the sample hold positive attitudes towards each mode, however by and large, the values reported are closer to zero than 10 suggesting significant improvements can be made for all modes. Also shown in the tables are the percentage of respondents who report a negative attitude towards particular model attributes, suggesting that even where the public hold positive values on average, significant proportions of respondents are negatively predisposed in terms of the attitudes they hold for certain modal attributes. With regards to taxis and hire cars, the attitudes towards taxis appear to be substantially higher on average for taxis than for hire cars. Care needs to be given however to this interpretation, as a large number of respondents appear not to have used a hire car in the past, and hence have a neutral attitude towards this mode of transport. As such, perhaps a better understanding of the attitudes towards taxis versus hire cars is to examine the percentage of negative attitudes held within the public. Based on the results of Table 9, it is clear that the number of negative attitudes held for taxis outstrips those for hire cars on several dimensions, again suggesting room for significant improvements of customer service. It is also possible to obtain an overall measure of attitudes towards each mode using the method described above. This is done by simply summing the values for each item or attribute within a mode. Table 10 presents the overall attitude towards each mode broken down by segment. Based on the results presented in Table 10, it is clear that on average, the tourist and night time travelling segments have a greater positive attitude towards all modes than all other segments, with the business, day time activity and MPTP card holder segments producing the smallest average attitude ratings across all segments. 9. Study Results

The main outputs are a set of estimated utility expressions (as above) for agreed market segments (e.g., corporate users, international and domestic tourists, late night travellers and those travelling for general non work activities) that can be used to obtain relevant elasticities and other useful outputs related to each user segments preferences for specific service an costs levels. In addition, we provide a range of informative outputs related to other data collected outside of the stated choice experiment. We estimate multinomial logit and mixed multinomial logit models so that we can get the behavioural benefits that these models offer. Table 11 presents the results for five separate econometric models; one MNL model for the MPTP market segment and four MMNL models estimated on the data collected from the remaining four market segments. As can be seen from this table, significantly different utility expressions were found to represent the best representation of preference structures of those belonging to each of the various travel segments. For this reason, we discuss the model results for each segment separately.

27

Table 10: Overall modal attitudes by segment* Tourism Segment Business Segment Day to Day Activity Segment Night Time Travel Segment MPTP Card Holder Segment Bus

Mean 27.938 18.123 21.369 47.938 7.785Std dev. 24.662 24.080 25.496 53.950 30.717(t-ratio) (9.13) (7.36) (11.14) (9.77) (4.47)% -ve 6.15% 13.85% 12.31% 1.54% 0.00%Min -20 -28 -42 -150 0 Max 104 100 100 150 140 Tram

Mean 12.867 7.602 10.180 46.492 24.602Std dev. 19.790 22.649 21.864 56.690 49.963(t-ratio) (6.07) (3.80) (5.81) (5.74) (2.93)% -ve 8.59% 21.09% 13.28% 12.50% 0.00% Min -26 -41 -60 -150 0 Max 75 120 120 150 150 Train

Mean 25.580 13.134 18.241 48.134 16.402Std dev. 24.308 23.917 24.616 47.886 42.977(t-ratio) (6.76) (5.27) (7.84) (7.11) (3.66)% -ve 9.82% 22.32% 12.50% 11.61% 0.00%Min -28 -44 -55 -98 0 Max 114 86 99 150 150 Taxi

Mean 22.109 13.286 14.975 51.739 17.908Std dev. 24.694 25.239 22.985 49.864 44.118(t-ratio) (7.16) (9.28) (10.64) (11.32) (5.58)% -ve 7.56% 23.53% 14.29% 9.24% 0.00%Min -80 -80 -80 -58 0 Max 92 87 85 150 150 Hire Car

Mean 23.436 16.231 19.103 63.231 16.872 Std dev. 32.724 34.596 32.557 70.740 47.004 (t-ratio) (2.04) (5.57) (4.04) (4.43) (2.24)% -ve 12.82% 17.95% 20.51% 23.08% 5.13%Min -27 -60 -24 -110 -50 Max 90 106 114 150 150 * For public transport modes, the attitude range is between -120 and 120. * For taxi and hire car modes, the attitude range is between -150 and 150

9.1 Tourism Segment For ease of discussion, the results for the tourist travel segment are separately reported in Table 12. The results reported are for a MMNL model in which several parameters are assumed to be randomly distributed across the population. In estimating the model, each random parameter is specified using a constrained triangular distribution1. Hensher and Greene (2003) have shown that for the triangular distribution, when the mean parameter is constrained to equal its spread (i.e., βjk = βk + |βk| Tj, where Tj is a triangular distribution ranging between -1 and +1), the density of the distribution rises linearly to the mean from zero before declining to zero again at twice the mean. Therefore, the distribution lies between zero and some estimated value (i.e., the βjk). As such, all parameter estimates are constrained to be of the same sign. 1 For example, the usual specification in terms of a normal distribution is to define βnk = βk + ηkvn where vn is the random variable. The constrained specification would be βnk = βk + βkvn when the standard deviation equals the mean or βnk = βk + hβkvn when h is the coefficient of variation taking any positive value. We would generally expect h to lie in the 0-1 range since a standard deviation greater than the mean estimate typically results in behaviourally unacceptable parameter estimates.

28

Table 11: All models* Tourism Segment Business Segment Day to Day Activity Segment Night Time Travel Segment MPTP Card Holder Segment Mode Par. (t-ratio) Par. (t-ratio) Par. (t-ratio) Par. (t-ratio) Par. (t-ratio) Alternative Specific Constants (w.r.t taxi) Constant Bus 7.588 (3.55) -11.047 (-3.58) 1.146 (0.36) 5.629 (2.78) - - Constant Tram 10.070 (4.49) -2.874 (-0.80) 7.420 (3.28) -0.007 (-0.01) 8.475 (4.74) Constant Train 8.655 (3.86) -6.053 (-1.79) 6.188 (2.83) 1.223 (0.83) - - Constant Walk 10.056 (2.22) - - 8.481 (1.97) 17.414 (4.04) - - Constant Car 0.440 (0.21) - - -5.380 (-1.08) 9.311 (3.71) - - Constant Hire car 2.867 (2.07) - - 8.341 (2.15) - - - -

Access TimeAccess Time All PT - - - - -0.228 (-2.05) -0.018 (-2.93) -1.026 (-4.04) ln(Access Time) Taxi and Hire car - - - - - - - - -0.527 (-1.87) Waiting TimeWait Time2 All PT -0.031 (-1.92) - - -0.033 (-1.68) - - - - ln(Wait Time) All PT - - -0.593 (-1.60) - - -1.208 (-1.99) - - Wait Time Taxi and Hire car -0.103 (-2.13) - - -0.061 (-1.79) - - - - ln(Wait Time) Taxi and Hire car - - -0.385 (-1.78) - - -0.809 (-2.99) -1.266 (-1.93)

Main Mode Travel Time Main Mode Travel Time All PT - - -0.087 (-2.57) -0.091 (-4.03) - - - - log(Main Mode Travel Time) All PT -2.057 (-4.54) - - - - - - - - Main Mode Travel Time Walk -0.301 (-2.40) - - - - - - - - Main Mode Travel Time2 Walk - - -0.018 (-1.99) -2.755 (-3.50) - - - - log(Main Mode Travel Time) Walk - - - - - - -7.688 (-5.02) - - log(Main Mode Travel Time) Car - - -2.135 (-4.91) -3.678 (-2.42) -4.127 (-6.11) - - Main Mode Travel Time2 All PT & car - - - - - - - - -0.002 (-3.43) Main Mode Travel Time2 Taxi and Hire car - - -0.0002 (-1.94) -0.0015 (-3.74) - - - - log(Main Mode Travel Time) Taxi and Hire car - - - - - - - - -1.604 (-2.34) Egress TimeEgress Time2 All PT - - - - -0.010 (-2.51) - - - - Egress Time2 Car - - -0.022 (-2.17) - - - - - - Egress Time All PT and car - - - - - - -0.214 (-3.70) - - ln(Egress Time) All PT and car - - - - - - - - -0.884 (-2.10)

CrowdingProportion of people sitting All PT -2.535 (-2.41) -3.611 (-3.76) -2.852 (-2.51) - - - - ln(Proportion of people sitting) All PT - - - - - - -1.74536 (-3.94) - - Number of people standing All PT - - -0.044 (-2.26) - - - - - - Number of people standing2 All PT - - - - -0.0004 (-2.83) -0.0001 (-2.63) - - Ln(Number of people standing) Bus -1.428 (-4.64) - - - - - - - - Ln(Number of people standing) Tram and Train -0.461 (-2.23) - - - - - - - - *Values in bold represent random parameters with constrained triangular distributions

29

Table 11: All models cont’d Tourism Segment Business Segment Day to Day Activity Segment Night Time Travel Segment MTPT Card Holder Segment Mode Par. (t-ratio) Par. (t-ratio) Par. (t-ratio) Par. (t-ratio) Par. Delay Average Delay Time (mins) All PT - - - - - - - - -0.230 (-2.39) Average Delay Time (mins)2 All PT - - - - - - -0.003 (-1.79) - - Frequency of Delay Time All PT - - - - - - - - -25.396 (-1.82)

Travel CostsFare2 All PT -0.011 (-3.58) - - - - - - - - ln(Fare) All PT - - -2.846 (-4.84) -3.394 (-5.03) - - - - Cost (Parking + toll + fuel) Car -0.394 (-2.01) -1.325 (-2.91) -0.140 (-5.81) -0.352 (-2.46) - - ln[Cost (Parking + toll + fuel)] Car - - - - - - - - -3.525 (-2.62) Fare Taxi and Hire car -0.156 (-4.98) - - - - - - - - Fare2 Car - - - - - - -0.0024 (-4.68) -0.093 (-3.88) ln(Fare) Taxi and Hire car - - -4.385 (-8.33) -3.747 (-6.12) - - - - Time & Cost Interactions Access Travel Time × Income All PT -0.002 (-1.97) - - - - - - - - Main Mode Travel Time × Income Car -0.001 (-2.55) - - - - - - - - Main Mode Travel Time × Fare2 All PT - - - - - - -0.0001 (-4.24) - - Main Mode Travel Time × Age All PT - - - - - - -0.001 (-1.98) - - Fare × Age All PT - - - - - - - - -0.019 (-1.91) Parking costs × Gender (1 = Female) Car - - - - - - - - 0.125 (1.89) Toll costs × ln(main mode travel time) Car - - - - -3.059 (-6.01) - - - - Main Mode Travel Time × Age Taxi and Hire car - - - - - - -0.009 (-7.09) - - Wait Time × Age Taxi and Hire car - - - - - - - - -0.003 (-1.99) Main Mode Travel Time × Income Taxi and Hire car -0.0004 (-2.26) - - -0.0002 (-4.00) - - - -

Socio-demographics and other variablesNumber of children present during trip All PT - - - - -6.830 (-2.72) - - - - Afternoon trip (1 = yes) All PT - - - - -6.971 (-3.37) - - - - Trip purpose - visit friends All PT - - - - 4.438 (1.88) -1.464 (-1.97) - - Trip purpose - shopping trip All PT - - - - 6.364 (2.27) -10.208 (-4.84) - - Gender (1 = Female) All PT - - - - - - - - 1.965 (2.06) Income Walk -0.074 (-2.53) -0.090 (-1.92) -0.091 (-3.00) - - - - Need to arrive on special time (1 = No) Car - - 3.127 (2.88) - - - - - - Trip purpose - shopping trip Car - - - - - - - - 3.001 (2.71) Need to arrive on special time (1 = No) Taxi and Hire car - - - - -3.136 (-3.37) - - - - Weather (overcast) Taxi and Hire car - - - - - - 4.276 (3.99) - - Weather (light rain) Taxi and Hire car - - - - -7.909 (-2.58) 7.267 (2.22) - - Model FitsLL(0) -1621.964 -3194.022 -2794.769 -2969.443 -910.686 LL(β) -259.910 -569.172 -428.194 -565.488 -240.618 ρ2 0.840 0.822 0.847 0.810 0.736 Adj. ρ2 0.836 0.820 0.844 0.806 0.726 Number of respondents 65 128 112 119 39 Number of choice observations 780 1536 1344 1428 468

30

Empirically, the distribution will be symmetrical about the mean, which not only allows for ease of interpretation, but also avoids the problem of long tails often associated with drawing from a lognormal distribution as in Bhat (1998, 2000) and in earlier studies in Sydney. To appreciate the behavioural plausibility of constraining the distribution, given that all analytical distributions are approximations to reality, there is growing concern by choice modellers that distributions like the lognormal force a very small number of observations to take on very large values in order to statistically fit the (analytical) distribution. These very large positive values inflate and behaviourally distort the average for the sample and hence should be removed. However, simply truncating a distribution through ignoring very high values is not acceptable to us. To estimate the model and the random parameters Simulated Maximum Likelihood is used with 1,000 Halton draws. Table 12: Tourism Segment Model

Par. (t-ratio) Alternative Specific Constants (w.r.t taxi)Constant Bus 7.588 (3.55) Constant Tram 10.070 (4.49) Constant Train 8.655 (3.86) Constant Walk 10.056 (2.22) Constant Car 0.440 (0.21) Constant Hire car 2.867 (2.07)

Waiting TimeWait Time2 All PT -0.031 (-1.92) Wait Time Taxi and Hire car -0.103 (-2.13) Main Mode Travel Timelog(Main Mode Travel Time) All PT -2.057 (-4.54) Main Mode Travel Time Walk -0.301 (-2.40)

CrowdingProportion of people sitting All PT -2.535 (-2.41) Ln(Number of people standing) Bus -1.428 (-4.64) Ln(Number of people standing) Tram and Train -0.461 (-2.23) Travel costsFare2 All PT -0.011 (-3.58) Fare Taxi and Hire car -0.156 (-4.98) Cost (Parking + toll + fuel) Car -0.394 (-2.01)

Time & Cost Interactions Access Travel Time × Income All PT -0.002 (-1.97) Main Mode Travel Time × Income Car -0.001 (-2.55) Main Mode Travel Time × Income Taxi and Hire car -0.0004 (-2.26) Socio-demographics and other variablesIncome Walk -0.074 (-2.53)

Model fitsLL(0) -1621.964 LL(β) -259.910 ρ2 0.840 Adj. ρ2 0.836 Number of respondents 65 Number of choice observations 780 Values in bold represent random parameters with constrained triangular distributions In order to obtain the final model, a number of specifications were tested for each attribute, including taking logs, squaring the attribute and estimating interaction effects with other attributes and socio-demographic variables. Attitudinal data were not tested, as it has been well recognised that such data cannot be used to forecast mode choice into the future. Overall, the model reported

31

in Table 12 provided an excellent model fit with an adjusted ρ2 value of 0.836. All (non-alternative specific constant) parameters reported are statistically significant at the 10 percent level of confidence and of the expected sign. For the tourist travel segment, access time was not found to be statistically significant as a standalone attribute, even after testing several transformations of the attribute. For public transport modes however, the attribute was found to have a statistically significant interaction with income, suggesting that higher income earners have larger marginal disutility for access time increases. For the taxi and hire car modes, access time was not found to be statistically significant even after testing for several possible interaction effects. As such, the attribute was removed from the final model for these modes. Waiting time in the current context represents time spent either waiting at a bus stop or train or tram station, or time spent waiting for a taxi or hire car after arriving at a taxi rank or street where a taxi can be caught, or time waiting for a taxi or hire car after the time it was due to arrive if booked. As such, for taxi or hire car, this attribute may act as a proxy for frequency of available services or tardiness on behalf of taxi or hire car company. The best model fit was obtained when the waiting time parameter for the public transport modes was treated as being generic, with a separate generic waiting time parameter for the taxi and hire car modes. In this instance, the square of waiting time for public transport and a linear waiting time for taxi and hire car modes was found to produce the best model results. This implies that for the public transport modes, that the marginal disutility for waiting time increases as a square route function of time whereas for the taxi and hire car alternatives, each additional minute of waiting time adds the same amount of marginal disutility as the previous and subsequent minutes spent waiting. With regards to main mode travel time, the log of public transport travel times were found to produce the best model fits assuming a generic parameter across all three public transport modes. A linear walking time attribute was found also to be statistically significant. For the car, taxi and hire car alternatives, the best model fit was obtained when travel time was interacted with income. Again, negative parameter estimates were obtained suggesting that higher income earners have larger marginal disutility for travel time using these modes relative to other modes. Egress time was not found to have any statistical influence on mode choice for any of the modes and hence was removed from the final model specified. Crowding on public transport was also found to have a statistically significant influence on mode choice. Two crowding attributes were included in the SC scenarios that respondents were asked to complete; one reflecting the number of seats occupied, and the other representing the number of people standing. For the seating attribute, a generic attribute across all modes was found to provide the best model fit, whereas an alternative specific parameter for bus, different to that of a generic tram and train parameter, was found to best represent the influence of the number of people standing. A generic parameter estimate was applied to the square of fare for public transport modes alongside a generic taxi/hire car parameter for a linear in the attributes treatment of fare. For car, an aggregation of all car related costs (fuel, parking and toll costs) was used in the final model. In all

32

cases, the influence of costs are statistically significant and negative, suggesting increasing costs will result in lower utility, and hence lower market shares for these modes. Finally, income was found to be statistically significant and negative as a main effect within the walking alternative. This suggests that all else being equal, higher income earners have a higher marginal disutility for walking and are more likely to select another mode of transport. 9.2 Business Traveller Segment Table 13 re-presents the model results for the business traveller segment of data. As with the tourist traveller segment, a MMNL model was estimated assuming all random parameter estimates were drawn from constrained triangular distributions. One thousand Halton draws are used in the model estimation. Overall, the final model fit was exceptional with an adjusted ρ2 value of 0.822.

Table 13: Business Segment Model Par. (t-ratio)

Alternative Specific Constants (w.r.t taxi)Constant Bus -11.047 (-3.58) Constant Tram -2.874 (-0.80) Constant Train -6.053 (-1.79) Waiting Timeln(Wait Time) All PT -0.593 (-1.60) ln(Wait Time) Taxi and Hire car -0.385 (-1.78)

Main Mode Travel TimeMain Mode Travel Time All PT -0.087 (-2.57) Main Mode Travel Time2 Walk -0.018 (-1.99) log(Main Mode Travel Time) Car -2.135 (-4.91) Main Mode Travel Time2 Taxi and Hire car -0.0002 (-1.94) Egress TimeEgress Time2 Car -0.022 (-2.17)

Crowding Proportion of people sitting All PT -3.611 (-3.76) Number of people standing All PT -0.044 (-2.26) Travel costs ln(Fare) All PT -2.846 (-4.84) ln(Fare) Taxi and Hire car -4.385 (-8.33) Cost (Parking + toll + fuel) Car -1.325 (-2.91)

Socio-demographics and other variablesNeed to arrive on special time (1 = No) Car 3.127 (2.88) Income Walk -0.090 (-1.92) Model fitsLL(0) -3194.022 LL(β) -569.172 ρ2 0.822 Adj. ρ2 0.820 Number of respondents 128 Number of choice observations 1536

Values in bold represent random parameters with constrained triangular distributions Access time for the business segment was not found to be a statistically significant attribute, even after testing several transformations of the attribute and interactions with other variables and attributes. As such, access time was removed completely from the model. The best model fits were obtained assuming a generic waiting time parameter for the log of waiting time for public transport modes. Likewise, a generic waiting time parameter for the log of waiting time for taxi and hire cars was found to best represent the data.

33

A generic parameter across a linear in the attribute main mode public transport travel time was found to provide the best model fit whilst for walk, the square of travel time produced the best model result. Likewise the square of travel time was used to enter the taxi and hire car utility functions. The log of travel time however was found to best represent the influence on utility for the car mode. Unlike the tourist segment, egress for the car mode was found to be statistically significant and negative, and hence was retained in the final model. Crowding in the form of the number of seats occupied and the number of people standing was also found to be a statistically significant driver of mode choice for the public transport modes. Unlike the tourist segment of data, the final model treats the influence of seats and standing as being generic across all public transport modes however. Travel costs were found to be statistically significant for all modes, with the natural log of public transport and taxi and hire car fares entering into the model utility functions. As with the tourist segment, all car related costs (fuel, parking and toll costs) were aggregated into a single cost variable in the final estimated model. 9.3 General Day to Day Activity Travel Segment The model results for the general day to day activity segment are reproduced in Table 14. The results reported are for a MMNL model assuming several randomly distributed parameters following constrained triangular distributions, and was estimated using 1,000 Halton draws. As with the previous two models discussed, the model in Table 14 has an exceptionally high model fit with an adjusted ρ2 value of 0.844. Within the model, access time was found to be statistically significant only for the public transport modes. As such, it does not appear within the utility functions for taxi and hire car alternatives. Waiting time squared and waiting time were found to be statistically significant drivers of mode choice for the public transport, and taxi and hire car alternatives respectively. Separate generic parameters were used for the public transport modes and taxi and hire car alternatives. A generic parameter applied to the main mode travel time of the public transport alternatives was found to best represent the data. For the walk and car alternatives, alternative specific parameters were used for the log of the main mode travel times associated with each of these attributes. For the taxi and hire car alternatives, a generic parameter applied to the square of travel time was used in the final model specification. An interaction term between income and the main mode taxi and hire car travel times was also found to be statistically significant with the negative parameter estimate suggesting that higher income travellers have a higher marginal disutility for travel time than lower income business travellers. Like access time, egress time was found to be statistically significant for the public transport modes only, however rather than entering the utility function as a linear in the attribute manner, the square of egress time was found to best represent the data in the final model. As with the previous models, crowding on public transport modes was also found to be a key determinant of mode choice. No differences were found in the marginal disutilities for seating and standing across the modes; however the square of the standing attribute tended to produce a better model fit than if it entered utility in any other fashion.

34

Table 14: Day to Day Activity Segment Par. (t-ratio)

Alternative Specific Constants (w.r.t taxi)Constant Bus 1.146 (0.36) Constant Tram 7.420 (3.28) Constant Train 6.188 (2.83) Constant Walk 8.481 (1.97) Constant Car -5.380 (-1.08) Constant Hire car 8.341 (2.15) Access TimeAccess Time All PT -0.228 (-2.05)

Waiting Time Wait Time2 All PT -0.033 (-1.68) Wait Time Taxi and Hire car -0.061 (-1.79) Main Mode Travel Time Main Mode Travel Time All PT -0.091 (-4.03) log(Main Mode Travel Time) Walk -2.755 (-3.50) log(Main Mode Travel Time) Car -3.678 (-2.42) Main Mode Travel Time2 Taxi and Hire car -0.002 (-3.74)

Egress TimeEgress Time2 All PT -0.010 (-2.51) CrowdingProportion of people sitting All PT -2.852 (-2.51) Number of people standing2 All PT -0.0004 (-2.83)

Travel costsln(Fare) All PT -3.394 (-5.03) ln(Fare) Taxi and Hire car -3.747 (-6.12) Cost (Parking + toll + fuel) Car -0.140 (-5.81) Time & Cost InteractionsToll costs × ln(main mode travel time) Car -3.059 (-6.01) Main Mode Travel Time × Income Taxi and Hire car -0.0002 (-4.00)

Socio-demographics and other variablesNumber of children present during trip All PT -6.830 (-2.72) Afternoon trip (1 = yes) All PT -6.971 (-3.37) Trip purpose - visit friends All PT 4.438 (1.88) Trip purpose - shopping trip All PT 6.364 (2.27) Income Walk -0.091 (-3.00) Need to arrive on special time (1 = No) Taxi and Hire car -3.136 (-3.37) Weather (light rain) Taxi and Hire car -7.909 (-2.58) Model fitsLL(0) -2794.769 LL(β) -428.194 ρ2 0.847 Adj. ρ2 0.844 Number of respondents 112 Number of choice observations 1344

Values in bold represent random parameters with constrained triangular distributions As with the business segment, travel costs were found to be statistically significant for all modes, with the natural log of public transport and taxi and hire car fares entering into the model utility functions. Further, as with both previous models, all car related costs (i.e., fuel, parking and toll costs) were aggregated into a single cost variable in the final estimated model. For the car alternative however, an interaction between toll costs and the natural log of in-vehicle main mode travel time was also found to be statistically significant, suggesting a relationship between the marginal utility for paying tolls and travel time exists for the sampled population.

35

Several covariates were also found to be statistically significant within the model. With regards to the public transport modes, as the number of children present during the trip increased or if the trip was in the afternoon, this resulted in a, increased disutility for using these modes, whereas for trips involving travelling to friends or to the shops, a higher utility was obtained. Income was found to be statistically significant and negative for walking trips implying higher income earners have a greater disutility for walking as a transport mode relative to other modes, all else being equal. Finally, the model results suggest that respondents who did not have a specific requirement to be at their destination at a given time are less inclined to take a taxi relative to those who stated that they had a deadline to meet, whilst those travelling during light showers were also less likely to take a taxi or hire car than other modes, all else being equal. 9.4 Night Time Travel Segment As with the three previous models reported, a MMNL model assuming constrained triangular distributions and estimated using 1,000 Halton draws was estimated on the night time travel segment. The results of this model are presented in Table 15 below. An adjusted ρ2 value of 0.806 was obtained for the model. As with the general day to day activity segment model reported in Table 14, access time was found to be statistically significant only for the public transport modes and hence does not appear within the utility functions for taxi and hire car alternatives. The natural log of waiting time was found to be a statistically significant driver of mode choice for the public transport and taxi and hire car alternatives with separate generic parameters estimated for each. Likewise, the log of main mode travel time for walk and car were found to be statistically significant drivers of mode choice. For the public transport modes however, the best model fit was obtained when main mode travel time was interacted with the square of public transport fares, suggesting an intricate relationship between these two attributes. Further, a second interaction between main mode travel time and age was also found to be statistically significant for both the public transport modes and taxi and hire car alternatives. In both instances, the negative parameter estimate suggests that older respondents had a greater disutility for travel time than younger respondents for these modes, all else being equal. Travel costs were found to be statistically significant for all modes, with the square of the trip specific fare entering into the taxi and hire car utility functions. As with the previous models, all car-related costs (i.e., fuel, parking and toll costs) were aggregated into a single cost variable in the final estimated model. For the public transport modes, fare squared is interacted with the main mode in-vehicle travel time as opposed to entering into the utility functions as a singular attribute. Finally, similar to the general day to day activity segment model, several covariates appear to be significant drivers of mode choice within the night time travel segment population. Contrary to the general day to day activity segment model, the night time model results suggest that trips involving travelling to friends or to the shops produce a lower utility or generate a disutility for public transport modes relative to other modes, suggesting that such trips undertaken during the day are more likely to result in public transport use, but less likely during the night. Further, opposite again to the general day to day activity segment model, light rain or the possibility of rain are more likely to generate a taxi or hire car trip at night than other weather patterns, all else being equal.

36

Table 15: Night Time Travel Segment Par. (t-ratio)

Alternative Specific Constants (w.r.t taxi)Constant Bus 5.629 (2.78) Constant Tram -0.007 (-0.01) Constant Train 1.223 (0.83) Constant Walk 17.414 (4.04) Constant Car 9.311 (3.71) Access Time Access Time All PT -0.018 (-2.93)

Waiting TimeLn(Wait Time) All PT -1.208 (-1.99) Ln(Wait Time) Taxi and Hire car -0.809 (-2.99) Main Mode Travel Timelog(Main Mode Travel Time) Walk -7.688 (-5.02) log(Main Mode Travel Time) Car -4.127 (-6.11)

Egress TimeEgress Time All PT and car -0.214 (-3.70) Crowdingln(Proportion of people sitting) All PT -1.745 (-3.94) Number of people standing2 All PT -0.0001 (-2.63)

DelayAverage Delay Time (mins)2 All PT -0.003 (-1.79) Travel costsFare2 Taxi and Hire car -0.002 (-4.68) Cost (Parking + toll + fuel) Car -0.352 (-2.46)

Time & Cost InteractionsMain Mode Travel Time × Fare2 All PT -0.0001 (-4.24) Main Mode Travel Time × Age All PT -0.001 (-1.98) Main Mode Travel Time × Age Taxi and Hire car -0.009 (-7.09) Socio-demographics and other variablesTrip purpose - visit friends All PT -1.464 (-1.97) Trip purpose - shopping trip All PT -10.208 (-4.84) Weather (overcast) Taxi and Hire car 4.276 (3.99) Weather (light rain) Taxi and Hire car 7.267 (2.22)

Model fitsLL(0) -2969.443 LL(β) -565.488 ρ2 0.810 Adj. ρ2 0.806 Number of respondents 119 Number of choice observations 1428 Values in bold represent random parameters with constrained triangular distributions

9.5 MPTP Card Holder Travel Segment The model results for the final MTPT card holder segment are presented in Table 16. Unlike the models reported for the other segments, a MNL model was fitted to the MTPT card holder segment as opposed to a MMNL model. This was necessitated by the small MTPT card holder sample acquired. Despite the known limitations of the MNL model relative the MMNL model, a more than respectable adjusted ρ2 value of 0.726 was obtained for the model (typically, a ρ2 value of 0.3 is suggested as being the benchmark for this model – see Hensher et al. 2005).

37

Table 16: MPTP Travel Segment Par. (t-ratio)

Alternative Specific Constants (w.r.t taxi)Constant Tram 8.475 (4.74) Access TimeAccess Time All PT -1.026 (-4.04) ln(Access Time) Taxi and Hire car -0.527 (-1.87)

Waiting TimeLn(Wait Time) Taxi and Hire car -1.266 (-1.93) Main Mode Travel TimeMain Mode Travel Time2 All PT & car -0.002 (-3.43) ln(Main Mode Travel Time) Taxi and Hire car -1.604 (-2.34)

Egress Timeln(Egress Time) All PT and car -0.884 (-2.10) DelayAverage Delay Time (mins) All PT -0.230 (-2.39) Frequency of Delay Time All PT -25.396 (-1.82)

Travel costsFare2 Taxi and Hire car -0.093 (-3.88) ln[Cost (Parking + toll + fuel)] Car -3.525 (-2.62) Time & Cost InteractionsFare × Age All PT -0.019 (-1.91) Parking costs × Gender (1 = Female) Car 0.125 (1.89) Wait Time × Age Taxi and Hire car -0.003 (-1.99)

Socio-demographics and other variablesGender (1 = Female) All PT 1.965 (2.06) Trip purpose - shopping trip Car 3.001 (2.71) Model fitsLL(0) -910.686 LL(β) -240.618 ρ2 0.736 Adj. ρ2 0.726 Number of respondents 39 Number of choice observations 468 Unlike the models discussed previously, access time was found to be statistically significant at the five percent level of public transport modes and the natural log of access time at 10 percent for the taxi and hire car alternatives. Unlike the previous models however, wait time for the public transport modes were not statistically significant; however the natural log of wait time for the taxi and hire car alternatives was statistically significant. This suggests that for MTPT card holders, each additional minute of wait time results in an increase in marginal disutility for taxi and hire car that increases at a decreasing rate. Nevertheless, a statistically significant interaction term between wait time and age was also found to exist for the data, suggesting that older MTPT card holders have a greater marginal disutility for waiting for taxi and hire cars than younger MTPT card holders. The final MTPT model has a generic parameter for the public transport and car alternatives associated with the square of the main mode in-vehicle time attribute. A generic parameter associated with the natural log of the main mode in-vehicle time attributes of the taxi and hire car alternatives is also used within the model. The final model also includes a generic parameter for the public transport and car alternatives associated with the natural log of egress travel time. With regards to transport costs, an interaction appears to exist between age and fare with regards to the public transport modes. The negative sign of the interaction suggests that older MTPT card holders have a larger disutility for fares than do younger MTPT card holders, all else being equal.

38

For the car mode, all car related costs (i.e., fuel, parking and toll costs) were once more aggregated into a single cost variable in the final estimated model, however in this case the natural log of this aggregated costs attribute is used in the final model. There also exists an interaction term between parking costs and gender, the results of which suggest that males are more sensitive to parking costs than females, all else being equal. Finally, for the taxi and hire car alternatives, the square of the fare enters into the utility function of the final model. Two covariates also were found to be predictors of mode choice for the MTPT card holder segment. Firstly, females were found to prefer the use of public transport relative to males. Secondly, MTPT card holders were more likely to choose the car alternative for trips involving travelling to the shops, relative to other modes, all else being equal. 10. A Decision Support System (DSS) In order to operationalise the models in a manner that is meaningful for the Victorian Taxi Industry Inquiry, a Decision Support System (DSS) is provided alongside this report. The DSS which accompanies the report is designed to allow users to test how changes to attributes associated with the modes studied will likely affect the market shares, including providing evidence on the implied direct and cross elasticities. The DSS is constructed in Microsoft Excel and provides a simple user interface. The construction and inclusion of a DSS for the current study is necessitated for a number of reasons. Firstly, the sampling strategy required for estimating the mode choice models was such that the sample of trips collected and used to generate the SC experiment are not reflective of the overall travel patterns existing. Based on data available from the Victorian Department of Transport on modal shares, taxi trips represent only 0.42 percent of all trips undertaken in the Melbourne metropolitan area. As such, a random sample of 505 respondents would be expected to yield less than three recent trips undertaken in which a taxi was used, and possibly a not much larger sample of trips where a taxi was considered to be an alternative mode of transport, even when not chosen. For hire cars, the problem of randomly sampling trips from the general population is likely to represent a significantly larger problem. Given such a small number of taxi and/or hire car trips, it would not be feasible to estimate a mode choice model for these modes given such a sampling strategy. As such, we deliberately oversampled the number of taxi and hire car trips relative to the general number of trips by these modes in our sample so as to be able to estimate the models of interest. Combined with the fact that we have utilised a SC data set as opposed to an RP data set, the oversampling of certain modes has a number of implications in terms of how the estimated models may be used in practice. First and foremost, the SC experiment utilised herein involves each respondent making choices across 12 choice scenarios. This differs to what one would expect in real market data, where each respondent is likely to be observed to make one or two mode choice decisions in the survey period covered. Secondly, the SC experiment varies the attribute levels of the alternatives so that the levels shown to respondents, whilst within a realistic range, may not necessarily fully reflect market reality. As such, the choices observed are based on the SC choice scenarios and not real markets. Given the above, the mode shares in SC data will typically not match those of real markets. Ordinarily, this need not be a significant problem as SC experiments are

39

specifically designed to force respondents to make trade-offs between attributes, and hence produce robust parameter estimates for the attributes of interest. Nevertheless, given the oversampling of taxi and hire car trips relative to such trips within real markets, the mode specific constants for the models are likely to be far less reflective of those produced from RP data. Given the impact on mode specific constants, it is typical to calibrate the constants after fixing the remaining parameter estimates. After calibrating the model constants, the mode shares of the model should reflect the known market shares. The process of calibrating the modal constants however requires that the estimated model be fed through the data. In the current context however, a number of the sampled trips within the data involved trips where respondents reported being captive to taxi as the only mode available for the trip. Such trips were particularly prevalent for the MPTP card holder market segment (see Table 17 for the percentage of taxi captive trips broken down by segment). In the SC experiment, these respondents were shown scenarios in which they were asked to choose between two hypothetical taxis. The high number of such trips in the SC data presents problems when aggregating the choice shares as well as when calibrating the modal constants. For modal captive respondents, the probability of selecting a taxi is always one, and hence the proportion of such respondents within a data set represents the minimum mode share that could be obtained for that mode given the data set, independent of any constants or parameter estimates. Not even the imposition of an infinite (or negative infinite) modal constant will produce a lower market share than presented in Table 17, given such a data set. Table 17: Percentage of taxi captive trips by segment

Segment Proportion of captive respondents Tourism Segment 9.92%Business Segment 11.51% Day to Day Activity Segment 16.17%Night Time Travel Segment 19.51%MPTP card holders 51.22%The modal constants are important for a number of reasons, none more so than to obtain elasticity estimates. In discrete choice models, elasticities are a function of not just the parameter estimates and the data, but also the choice probabilities. As such, it is important that the mode specific constants reproduce the known market shares, otherwise any elasticities generated from the model will be biased. To overcome all of the above, we simulate data for each mode for 2,500 respondents. In simulating the data, we use data from the 2011 Metro NSP data set to establish the average taxi travel times (16.04 minutes) and fares ($23.57). For each trip segment, we then draw from a log-normal distribution taxi travel times with the same mean travel time as that obtained from the 2011 Metro NSP data. Next, fares are drawn for each travel time based on the fare formula currently used in Victoria (http://www.taxifare.com.au/rates/australia/melbourne/) with a stochastic term to provide some variation. The fare and travel time distributions for the general day to day activity segment are shown in Figure 7. The remaining attribute levels are then drawn from similar distributions for the remaining modes based on the values obtained from the sample, but correlated with the taxi times and costs (so that for example, shorter taxi trips are matched with shorter train trips).

40

Figure 7: Simulated travel time and fare distributions for the general day to day travel segment We then calibrate the mode-specific constants using the simulated data based on data obtained from the Victorian Department of Transport. We discuss this process now.

10.1 Constructing a Population-Level Modal Share The survey data is a sample of current modal trips, designed to ensure we get a good representation of all modes of interest as a chosen main mode for the land-based trip component in Melbourne. This includes situations where a mode such as a taxi might be an access mode to the airport, which we define as the main mode for our study. The data available from the Victorian Department of Transport on modal shares is limited to households surveyed as part of the VISTA 09-10 study. Any individual who was in a Melbourne house at the time of the interviews was captured, including persons visiting from overseas or from Australia but outside of Melbourne; however tourists living in other accommodation such as a hotel are not captured. In addition, no hire car information is explicitly collected as part of VISTA. Thus, although we use the VISTA 09-10 modal share data (based on an average day in Melbourne) as a guide, in all of our categories of trips, we have the potential for the presence of respondents who were not in a household situation at the time of the interview. The VISTA 09-10 data was re-processed by THG to obtain ‘indicative’ mode shares for each trip purpose, as given in Table 18a. Table 18a: Aggregate Modal Shares extracted from VISTA Average day household survey 2009-10 Tourism Segment Business Segment Day to Day Activity Segment Number Percentage Number Percentage Number Percentage Public Bus 51,900 0.92% 21,700 0.88% 112,300 1.46% Tram 44,200 0.78% 77,600 3.16% 101,500 1.32% Train 63,300 1.12% 240,500 9.80% 193,100 2.51% Car 4,785,500 84.52% 1,949,100 79.44% 6,251,900 81.21% Walking 708,200 12.51% 149,600 6.10% 1,014,100 13.17% Taxi 8,849 0.16% 15,169 0.62% 25,282 0.33% Total 5,661,949 100.00% 2,453,669 100.00% 7,698,182 100.00%

41

Table 18b: Assumed Day to Day activity and Night Time travel shares Day to Day Activity Segment Night Time Travel Segment Number Percentage Number Percentage Public Bus 84,225 1.46% 28,075 1.46% Tram 76,125 1.32% 25,375 1.32% Train 144,825 2.51% 48,275 2.50% Car 4,688,925 81.27% 1,562,975 81.05% Walking 760,575 13.18% 253,525 13.15% Taxi 15,169 0.26% 10,113 0.52% Total 5,769,844 100.00% 1,928,338 100.00% No data were available for night time travel or MPTP card holder travel segments. Given this limitation, we have made a simple assumption that 75 percent of all day to day activities occur during the day time and 25 percent late at night. The mode shares for these two segments are therefore assumed to be the same across these two segments; however the total number of trips differs. Given that the calibration of the modal constants is based on the market shares and not the total number of trips, the 75/25 percent assumption is not of importance, however the assumption that the mode shares are similar for these two segments is. The assumed number of trips for the all day to day activities and night time travel segments are shown in Table 18b. Table 19 shows the modal shares by segment obtained within the SC data. For each mode, the number of times an alternative is present and the number of times it was chosen, given that it was present, is shown. A close examination of this data suggests that there is an under-estimation of taxi trips associated with tourists, for the reasons given above. In order to add additional evidence on this, we sourced data from the National Visitor Survey (NVS) conducted annually by Tourism Research Australia (TRA2008). Each year, 120,000 Australian residents are surveyed using random digit dialling and a computer aided telephone interview (CATI). This survey is the principal source of information on domestic tourism movements in Australia. Information about the movements of international visitors is collected separately. As “tourism”' in this context is defined to include holidaying, visiting friends and relatives and travel for business, education or employment, the survey covers a broad range of passenger movements. There are unfortunately problems with this source. It excludes trips associated with individuals who do not stay overnight in the Melbourne metropolitan area, and hence day visitors are excluded. The largest gap is the daily business visitors who are frequent users of taxis and hire cars to and from the airport. Table 19: Sample Modal Shares by Segment

Tourism Segment Business Segment Day to Day Activity Segment Night Time Travel Segment MPTP card holder Segment # Chosen # Chosen # Chosen # Chosen # Chosen Bus 312 134 456 98 72 58 72 54 36 0 Tram 108 70 84 41 300 79 300 143 24 16 Train 72 61 24 1 192 90 192 115 72 28 Car 216 190 540 393 420 445 420 319 108 58 Walk 72 37 60 25 1644 95 1644 658 0 0 Taxi 876 281 1680 858 288 554 288 132 720 364 Hire car 12 7 300 120 36 23 36 7 24 2

Total 1668 780 3144 1536 2952 1344 2952 1428 984 468

42

If we were to assume that the VISTA data was our population level modal shares for the tourist, corporate, and day to day activity segments, then the scaling required for calibration in the DSS would be as indicated in Table 20. It is reasonable to assume, given what we know about the VISTA data and the likely element of tourists who do not reside in a household residence, together with the suggestion that many business travellers visit Melbourne for the day and use taxis a lot (to and from the airport), that the population to sample weights for taxi and hire in particular are far too low. Table 20: Ratio of Population to Sample Shares excluding where known Tourism Segment Business Segment Day to Day Activity Segment Population Shares Sample Shares Ratios Population Shares Sample Shares Ratios Population Shares Sample Shares Ratios Bus 0.92% 8.03% 8.76 0.88% 3.12% 3.53 1.46% 1.96% 1.34 Tram 0.78% 4.20% 5.38 3.16% 1.30% 0.41 1.32% 2.68% 2.03 Train 1.12% 3.66% 3.27 9.80% 0.03% 0.00 2.51% 3.05% 1.22 Car 84.52% 11.39% 0.13 79.44% 12.50% 0.16 81.27% 15.07% 0.19 Walk 12.51% 2.22% 0.18 6.10% 0.80% 0.13 13.18% 3.22% 0.24 Taxi 0.16% 16.85% 107.81 0.62% 27.29% 44.14 0.26% 18.77% 71.40 Hire car 0.00% 0.42% - 0.00% 3.82% - 0.00% 0.78% - Given that we have no market shares for hire cars for any segment, a number of assumptions have had to be made with regards to hire car usage. For each segment, we have assumed that a certain proportion of the taxi market share belongs to hire cars. We have assumed 0.01 percent and 0.08 percent of the market shares for the tourism and business markets are associated with hire cars, whilst 0.05 percent of the market share belongs to hire cars for the day to day general travel and night time travel segments. For the MPTP segment, we have assumed 0.016 percent of the market is associated with hire cars. As such, the total numbers of taxis do not align with the numbers shown in the tables above, as some of the taxi shares have been re-distributed to hire cars. Hence, summing the number of taxis and hire cars will reproduce the numbers shown for taxis above.

10.2 The DSS The DSS consists of three screens, the first of which represents an introduction screen. The two screens of interest are the Input and Scenario output screens (see Figures 8 to 12 for screen captures of these screens). The input screen allows users to enter as either percentage or absolute changes, for changes to the attribute levels of the seven modes used as part of this study. For the main in-vehicle or main mode walk times and fares, the DSS has separated the data into trips less than 20 minutes, between 20 and 40 minutes, and greater than 40 minutes. This allows the user to change as an absolute number, the fares or costs for subsets of the data. For example, the user can change the fare only for trips that are under 20 minutes, or increase the travel times for trips over 40 minutes. The DSS also allows for several attributes to be changed at once or only a single attribute to be changed, depending on the specific scenario being tested.

43

Figure 8: DSS input screen As shown in Figure 8, to the right of the scenario section where attribute level changes can be made, are the results of calculations that convert for the taxi mode, the absolute changes for main mode times and fares into percentage changes. For example, if the user wishes to impose a $2.50 increase in taxi fares for trips that are less than 20 minutes, the numbers shown in these boxes calculate and show what the percentage change in fares by segment is. Note that percentage changes will be greater for shorter trips than longer trip lengths, for a given fare change. Two buttons are also available that will switch the main mode or in-vehicle travel times and fares between percentage and absolute value changes. Not accessible to the user, the DSS makes use of the discrete choice models estimated and reported in Section 9 of this report that are linked to the simulated respondents as discussed in Section 10 above. Given the use of MMNL models for four of the five segments, simulation is required for these models to obtain the predicted market shares. The DSS utilises 1,000 Halton draws per each of the 2,500 respondents to obtain the predicted market shares for a given scenario. As such, each scenario run requires 17,500,000 (2,500 respondents × 1,000 draws × 7 (modes)) calculations per market segment. Given the large number of simulated draws, the running of each scenario may be quite time consuming. For this reason, a progress bar is provided as part of the DSS. Once the user has inserted the required attribute level percentage changes for a scenario, it is necessary to press the run button on the right hand side of the input screen. This will initiate the simulation process which should not be stopped during the simulation. If necessary, the user may stop the simulation part way through a simulation run by pressing the escape key. Given the time taken to perform simulations, a number of preloaded simulations have been stored for the user. These may be accessed via the buttons provided on the right hand side of the DSS input screen. Also provided is a button that can quickly reset the simulation to the base model.

44

Figure 9: DSS output screen I

Figure 10: DSS output screen II The output screen, as shown in Figures 9 and 10, provides a number of outputs for the user. Of particular interest are the predicted changes to the modal shares by segment and the implied elasticities generated from the model (Figure 9). Figure 10 shows firstly what the average market changes are for all segments (excluding the MTPT segment) and secondly, how, by travel segment, the market shares change by trip length. Figure 11 demonstrates the output for a ten percent increase in the main mode travel time of taxis applied to all trip lengths.

45

Figure 11: DSS output screen for a ten percent taxi main mode travel time increase for trips all trip lengths The results shown in Figure 11 suggest such an increase will have on average a larger impact on night time travel and MPTM card holders whilst having only a marginal impact on business and day to day general activity trips. Figure 12 shows the results for a taxi travel time increase of 5 minutes for trips under 20 minutes. It is important to note that the fare increase placed some respondents into a higher travel time, hence the small impacts shown for the 20 to 40 minute travel time segments. The price elasticities for such a change can be seen also.

Figure 12: DSS output screen for a five minute taxi main mode travel time increase for trips under 20

minutes

46

Table 21 presents some indicative taxi direct elasticities obtained from the DSS. Shown are the elasticities obtained for a two minute increase in the in-vehicle times for all trips irrespective of trip length and a $2.00 fare increase for taxis irrespective of trip length. In brackets are the percentage changes that are generated given the absolute time or fare increases. Also shown are the elasticities for waiting time, main mode or in-vehicle travel time and fares (based on ten percent changes to waiting time, travel time or fare respectively). Table 21: Some Taxi Elasticities obtained from the DSS Tourism Segment BusinessSegment Day to Day Activity Segment Night Time Travel Segment MPTP card holder Segment Weighted Average Attribute Absolute change In-vehicle time (2 min inc.) -0.387 (14.03%) -0.079(14.10%) -0.471(14.08%) -1.263(14.03%) -0.964 (14.10%) -0.477(14.06%) Fare ($2.00 inc.) -1.437 (9.72%) -0.556 (9.92%) -0.671 (9.85%) -1.079 (9.84%) -0.578 (19.91%) -0.977 (9.81%) Attribute Percentage change Waiting time (10% inc.) -0.603 -0.226 -0.273 -0.393 -1.533 -0.340 In-vehicle time (10% inc.) -0.430 -0.123 -0.657 -1.314 -0.954 -0.573 Fare (10% inc.) -1.478 -0.645 -0.753 -1.132 -0.605 -1.042 It should be noted that there are very few empirical studies available of sufficient quality (see Table 22) to be able to be used as a set of reference taxi elasticities. We might reasonably claim that the current study is behaviourally, the most detailed study ever conducted, including greater market segmentation than previous studies. Furthermore the equations used are highly non-linear in the influencing attributes such as fares and travel times (including logarithmic, quadratic and interaction forms), such that a set of average elasticities within each trip purpose segment are not meaningful. The evidence in Table 21 is hence illustrative, but substantive, for the specific policy changes in fares and travel times. Interpreting the Table 21 elasticities, as expected, the taxi direct elasticities for the business segment are lower than those of the other segments, with the exception of the fare elasticity for the MPTP card holder segment. MPTP card holders receive heavily discounted fares when travelling by taxi. The relatively small travel time elasticity for business travellers is an interesting finding, suggesting that the convenience of taxi use (door to door) is being built into the travel time response, in a context where many business trips are also not paid by the actual taxi user, but by the traveller’s employer or client. Tourists and late night travellers have the highest fare elasticities. Given the current exchange rate for the Australian dollar, the purchasing power for many tourists has been significantly diminished of late, and hence are relatively price sensitive with lower disposable income to spend on activities and travel. Those travelling late at night might be expected to have a relatively higher fare elasticity compared to other segments, given that much of the segment includes trips travelling to and from hotels and nightclubs or from other such expensive activities, where extra expenditure on travel might be seen as a significant impost on the night budget. Night time travellers and MPTP card holders appear to be relatively more sensitive to changes in taxi travel time than other segments.

47

11. Existing Evidence on Taxi Elasticities In order to provide some external validity to the elasticities obtained from the DSS, a literature review of existing studies dealing with the taxi industry was undertaken. Table 21 reports several elasticities based on this literature review. For example, Schaller (1999) finds that in New York City, the elasticity of taxi demand with respect to fares is –0.22, the elasticity of service availability with respect to fares is 0.28, and the elasticity of service availability with respect to total supply of service is 1.0. Based on these values he concludes that fare increases tend to increase total industry revenues and service availability, and that the number of taxi licenses can often be expanded without reducing the revenue of existing operators. Comparing the elasticities reported in Table 22 to those obtained from the current study provides some confidence that the results from the current study are plausible and consistent with the existing research. Indeed, all elasticities, with the exception of MPTP card holders (which have not been studied before), fall within the ranges of elasticities found from our literature review. For example, the taxi fare elasticities range from -0.22 to -1.75 with the largest reported value associated with trips associated with “going out”. This is similar to our late night travel segment which also has the largest fare elasticity magnitude. Further, our waiting time elasticities appear to be in line with those of the reported literature, as to is the fact that the relative rankings of the elasticities in terms of magnitude are fare, travel time followed by waiting time. Indeed, our values are very similar to those reported by Rouwendal (1998). 12. Conclusions

This report has investigated the behavioural influences on traveller choice of mode for specific trips in the Melbourne metropolitan area, with a special focus on understanding the factors that influence the choice of, and hence demand for, taxis and hire car services. Given the importance of positioning preferences for taxi and car hire services within the broader set of modal options (including car, train, tram and bus), we have developed a modal choice model capability for all available modes of transport for trips undertaken by individuals or groups of individuals in the broad categories of corporate travellers, international and domestic tourists, late night social (including visiting friends and relatives) users, locals undertaking social outings, and users that hold multi-purpose trip program (MPTP) cards, which provide for a 50 percent discount on taxi fares.

48

Table 22: Direct Taxi Demand Elasticities with respect to Fare and Service

Notes: SP: Stated preference; TP: Transfer Price2 ^: The original elasticity provided in Schaller (1999) is a

revenue elasticity (-0.22) with respect to fare, which is equivalent to a demand (kilometre) elasticity of -1.22. BITRE Database*: http://www.bitre.gov.au/tedb/index.aspx New data has been collected using state of art choice experiments referenced around recent modal trip activity of a sample of individuals undertaking travel within the Melbourne Metropolitan Area in 2012. Combined with estimated modal choice models of the multinomial or mixed multinomial logit form, we have identified the key drivers of choices made amongst available modes of transport, with a specific focus on the role that taxis and hire cars play in the modal mix. The findings for each trip purpose segment, especially in respect of fares and service levels, have been integrated into a Decision Support System, calibrated to known population modal market shares, to give the Victorian Taxi Inquiry a capability of identifying behavioural responses to changes in fares and service levels (represented as direct and cross elasticities) as well as

2 “SP methods present individuals with hypothetical scenarios and use the responses supplied to reveal information about the preferences underlying the choices made. TP methods elicit from each respondent the change in an attribute level of their chosen mode which would be just sufficient to cause a change in behaviour” (Toner 2010, p.306).

Study Location Data Elasticity measure

Fare In-vehicle time Waiting time

Schaller (1999) New York, USA

Time series (1990-96)

Revenue -0.22^

Flores-Guri (2003) New York, USA

Time series (1990-99)

Kilometre driven

-1.05

Toner (2010) Four UK cities

SP/TP (collected in 1989-91)

Number of trips

-1.0 -0.10 -0.07

Rouwendal (1998) the Netherlands

SP (collected in 1997)

Number of trips

All taxi users: -1.14 Business: -0.76 Going out: -1.75 Going to the railway station: -0.69

Business: -0.44 Going out: -0 52 Going to the railway station: -0.35

Business: -0.58 Going out: -0.62 Going to the railway station: -0.48

Beesley (1979) London, UK

Time series (1951-52)

Kilometre driven

-0.35

Wong (1971) cited in Frankena and Pautler (1984)

Washington DC, USA

N/A Number of trips

-1.4

Applied Economics Associates (1978) cited in Frankena and Pautler (1984)

Seattle, USA

N/A Number of trips

-1.0

Kitch et al. (1979) cited in Frankena and Pautler (1984)

Chicago N/A Number of trips

-0.80

McGillivray (1979), cited in Frankena and Pautler (1984)

Danville, USA

Time series: 1975-77

Number of trips

-0.60

Brown and Fitzmaurice (1978) cited in Frankena and Pautler (1984)

21 cities, USA

N/A Number of trips

-0.80

Orfeuil and Hivert (1989), cited in BITRE Database*

Paris, France

N/A N/A -0.50

Queensland Transport (2000)

Queensland, Australian

N/A Number of trips

Brisbane: -0.36 Other cities: -0.50

Booz Allen Hamilton (2003)

Canberra, Australia

SP (collected in 2002)

Number of trips

All taxi users: -0.36 Peak hour: -0.23 Off peak: -0.41

All taxi users: -0.10

49

predictions of changes in market modal shares. This desktop capability enables the Inquiry to investigate numerous demand-response scenarios in respect of reform options for the taxi and hire car sector.

50

Appendix A: Statistical Properties of Stated Choice Designs 1. Efficient Choice Designs A statistically efficient design is a design that minimises the elements of the asymptotic (co)variance matrix, resulting in more reliable parameter estimates for a design with a fixed number of choice observations. The generation of an efficient experimental design therefore requires a priori knowledge as to the elements within the asymptotic (co)variance matrix of the model to be estimated, which in most instances will not be known by the analyst prior to model estimation. It therefore becomes necessary for the analyst to conjecture a set of priors that may be used to construct the asymptotic (co)variance matrix of different designs, which may then be compared in order to determine which will provide the greatest level of statistical efficiency. A number of different measures of design efficiency have been postulated within the literature (see Huber and Zwerina 1996 for example), all of which are derived from the work of McFadden (1974) on random utility theory (RUT) and summarised in a number of sources (e.g., Louviere, et al. 2000; Train 2009; Hensher et al. 2005). To explain the concept of RUT, consider a situation in which an individual is faced with multiple choice tasks involving a series of discrete choices made from a universal but finite number of alternatives. Let subscripts n and j refer to choice task n = 1, 2, …, N, and alternative j = 1, 2, …, J. The utility possessed by an individual for alternative j in choice task n may be expressed as:

,jn jn jnU V ε= + (A.1) where Ujn is the utility associated with alternative j in choice set n consisting of an observed component of utility for each alternative j in choice task n, Vjn, as well as a component, εjn, that is unobserved by the analyst. The observed component of utility is assumed to be a linear additive function of several attributes with corresponding weights. These weights are the unknown parameters to be estimated. We distinguish between generic parameters and alternative-specific parameters. Generic parameters have the same value for all alternatives that the corresponding attribute appears in, in contrast to alternative-specific parameters that may be different for each alternative. Let the generic and alternative-specific parameters be denoted by *,kβ *1, , ,k K= and ,jkβ 1, , ,jk K= respectively, with their associated attribute levels *

jknx and jknx for each choice situation n. The observed utility including both generic and alternative-specific attributes may be represented as equation (A.2). *

* *

1 1

, 1, , , 1, , .jKK

jn k jkn jk jknk k

V x x j J n Nβ β= =

= + ∀ = ∀ = (A.2) The presence of subscript j in jkβ allows for the estimation of alterative-specific parameter estimates across the j utility specifications in a labelled choice experiment. The total number of parameters to be estimated is equal to * .jj

K K K= + Assuming that the unobserved components

51

of utility, ,jnε are independently and identically extreme value type I distributed, the probability,,inP of choosing alternative i in choice set n becomes:

( )( )

1

exp, 1, , , 1, , ,

exp

inin J

jnj

VP j J n N

V=

= ∀ = ∀ =

(A.3) and the log-likelihood as a function of the parameters, becomes (assuming a single respondent)

* *

*

1 1

* * * *

1 1 1 1 1 1 1

( , ) log

log expj j

N J

jn jnn j

K KN J K J K

jn k jkn jk jkn k ikn ik jknn j k k i k k

L y P

y x x x x

β β

β β β β

= =

= = = = = = =

=

= + − +

(A.4)

where the vector y describes the binary outcome of all choice tasks, such that yjn equals one if alternative j is chosen in choice task n and is zero otherwise. The asymptotic (co)variance matrix is related to the second derivative of the log-likelihood function. Allowing for alternative-specific and generic parameter estimates, this is set out in equation (A.5) (see Rose and Bliemer 2009): 1 2 2

1 2

2 ** * * *

1 2* *1 1 1

( , ), , 1, , ,

N J J

jk n jn jk n in ik nn j ik k

L x P x P x k k Kβ ββ β = = =

∂ = − − ∀ = ∂ ∂ (A.5a)

1 1 1 1 2 2 1

1 1 2

2 ** * *

1 1 2*1 1

( , ), 1, , , 1, , , 1, , ,

N J

j k n j n j k n ik n in jn ij k k

L x P x x P j J k K k Kβ ββ β = =

∂ = − − ∀ = = = ∂ ∂ (A.5b)

( )

1 1 2 2 1 2

1 1 2 2

1 1 2 2 1 2

1 22 *1

1 21

, if ;( , )

1, , , 1, , .

1 , if .i

N

j k n j k n j n j nn

i i jNj k j k

j k n j k n j n j nn

x x P P j jL j J k K

x x P P j j

β ββ β

=

=

≠∂ = ∀ = =∂ ∂ − − =

(A.5c)

Equations (A.5a-c) do not rely on the outcomes, y. In addition, assuming M respondents, these second derivatives are multiplied by M. In case only generic parameters exist, only equation (A.5a) remains, which is similar to the result in McFadden (1974). In the case of only alternative-specific parameters, only Equation (A.5c) remains. The maximum likelihood (ML) estimates of *( , )β β are found by maximizing the log-likelihood function, or alternatively, by setting the first derivatives equal to zero (since the log-likelihood function is concave). Denoting the ML estimates as *ˆ ˆ( , )β β we have: *

* *

( , )

ˆ ˆ( , ) arg max ( , ).Lβ β

β β β β= (A.6)

52

Let *( , )β β denote the true values of the parameters. McFadden (1974) showed that for the generic case, that the ML estimates *β̂ are asymptotically normally distributed with mean *β and asymptotic (co)variance matrix, Ω , which is equivalent to the negative inverse of the Fisher information matrix. This result can be extended to the case of (a combination of generic and) alternative-specific parameters. The Fisher information matrix I is defined as the expected values of the second derivative of the log-likelihood function, that is: 2 *

* ( , )( , ) .

'

LI M β ββ ββ β

∂= ⋅∂ ∂

(A.7) Hence, the asymptotic (co)variance matrix becomes the following K K× matrix:

12 *1* 1 ( , )

( , ) .'

LIM

β ββ ββ β

−− ∂ Ω = − = − ∂ ∂

(A.8) Within equation (A.8), the presence of M suggests that the (co)variances become smaller with larger sample sizes. This also follows for the asymptotic standard errors, obtained by taking the square root of the diagonal elements (including M) of this matrix (i.e., variances). By taking the square root of M, one will observe diminishing improvements to the standard errors over increases in sample size. The asymptotic (co)variance matrix plays an important role when determining efficient experimental designs, as will shown in the next section. 2. Measuring the Statistical Efficiency of SC Experimental Designs To compare the statistical efficiency of SC experimental designs, a number of alternative approaches have been proposed within the literature (see e.g., Rose and Bliemer 2009). The preferred measure within the literature is D-error, computed by taking the determinant of the asymptotic (co)variance matrix and applying a scaling factor to account for the number of parameters, .K It is common in generating efficient designs to do so assuming a single respondent (i.e., M = 1) representative of all respondents, an assumption consistent with the MNL model form. Hence, we the D-error is computed as:

( )1/

2 *1/ ( , )

D-error det det .'

KK L β β

β β ∂= Ω = − ∂ ∂

(A.9) The D-error measure of design efficiency may be used to distinguish between designs of the same dimension so that if the D-error is low, the (co)variances of the parameter estimates are also low. Two popular approaches exist within the literature for computing the D-error of a design. The first approach assumes that the analyst has no information, not even likely sign, of the true parameter values. This assumption results in what the literature has termed Dz-error in which the parameters are assumed to be all equal to zero. The Dz-error can be computed as:

53

1/2

z

(0,0)D -error det ,

'

KLβ β

− ∂= − ∂ ∂

(A.10) An alternative measure is the Dp-error, which assumes that the analyst has some prior knowledge in the form of prior parameter values *( , )β β , for at least one parameter. The Dp-error measure may therefore be computed as:

1/2 *

p

( , )D -error det .

'

KL β β

β β

− ∂= ∂ ∂

(A.11) For the purposes of this paper, we will assume non-zero priors and hence rely on the Dp-error criterion. For designs of the same dimensions (i.e., number of choice sets, alternatives, attributes and attribute levels), the design(s) with the lowest D-error is (are) termed the D-optimal design(s). Determining whether a design is D-optimal is difficult however, given the large number of possible attribute level combinations that may exist for a design of fixed dimensions. In general, for all but the smallest of designs, it will be unlikely that an analyst will be able to calculate the D-error for all possible design permutations. It is therefore more appropriate to label a design as D-efficient as opposed to D-optimal. Assuming not all possible permutations are to be examined, the literature has suggested a number of different strategies to determine which permutations should be examined and how (see e.g., Huber and Zwerina 1996).

54

Appendix B: The CAPI Screens Introduction

55

Trip Explanation

56

Taxi / Hire Car Usage

57

Recent Trip, Part 1

58

Recent Trip, Part 1 (Full Fields)

59

Recent Trip, Part 2

60

Your Trip - Bus

61

Your Trip – Bus (Full Fields)

* The second part of Question ii is also included in Tram & Train modes.

62

Your Trip - Tram

63

Your Trip - Train

64

Your Trip - Car

65

Your Trip - Walk

66

Your Trip - Taxi

67

Your Trip – Hire Car

68

Game Introduction

69

Practice Game – Bus, Car, Taxi

70

Practice Game – Car, Taxi

71

Waiting for game start

72

Game – Bus, Car, Taxi

73

Game– Car, Taxi

74

Attitudes to Public Transport

75

Attitudes to Taxis/Hire Cars

76

Socioeconomic Characteristics

77

Thanks

78

Appendix C: Recruitment Screener

Taxi and Hire Car Study Recruitment Screener – TRC 4303 Refusals: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Terminations: 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Good morning/afternoon/evening.... My name is ____________ from Taverner Research on behalf of the Victorian Taxi Industry Inquiry. We are undertaking a study about Taxis and Hire Cars in Victoria to establish current transport use and preferred options for travel. A hire car is similar to a taxi, however a hire car requires that you make a booking in advance and you have a fixed agreed fee for the trip rather than use a meter. Q1. First do you live in greater Melbourne or are you visiting? Clarify if visiting from inside Australia or Overseas? 1. Greater Melbourne-Skip to Q3 2. Australian Visitor-Classify as Tourist 3. Overseas Visitor-Classify as Tourist Q2. Have you made any trips in or around Melbourne recently, where, for some or all of the trip, you either could have chosen to take a taxi or hire car, or you actually did take a taxi or hire car for non-business related activity? 1. Yes – Check quota requirements and skip to recruitment if needed 2. No – Continue Q3. Have you made any trips in or around Melbourne recently (including to or from the airport), where, for some or all of the trip, you either could have chosen to take a taxi or hire car, or you actually did take a taxi or hire car for (READ OUT)? 1. A business-related travel at any time of day (24hrs) 7 days a week, or –Classify as Business 2. For any reason after 10pm and up to 4am in morning, or –Classify as Late Night

79

3. For any reason after 4am in the morning and up to 10pm – Include as Tourist if no to 1 & 2 4. Neither - terminate Check quotas and recruit for trip required Quota Group Quota Achieved (Interviewer to update) Tourist 135Business Travel 135Late Night Travel 135

80

Interview Recruitment As I mentioned earlier, the Victorian Taxi Industry Inquiry is conducting this study to collect information on the communities’ views on taxis, hire cars and various transport options. As someone who travels by taxi (or hire car), we would like to include your views in the study. By agreeing to participate in this study, you will assist the Inquiry in improving options for travelling by taxi and/or hire car. The study will only take about 20 minutes of your time, but as it includes seeking your views on several travel options, we have set-up the survey on an iPad2 much like a computer game, so you are able to study the options and carefully consider your responses. To compensate you for your time, we will send you a $20 cheque or you can donate the money to one of these charities. Cancer Council Australia Red Cross

Royal Flying Doctors RSPCA

Cheque to: Name _____________________________________________ Address: _____________________________________________ City: _____________________________________________ State & Post Code:___________________________________________ Country: ___________________________________________________ If you can’t do it right now, we can make an appointment for you to come back later today or on another day.

81

Respondent Number:________ In the very unlikely case my supervisor needs to contact you for validation purposes, may I have your first name and telephone number please? Your details will not be used for any other purposes.

Name:

Phone:

I certify that this is a true, accurate and complete interview, conducted in accordance with the IQCA standards and the MRSA Code of Professional Behaviour (ICC/ESOMAR). I will not disclose to any other person the content of this questionnaire or any other information relating to this project.

Interviewer: ______________________________________________________________________________________________

Signed: _______________________________________ Date: ____/02/2012

Time:___:___

82

References Bhat, C.R. (1998) Accommodating flexible substitution patterns in multidimensional choice modeling: formulation and application to travel mode and departure time choice, Transportation Research Part B, 32, 425-440. Bhat, C. R. (2000) A multi-level cross-classified model for discrete response variables, Transportation Research Part B, 34 (7), pp. 567-582. Hensher, D.A. (2010) The Estimation of Travel Demand through SP/RP, invited chapter for A. Cappelli and S. Nocera (eds.) Feasibility Decisions in Transportation Engineering, McGraw Hill, USA, 85-110. Hensher, D.A. (1994) Stated preference analysis of travel choices: the state of practice, Transportation, 21 (2), 107-134. Hensher, D.A. and Greene, W.H. (2003) Mixed logit models: state of practice, Transportation, 30 (2), 133-176. Hensher, D.A., Rose, J.M. and Greene, W.H. (2005) Applied Choice Analysis: A Primer, Cambridge University Press, Cambridge. Huber, J. and Zwerina K. (1996) The Importance of utility Balance and Efficient Choice Designs, Journal of Marketing Research, 33 (August), 307-317. Gilboa, I., Schmeidler, D. and Wakker, P. (2002) Utility in case-based decision theory, Journal of Economic Theory, 105, 483-502. Kahnemann, D. and Tversky, A. (1979) Prospect theory: an analysis of decisions under risk, Econometrica, 47 (2), 263-91. Louviere, J.J., Hensher, D.A. and Swait, J. (2000) Stated Choice Methods: Analysis and Applications in Marketing, Transportation and Environmental Valuation, Cambridge University Press, Cambridge. McFadden, Dan (1974) Conditional Logit Analysis of Qualitative Choice Behaviour. In Zarembka, P. (ed.), Frontiers of Econometrics, Academic Press, New York, 105-142. McFadden, D. and K. Train (2000) Mixed MNL models for discrete response, Journal of Applied Econometrics, 15, 447-470. Revelt, D. and Train, K. (1998) Mixed logit with repeated choices: households’ choices of appliance efficiency level. Review of Economics and Statistics. Rose, J.M., Bliemer, M.C., Hensher and Collins, A. T. (2008) Designing efficient stated choice experiments in the presence of reference alternatives, Transportation Research Part B, 42 (4), 395-406. Rose, J.M. and Bliemer, M.C.J. (2009) Constructing Efficient Stated Choice Experimental Designs, Transport Reviews, 29(5), 587-617. Schaller, B. (1999), Elasticities for taxicab fares and service availability, Transportation, 26(3), 283-297. Starmer, C. (2000) Developments in non-expected utility theory: the hunt for a descriptive theory of choice under risk, Journal of Economic Literature, XXXVIII, 332-382. Train, K. (2009) Discrete Choice Methods with Simulation 2nd Ed. Cambridge University Press, Cambridge.

83

References on Previous Elasticity Studies Associated with Taxis Basu, D. and Maitra, B. (2010) Stated Preference Approach for Valuation of Travel Time Displayed as Traffic Information on a VMS Board, Journal of Urban Planning and Development, 136(2), 214-224. Beesley, M. E. (1979) Competition and Supply in London Taxis, Journal of Transport Economics and Policy, 13(1), 102–131. Booz Allen Hamilton (2003) ACT Transport Demand Elasticities Study, Report for Department of Urban Services, ACT, Australia. Chintakayala, P.K. and Maitra, B. (2010) Modeling Generalized Cost of Travel and Its Application for Improvement of Taxies in Kolkata, 13(1), 42-49. Flores-Guri, D. (2003) An economic analysis of regulated taxicab markets, Review of Industrial Organisation, 23(3), 255-266. Frankena, M.W. and Pautler, P.A. (1984) An Economic Analysis of Taxicab Regulation, US Federal Trade Commission, Bureau of Economics Staff Report. Pells, S.R. (1990) The Demand for Taxi Services in Sheffield: An Empirical Study of the Value of Waiting Time and the Price Elasticity of Demand, Working Paper, Institute of Transport Studies, University of Leeds. Queensland Transport (2000) National Competition Policy Review of the Transport Operations (Passenger Transport) Act 1994, September 2000. Rouwendal, J., Meurs, H. and Jorritsma, P. (1998) Deregulation of the Dutch taxi sector, Proceedings of Seminar F, European Transport Conference, 37-49. Schaller, B. (1999), Elasticities for taxicab fares and service availability, Transportation, 26(3), 283-297. Toner, J.P. (1991) The Demand for Taxis and the Value of Time – A Welfare Analysis, Working Paper, Institute of Transport Studies, University of Leeds. Toner, J.P. (2010) The Welfare Effects of Taxicab Regulation in English Towns, Economic Analysis and Policy, 40(3), 299-312. Yang, L., Choudhury, C.F., Ben-Akiva, M., Silva, J.A. and Carvalho, D. (2009) Stated Preference Survey for New Smart Transport Modes and Services, Working paper, MIT Portugal.