Exploring Susceptibility to Use Demand Responsive Transport (DRT)

Shubham Jain (ORCiD: 0000-0002-2719-586X)

SUBMITTED IN TOTAL FULFILMENT OF THE REQUIREMENTS OF THE DEGREE OF MASTER OF PHILOSOPHY – ENGINEERING

Department of Infrastructure Engineering

The University of Melbourne

Australia

July 2016

Supervisors: Prof. Stephan Winter

Dr. Nicole Ronald

A/Prof. Russell Thompson

Abstract

Shared transportation providing point-to-point services on demand, although not an unknown element in urban mobility, has started gaining more presence with growth of information technology in the transport sector. These forms of transport modes will supplement or compete with existing public and private transport. Their mixed reception in the past is a matter of concern especially before making investment decisions. To find feasible opportunities of implementation, an estimation of the demand patterns in the target city is desirable.

This research provides and evaluates a methodology for this estimation that avoids ambivalent and expensive user surveys. Demand patterns are caused by the spatial variation of socio-economic and demographic characteristics, family structures, and travel behavior over the city. Thus, the new methodology takes into account the use of socio-economic and demographic data and current trip data from travel surveys of a sample of the population, along with usage patterns of existing similar services elsewhere in the world. Demographic factors such as gender, age, occupation, income, household structure, motor vehicle ownership, and driving licence availability together with trip characteristics such as current trip purpose, walking time, and waiting time can be analyzed to come up with demand patterns, and their variation in the city. Usage patterns from existing similar services worldwide are then used to explore the overall spatial pattern of susceptibility of DRT in a target city. The outcomes identify more favorable areas for implementation of DRT.

The methodology can be validated by applying it on existing transport modes in the target city which will also help in understanding the nature of competition among the proposed and existing transport modes. As the review of operating services is generic, it can be used in conjunction with respective travel surveys in different places. Similarly, review can be done for any proposed transport mode, and provided methodology can be applied for exploring demand patterns. The methodology is tested for Greater Melbourne in this study.

Further, synthetic population is created at household and person levels for Greater Melbourne. PopSynWin, which is based on Iterative Proportional Fitting (IPF) algorithm, and PopGen, which is based on Iterative Proportional Update (IPU) algorithm, are used

as tools for this purpose, generating two different synthetic populations. Both the generated populations are compared statistically, and the better one is used to assign travel diaries from sample travel surveys for a study region. The methodology to explore the susceptibility to use DRT, provided in this research, is applied on individual travel diaries to further explore demand patterns at a finer granularity.

The content of this thesis has been presented in following research forums:

• Jain, S., N. Ronald, R. Thompson and S. Winter (2016). "Predicting susceptibility to use demand responsive transport using demographic and trip characteristics of the population." Travel Behavior and Society, Volume 6, Pages 44–56, doi:10.1016/j.tbs.2016.06.001.

• Jain, S., N. Ronald and S. Winter (2015). Creating a Synthetic Population: A Comparison of Tools. 3rd Conference of the Transportation Research Group of India, Kolkata, India.

• Jain, S., N. Ronald, R. Thompson, R. Kutadinata and S. Winter (2015). Exploring susceptibility of shared mobility in urban space. Disruptive Mobility - A Global Summit Investigating Sustainable Futures, Cambridge. US.

• Jain, S., N. Ronald, R. Thompson and S. Winter (2015). Towards Agent Based Activity–Travel Pattern Modelling to Predict Travel Demand for Demand Responsive Transport in Melbourne. Conference of Australian Institutes of Transport Research, Melbourne, Australia.

• Ronald, N., Z. Navidi, Y. Wang, M. Rigby, S. Jain, R. Kutadinata, R. Thompson and S. Winter (2016). "Mobility Patterns in Shared, Autonomous, and Connected Urban Transport."Book chapter, Disruptive Mobility (under review).

Declaration

This is to certify that

i. the thesis comprises only my original work towards the Masters except where indicated, ii. due acknowledgement has been made in the text to all other material used, iii. the thesis is less than 50000 words in length, inclusive of footnotes, but exclusive of tables, maps and bibliography.

July 22nd, 2016

Shubham Jain

Acknowledgements

I wish to express special gratitude to Prof. Stephan Winter that he provided me the opportunity to do this research under his supervision, showed me how to observe, plan, and think like a researcher, and guided me through all scientific obstacles. I would like to thank my co-supervisors Dr. Nicole Ronald and A/Prof. Russell Thompson for guiding me during this research, and their contribution in terms of time and knowledge.

I would like to express my deepest gratitude to a few very special people - my parents and sister, who were always encouraging and supportive from all possible aspects, so that I could focus on my research. Special thanks to Zahra, Rahul, Aiswarya, Sudeep, Vineet, Hardik, and all my friends, colleagues, and housemates in Melbourne, who assisted me with many things, so that I could keep going.

Table of Contents

Chapter 1: Introduction ...... 1 1.1 Research Motivation ...... 1 1.2 Research Hypothesis ...... 3 1.3 Research Approach ...... 3 1.4 Study Area ...... 4 1.5 Research Design ...... 5 1.6 Thesis Structure ...... 7 Chapter 2: Literature Review ...... 8 2.1 Demand Responsive Transport (DRT) ...... 8 2.2 Understanding Travel Demand ...... 11 2.3 Preference Surveys ...... 14 2.4 The Effect of Socio-Economic, Demographic and Trip Characteristics on Travel Behavior ...... 17 Chapter 3: Identifying Parameters Favorable to DRT ...... 20 Chapter 4: Predicting Susceptibility of Regions to Use DRT ...... 30 4.1 Data and Methodology ...... 30 4.2 Analysis of Travel Survey Data ...... 36 4.3 Findings ...... 43 4.4 Validation ...... 47 4.5 Conclusions ...... 60 Chapter 5: Predicting Likeliness of Trips to Convert to DRT ...... 62 5.1 Synthetic Population ...... 63 5.1.1 Background ...... 64 5.1.2 Data Preparation ...... 66 5.1.3 Synthesis ...... 67 5.1.4 Results ...... 67 5.1.5 Interpretation of Results ...... 76 5.2 Travel Diary Assignment ...... 77 5.3 Exploring DRT Susceptibility of Trips in Yarra Ranges ...... 82 5.4 Conclusions ...... 92 Chapter 6: Conclusions ...... 94 Appendix ...... 98 Bibliography ...... 107

List of Tables

Table 1.1: Research design ...... 6 Table 3.1: Important findings from review of existing DRT services ...... 24 Table 3.2: Parameters identified from literature review ...... 27 Table 4.1: Summary statistics of all parameters favorable to DRT ...... 34 Table 4.2: Classification of SA3s into SA4s ...... 35 Table 4.3: DRT favorable parameters at SA4 level ...... 36 Table 4.4: Correlation of parameters affecting susceptibility of DRT ...... 38 Table 4.5: Total standardized value of DRT susceptibility for each SA3 regions ...... 40 Table 4.6: DRT susceptibility for SA4 regions ...... 43 Table 4.7: Summary statistics for Public Transport favorable parameters ...... 49 Table 4.8: Public transport favorable parameters at SA4 level ...... 50 Table 4.9: Correlation of parameters affecting susceptibility of public transport ...... 51 Table 4.10: Total standardized value of public transport susceptibility for each SA3 region .... 52 Table 4.11: Share of Public transport trips in SA3 regions of Melbourne ...... 55 Table 4.12: Difference of standardized values of predicted DRT and public transport susceptibility with existing public transport usage ...... 58 Table 5.1: Overall comparison of synthesized and actual population ...... 68 Table 5.2: Distribution of household level control variables ...... 68 Table 5.3: Distribution of person level control variables ...... 69 Table 5.4: Weighted average of absolute percentage point difference in distribution ...... 71 Table 5.5: Delta1 for all control variables ...... 73 Table 5.6: Delta4 for all control variables across all SA4 ...... 76

Appendix

Table A.1: Raw data from VISTA sample of 40 SA3 regions for DRT favorable parameters.…….98 Table A.2: Standardized data of 40 SA3 regions for DRT favorable parameters………….………...100 Table A.3: Raw data from VISTA sample of 40 SA3 regions for public transport favorable parameters...... …...102 Table A.4: Standardized data of 40 SA3 regions for public transport favorable parameters.....104

List of Figures

Figure 1.1: Map of Greater Melbourne and its SA3 regions...... 5 Figure 4.1: Map of Greater Melbourne and its SA3 regions...... 32 Figure 4.2: Methodology to explore susceptibility of DRT...... 33 Figure 4.3: Susceptibility of DRT in Greater Melbourne...... 44 Figure 4.4: DRT favorable parameters comparison for Melbourne City and Yarra Ranges...... 46 Figure 4.5: DRT favorable parameters comparison for Melbourne City and Brunswick – Coburg...... 46 Figure 4.6: DRT favorable parameters comparison for Yarra Ranges and Melton - Bacchus Marsh...... 47 Figure 4.7: Susceptibility of public transport in Greater Melbourne...... 54 Figure 4.8: Existing usage pattern of public transport in Greater Melbourne...... 56 Figure 4.9: Predicted public transport and DRT susceptibility vs. existing public transport usage...... 57 Figure 5.1: (Number of motor vehicles, Dwelling structure) variation...... 72 Figure 5.2: (Labor force status, Gender) variation...... 72 Figure 5.3: Delta2 against number of SA1...... 74 Figure 5.4: Visualizing Delta3 at SA3 level for PopSynWin and PopGen...... 75 Figure 5.5: Visualizing Delta3 at SA4 level for PopSynWin and PopGen...... 75 Figure 5.6: Map of Yarra Ranges in Greater Melbourne...... 78 Figure 5.7: Travel diary assignment to synthetic persons using VISTA sample...... 80 Figure 5.8: Trip count proportions by trip purpose...... 81 Figure 5.9: Trip count proportions by travel mode...... 81 Figure 5.10: Population proportion by the number of trips in a day...... 82 Figure 5.11: Number of trips satisfying a particular parameter...... 84 Figure 5.12: Number of trips satisfying as number of parameters...... 84 Figure 5.13: Likeliness of trips to convert to DRT...... 85 Figure 5.14: Proportion of trip purposes as per likeliness to convert to DRT...... 85 Figure 5.15: Proportion of travel modes as per likeliness to convert to DRT...... 86 Figure 5.16: Temporal distribution of trips across categories of likeliness...... 88 Figure 5.17: Distribution of individuals who are less likely to convert to DRT in Yarra Ranges...... 89 Figure 5.18: Distribution of individuals who are likely to convert to DRT in Yarra Ranges. .... 90 Figure 5.19: Distribution of individuals who are highly likely to convert to DRT in Yarra Ranges...... 91 Figure 5.20: SA1 Regions in Yarra Ranges as per percentage of individuals who are highly likely to convert to DRT...... 92

Chapter 1: Introduction

Population mobility and provision of convenient transport options for accessing resources and the places of activities are problems for growing cities. To meet these challenges in a sustainable manner, taking into account congestion, environmental impacts, and fuel consumption, new forms of transport are needed to be explored. One possible solution is to encourage demand responsive transport (DRT) modes which provide flexible point-to- point service on casual demand. Different types of public and private on-demand transport services operate in various parts of the world that provide shared forms of transport. There are instances of both failure and success of DRT. Also, these forms of transport modes will supplement or compete with existing public and private transport. Their mixed reception in the past is a matter of concern especially before making investment decisions. Depending on population characteristics, current trip patterns and available transport options, different cities and different regions of a city respond differently to newly suggested transport services. Therefore, before implementing any particular service, it is essential to identify the potential regions that are more susceptible to use DRT services, and also the potential trips that can be converted to DRT.

1.1 Research Motivation Demand patterns of travel in a city are needed to be predicted realistically and efficiently for better transport planning (Arampatzis et al., 2004). Many transport models (Damm 1980, Hirsh et al. 1984, Kitamura and Lam 1984, Ben-Akiva and Lerman 1985, Pas 1985, Hirsh et al. 1986, Horowitz 1991, Axhausen and Gärling 1992, Dagsvik 2000) have been presented for modeling travel demand of various regions, but it is always challenging to

1 incorporate changes in transportation environment and new modes of transportation (Train 1979, Swait and Adamowicz 2001, Shaheen and Rodier 2004, Lu and Gosling 2007, Habib et al. 2012). Travel demand patterns are the aggregates of individual behaviors and understanding the behavior changes require the development of transport models that are responsive to changes in the external environment and also characteristics of the travelers (Dix and Layzell 1986). Therefore, characteristics of households and individuals within a population are typically used to examine the travelers and their trip properties including time, mode, and origin-destination. (Axhausen and Gärling 1992). The travel diaries recorded in household transport surveys for a sample of the population contain details of trips such as purpose, time, origin, destination, distance, duration, waiting time, walking time, and mode of transportation. These travel diaries give the reflection of travel demand in the city, its distribution across the regions, and existing transportation mode distribution (Bose and Sharp 2005). This research, however, aims to predict travel demand patterns for a new transport mode, which is not captured in these household transport surveys as they only record the existing trip details of the population. Hence, the challenge of this research is to identify which trips among the existing demand have more potential to be converted to the new mode of transport after its implementation. Also, it is important to explore variation in their likeliness to convert across a city to make decisions about feasible regions of implementation.

Piatkowski and Marshall (2015) mention in the context of bicycling that appropriate strategies for encouraging a transport mode vary greatly across different population groups; thus, identifying groups that are more prone to change in travel behavior may act as an important strategy for rapidly and effectively impacting commute mode choice. The shift to a new mode of transportation depends on several factors such as its affordability, travel time, travel cost, convenience, flexibility, technology, and its level of service (LOS) compared to existing modes of transportation. As these factors are not known at the time of the inception of new services, and depend on their uptake in the market, they are not available in advance for an assessment of mode shift. So the alternative approach of this research is to look for the target market of these services and identify individuals who will be willing to use these new services and which of the trips they will be converting. In principle, one of the possible solutions for this is to conduct a stated preference (SP) survey (Jordan 1988, Louviere 1988, Timmermans and Golledge 1990, Hensher 1994) to explore the willingness to shift to new modes of transport. However, the results from

2 preference surveys can be misleading (Levin and Louviere 1981, Wardman 1988, Wardman 1991, Gärling et al. 1996, Gärling et al. 1998, Clifton and Handy 2001) because the surveyed people lack awareness about the features and service details of the new mode, and had no time to adapt to the existence of a new mode available to them. This argument is even more applicable if the proposed services are novel in their character, which may be the case with the demand responsive forms of transport that are in the focus of this study (Takeuchi et al. 2003, Yang et al. 2009, Al-Ayyash et al. 2015). Moreover, the time required to carry out these preference surveys is huge (Yang et al. 2009).

Hence, this research develops an alternative methodology, which avoids a service- specific survey. This research proposes a methodology to predict travel demand patterns in a city for DRT using socio-economic and demographic characteristics of the population, current trip characteristics, and the usage patterns of existing similar services elsewhere in the world.

1.2 Research Hypothesis This study investigates the research hypothesis that usability patterns of (demand responsive) transport services in different regions of the world combined with demography and current trip characteristics of the target city can be used to determine the susceptibility for novel transport services, eliminating the requirement of service-specific surveys.

1.3 Research Approach The methodology is based on census data and data collected in conventional travel surveys. Travel surveys are used in metropolitan cities to understand the people’s (current) travel behavior and demand for transport planning. These travel surveys record socio-economic characteristics, demographic characteristics, household attributes and travel details/diaries. In these surveys, each individual in the sample has a travel diary comprising a sequence of trips in a representative day, trip attributes such as origin and destination region, travel mode, trip purpose, departure time, trip duration, waiting time, walking time and distance. Details of these travel diaries along with socio-economic and demographic characteristics of individuals and households in the population such as age, gender, relationship in household, vehicle ownership, employment type, household size and structure, household income, and driving license availability are available for predicting travel demand patterns for new modes of transportation.

To identify usability, user profile, usage pattern and dominant favorable characteristics of users and trips for DRT services, a review of features and usability pattern of some of the existing similar transport services in different regions of the world has been conducted, which allows identifying certain trip characteristics, individual and household characteristics that are likely to switch to DRT services. Extracted patterns from travel survey data will be summed up to explore the overall spatial pattern of susceptibility of DRT in the target city. These patterns can be mapped to find more favorable areas for implementation of DRT services. A similar exercise can be performed to find susceptibility for existing forms of transport in the target city, for example, conventional public transport, where the resulting estimated patterns can be compared with existing ones in order to validate the approach.

Hence, this research will provide a methodology on how demand for new modes of transport, demand responsive in particular, can be predicted. This methodology can be evaluated by applying it on some of the existing transport modes, and thus its predictive capacity is expected to be realistic and reliable.

This research is designed upon the idea of using population demographic and trip data (which can come from census and household travel survey data) for predicting usability of a proposed transport service. The challenge is that these kinds of available data have the details of the population’s current travel behavior or mode choices, but not about their willingness if a new transport service is proposed. This is exactly the challenge this thesis addresses. Traditionally, in such cases preference surveys are applied. However, this thesis, by highlighting the drawbacks of such surveys especially due to the innovative nature of any proposed service, proposes to use insights of user characteristics from similar services operating elsewhere in the world. It only picks up the parameters that are found to be dominant in various different studies reviewing such services. The suggested methodology thus depends on the existence of similar services elsewhere, and studies of their user characteristics; it would not be applicable without. But from there this research actually develops a model using household travel survey data along with additional insights from elsewhere.

1.4 Study Area The research methodology will be investigated in Greater Melbourne, Australia. The variation of the identified parameters from the review of existing similar services will be studied across Greater Melbourne, at the Statistical Areas 3 or SA3 level. Figure 1.1

4 presents the map of Greater Melbourne with demarcated SA3 census regions. This study will lead to the identification of SA3s that are more favorable for implementation of DRT services. These SA3 regions are very large in size and variations in susceptibility to DRT may exist within the region itself. Moreover, travel demand has to be predicted at a finer level for further modeling and simulation. Hence, one of the SA3s will be selected for further investigation of the trips that are more likely to convert to these services. A synthetic population of that SA3 region will be assigned travel diaries for a day from sample travel survey leading to existing travel demand and trip pattern of the region. The identified parameters favoring the DRT will then be used to assess the likeliness of those trips to be converted to DRT.

Figure 1.1: Map of Greater Melbourne and its SA3 regions. 1.5 Research Design Research design applicable to this study involving objectives and methods used to address the objectives has been presented in Table 1.1.

Table 1.1: Research design

Objective Method Chapter

To identify parameters Various DRT services in different parts of the 3 favorable to DRT world will be reviewed and dominant socio- demographic and trip characteristics of the population favoring the use of DRT will be identified.

To determine the Variation in the characteristics identified in the 4 susceptibility of various Chapter 3 will be used to predict the susceptibility regions in a target city to to use DRT services in urban space. It will involve use DRT analysis of the travel survey data of a sample of population in order to find the pattern of DRT favorable characteristics in a city. It will predict more feasible regions of implementation of DRT. The methodology will be tested using travel survey data of Greater Melbourne and corresponding feasible regions for implementation of DRT will be identified.

For a particular region, A synthetic population will be created for the 5 predict the number of entire Greater Melbourne area, and then travel trips which are more dairies will be assigned to a particular region using likely to convert to DRT sample travel survey data. The DRT favorable services, how they are parameters identified in Chapter 3 will be used to distributed temporally and predict which trips done by the population of that spatially, and what are region are more likely to shift to DRT. their current trip characteristics such as trip purpose and travel mode.

1.6 Thesis Structure This thesis has been structured into 6 chapters as follows:

Chapter 1: Introduction

Introduction chapter talks about the underlying problem and motivation for this research, its hypothesis, approach, and the study area it is applied to.

Chapter 2: Literature Review

Literature Review chapter talks about the background of DRT, research in travel demand prediction, preference surveys, and the Effect of socio-economic, demographic, and trip characteristics on travel behavior.

Chapter 3: Identifying Parameters Favorable to DRT

This chapter reviews various studies undertaken to know usability of DRT services in various regions of the world, and identifies dominant parameters which favor the use of DRT.

Chapter 4: Predicting Susceptibility of Regions to Use DRT

This chapter uses the parameters identified in Chapter 3 to develop a methodology to predict susceptibility of various regions in a city to use DRT. It uses travel survey data for a sample of population for this purpose. The methodology has been applied to predict susceptibility of census regions in Greater Melbourne to use DRT.

Chapter 5: Predicting Likeliness of Trips to Convert to DRT

This chapter talks about creating 100% travel diaries for population of a region. It involves creation of a synthetic population using sample micro-data and aggregate data of the actual population, and assignment of travel diaries to the synthetic population using sample travel survey data. The methodology has been applied to one of the census regions of Greater Melbourne.

Chapter 6: Conclusions

This chapter concludes the findings of this research, its implications, limitations, and scope of future work.

Chapter 2: Literature Review

This chapter reviews the research relevant for this thesis. Section 2.1 discusses the demand responsive transport, its features, applications, and variations. Section 2.2 talks about the importance of travel demand estimation in new transport services, background of research in travel demand modelling with focus on innovative transport services, current practices, and various challenges. Section 2.3 highlights the application of preference surveys in travel demand prediction and knowing willingness of people to use a proposed transport service, it also presents various shortcomings of these surveys. Section 2.4 signifies the effect of socio-economic, demographic and trip characteristics on travel behavior of the population by reviewing various researches which studied one or more of these characteristics.

2.1 Demand Responsive Transport (DRT) According to Ambrosino et al. (2004), DRT is an “intermediate form of transport, somewhere between bus and taxi which covers a wide range of transport services ranging from less formal community transport through to area-wide service networks” (p. 26). According to Bakker (1999), DRT is a “transportation option that falls between a private car and conventional public bus services. It is usually considered to be an option only for less developed countries and for niches like elderly and disabled people” (p. 109). The latter comment is clearly becoming outdated, observing the global success of Uber and other commercial ad-hoc DRT. DRT is also defined as a form of shared-ride public

8 transportation service that is responsive to the requests of passengers and deals with a number of customer requests served by a fleet of vehicles, while a number of constraints such as limited capacity (the capacity constraint) must be observed (Basnal et al. 2015).

The continuous expansion of metropolitan cities away from the city centers, have led to the growth of commuter trips which are cross-regional in nature (Data Management Group 1996, Data Management Group 2011). Cross-regional trips result in long travel times for commuters, especially for travel modes other than driving. These long distance trips are multimodal, may involve the use of multiple transit services or the interaction between two travel modes, and hence, complex. This complexity often result in delays due to the lack of coordination. DRT can be a solution in such circumstances.

Various transport modes such as buses, coaches, taxis, adapted taxis, and minibuses have been undertaken in several DRT services. In many instances, community transport operators, local authority, ambulance vehicles, and car-club members provide DRT services. DRT can be provided in free-standing and can also be integrated between different modes to provide feeder services for bus, tram and train services (Mageean and Nelson 2003). DRT can operate on demand, picking up people and dropping them off according to their needs. When referring to the flexible or non-fixed route of DRT, it does not mean that there are not any stop points, but the stops might be terminals, fixed points along the route (high-demand points), and non-predefined stop points (usually the doorstep of a user) (Silva et al. 2015).

DRT can solve the challenges of public transportation in low-density urban areas and also short distance pickup and the delivery problem (Wang and Winter 2010). Koffman (2004) revealed that DRT was generally used in small and difficult to serve areas, though, he also presented the cases where DRT was used in large areas, or else provided services at times of low-demand. Traditionally, DRT was implemented to provide services for the disabled, elderly, and where demand is too low for other public transports (Black 1995, Bakker 1999). Sihvola et al. (2010) investigated open-for-all DRT in high-density areas.

Flexibility, ability to meet demand in low or unpredicted demand areas where traditional transportation prices are high (Zografos et al. 2008), and the potential to reduce gas emissions (Rodier et al. 1998, Prud’homme et al. 2011) and traffic congestion (Rodier et al. 1998) are some of the attractive features of DRT. It has the potential to improve personal security and accessibility at a cheaper cost than taxis (Zografos et al. 2008), but

9 with a similar level of flexibility (Ronald et al. 2013). It reduces dependency on private vehicles (Ronald et al. 2013) and can be used to support mobility disabled people (ActiveAge 2008). Davison et al. (2014) conducted a survey of DRT providers in Great Britain, they identified DRT as the most cost effective way of ensuring services access to rural communities without a conventional bus service, and providing efficient transport coverage. DRT services can minimize the empty routes and use the capacities in a highly effective way, if planned properly, and hence, can be economically sustainable (Andrejszki and Török 2014).

Although, DRT is not an unknown element in urban mobility, it has started gaining more presence with the growth of information technology in the transport sector and several other reasons. For example, Mageean and Nelson (2003) mentions that the shortcomings of conventional regular bus, taxi and special transport services, new developments in community transport; and an interest to combat social exclusion in the transport sector are forces behind exploring the growing potential of DRT. DRT has been studied under multiple terminologies such as dial-a-ride (Cordeau and Laporte 2007), paratransit (Cervero 1997) or ad-hoc ride-sharing (Winter and Nittel 2006, Braun and Winter 2009).

Dial-a-ride systems were originally provided in the response of widely dispersed trip- patterns. They generally catered for non-work journeys in low-density suburban areas. Interested users telephoned in their requests in advance, and the operators planned the service the day before the trip. There were many limitations of traditional dial-a-ride systems such as, the lack of flexibility in route planning, relatively high cost of provision, and inability to manage high-demand (Mageean and Nelson 2003). Schemes, which catered to special needs transport, were successful in developing moderate customer bases and considerable international experience (Ashford 1978). Correspondingly, there is no consistent body of knowledge, and different communities have developed similar solutions for DRT services, and increasingly for real-time DRT systems. The common challenge, however, has always been optimization between the transport demand and supply, and thus, the demand patterns determine the constraints of the whole system.

Use of transport telematics in recent years has greatly influenced the ability of DRT concepts to provide efficient and viable transport services (Brake et al. 2004). With the development in technologies (Giannopoulos 2004) as well as advancement in optimization methods (Horn 2002), DRT has more potential to grow. The most important

10 technology induced progress in this regard is real-time or ad-hoc DRT. For example, Braun and Winter (2009) have demonstrated that collaborative transport can effectively solve classical transport planning problems in real-time. Ad-hoc DRT operates along flexible routes in order to provide point-to-point transportation, and does not have predefined schedules in order to react on demand in real-time (Ronald et al. 2013). While sharing existing infrastructure, it can offer shared forms of transport (e.g., by private ride- sharing, or by commercially operated vans), hence its fares are expected to be lower than taxis.

Currently, the DRT exists in various forms in many cities but mostly for fixed origins or destinations, or fixed routes, or with some form of pre-booking. This thesis concentrates on DRT in general. It can be seen as a mini-bus providing shared, point-to-point transportation on demand with flexible routes. It is an alternative way to deliver the demand of public transport. As this is an early stage study and aims to predict susceptibility to DRT by learning from the experience of similar services operating elsewhere, its scope is kept open and broad, so as to have enough number of studies in identifying the dominant parameters favoring the use of DRT.

DRT has a number of objectives, constraints, and various decisions to undergo for its functionality, which makes it very complex in terms of computation and implementation. This research is limiting its focus on the potential uptake, or demand patterns, rather than the technology, optimization, or other operational strategies. There are only a few operational results relating to DRT are available which provide numerical studies relating demand factors with the DRT use (Wang et al. 2014, Wang et al. 2015). There is relatively little quantitative information about how factors at the personal or household level influence the use of DRT in the available literature. This research is useful to explore inclination of travelers towards proposed transport service while making policy, and investment decisions.

2.2 Understanding Travel Demand Transportation systems are an important component of urban development and public transport projects are often massive with widespread social impact. Hence, before implementation of transportation facilities and policy decision making, it is required to predict travel demand pattern and understand travel behavior of the population. In many cases, mathematical models can be developed for travel demand analysis and sometimes, substantial insight can be generated without any such models (Pas 1985). Discrete choice

11 random utility models (Horowitz 1991) and activity-based approaches (Axhausen and Gärling 1992) have widely been used for travel demand modeling. These two streams of travel demand research have been complementary in nature (Pas 1985). The development of random utility model formulation has operationalized some activity-based research. (Damm 1980, Hirsh et al. 1984, Hirsh et al. 1986) and simultaneously, the ideas and results of the activity-based research have enhanced the random utility models (Kitamura and Lam 1984).

The activity-based approaches have made significant contributions in understanding travel behavior. They require the transport planners to pay more attention to the socio- economic and demographic characteristics of the population as they affect the demand for participating in activities, and hence, travel and often constrain travel choices (Pas 1985). In these approaches, the basic principle in predicting travel demand is that it is derived from travelers’ willingness to participate in spatially dispersed activities such as work, shopping, recreation, and personal business. These decisions are complex and involve trip purpose, frequency, timing, destination, and mode of travel. Also, these choices are simultaneously affected by the choices of vehicle ownership, the location of accommodation, and end-of-trip activities. Characteristics of households and individuals within a population are used to examine who, when, how and where the individuals travel. An individual decides his activity program and allocates time and resources to these activities. These decisions depend on the socio-economic characteristics of the individual including income, household structure, and occupation, and characteristics of the environment including transportation systems (McFadden 1974, Axhausen and Gärling 1992, Ben-Akiva et al. 1996).

The transport environment of an individual or household is influenced by the choice of a new transport service. Therefore, to predict travel demand associated with new transport service, it is important to understand how they adapt to these changes. The models developed to understand the travel behavior of the population are required to be responsive to changes in the external environment and also socio-economic and demographic characteristics of the travelers (Dix and Layzell 1986). It is also important to examine the effects of socio-demographic characteristics and current trip behavior on travel decisions of the individuals.

Identifying patterns of urban mobility and travel demand are important elements in transportation planning and behavior modeling. Nowadays, the effects of individual and household socio-demographics on travel related decisions and mode choices are considered in disaggregated models of travel demand. Sometimes, disaggregated trip- based models are implemented in an aggregated manner with aggregated socio- demographics data. Although, due to the extreme complexity of human behavior, it is not possible to determine accurately what lies behind individual’s travel decisions, but reliable inferences can be drawn from the patterns of choices that groups of people make (Silva et al. 2015). This is the principle behind discrete choice models (Ben-Akiva and Lerman , Dagsvik 2000). They define the decision maker and its characteristics, determine the alternatives available, measure the attributes such as benefit and cost of alternatives, and describe the process used by the decision-maker to choose an alternative.

The modeling of a large scale DRT is challenging. Traditional transport models have limits in assessing the potential of new transport modes in general, and of innovative transport systems in particular (Shaheen and Rodier 2004). Previous research has utilized discrete choice models in order to understand the market response to DRT service improvements. These models quantify the impacts of attributes such as travel cost, travel time, in-vehicle time, the time window for pickup and delivery, and individual characteristics and attitudes. Ciari (2009) showed that the representation of individual travel needs such as trip purpose (shopping, work, leisure, etc.) increased the precision of such models and helped in detecting who could meaningfully use the taxi service in his/her out-of-home tour.

Al-Ayyash et al. (2015) presented a model to predict the demand of a shared-ride taxi service among students. The model captured the effect of several attributes on the average number of trips per week undertaken by the students using the new taxi service. They found that the factors such as traveler’s attitude, highly impact the ridership of a mode. Students’ response to changes in the LOS of the new taxi service varied depending on their current commuting mode. In general, it was observed that the new shared-ride taxi service is not an attractive alternative for bus users.

Whenever a new transport service is proposed, it is seen to lead to some changes in the transportation system of the concerned environment. Traditionally, planners focus on modeling the change in traveler mode choice behavior in response to the proposed

13 changes. Lu and Gosling (2007) mentioned that discrete choice models are widely used for this purpose where the coefficients of these models are generally determined from survey data collected in the relevant region at a specific time. Many times, proposed change in transport environment involves the introduction of a new mode or service that was not available when the data on which the model was estimated was collected. In that scenario, it is difficult to include the new mode or service to the mode choice model. Solving this problem generally involves adding a new alternative to the choice set in the mode choice model. Then, a utility function for that alternative can be defined. (Lu and Gosling 2007). However, in this process, the choice of an appropriate value for the alternative-specific constant in the utility function is still an unresolved issue.

According to Lu and Gosling (2007), three approaches are possible to choose an appropriate value for the alternative-specific constant in the utility function. The first is to predict the value by using values of the alternative-specific constants of the existing modes by judging what difference in cost would make a traveler indifferent between the new mode and an existing mode if the travel and other times involved in using the two modes were identical. The second approach is to look for a place where the new mode or service, in fact, was already available and a mode choice model which incorporates that particular service exists there. Using that mode choice model, the alternative-specific constants for the various modes can be compared. The third approach is to perform an SP experiment, in which respondents would be asked about which mode they would choose, or whether they would modify an existing choice if the new alternative was available. It can be understood that this kind of modeling can be desirable only when service parameters and features have been decided, which can be used to predict the coefficients for the mode choice model. In cases where prospective planners, modelers, and users are not aware of exact service features and quality, it is difficult to estimate the model coefficients by inferring from the judgment. Also, surveys have their own shortcomings, which are presented in the next section.

2.3 Preference Surveys In transport planning, predicting travel behavior has always faced methodological challenges (Goodwin et al. 1990). To meet these challenges, initially, transport researchers adopted methods to measure travel attitudes (Golob et al. 1977, Koppelman and Lyon 1981, Pas 1990). In the later stages, the stated-preference, stated-choice, and stated-adaptation methods are prevalent (Jordan 1988, Louviere 1988, Timmermans and

Golledge 1990, Hensher 1994, van Bladel et al. 2008). These methods predicting travel demand on the basis of responses of users lack substantial theoretical underpinning (Levin and Louviere 1981, Gärling et al. 1996). Hence, it is difficult to provide validation for predicted results (Gärling et al. 1998).

The reliability of survey data for travel demand prediction and decision making is affected by various shortcomings. Surveys are preferred when the issues under research are clearly defined and participant responses can be anticipated. In exploratory areas of research, various issues remain unidentified and the researchers seek to answer some questions, hence, the survey methods may not be suitable (Clifton and Handy 2001). Poulenez- Donovan and Ulberg (1994) argued, “The world of the survey, however, is bounded by the perspectives and goals of the survey writers. The survey restricts not only the question frame but the answer frame as well, anticipating the important issues and questions and the responses."

Mahmoud et al. (2015) mentioned that a proper understanding of individuals’ travel choices is required to evaluate the effectiveness of innovative transport initiatives such as DRT. This is more applicable especially for a hard-to-reach target population with unique travel characteristics as cross-regional commuters. Previous research ( Louviere et al. 2000, Osman et al. 2012) showed that the revealed preference (RP) data does not capture the behavioral trade-offs involved in the travelers’ decision making process adequately. Therefore, demand models developed based only on RP data cannot accurately forecast the individual choices in response to the introduction of new modes or new transportation policies.

SP data has increasingly replaced the RP data in order to overcome the limitations of RP models. SP surveys measure the individuals’ preferences towards hypothetical scenarios by questioning the respondents on services or policies that do not exist (Hensher 1994, Hensher et al. 2008, Louviere et al. 2008). Takeuchi et al. (2003) developed a demand model for DRTs through SP survey to observe the relationship between LOS for DRTs and each user’s behaviors. Because few DRTs had been installed in Japan, it was difficult to collect data about the actual and observed choice made by DRT users. And, if RP data was collected from areas where no DRTs have been implemented, it was not possible to observe LOS variation influenced by trip reservation, which is the most important characteristic of DRTs. Hence, RP survey was not applied in the analysis stage. In SP

15 survey, the authors focused on the situation that each user’s reservations may change LOS, and that influence choice behavior. Access time, waiting time, in-vehicle time, and its variation were chosen as question variables of DRT’s LOS, and age, mode, travel time to station in daily use, and bus headway near residences were chosen as characteristics of individuals.

However, SP data also has its own drawbacks. According to Wardman (1988), a systematic bias is induced in the data as individuals’ stated preferences are not always consistent with their actual choices. The SP experiment is performed to quantify the independent effects of the design attributes on choices of the respondents (Louviere and Hensher 1983, Louviere and Woodworth 1983, Carson et al. 1994).With the increase in the number of attributes, the experiment design for SP survey keeps getting complex. Our prediction for DRT may require many more attributes, hence, using SP survey is challenging. It is also possible that some of the groups are underrepresented in the SP survey and it would be even worse if those groups are more in need of the proposed transport service. The main criticism levelled at SP method is that individuals’ stated and actual preferences do not necessarily coincide, and there are more serious consequences for demand prediction of even random error in SP responses as a result of scale factor problem (Wardman 1991).

Al-Ayyash et al. (2015) considered the use of SP data as a potential shortcoming in their model to predict the demand for of a shared-ride taxi service among students. They emphasized that when a new modal alternative is to be introduced to the market, the results of SP survey might be associated with bias and the new service demand might be overestimated.

Yang et al. (2009) highlighted the challenges and complexity in the design of SP surveys for new and innovative transport modes. They mention that the smart features and use of information technology in new transport modes affect the accuracy of travel time prediction and people’s travel preferences. Hence, SP surveys must include the attributes to capture such effects. Additionally, the transport alternatives and required attributes are not uniform in format, and thus, their organization and presentation in SP surveys are found challenging. Also, they find it important to present the features of new transport modes realistically in the surveys to minimize the potential biases. They conclude that the

16 practical design of an SP survey under complex and multiple scenarios is an extremely time-consuming process.

Sometimes, qualitative methods (Clifton and Handy 2001) and stated adaptation surveys (van Bladel et al. 2008) can give better results compared to traditional preference surveys, but these methods are expensive and require special training and skills. Due to these limitations of preference surveys, this research provides a methodology to predict susceptibility to use a proposed transport service without doing such surveys. It uses socio-economic, demographic and trip characteristics of the population. The effect of these characteristics on the travel behavior of the population is presented in the next section.

2.4 The Effect of Socio-Economic, Demographic and Trip Characteristics on Travel Behavior Litman (2013) defines travel demand as “the amount and type of travel that people would choose in particular situations” (p. 2). He mentions that various demographic, geographic and economic factors such as the number of people (residents, employees and visitors), employment rate, wealth/incomes, age/lifecycle, lifestyles and preferences can affect travel demand. Scheiner and Holz-Rau (2007) argue that an individual’s travel behavior is affected by life situation both directly and indirectly and life situation and travel behavior are classical determinants of travel demand. According to Berger and Hradil (1990), life situations are structural differences that can be described by socio-economic and demographic variables such as income, education level, sex, age, and nationality. This socio-economic demography points to an individual’s personal circumstances relevant to his or her travel. According to Simma and Axhausen (2001), mode choice and other aspects of travel behavior are affected by the availability of certain means of transport and are a part of the decision process depending on socio-economic and demographic factors of life situation.

Kattiyapornpong and Miller (2006) found that socio-economic and demographic variables such as age, income, and life cycle have significant differential and interactive effects on travel behavior. Curtis and Perkins (2006) reviewed a large number of papers, which studied the impact of socio-economic and demographic variables on travel behavior. They found some significant relationships between travel behavior including mode choice and variables such as age, gender, car ownership, household composition, and income. Best and Lanzendorf (2005) and Boarnet and Sarmiento (1998) cite gender,

17 household composition and income, habit, and car ownership amongst others as significant factors in influencing travel behavior. Choi et al. (2014) analyzed the changes in travel behavior using household travel survey data collected in Seoul Metropolitan Area and showed that the temporal and spatial characteristics of changes in travel behavior were interpretable regarding socio-economic, regional, and trip characteristics. Olaru et al. (2005) studied travel behavior in the Sydney metropolitan area and found a number of socio-economic and demographic variables influencing travel behavior. Rasouli and Timmermans (2014) emphasize that understanding how mode choice decisions and travel behavior vary with gender, income categories, age groups, and other socio-economic and demographic characteristics are crucial in evaluating social effects and equity of transport investment. Kaltheier (2002), Rosenbloom (2004) and Contrino and McGuckin (2009) have studied the impact of socio-economic and demographic variables on travel behavior and found a significant relationship between travel behavior and variables such as income, age, gender, and ethnicity.

Age and gender are found to be crucial factors in affecting travel behavior and patterns. Tranter and Whitelegg (1994), Fyhri and Hjorthol (2009) and Su and Bell (2009) suggest that people in different age groups have differences in their travel behavior. These differences mainly occur because they tend to engage in different types of activities. For elder people, physical constraints also affect their travel behavior. Curtis and Perkins (2006) mention that women are recognized as being more likely to adopt sustainable travel behaviors compared with men. Polk (2003) and Polk (2004) also found a significant relationship between sustainable travel patterns and gender. He concluded that women are more positive towards reducing the environmental impact of travel modes, and hence, more willing to reduce their use of the car than men. Best and Lanzendorf (2005) and Boarnet and Sarmiento (1998) found that the type or destination of trip shows gender differences and women have made fewer journeys to work by car and more journeys for activities such as shopping and child-care.

Employment status is also an important factor in travel behavior. Best and Lanzendorf (2005) showed that part-time workers are more likely to travel often than full-time workers because they are also engaged in other activities such as shopping or child escorting. McGuckin and Nakamoto (2004) and Vance et al. (2005) found that women are less likely to use a car than men, and the difference is also influenced by socio- economic and demographic characteristics such as the presence of the children. Ryley

(2005) showed that households with children have distinct travel behavior characteristics and travel mode decisions. He concluded that households consisting of students, the unemployed and part-time workers without children are most likely to use non-motorized forms of transport, whereas families consisting of retirees, high-income owners, and households with children are least likely to use non-motorized forms of transport. Dieleman et al. (2002) showed that high-income households have higher car ownership and usage, and households with children are more likely to use the car than one or two person households. They also mentioned that the purpose of the trip also influences the travel mode selected.

From this review, it is understood that before implementation of transportation facilities and policy decision making, it is required to predict travel demand pattern and understand travel behavior of the population. Preference surveys which ask willingness of users to use a proposed transport service are commonly used to estimate the demand patterns but they have their own shortcomings which are identified in this literature review. Also, from this review of the literature, it can be concluded that various socio-economic and demographic characteristics of the population and trip characteristics affect travel decisions. Hence, this research predicts the usability of a proposed transport service by using characteristics of the population. The requirement of a preference survey is eliminated by identifying the characteristics that are expected to encourage a mode shift towards proposed transport service. It involves reviewing usage pattern of some of the existing similar services in the world. Variation of these identified characteristics in a city derived from a sample travel survey data gives the susceptibility of proposed transport service.

Chapter 3: Identifying Parameters Favorable to DRT

From the last chapter, it can be understood that it is important to find out usability of a transport service before its implementation. The literature review also highlighted the limitations of preference surveys in assessing the usability. Research on travel behavior confirms that various socio-economic and demographic characteristics of the population affect travel decisions (Simma and Axhausen 2001, Curtis and Perkins 2006, Kattiyapornpong and Miller 2006). The study in this thesis provides a methodology which will identify demographic characteristics favorable to DRT by reviewing DRT services in different parts of the world and analyze their variation in a target city (tested here for Greater Melbourne) to find more feasible regions of implementation. This chapter presents the findings of the review of various studies about usability of DRT services.

One DRT system is Telebus. Telebus services operate in the outer eastern suburbs of Melbourne, Australia. These areas are often quite hilly and have narrow streets. The operation of conventional normal-size buses is quite difficult in these areas due to spatial configuration of the population. Hence, telebus services offer a combination of fixed route and demand responsive services. Passengers can be picked up or dropped off anywhere in the Telebus zone. They can also be picked up or set down at designated bus stops along the route at the normal bus fare. The Telebus operates from 5am to 7pm on weekdays,

20 running every 20-40 minutes in peak times and hourly in off-peak. The Telebus services have stops at the local train stations, local shopping precincts, primary schools, and secondary schools along the routes. The Telebus Mobility and Access Benefits Project (Maddern and Jenner 2007)reviewed the Telebus services. This study conducted with users over 15 years of age found that females comprise 66% of Telebus users. 38% and 36% of users were in the age groups of 15-24 year olds and 55 years and over respectively. 78% of users did not hold a driver licence, and shopping (31%) was the major travel purpose. Among the Telebus users, females were the dominant users, representing 74% of the users and 81% of the frequent users.

Other studies confirm these observations qualitatively. In the survey of demand responsive transit shuttles operated by the San Francisco Bay Area Rapid Transit, BART, respondents were asked for their likeliness to use the shuttle service to get to and from the transit station (Anspacher et al. 2004). The survey indicated that there was a moderate willingness to use shuttles in both urban and suburban neighborhoods. On an average, suburbanites were found to be willing to pay higher fares, and showed higher acceptance of waiting times. The distance individuals live from the nearest transit station is identified as an important factor in predicting their willingness to use a shuttle. As per this study, individuals who live within 0.25 miles (0.4 kms) and between 0.25 and 0.5 miles (0.4 and 0.8 kms) of the nearest transit station are 8.6% and 7.3% less likely to be ‘very willing’ to use a shuttle. This result coincides with the fact that the walkable distance to a transit station is typically considered to be less than 0.8 kms. This survey also found that greater the numbers of vehicles per household, less willing the individuals are to use a shuttle.

Statistics from the 49Link DRT service in South Shropshire, UK (ActiveAge 2008) show that the people use the service mainly to do their shopping, to gain access to healthcare facilities, and to socialize. Ring and Ride service in Fife, Scotland, offers the door-to- door accessible transport for people with reduced mobility. It operates six vehicles and undertakes approximately 1000 passenger trips per week. 25% of these trips cater for shopping, and 50% for social visits (ActiveAge 2008).

DRT services can be used in a variety of situations and can address a wide range of user needs. They have applications in low-demand situations as well as for the access needs of the transport disadvantaged (Scott and Booz & Company 2010). The New Zealand Transport Agency’s planning, programming and funding manual, PPFM (NZTransport

Agency 2008)and The New Zealand Public Transport Management Act (Public Transport Management Act 2008) have classified following groups as more prone to being transport disadvantaged:

1. Elderly people

2. Young people and children

3. Those without access to a motor vehicle, either involuntarily or by choice

4. Disabled people (mobility and/or communication impaired)

5. Those on low incomes

6. Minority groups and new migrants

7. People in areas not served by public transport, including those in geographically isolated areas.

The ‘DRT for DRT’ project funded by the “British Engineering and Physical Sciences Research Council 2010-2013” was started to determine the potential contribution of DRT to meet transport and public policy objectives in Great Britain. Ryley et al. (2013) explored the potential contribution of DRT to meet transport and public policy objectives in Great Britain. They developed a series of basic demand models to explore the effects of various socio‐economic and demographic factors on the demand for DRT. A questionnaire survey was conducted to determine the propensity to use DRT from the general population. Most respondents were favorable towards the DRT concept, particularly the aspects of pre‐booking and door‐to‐door services. Trips to health facilities, shopping and group leisure activities were found to be having the highest potential to convert to DRT.

Wang et al. (2014) explored the effects of various area-wide factors on the demand of DRT in the metropolitan region of Greater Manchester. They found that the demand for DRT services was higher in areas with low car ownership and population density. Also, high levels of social deprivation (in terms of income, employment, education, housing and services, health, and disability), and living environment favored the use of DRT services. Lerman et al. (1980) also found car ownership to be inversely associated with the number of DRT trips.

Another study was conducted by Loughborough University in 2006 for Wiltshire County Council to evaluate DRT services in the county. It identified target markets for DRT as: people who cannot access public transport, people living at places where public transport is not available, people without their own transport, unemployed people, economically inactive people, single pensioner households, people with a limiting long-term illness, ethnic minority households, and people aged 14-19 years (Enoch et al. 2006). Laws (2009) found that the school children and retired people were dominant users of the Wiltshire Wigglybus (a DRT service) in Pewsey, UK, and 33% passengers used the service for shopping trips. Also, he found that in Calne, retired and disadvantaged people were majority of the users.

According to Häme (2013), one of the main present goals of planning DRT is seen to be the developing of functional public transport services able to compete with private car and taxicab transportation. In order to compete with private cars, public transport should offer direct and fast connections, minimize the walking and waiting times, offer a door- to-door service, and the amount of transfers between modes should be minimal.

According to Takeuchi et al. (2003), results of a demand model showed that users prefer shorter waiting time and shorter in-vehicle time. According to Mageean and Nelson (2003), majority of the users of DRT services were female (over 80% in Belgium and 70% in Gothenburg) and two thirds of users in Belgium are retired, house persons, and students. Also, they showed that trip purpose reflects the type of users (shopping dominates in Limbourg and West Flanders, social visits in East Flanders, work and shopping in Campi and Porta Romana). TCRP (1995) and TCRP (2004) identified the elderly, mobility limited, and those on low incomes as potential markets for DRT in rural areas. Nelson and Phonphitakchai (2012) reported that in the metropolitan area of Tyne and Wear in the UK more than 50% of the users of the DRT services were female and retired. According to Rosenbloom and Fielding (1998), 61% of users of OmniLink DRT service in Prince William County, Virginia, USA were female and 64% users earned less than USD 25,000 per annum. Koffman (2004) studied the dial-a-ride transit (DART) scheme in Winnipeg, Manitoba, and found that 53% of users were female and 29% were aged under 18. Bearse et al. (2004) used a time-series model and found that women took about 30% more DRT trips per month than men. SG Associates Inc. et al. (1995) and Spielberg and Pratt (2004) noted that the poor, elderly, disadvantaged, or disabled population are more likely to use the DRT services. Bradley and Bowman (2006) proved

23 that the travel forecasting models are significantly affected by the income, gender, age, and household size, and that the shopping, social and recreational tours are more likely to be done by shared-ride.

Joewono (2009) conducted a survey to explore the willingness and ability to pay of the paratransit users in Indonesia. He found that not owning a car is the most prominent reason for using paratransit services. Also, different user characteristics affect their willingness to use and decision depends on users’ perceptions regarding the service quality, the characteristics of trips, and their financial capability. This decision making has policy implications and helps to achieve objectives of increasing the mobility of specific groups. Rayle et al. (2014) collected 380 complete surveys in San Francisco to explore usability of ridesourcing. The survey asked about trip origin and destination, trip purpose, previous and alternative modal choice, car ownership, and basic demographics. Results showed that a major portion of ridesourcing trips were spatially and temporally not well served by public transit. Ridesourcing users also appeared to be less likely to own an automobile. Share of social and leisure trips was very high and many used ridesourcing to access transit. Estimated total travel times, including waiting and in- vehicle times, were consistently higher for transit than ridesourcing. The review of existing DRT services has revealed the dominant characteristics affecting its usability. The important findings of the review have been included in Table 3.1, along with the parameters which can be identified to affect the usability of DRT.

Table 3.1: Important findings from review of existing DRT services

Study Findings Parameters identified Rayle et al., 2014 Ridesourcing users appeared to be less likely  Vehicle to own an automobile. Share of social and ownership

leisure trips was very high and many used it  Social trips to access transit. Estimated total travel times,  Train including waiting and in-vehicle times, were station consistently lesser compared to transit. proximity  Waiting time Wang et al., 2014 Demand for DRT services was higher in  Vehicle areas with low car ownership and population ownership

density. Also high levels of social  Income deprivation which can be measured in terms of income, employment, education, housing and services, health and disability, and living environment favored the use of DRT.

Häme, 2013 In order to compete with private cars and  Waiting taxicab transportation, public transport and should offer direct and fast connections and walking minimize the walking and waiting times time  Train station proximity Ryley et al., 2013 Trips to healthcare, shopping and group  Shopping leisure activities were found to be with and social highest potential to convert to DRT. trips Nelson and In Tyne and Wear in the UK, more than 50%  Gender Phonphitakchai, 2012 of the users of the DRT services are female  Workforce and retired. status Scott and Booz & DRT can solve access needs of transport  Age group Company , 2010; NZ disadvantaged: elderly people, young people  Vehicle Transport Agency, and children, those without access to a motor ownership 2008; Public Transport vehicle, disabled people, those on low  Income Management Act, 2008 incomes, minority groups and new migrants,  Train people in areas not served by public transport station proximity Joewono, 2009 Not owning a car is the most prominent  Vehicle reason for using paratransit services. Also, ownership

different user characteristics affect their  Income willingness to use and decision depends on users’ perceptions regarding the service quality, the characteristics of trips, and their financial capability. Laws, 2009 School children and retired people are  Workforce dominant users of the Wiltshire Wigglybus status

(a DRT service) in Pewsey, UK, and 33%  Shopping passengers used the service for shopping trips trips. In Calne, retired and disadvantaged  Age group people are dominant users. ActiveAge, 2008 The main reasons for using the 49Link DRT  Shopping service were shopping, healthcare access and and social to socialize. 25% of Ring and Ride trips cater trips for shopping, and 50% for social visits. Maddern and Jenner, 74% Telebus users were in the age groups of  Age group 2007 15-24 years and over 55 years. 78% of users  Gender did not hold a driver licence, and 31% used  Driving

for shopping. Females were 74% of the users licence and 81% of the frequent users.  Shopping trips Bradley and Bowman, Income, gender, age and household size have  Age group 2006 significant effects in travel forecasting  Gender models and also shopping, social and  Income

recreational tours are more likely to be done  Single by shared-ride. person household

 Shopping and social trips Enoch et al., 2006 Target markets for DRT are: people who  Age group cannot access public transport, people living  Vehicle

at places where public transport is not ownership available, people without their own  Income transport, unemployed people, economically  Single inactive people, single pensioner person households, people with a limiting long-term household illness, ethnic minority households, and  Workforce people aged 14-19 years. status Anspacher et al., 2004 The distance from the nearest transit station  Vehicle is an important factor in predicting ownership

willingness to use a demand responsive  Train shuttle. The greater the numbers of vehicles station per household, the less willing individuals proximity are to use a shuttle. Bearse et al., 2004 Women take about 30% more DRT trips than  Gender men. Koffman, 2004 For the dial-a-ride transit scheme in  Gender Winnipeg, Manitoba, 53% of users are  Age group female and 29% are aged under 18. Spielberg and Pratt, Typical DRT rider is likely to be poor,  Income 2004 elderly, and disadvantaged  Age group TCRP, 2004; TCRP, Identified the elderly, mobility limited, and  Age group 1995 those on low incomes as potential markets  Income for DRT in rural areas. Mageean and Nelson, Females are the dominant users of DRT  Gender 2003 services (over 80% in Belgium and 70% in  Age group Gothenburg). Two thirds of users in Belgium  Workforce

are retired, house persons and students. Trip status purpose reflects the type of users (shopping  Shopping dominates in Limbourg and West Flanders, and social social visits in East Flanders, work and trips shopping in Campi and Porta Romana). Takeuchi et al., 2003 Results of the demand model for DRT  Trip showed that users prefer shorter waiting time waiting and shorter in-vehicle time. time Rosenbloom and For the OmniLink DRT service in Prince  Gender Fielding, 1998 William County, Virginia, USA, 61% of  Income users are female and 64% earn less than USD 25,000 per annum. SG Associates Inc. et Typical DRT rider is likely to be poor,  Income al., 1995 elderly, and disadvantaged  Age group

Lerman et al., 1980 Car ownership is inversely associated with  Vehicle the number of DRT trips. ownership

Some of these identified parameters are individual characteristics, some are household characteristics, and some are trip characteristics or public transport related characteristics. Identified parameters are classified in these 3 groups in Table 3.2.

Table 3.2: Parameters identified from literature review

Parameters Findings identified Age group 74% Telebus users in Melbourne were in the age groups of 15-24 years and over 55 years. Individual characteristics According to Scott and Booz & Company, DRT can solve access needs of elderly people, young people and children. According to Spielberg and Pratt, 2004 and TCRP, 2004; TCRP, 1995, typical DRT users are likely to be elderly. For the dial-a-ride transit scheme in Winnipeg, Manitoba, 29% users are aged under 18. Enoch et al., 2006 found people aged 14-19 as target market for DRT. Gender In Tyne and Wear in the UK, more than 50% of the users of the DRT services are female. Females were 74% of the users and 81% of the frequent users of Telebus in Melbourne. According to Bearse et al., 2004, women take about 30% more DRT trips than men. For the OmniLink DRT service in Prince William County, Virginia, USA, 61% of users are female. For the dial-a-ride transit scheme in Winnipeg, Manitoba, 53% of users are female. Females are the dominant users of DRT services (over 80% in Belgium and 70% in Gothenburg). Workforce In Tyne and Wear in the UK, more than 50% of the users of status the DRT services are female and retired. Two thirds of DRT users in Belgium are retired, house persons and students. Enoch et al., 2006 found unemployed people and economically inactive people as target markets for DRT. Driving According to Maddern and Jenner, 2007, 78% of users of licence Telebus in Melbourne did not hold a driver licence. Vehicle According to Lerman et al., 1980, car ownership is inversely ownership associated with the number of DRT trips.

Household Anspacher et al., 2004 found that the greater the numbers of characteristics vehicles per household, the less willing individuals are to use a demand responsive shuttle. Enoch et al., 2006 and Scott and Booz & Company, 2010 found that people without their own transport are likely to be DRT riders. According to Wang et al., 2014, Demand for DRT services was higher in areas with low car ownership. According to Joewono, 2009, not owning a car is the most prominent reason for using paratransit services. According to Rayle et al., 2014, ridesourcing users appeared to be less likely to own an automobile Income According to Wang et al., 2014, social deprivation in terms of income favored the use of DRT. According to Joewono, 2009 and Scott and Booz & Company, 2010, people on low household income are likely to use DRT. Spielberg and Pratt, 2004, SG Associates Inc. et al., 1995, TCRP, 2004 and TCRP, 1995 identified poor and low income households as target markets for DRT. For the OmniLink DRT service in Prince William County, Virginia, USA, 64% users earn less than USD 25,000 per annum. Single According to Enoch et al., 2006 and Bradley and Bowman, person 2006, single pensioner households are target market for DRT. household Train According to Anspacher et al., 2004, the distance from the station nearest transit station is an important factor in predicting proximity willingness to use a demand responsive shuttle. According to Scott and Booz & Company, 2010, DRT can solve access needs of people in areas not served by public transport. According to Rayle et al., 2014, many used ridesourcing to access transit. Shopping According to ActiveAge, 2008, the main reasons for using the and social 49Link DRT service were shopping, healthcare access and to Trip trips socialize. 25% of Ring and Ride trips cater for shopping, and characteristics 50% for social visits. 31% users used Telebus in Melbourne for shopping. According to Ryley et al., 2013, trips to healthcare, shopping and group leisure activities were found to be with highest potential to convert to DRT. 33% passengers used the Wiltshire Wigglybus (a DRT service) in Pewsey, UK for shopping trips. According to Rayle et al., 2014, among ridesourcing trips, share of social and leisure trips was very high.

According to Bradley and Bowman, 2006, shopping, social and recreational tours are more likely to be done by shared- ride. According to Mageean and Nelson, 2003, trip purpose reflects the type of users (shopping dominates in Limbourg and West Flanders, social visits in East Flanders, work and shopping in Campi and Porta Romana). Trip According to Takeuchi et al., 2003, results of the demand waiting model for DRT showed that users prefer shorter waiting time time and shorter in-vehicle time. According to Häme, 2013, in order to compete with private cars and taxicab transportation, public transport should offer direct and fast connections and minimize the walking and waiting times. According to Rayle et al., 2014, estimated total travel times, including waiting and in-vehicle times, were consistently lesser in Ridesourcing compared to transit. Trip According to Häme, 2013, in order to compete with private walking cars and taxicab transportation, public transport should offer time direct and fast connections and minimize the walking and waiting times. According to Rayle et al., 2014, estimated total travel times were consistently lesser in Ridesourcing compared to transit.

From this review of the existing DRT services, it can be concluded that there are various socio-economic, demographic, and trip characteristics of individuals and households, which affect usability of DRT services anywhere in the world. Some of these characteristics, which have been found to affect the usability of DRT services consistently across the studies, will be used to predict the susceptibility to use DRT in the next chapter.

Chapter 4: Predicting Susceptibility of Regions to Use DRT

In the previous chapter, various studies were reviewed to gain the characteristics of the population and their trips that favor the use of DRT. This chapter provides a methodology to predict the susceptibility to use DRT services in urban space using the variation in the characteristics identified in the previous chapter. It involves analysis of the travel survey data in order to find the pattern of DRT favorable characteristics in a city. It predicts the more feasible regions of implementation of DRT. The methodology is tested using travel survey data of Greater Melbourne and corresponding feasible regions for implementation of DRT are identified.

4.1 Data and Methodology The characteristics of population and trips that are more favorable to be converted to DRT are identified by doing a review of studies of usability of DRT services in various regions of the world. The review in Chapter 3 has revealed the dominant characteristics affecting its usability. On the basis of this review, following eleven parameters are identified to affect the usability of DRT.

1. (15-24) and (55 or more) age groups

2. Trips of female

3. Not in workforce

4. No driving licence

5. Low household vehicle ownership (0 or 1)

6. Low household income

7. Single person household

8. Lack of train station proximity

9. Shopping and social trips

10. Higher trip waiting time

11. Higher trip walking time

Parameters 1 to 4 are individual characteristics, parameters 5 to 8 are household characteristics, and parameters 9 to 11 are trip characteristics. Parameter 10 and 11 test the performance of existing public transport in a way. Parameters 1, 2, 3 and 4 are respectively, the percentage of trips which were done by persons of age groups (15-24) and (55 or more), females, persons who were not in workforce, and persons who did not possess a driving licence. Parameters 5, 6, 7 and 8 are respectively, the percentage of trips which were done by persons belonging to households having none or single car ownership, having low income, having single person, and far from train station. Parameter 9 is the percentage of trips which were done for either shopping or social visit. Parameter 10 and 11 are respectively, the average value of waiting time and walking time of the trips involving public transport.

These parameters are investigated in Greater Melbourne, Australia. Melbourne is the second most populated city of Australia and capital of state of Victoria. It has population of 4,347,955 (in 2013) spread across area of 9990 km². Australian Bureau of Statistics (ABS) has divided Australia in statistical areas at different levels such as SA1, SA2, SA3, SA4 with SA1 and SA4 being the smallest and largest geographical units respectively. There are 351 Statistical Area Level 3 (SA3) regions with populations in the range of 30,000 to 130,000 in Australia with Greater Melbourne having 40 of them.

Figure 4.1 presents the map of Greater Melbourne and its SA3 regions.

Figure 4.1: Map of Greater Melbourne and its SA3 regions. The value of each parameter is calculated using VISTA (Victorian Integrated Survey of Travel and Activity 2009-10 (The Victorian Department of Transport 2011), a self- conducted household travel survey for each census region considering them as trip origins. Household income is divided in five equal categories in VISTA, the lower two are considered as low income in this study. Similarly, train station proximity is divided in four equal categories in VISTA, the upper two are considered as lack of proximity in this study. While finding the required values, sampled trip weights provided in VISTA are used to make the data representative of entire population.

A flow-chart of the overall methodology investigated in this chapter is presented in Figure 4.2.

Review of existing similar services elsewhere in the world to identify dominant parameters of users and

trips favoring use of DRT.

Extracting patterns of identified parameters across

census regions in the target city using travel surveys (viz. VISTA in Melbourne).

Standardizing values of parameters and then averaging for each region to explore overall susceptibility to use DRT.

Validating the methodology by repeating the above process for public transport and comparing predicted and existing results.

Figure 4.2: Methodology to explore susceptibility of DRT.

Table A.1 in Appendix presents the raw data of all 40 SA3 regions for eleven DRT favorable parameters extracted and calculated from VISTA sample.

Table 4.1 presents the summary statistics of the raw data of all 40 SA3 regions for eleven DRT favorable parameters extracted and calculated from VISTA sample.

Table 4.1: Summary statistics of all parameters favorable to DRT

Parameters Minimum Maximum Mean Std. deviation Shopping and social trips (%) 22.395 43.127 34.565 3.963

(15-24) and (55 or more) age groups 22.773 43.966 31.570 4.808 (%)

Trips of female (%) 46.847 58.046 53.088 2.591

No driving licence (%) 11.914 31.860 23.544 4.208

Low household vehicle ownership 9.441 54.751 29.049 10.414 (0 or 1) (%)

Lack of train station proximity (%) 6.319 83.063 34.936 19.741

Low household income (%) 11.430 34.896 23.484 6.115

Higher trip waiting time (sec) 6.826 16.001 11.001 1.871

Higher trip walking time (sec) 7.611 16.096 13.060 1.993

Not in workforce (%) 19.857 52.584 39.993 6.397

Single person household (%) 1.660 16.859 7.586 3.583

Table 4.2: Classification of SA3s into SA4s

SA4 SA3 SA4 SA3 Brunswick - Coburg Keilor Darebin - South Macedon Ranges Essendon Melbourne - Moreland - North Melbourne - Melbourne City North West Sunbury Inner Tullamarine - Port Phillip Broadmeadows Stonnington - West Yarra Knox Manningham -

East Boroondara Melbourne - Maroondah Manningham - Outer East Melbourne - West Whitehorse - East Inner East Whitehorse - West Yarra Ranges

Bayside Cardinia Glen Eira Casey - North Melbourne - Melbourne - Inner South Kingston Casey - South Stonnington - South East East Dandenong Monash Banyule Darebin - North Melbourne - Nillumbik - North East Kinglake Brimbank Whittlesea - Wallan Hobsons Bay Melbourne - Maribyrnong West Melton - Bacchus Mornington Frankston Marsh Peninsula Mornington Peninsula Wyndham

Table 4.3 presents the data of all nine SA4 regions for eleven DRT favorable parameters extracted and calculated from VISTA sample. It can be seen that the values for some of the parameters are significantly different for some of the regions. This is the most applicable for Melbourne-Inner region which includes Melbourne CBD. Lack of train station proximity, 15-24 and 55 or more age groups, and no driving licence proportion is considerably low for this region while low household vehicle ownership and single person

35 household proportion is considerably high. This is because more working population is living in CBD and better public transportation leads to lesser car ownership.

Table 4.3: DRT favorable parameters at SA4 level

Low (15-24) Lack househ Shoppi and No of Low Higher Highe Trips old Single ng and (55 or drivin train househ trip r trip Not in of vehicle person SA4 social more) g station old waitin walkin workfor female owners househo trips age licence proxi income g time g time ce (%) (%) hip (0 ld (%) (%) groups (%) mity (%) (sec) (sec) or 1) (%) (%) (%)

Melbourne 37.35 27.20 52.77 19.76 46.39 24.77 19.68 10.70 14.10 30.91 13.12 - Inner Melbourne - Inner East 36.10 36.48 54.65 22.16 27.30 46.43 19.05 8.95 10.85 40.74 6.08

Melbourne - Inner South 36.87 34.97 55.16 21.59 30.43 12.82 19.57 10.55 12.73 40.59 9.85

Melbourne - North East 35.96 31.76 54.15 24.98 24.47 39.61 20.69 10.60 13.01 40.93 4.04

Melbourne - North West 29.49 31.81 50.29 25.37 22.45 28.64 27.79 11.87 13.17 42.36 5.91

Melbourne - Outer East 34.47 34.84 53.99 23.11 19.93 46.85 21.11 11.50 13.63 41.52 5.42

Melbourne - South East 34.19 29.95 52.23 25.96 25.95 41.94 29.30 10.80 13.08 43.77 5.94

Melbourne - West 31.76 27.01 51.85 27.45 28.26 35.41 26.85 11.79 12.42 41.45 6.68

Mornington Peninsula 36.00 39.03 55.58 20.61 26.41 55.43 29.08 11.97 13.34 44.82 9.02

4.2 Analysis of Travel Survey Data In this study, the variation of identified parameters has been studied across Greater Melbourne at the level of its 40 census regions, the Statistical Areas 3 or SA3. The values of all parameters do not have the same units and also their numerical scale is different. But they have to be aggregated to explore overall effects on DRT susceptibility. Also, the study aims to find the regions that are more susceptible to DRT than others. Hence, the relative values of parameters are more important than the absolute numbers. This

36 requires standardization of the existing raw values for each parameter for all regions to make them dimensionless and mutually compatible. Standardization is performed using ‘Additive’, ‘Maximum’ and ‘Extreme Value’ methods. Additive value is the relative position of a value with respect to average of all values and is formulated as:

xij x'ij = j xij where, j x'ij = 1 Maximum, (Max) value is the relative distance between the origin and the maximum. It can be formulated as:

xij x'ij = maxj{xij} Extreme Value (EV) is the relative position of a value on interval between the lowest and highest values and can be formulated as:

xij - minj{xij} x'ij = maxj{xij} - minj{xij}

The results after standardization have little variations using these 3 methods. For the purpose of this study, standardized values using ‘Extreme Value’ method are used. Table A.2 in Appendix presents the standardized version of raw data of all 40 SA3 regions for eleven DRT favorable parameters using ‘Extreme Value’ standardization method.

After standardization, each region has values for all the parameters ranging from 0 to 1. The region with the highest raw value of a particular parameter has standardized value of 1 for that parameter and similarly, the region with the lowest raw value will have standardized value of 0. As each of these parameters is favoring the use of DRT, the higher the value of the parameter, the higher is the susceptibility to use DRT, and hence their effect is considered additive. Also, each parameter is assumed to have the same weight in affecting susceptibility to use DRT. Thus, if any two or more parameters are highly dependent on each other, then, their effect on the results is counted repetitively if they are all taken into the account. Therefore, correlation analysis is performed on all the eleven parameters to find parameters with high correlation. Table 4.4 presents the results of correlation analysis conducted on the values of eleven parameters affecting the susceptibility to use DRT.

Table 4.4: Correlation of parameters affecting susceptibility of DRT

(15- 24) Low Shop No Lack of and Trips househol Low Highe Highe Single ping drivin train Not in (55 or of d vehicle househol r trip r trip person Parameters and g station workfor more) femal ownershi d waitin walkin househol social licenc proximi ce age e p (0 or income g time g time d trips e ty group 1) s

Shopping and social 1.00 0.30 0.35 -0.38 0.49 -0.21 -0.18 -0.20 0.13 -0.24 0.45 trips

(15-24) and (55 or 0.30 1.00 0.18 -0.31 -0.25 0.04 0.04 -0.16 0.04 0.29 -0.11 more) age groups

Trips of 0.35 0.18 1.00 0.03 -0.09 0.10 -0.21 -0.19 -0.20 0.10 -0.01 female

No driving -0.38 -0.31 0.03 1.00 -0.38 0.19 0.41 0.28 -0.19 0.72 -0.66 licence

Low household vehicle 0.49 -0.25 -0.09 -0.38 1.00 -0.46 -0.06 -0.01 0.13 -0.56 0.77 ownership (0 or 1)

Lack of train -0.21 0.04 0.10 0.19 -0.46 1.00 0.02 -0.02 -0.04 0.30 -0.55 station proximity

Low household -0.18 0.04 -0.21 0.41 -0.06 0.02 1.00 0.23 0.28 0.54 -0.16 income

Higher trip waiting -0.20 -0.16 -0.19 0.28 -0.01 -0.02 0.23 1.00 0.26 0.24 -0.13 time

Higher trip walking 0.13 0.04 -0.20 -0.19 0.13 -0.04 0.28 0.26 1.00 0.04 0.24 time

Not in -0.24 0.29 0.10 0.72 -0.56 0.30 0.54 0.24 0.04 1.00 -0.68 workforce

Single person 0.45 -0.11 -0.01 -0.66 0.77 -0.55 -0.16 -0.13 0.24 -0.68 1.00 household

Evans (1996) suggests for the absolute value of correlation coefficient:

• 0.00-0.19 “very weak”

• 0.20-0.39 “weak”

• 0.40-0.59 “moderate”

• 0.60-.079 “strong”

• 0.80-1.0 “very strong”

It can be seen in the Table 4.4 that the parameter ‘single person household’ has relatively higher values of correlation with ‘no driving licence availability’ (correlation coefficient = -0.66 with p-value = 0.000004 < 0.05), ‘low household vehicle ownership’ (correlation coefficient = 0.77 with p-value = 0.000000005 < 0.05), and ‘not in workforce’ (correlation coefficient = -0.68 with p-value = 0.000001 < 0.05) and the parameter ‘no driving licence availability’ is also having high correlation with ‘not in workforce’ (correlation coefficient = 0.72 with p-value = 0.0000001 < 0.05) . Hence, the parameters ‘single person household’ and ‘no driving licence availability’ are removed from the model and all other parameters are assumed to be affecting the results independently.

After doing correlation analysis, factor analysis is performed on the data. The first three generated factors having eigenvalues greater than 1 account for cumulative variation of less than 55% and have significant effect of all parameters in the factor pattern. Also, some parameters have negative effect on the factor pattern but for our study, parameters have been defined in a way that the higher the value of the parameter, the higher is the

susceptibility to use DRT. Also, Kaiser-Meyer-Olkin (KMO) measure of sampling adequacy was 0.544 and hence, the data is not considered suitable for factor analysis.

A simple regression analysis is conducted on the data of raw values of independent parameters to find the goodness of fit with the model. The analysis resulted in Multiple R value of 0.99, R Square value of 0.98, and Adjusted R Square value of 0.94, hence, it can be said that the model fits the data. Finally, the standardized values of all independent parameters are added for each region to explore the overall susceptibility to use DRT. Hence, the region with the highest total standardized value of parameters has the highest susceptibility to use DRT.

Table 4.5 presents the standardized values of all independent parameters and their total for each region.

Table 4.5: Total standardized value of DRT susceptibility for each SA3 regions

(15- 24) Low Shopping Lack of Higher Higher and Trips household Low Total and train trip trip Not in SA3 (55 or of vehicle household standardized social station waiting walking workforce more) female ownership income value trips proximity time time age (0 or 1) groups

Frankston 0.57 0.53 0.85 0.41 0.59 0.75 0.88 0.74 0.68 6.01

Moreland - 0.58 0.60 0.34 0.42 0.05 1.00 0.90 0.98 1.00 5.87 North

Brunswick - 0.66 0.07 0.61 0.81 0.62 0.40 1.00 0.91 0.61 5.69 Coburg

Mornington 0.74 1.00 0.71 0.33 0.69 0.75 0.24 0.25 0.84 5.56 Peninsula

Yarra Ranges 0.52 0.58 0.71 0.33 0.44 0.69 0.71 0.79 0.77 5.55

Darebin - 0.82 0.64 0.57 0.70 0.25 0.65 0.32 0.85 0.63 5.42 North

Casey - South 0.47 0.08 0.71 0.50 0.63 0.85 0.62 0.61 0.90 5.38

Brimbank 0.59 0.38 0.38 0.52 0.10 1.00 0.81 0.59 0.84 5.21

Casey - North 0.73 0.26 1.00 0.22 0.59 0.86 0.36 0.51 0.65 5.18

Manningham - 0.75 0.70 0.64 0.22 1.00 0.37 0.20 0.55 0.73 5.15 West

Maroondah 0.61 0.63 0.80 0.33 0.44 0.50 0.50 0.65 0.65 5.12

Whitehorse - 0.61 0.66 0.71 0.49 0.36 0.53 0.30 0.75 0.65 5.05 West

Darebin - 0.83 0.26 0.80 0.80 0.52 0.60 0.00 0.56 0.66 5.03 South

Knox 0.58 0.55 0.76 0.16 0.59 0.45 0.35 0.97 0.59 5.00

Maribyrnong 0.70 0.20 0.50 0.55 0.13 0.70 0.60 1.00 0.47 4.85

Kingston 0.68 0.58 0.63 0.52 0.23 0.39 0.41 0.86 0.56 4.85

Banyule 0.73 0.52 0.78 0.35 0.28 0.32 0.36 0.81 0.63 4.79

Bayside 0.55 0.56 0.93 0.39 0.00 0.10 0.61 0.98 0.67 4.79

Glen Eira 0.72 0.60 0.52 0.54 0.06 0.44 0.40 0.77 0.69 4.74

Dandenong 0.60 0.50 0.00 0.46 0.55 0.84 0.40 0.65 0.70 4.70

Monash 0.60 0.67 0.17 0.37 0.41 0.59 0.55 0.70 0.59 4.65

Stonnington - 0.84 0.56 0.89 0.40 0.05 0.46 0.21 0.58 0.62 4.61 East

Whittlesea - 0.59 0.30 0.55 0.24 0.69 0.48 0.46 0.55 0.69 4.55 Wallan

Whitehorse - 0.75 0.58 0.43 0.34 0.33 0.41 0.43 0.52 0.62 4.42 East

Stonnington - 1.00 0.35 0.59 0.73 0.03 0.36 0.38 0.73 0.24 4.41 West

Hobsons Bay 0.55 0.36 0.45 0.52 0.36 0.55 0.42 0.59 0.60 4.40

Essendon 0.60 0.31 0.59 0.67 0.08 0.44 0.39 0.78 0.49 4.37

Yarra 0.79 0.16 0.39 1.00 0.15 0.28 0.46 0.77 0.21 4.22

Keilor 0.46 0.43 0.29 0.30 0.67 0.31 0.51 0.67 0.54 4.19

Boroondara 0.62 0.58 0.75 0.48 0.21 0.07 0.20 0.69 0.54 4.14

Manningham - 0.44 0.51 0.49 0.00 0.83 0.00 0.56 0.61 0.68 4.12 East

Wyndham 0.23 0.03 0.65 0.29 0.75 0.45 0.35 0.57 0.49 3.82

Nillumbik - 0.48 0.24 0.71 0.04 0.51 0.13 0.51 0.59 0.62 3.81 Kinglake

Cardinia 0.44 0.19 0.51 0.27 0.14 0.67 0.24 0.47 0.81 3.74

Tullamarine - 0.24 0.14 0.04 0.31 0.25 0.91 0.58 0.48 0.64 3.60 Broadmeadows

Melbourne 0.57 0.31 0.25 0.84 0.18 0.24 0.37 0.71 0.00 3.47 City

Macedon 0.00 0.55 0.40 0.20 0.37 0.94 0.30 0.00 0.67 3.43 Ranges

Port Phillip 0.59 0.00 0.47 0.85 0.09 0.14 0.36 0.75 0.15 3.40

Melton - Bacchus 0.19 0.03 0.26 0.20 0.56 0.59 0.52 0.11 0.90 3.35 Marsh

Sunbury 0.42 0.40 0.46 0.21 0.11 0.33 0.46 0.02 0.59 3.01

For better visualization of these total standardized values, similar analysis was conducted on SA4 regions using classified data from Table 4.3. First, the values were standardized, correlated parameters were removed, and total standardized values were calculated. Results of this analysis is presented in Table 4.6.

Table 4.6: DRT susceptibility for SA4 regions

Predicted DRT Susceptibility (Total SA4 Standardised Value) Mornington Peninsula 7.82 Melbourne - Outer East 5.45 Melbourne - South East 5.34 Melbourne - North East 4.84 Melbourne - Inner South 4.77 Melbourne - Inner 4.41 Melbourne - West 4.37 Melbourne - Inner East 4.23 Melbourne - North West 4.22

It can be seen that there is a pattern significantly higher susceptibility to convert to DRT in the outer regions of Greater Melbourne than inner regions. But, this cannot be generalized as there are both outer and inner regions which are found to be more or less susceptible. These differences arise due to effect of different parameters. It has been explained in the next section with the help of findings.

The resulting spatial patterns of remaining parameters can be used to explore the susceptibility of DRT in Greater Melbourne and mapped across SA3 regions to find the more favorable areas for implementation. The findings are presented in Section 4.3.

4.3 Findings To visualize the overall susceptibility value calculated for each region, the values are mapped across Greater Melbourne. Figure 4.3 highlights the findings.

Figure 4.3: Susceptibility of DRT in Greater Melbourne. It can be seen that some regions have distinctly higher susceptibility to be origins of DRT trips and hence, provide better opportunity for implementation of the proposed type of service. These regions, interestingly, can be seen both around Melbourne City (CBD) and outer regions. The effect of various parameters on susceptibility of different regions is mixed. The reasons for higher susceptibility to DRT are different for various regions. It can be either due to household and person characteristics or due to their current trip pattern and public transport performance in their regions. Regions having higher susceptibility near Melbourne City are Moreland - North, Brunswick - Coburg, Darebin – North, and Brimbank. The contributing parameters for Moreland - North are low performance of the existing public transport resembling in higher waiting and walking times in current trips, workforce status, and low income. The neighboring region of

Brunswick - Coburg has higher susceptibility because of higher waiting and walking times in current trips and low vehicle ownership. The contributing parameters in the case of Darebin - North are higher shopping trips, higher walking time in current trips, and low vehicle ownership. Similarly, for Brimbank, contributing parameters are low income, higher waiting time in current trips, and workforce status. Among outer regions, Frankston, Mornington Peninsula, Yarra Ranges, Casey - South, and Casey - North are having higher susceptibility. For Frankston and Yarra Ranges, the contributing parameters are higher waiting and walking times in current trips, higher rates of trips of females, low income, and workforce status. For Casey - South and Casey - North, the contributing parameters are low income, trips of females, and workforce status. For Mornington Peninsula, the contributing parameters are age-group, workforce status, low income, and higher rates of trips of females. The susceptibility of Melbourne City is among the lowest. Reasons for this are lowest proportion of ‘not in workforce’ employment status, better train station proximity, and higher income.

Some of the SA3 regions with low susceptibility (in inner region and outer region) and high susceptibility (in inner region and outer region) are selected for better visualization of parameters affecting their susceptibility values. Graphs are drawn by comparing values of all the parameters for different combination of these regions. Figure 4.4, Figure 4.5, and Figure 4.6 presents the findings. In Figure 4.4 Melbourne City, an inner region with low susceptibility to DRT has been compared with Yarra Ranges, an outer region with high susceptibility to DRT. It can be seen that low household vehicle ownership, lack of train station proximity, low household income, and not in workforce are the parameters which are leading to high difference in susceptibility for these regions. Vehicle ownership is low in Melbourne City due to better provision of public transport, train stations are closer, household income is higher, and more people are in workforce. In Figure 4.5 Melbourne City, an inner region with low susceptibility to DRT has been compared with Brunswick - Coburg, an inner region with high susceptibility to DRT. It can be seen that lack of train station proximity, and not in workforce are the parameters which are leading to high difference in susceptibility for these regions. Train stations are closer, and more people are in workforce in Melbourne City region. In Figure 4.6 Yarra Ranges, an outer region with high susceptibility to DRT has been compared with Melton - Bacchus Marsh, an outer region with low susceptibility to DRT. It can be seen that various parameters are causing difference in susceptibility value for these regions. So, it can be said that regions

45 with low or high susceptibility exist in different parts of the city, and the difference can be due to various different parameters.

20 (in respective units) respective (in Value of parameters parameters of Value 10 Melbourne City 0 Yarra Ranges

Parameters

Figure 4.4: DRT favorable parameters comparison for Melbourne City and Yarra Ranges.

20 (in respective units) respective (in Value of parameters of Value 10

0 Brunswick - Coburg Melbourne City

Parameters

Figure 4.5: DRT favorable parameters comparison for Melbourne City and Brunswick – Coburg.

20 (in respective units) respective (in Value of parameters parameters of Value 10 Yarra Ranges 0 Melton - Bacchus Marsh

Parameters

Figure 4.6: DRT favorable parameters comparison for Yarra Ranges and Melton - Bacchus Marsh. The results presented in this section cannot be verified due to absence of city wide similar services, the methodology will be validated by applying it on the existing public transport in Melbourne which will also help in understanding the nature of competition among the DRT and existing public transport. The validation findings are presented in the next section.

4.4 Validation The aim of this research is to predict the demand patterns of DRT, but with the absence of city wide similar services, validating the results is a challenge. However, the methodology can be validated by applying it on an existing mode and then comparing the results with current demand patterns of that mode. Thus, the same methodology is performed for an existing mode, ‘Public Transport’ (using its own set of parameters), and results are compared with known usage pattern of that mode from VISTA. For this purpose, the characteristics of population and trips that are more favorable to be using public transport services are required. A review of some of the studies on demand pattern of public transport in different regions of the world has been conducted in order to identify socio-economic, demographic, and trip characteristics which are susceptible to use conventional public transport.

Ryley et al. (2014) found in a survey that the household car ownership of car users was 1.6 cars per household, but only 0.45 for bus users. Also, bus users had high proportions of the students, unemployed persons, carers and homemakers. The survey conducted in the Telebus Mobility and Access Benefits Project (Maddern and Jenner, 2007) mentions that compared with the Telebus service, the users of fixed bus service are more likely to travel for education purposes (36%) followed by work (26%) and shopping (19%). Daily bus users, who travel for work and education purposes make up 40% of the fixed route bus users and 41% of fixed route bus users are students. Annual statistics on the travel behavior of Sydney residents derived from the continuous Household Travel Survey conducted in 2008/09 (The Transport Data Centre (TDC), of Transport NSW 2010) can be reviewed to identify the dominant population and trip characteristics favoring the use of public transport. The highest proportion of use of bus and train is seen in the age group of 11-20 and 21-30. Also, the highest proportion of bus and train use is seen for trip purposes such as commute, education, and childcare.

Corpuz (2007) suggests that the vehicle ownership has the greatest impact of enhancing car use. The lack of access to the vehicle is associated with the higher public transport use. Public transport use is also found to be the highest for educational and commute trips. Those in combination with children in the households tend to use car more than public transport. Khan et al. (2007) highlight the positive perception of tertiary students and white collar workers towards public transport. Favorable mode choice of public transport for home based education (tertiary) trips is influenced by the household variable of tertiary students per household. They conclude that the modal split of public transport is likely to increase substantially with the increase in the number of tertiary education students. Annual bus statistics of England 2013/14 (Department of Transport, England 2014) show that the bus use is generally higher in areas with a higher proportion of no- car households. Paulley et al. (2006) show the statistics to highlight that the bus use (both in trips and person-km) falls substantially as the car ownership per household rises.

On the basis of this review, the following five parameters are identified to affect usability of public transport:

1. Education and work trips

2. (11-30) age group

3. Tertiary student trips

4. Low household vehicle ownership (0 or 1)

5. Single person household

Parameter 1 is the percentage of trips, which were done for either education or work purpose. Parameters 2 and 3 are respectively, the percentage of trips, which were done by persons of age groups (11-30) and tertiary students. Parameters 4 and 5 are respectively, the percentage of trips, which were done by persons belonging to households with none or single car ownership, and having single person. Table A.3 in Appendix presents the raw data of all 40 SA3 regions for 5 public transport favorable parameters extracted and calculated from VISTA sample.

Table 4.7 presents the summary statistics of the raw data of all 40 SA3 regions for five public transport favorable parameters extracted and calculated from VISTA sample.

Table 4.7: Summary statistics for Public Transport favorable parameters

Parameters Minimum Maximum Mean Std. deviation 14.08 7.45775 2.562082 Tertiary student trips (%) 2.57 (11-30) age groups (%) 17.15 32.66 24.185 3.181469

Education and work trips (%) 19.68 43.63 27.197 4.293304

Single person household (%) 1.66 16.86 7.58575 3.581817

Low household vehicle ownership (0 or 1) (%) 9.44 54.75 29.0495 10.41461

For better visualization, the raw data of 40 SA3 regions can be grouped into SA4 regions which are larger in size than SA3 regions. There are nine SA4 regions in Greater Melbourne. Table 4.8 presents the data of all nine SA4 regions for five public transport favorable parameters extracted and calculated from VISTA sample. It can be seen that the values for some of the parameters are significantly different for some of the regions. This is the most applicable for Melbourne-Inner region which includes Melbourne CBD. Low household vehicle ownership and single person household proportion is considerably high for this region.

Table 4.8: Public transport favorable parameters at SA4 level

SA4 Tertiary (11-30) Educatio Single Low student age n and person househol trips (%) groups work househol d vehicle (%) trips (%) d (%) ownershi p (0 or 1) (%) Melbourne - Inner 5.91 24.06 28.11 13.12 46.39 Melbourne - Inner East 6.75 22.19 25.64 6.08 27.30 Melbourne - Inner 7.38 26.81 28.08 9.85 30.43 South Melbourne - North East 6.65 25.47 25.69 4.04 24.47 Melbourne - North 7.89 22.74 27.15 5.91 22.45 West Melbourne - Outer East 6.30 22.47 25.65 5.42 19.93 Melbourne - South East 8.84 22.95 25.57 5.94 25.95 Melbourne - West 9.28 26.55 29.83 6.68 28.26 Mornington Peninsula 9.51 24.86 29.08 9.02 26.41

Here also, the raw values are needed to be standardized for further analysis. These values are standardized using the ‘Extreme Value’ method. Table A.4 in Appendix presents the standardized version of raw data of all 40 SA3 regions for five public transport favorable parameters using ‘Extreme Value’ standardization method.

In this case also, if any two or more parameters are highly dependent on each other, then their effect on the results would be counted repetitively if they are all taken in account. Therefore, correlation analysis is performed on all the five parameters to find the parameters with very high correlation. Table 4.9 presents the results of correlation analysis.

Table 4.9: Correlation of parameters affecting susceptibility of public transport

Low (11-30) Education Single household Tertiary age and work person vehicle student groups trips household ownership (0 or 1)

Tertiary student 1.00 0.12 0.21 0.51 0.62

11-30 age groups 0.12 1.00 0.55 0.03 0.11

Education and work 0.21 0.55 1.00 0.26 0.23 trips

Single person 0.51 0.03 0.26 1.00 0.77 household

Low household vehicle 0.62 0.11 0.23 0.77 1.00 ownership (0 or 1)

It can be seen that the parameter ‘low household vehicle ownership’ has relatively higher values of correlation with ‘tertiary student’ (correlation coefficient = 0.62 with p-value = 0.00002 < 0.05) and ‘single person household’ (correlation coefficient = 0.77 with p- value = 0.000000005 < 0.05). Hence, the parameter ‘low household vehicle ownership’ is removed from the model and all other parameters are assumed to be affecting the results independently. Finally, the standardized values of all independent parameters are added for each region to explore the overall susceptibility to use public transport. Hence, the region with the highest total standardized value of parameters has the highest susceptibility to public transport. Table 4.10 presents the standardized values of all the independent parameters and their total for each region.

Table 4.10: Total standardized value of public transport susceptibility for each SA3 region

(11- 30) Single Tertiary age Education person Total student group and work household standardized SA3 trips (%) (%) trips (%) (%) value

Melbourne City 1.00 1.00 1.00 0.93 3.93

Yarra 0.79 0.64 0.46 0.68 2.57

Port Phillip 0.40 0.37 0.52 1.00 2.29

Stonnington - West 0.48 0.50 0.28 1.00 2.25

Maribyrnong 0.71 0.54 0.31 0.49 2.06

Stonnington - East 0.66 0.59 0.17 0.60 2.01

Whitehorse - West 0.75 0.45 0.36 0.40 1.95

Boroondara 0.70 0.59 0.32 0.33 1.94

Monash 0.53 0.49 0.51 0.38 1.91

Brimbank 0.54 0.65 0.37 0.31 1.87

Darebin - North 0.73 0.58 0.25 0.25 1.81

Casey - North 0.47 0.62 0.38 0.29 1.76

Essendon 0.41 0.35 0.33 0.66 1.76

Frankston 0.36 0.51 0.33 0.55 1.76

Tullamarine - Broadmeadows 0.31 0.53 0.57 0.30 1.71

Kingston 0.47 0.35 0.31 0.54 1.67

Dandenong 0.35 0.49 0.53 0.30 1.66

Darebin - South 0.72 0.16 0.14 0.63 1.64

Sunbury 0.20 1.00 0.27 0.16 1.63

Keilor 0.18 0.49 0.52 0.36 1.56

Knox 0.49 0.49 0.30 0.27 1.55

Casey - South 0.21 0.79 0.35 0.16 1.50

Hobsons Bay 0.41 0.23 0.34 0.50 1.48

Glen Eira 0.49 0.29 0.08 0.58 1.45

Brunswick - Coburg 0.56 0.32 0.16 0.38 1.41

Whittlesea - Wallan 0.23 0.56 0.47 0.06 1.32

Maroondah 0.34 0.43 0.18 0.33 1.28

Yarra Ranges 0.29 0.36 0.23 0.37 1.25

Macedon Ranges 0.60 0.00 0.39 0.23 1.23

Banyule 0.39 0.38 0.17 0.23 1.18

Cardinia 0.00 0.50 0.34 0.28 1.13

Wyndham 0.09 0.48 0.39 0.17 1.12

Melton - Bacchus Marsh 0.05 0.54 0.33 0.19 1.10

Whitehorse - East 0.32 0.29 0.14 0.27 1.03

Moreland - North 0.35 0.19 0.10 0.34 0.99

Nillumbik - Kinglake 0.35 0.37 0.17 0.09 0.98

Bayside 0.43 0.09 0.00 0.44 0.96

Manningham - West 0.37 0.35 0.05 0.15 0.92

Mornington Peninsula 0.13 0.23 0.15 0.41 0.91

Manningham - East 0.13 0.35 0.29 0.00 0.76

To visualize the overall susceptibility value calculated for each region, the values are mapped across Greater Melbourne. Figure 4.7 highlights the findings of the predicted susceptibility of public transport in Greater Melbourne.

Figure 4.7: Susceptibility of public transport in Greater Melbourne. However, the existing pattern of usage of public transport in Greater Melbourne is already known from the household travel survey VISTA. The percentage of trips originating from each census region using public transport, i.e. train, tram and public bus, is calculated for each region. Table 4.11 presents the percentage of public transport usage for each SA3 region.

Table 4.11: Share of Public transport trips in SA3 regions of Melbourne

Public Public transport transport SA3 usage (%) SA3 usage (%)

Melbourne City 29.41 Manningham - West 4.17

Yarra 12.57 Banyule 3.76

Stonnington - West 11.20 Wyndham 3.67

Port Phillip 10.80 Keilor 3.61

Essendon 9.95 Nillumbik - Kinglake 3.35

Darebin - South 9.04 Maroondah 3.28

Maribyrnong 8.96 Whittlesea - Wallan 3.13

Brunswick - Coburg 8.07 Casey - South 2.68

Stonnington - East 7.79 Whitehorse - East 2.65

Boroondara 7.75 Casey - North 2.59

Darebin - North 7.56 Knox 2.54

Whitehorse - West 7.12 Sunbury 2.40

Melton - Bacchus Glen Eira 6.60 Marsh 2.39

Tullamarine - Monash 5.53 Broadmeadows 2.37

Brimbank 5.25 Manningham - East 2.35

Dandenong 5.18 Yarra Ranges 2.24

Bayside 4.97 Frankston 2.23

Hobsons Bay 4.72 Cardinia 1.68

Kingston 4.28 Macedon Ranges 1.34

Moreland - North 4.21 Mornington Peninsula 1.24

To visualize and compare with the predicted pattern, the values are mapped across Greater Melbourne. Figure 4.8 highlights the existing usage pattern of public transport in Greater Melbourne.

Figure 4.8: Existing usage pattern of public transport in Greater Melbourne. By visual inspection alone it can be seen already that the predicted and existing usage pattern of public transport is matching considerably, which validates this methodology. Hence, it can be said that our predicted demand pattern for DRT in Greater Melbourne is trustworthy and can be considered suitable for deciding opportunities of implementation of proposed DRT services. To find if the predicted and existing usage patterns are fitting statistically also, the R-squared value of cross-correlation is calculated. The predicted values of public transport susceptibility are plotted on y-axis against the existing public transport usage values on x-axis for all census regions. Also, the predicted values of susceptibility of DRT are plotted on y-axis in the same graph for same x-axis, to see

56 whether the public transport and DRT are competing or not. The graph is plotted after standardizing the final susceptibility values of predicted DRT, predicted public transport and existing public transport usage on the scale of 0 to 1 using extreme value method. The resulting graph is presented in Figure 4.9.

1.2

1 R² = 0.7355

0.8

0.6 Predicted DRT Predicted Public Transport 0.4

R² = 0.0342 Predicted susceptibility values susceptibility Predicted 0.2

0 0 0.2 0.4 0.6 0.8 1 1.2 Exisitng public transport usage

Figure 4.9: Predicted public transport and DRT susceptibility vs. existing public transport usage. It can be seen in the graph that the predicted and existing values of public transport usage are in good fit with an R-squared value of 0.73. Also, the predicted and existing values of public transport usage are correlated with coefficient of 0.86. It proves statistically what the visual inspection already indicated that the methodology is predicting the susceptibility results realistically. Additionally, we can see that the predicted susceptibility of DRT has a significantly distinct pattern compared to the existing public transport usage, with an R-squared value of 0.03. The outlier in the graph corresponds to the region of Melbourne City (CBD), which has exceptionally higher predicted and existing usage of public transport compared to all other regions.

Also, the difference between standardized values of predicted and existing public transport usage, and existing public transport usage and predicted DRT susceptibility is calculated. Results are presented in Table 4.12.

Table 4.12: Difference of standardized values of predicted DRT and public transport susceptibility with existing public transport usage

Predicted Public Existing Predicted Transpor Public DRT t (Total Transp (Total Standardi ort Standardiz zed Usage Standar Standar Standardi ed Value), Value), (%), dized dized zed V3, Absolute Absolute SA3 (V1) (V2) (V3) V1, (S1) V2, (S2) (S3) (S1-S3) (S2-S3)

Frankston 6.01 1.76 2.23 1.00 0.31 0.04 0.96 0.28

Mornington Peninsula 5.56 0.91 1.24 0.85 0.05 0.00 0.85 0.05

Moreland - North 5.87 0.99 4.21 0.95 0.07 0.11 0.85 0.04

Melbourne City 3.47 3.93 29.41 0.15 1.00 1.00 0.85 0.00

Yarra Ranges 5.55 1.25 2.24 0.85 0.15 0.04 0.81 0.12

Casey - South 5.38 1.50 2.68 0.79 0.23 0.05 0.74 0.18

Casey - North 5.18 1.76 2.59 0.72 0.32 0.05 0.67 0.27

Brunswick - Coburg 5.69 1.41 8.07 0.89 0.20 0.24 0.65 0.04

Maroondah 5.12 1.28 3.28 0.70 0.16 0.07 0.63 0.09

Knox 5.00 1.55 2.54 0.67 0.25 0.05 0.62 0.20

Manningham - West 5.15 0.92 4.17 0.71 0.05 0.10 0.61 0.06

Brimbank 5.21 1.87 5.25 0.73 0.35 0.14 0.59 0.21

Darebin - North 5.42 1.81 7.56 0.81 0.33 0.22 0.58 0.11

Kingston 4.85 1.67 4.28 0.62 0.29 0.11 0.51 0.18

Banyule 4.79 1.18 3.76 0.59 0.13 0.09 0.51 0.04

Whitehorse - West 5.05 1.95 7.12 0.68 0.38 0.21 0.47 0.17

Bayside 4.79 0.96 4.97 0.59 0.06 0.13 0.46 0.07

Whittlesea - Wallan 4.55 1.32 3.13 0.51 0.17 0.07 0.45 0.11

Dandenong 4.70 1.66 5.18 0.57 0.28 0.14 0.43 0.14

Whitehorse - East 4.42 1.03 2.65 0.47 0.08 0.05 0.42 0.03

Darebin - South 5.03 1.64 9.04 0.67 0.28 0.28 0.40 0.00

Monash 4.65 1.91 5.53 0.55 0.36 0.15 0.40 0.21

Glen Eira 4.74 1.45 6.60 0.58 0.22 0.19 0.39 0.03

Maribyrnong 4.85 2.06 8.96 0.62 0.41 0.27 0.34 0.13

Hobsons Bay 4.40 1.48 4.72 0.46 0.23 0.12 0.34 0.10

Manningham - East 4.12 0.76 2.35 0.37 0.00 0.04 0.33 0.04

Keilor 4.19 1.56 3.61 0.39 0.25 0.08 0.31 0.17

Stonnington - East 4.61 2.01 7.79 0.53 0.39 0.23 0.30 0.16

Cardinia 3.74 1.13 1.68 0.24 0.11 0.02 0.23 0.10

Port Phillip 3.40 2.29 10.80 0.13 0.48 0.34 0.21 0.14

Nillumbik - Kinglake 3.81 0.98 3.35 0.27 0.07 0.07 0.19 0.01

Wyndham 3.82 1.12 3.67 0.27 0.11 0.09 0.18 0.03

Tullamarine - Broadmeadows 3.60 1.71 2.37 0.20 0.30 0.04 0.16 0.26

Boroondara 4.14 1.94 7.75 0.38 0.37 0.23 0.15 0.14

Essendon 4.37 1.76 9.95 0.45 0.31 0.31 0.14 0.00

Macedon Ranges 3.43 1.23 1.34 0.14 0.15 0.00 0.14 0.14

Stonnington - West 4.41 2.25 11.20 0.47 0.47 0.35 0.11 0.12

Melton - Bacchus Marsh 3.35 1.10 2.39 0.12 0.10 0.04 0.07 0.06

Sunbury 3.01 1.63 2.40 0.00 0.27 0.04 0.04 0.23

Yarra 4.22 2.57 12.57 0.40 0.57 0.40 0.00 0.17

On an average, across all 40 census regions, there is a difference of only 0.11 between predicted susceptibility values of public transport and existing public transport usage. On the other hand, there is a difference of 0.43 between predicted susceptibility values of DRT and existing public transport usage. This again confirms the finding from the graph in Figure 4.9 that the predicted and existing values of public transport usage are in good fit, and the predicted susceptibility of DRT has a significantly distinct pattern compared to the existing public transport usage. It can be seen in Table 4.12 that the difference between the predicted susceptibility of DRT and existing usage of public transport is the highest for regions such as Frankston (difference = 0.96), Mornington Peninsula (difference = 0.85) and Moreland - North (difference = 0.85). These regions have very high susceptibility to DRT and very low existing public transport usage, and hence, provide the most favorable opportunities for the implementation of DRT. These results show that the DRT is not competing with existing public transport and provides better opportunities of implementation in the regions with high susceptibility of DRT but not conventional public transport.

4.5 Conclusions This chapter presented a survey-independent methodology to estimate the susceptibility to DRT in an urban space. Preference surveys data is widely used for demand estimation but due to identified challenges and complexity (Yang et al. 2009) in the design of these surveys for new and innovative transport modes such as DRT, an alternative is presented in this Chapter. The model is based on the well-established findings that the travel choice

60 is affected by population demographic and current travel characteristics. Hence, this research used those characteristics available from VISTA and analyzed the variation of population demographic and current travel characteristics in the city. An extensive review of similar services elsewhere identified the dominant population characteristics that affect the usability of proposed transport service. With these inputs it is predicted which regions in the target city are more likely to use the proposed transport service.

For validation, the case of public transport is taken. It found corresponding favorable parameters for public transport by a similar review and predicted the usability of public transport in the city by using review inputs and population characteristics from VISTA data. As the pattern of public transport usage in the target city is already known (again using VISTA data), a strong match between predicted and known results validated the methodology for prediction purpose. As laid out before, the same validation method can be applied to other existing modes as well. Preference data has no other validation method in comparison. The methodology has been applied to find the more feasible areas of implementation of DRT services in Melbourne. Looking at the results, it can be concluded that there is indeed a significant spatial difference in susceptibility to use DRT over the area of a city (here: Greater Melbourne) and this form of mobility is not competing with the existing public transport. Regions with higher susceptibility to DRT are found both in outer regions and around the center of city. But, both the existing and predicted usage of public transport is higher around the center of city. There are regions with very high susceptibility to DRT and very low existing public transport usage. Hence, these regions which provide the most favorable opportunities for the implementation of DRT can be chosen to implement the proposed DRT service and can be studied further.

Chapter 5: Predicting Likeliness of Trips to Convert to DRT

In the last chapter, the methodology to predict the susceptibility of regions to DRT in a city has been presented. This has been investigated in the Greater Melbourne study area using Statistical Areas 3 (SA3) as per the ABS census. Yet, these statistical regions are of a size such that variations in susceptibility may exist within them, requiring careful attention. Travel demand has to be predicted at a finer level granularity for further modeling and simulation. Generally, travel surveys contain travel diaries of a day for a small sample of the population. This is true with the travel survey data used in this study too. Hence, the data is not enough for comparisons and modeling when smaller regions are considered. This chapter focusses on creating travel diaries of the study area requiring the creation of a synthetic population to predict the likeliness of travelers to shift to DRT. To demonstrate this process, a synthetic population is created for the entire Greater Melbourne area, and then travel dairies are assigned using sample VISTA data. The SA3 regions are used to predict travel demand, and the DRT favorable parameters identified in Chapter 3 are used to predict which trips done by the population of that region are more likely to shift to DRT. Section 5.1 presents methodology and results for synthetic population creation, Section 5.2 explains assignment of travel diaries to the synthetic population, and then susceptibility to DRT is explored in Section 5.3.

5.1 Synthetic Population This study has undertaken travel demand modeling to explore the propensity for the use of demand-responsive transport services. Existing transport models treat individuals in an aggregate manner, but evaluation of newer, demand-responsive transport modes requires details of individual behavior. For example, demand-responsive vehicles adapt their routes and timings depending on who is riding the vehicle; hence, when assessing demand for DRT, individual simulation techniques are of interest. However, to perform activity based micro-simulation models, details of individual and household attributes are needed for the entire population. Information at this level of detail, however, is publically unavailable due to privacy reasons. This creates a lack of completeness in micro-data, ultimately influencing micro-simulation.

To fulfil this fundamental data requirement, a synthetic population can be created using various techniques and many software applications have been developed over the years. Auld et al. (2008) presented a population synthesizer called PopSynWin. It was developed at University of Illinois at Chicago. It was successfully used for synthesizing the population of Chicago, Illinois and can be applied to scenarios in other study areas with different input data. Ye et al. (2009) presented an alternate approach called PopGen. It was developed at Arizona State University. Its primary application was Maricopa County, Arizona, which can also be used for other areas and input data after making suitable adaptation as per geographic classification and data of census. PopSynWin is based on an Iterative Proportional Fitting (IPF) algorithm, and PopGen uses an Iterative Proportional Update (IPU) algorithm. ILUTE (Salvini and Miller 2005), FSUMTS (Srinivasan and Ma 2009), CEMDAP (Pinjari et al. 2006), ALBATROSS (Arentze et al 2014) are some of the micro-simulation model systems which use outputs of the synthetic population developed using the IPF approach. The PopGen software was also successfully used to create a synthetic population for the Greater Sydney Metropolitan area using the ABS2006 census (Lim and Gargett 2013).

In this chapter, both PopSynWin and PopGen are used to create a synthetic population for another metropolis, Greater Melbourne, both at household and person levels. This involves testing and comparing either application’s performance investigating the advantages or disadvantages of each. Both output populations will be compared and visualized, and the more accurate approach will be used in further work.

5.1.1 Background Disaggregate modeling is gaining popularity in travel demand modeling and traffic micro-simulation modeling, and requires micro-level data of population at household and person level as a key input. This popularity is increasing demand for micro-level details of the population (Ryan et al. 2009). But these details are generally made unavailable for public use by suppressing due to privacy and confidentiality issues or are expensive due to a high cost of acquisition (Moeckel et al. 2003). To fulfil this lack of data, the population can be synthesized to represent actual demographics of the study area as per population census. This resulted in the development of synthetic population generation techniques for creating micro-level details of the population for disaggregate modeling(Beckman et al. 1996). Different data sources can be combined and marginal sums and correlation structure can be matched to derive a disaggregated representation of the households and persons. (Müller et al. 2010). Synthetic population matches the distribution of persons and households as per the demographics from census data to create a valid computational representation of the population in study area (Huynh et al. 2014). Aggregate data and sample micro-level data is taken from census and extended to create micro-level details for entire population. It is ensured that generated population is conforming as much as possible to the actual population (Lim and Gargett 2013). So in whole, it involves algorithms which are applied to a sample micro-data and its aggregate data, to result a synthetic population which is statistically representative of actual population and represents the best estimates (Ryan et al. 2009).

Quality of synthetic population depends on the algorithm used (Lim and Gargett 2013). Synthetic Reconstruction (SR) (Wilson and Pownall 1976) and Combinatorial Optimization (CO) (Williamson et al. 1998) are two major approaches, which have been used to create synthetic populations. In this study, the software used is based on SR methods which generally involves two principal stages of fitting and generation (Bowman 2009). SR methods are generally based on the IPF procedure, which creates a multiway table of conditional probabilities pertaining to population characteristics (Ryan et al. 2009). In fitting stage, micro-level sample data is fitted to subtotals in the aggregate data. Then, micro-level sample data is used to create observations, such that the generated distributions are collectively consistent with the cross tabulations provided by the actual population data. Then, in generation stage, joint distribution probabilities generated in the fitting stage are used to expand micro-level sample data to full population (Lim and

Gargett 2013). In the IPF procedure, the number of observations created synthetically in a given category of the selected control variable matches the corresponding aggregated subtotals in the population data and also the correlation structure of the disaggregated sample data is retained (Müller et al. 2010).

Traditional IPF procedure is capable of matching the observed and simulated control variables either based on household joint-distributions or person level joint-distributions, but it does not perform well in closely matching both household and person level distributions simultaneously (Lim and Gargett 2013). To overcome this limitation, the IPU procedure can be used for the population synthesis (Ye et al. 2009). The IPU procedure is an extended version of the IPF procedure and is developed to improve the fit of household level distributions and person level distributions simultaneously. In the IPU algorithm, household weights are adjusted to match both household and person level distributions as closely as possible. The IPF procedure is first applied to household level and then to person level to produce two separate and independent set of multiway constraints (Lim and Gargett 2013). The joint distributions of both household and person control variables are used to adjust and assign weights for each household/person type iteratively until marginal totals at both household and person level are closely matched to the known marginal distribution (Ye et al. 2009). Bar-Gera et al. (2009) presented an entropy optimization approach and Müller et al. (2010) presented Hierarchical IPF (HIPF) to achieve the same goal as IPU.

CO methods involve random selection of a group of individuals from a disaggregate data so that it matches the population size of the small area. Statistical comparison is carried out on this observed sample against a pre-defined set of demographic characteristics of the small area to generate a new observed sample. This process is repeated to gradually improve the goodness of fit and is stopped when a critical value of the comparison statistics is achieved (Huynh et al. 2013). Both SR and CO methods are found to be capable of generating synthetic population, which according to statistical tests, is similar to the actual population. But, both of them have their own set of superiorities and limitations. Similarly, various synthesizers using IPF algorithm are used for a different region and requires different input data. As per requirement of different projects, choice of method and model can be made, and results can be tested and compared as explained in this chapter. Choice of PopSynWin and PopGen for this research has been motivated by their use of two different variations of SR approach (namely IPF and IPU) and their

65 free availability as software packages, which are user-friendly, easy to set up, and faster in generating output. Furthermore, results are found to be satisfactory.

The next sub-section presents the process of creating a synthetic population for the Greater Melbourne study area using the ABS 2011 Census data. Micro-data is created for households and persons in Greater Melbourne using the fine granularity Statistical Area Level 1 (SA1) data. PopSynWin and PopGen are used as tools to generate two different synthetic populations. The micro-data within these populations is then aggregated to validate against the actual aggregate census data to evaluate the representations. Finally, both populations are compared to use the more accurate of them for further work.

5.1.2 Data Preparation This section presents the data sources and methodology to prepare the input data required for the tools and determination of control variables for the population synthesis. The input data is taken from the ABS 2011 population census. Data from the census of population and housing confidentialized unit record files (1% CURFs) will be referred as micro-data, and the census data from the Basic Community Profile (BCP) DataPacks at the SA1 level is referred as aggregate data in this chapter.

The micro-data has geographic classification at SA4 level. This is linked to the SA1 level data using a correspondence table from the Data Cubes of ASGS (Australian Statistical Geography Standard). In the ABS census, SA1s are designed as the smallest unit and generally have a population of 200 to 800 persons, (averaging approximately 400 persons). They are built from whole Mesh Blocks and approximately 55,000 SA1s cover the whole of Australia. The next level up are SA2s, which are a general-purpose medium- sized area built from SA1s. Next are SA3s which are built from whole SA2s, which provide a standardized regional breakup of Australia. Finally, SA4s are the largest sub- State regions in the Main Structure of the ASGS. SA4s are built from whole SA3s and there are 88 SA4 spatial units across Australia. Reviewing the data within the study area, there are 9658 SA1s, spreading across all 9 SA4s of Greater Melbourne.

At the first step, three control variables at household level and three control variables at person level need to be chosen for the population synthesis. It is important to note that the number of categories for a particular control variable may not always match the micro- data and aggregate data. The next step of this process is to link the selected control variables between both of the datasets. All the control variables have to be regrouped in

66 categories following similar pattern for both of the datasets. Some of these categories need to be collapsed or merged, from one of the data, to ensure that both datasets contain similar control variables with synchronized categories.

All sub totals and totals in the census aggregate data from BCP DataPacks are required to be made consistent for all the control variables considered. Introduction of confidentiality process to the aggregate data due to privacy rights leads to inconsistencies across sub-totals and totals of the aggregate data for different control variables. The synthesis process in both of the software is not initiated if there is any inconsistency in totals or sub totals of the control variables. Hence, these inconsistencies had to be resolved and totals and sub-totals had to be matched. This process of overcoming these inconsistencies is widely recognized as census data balancing (Chin and Harding 2006). This process is very time consuming process because inconsistency does not follow a particular pattern and dataset for entire Greater Melbourne is large. Minimal adjustments and uniform redistributions are made to achieve consistencies across sub-totals and totals of all six control variables in each geographical area. It involved collation and rearrangement of data for all control variables.

5.1.3 Synthesis Three control variables at household level: Dwelling Structure, Number of Persons Usually Resident in Dwelling, Number of Motor Vehicles Owned by Household and three control variables at person level: Gender, Age, Labor Force Status are used to generate synthetic population both from PopGen and PopSynWin software. Corresponding tables are extracted from micro-data and aggregate data and prepared according to software input requirements. After preparing and balancing data as per requirements of PopGen and PopSynWin software separately, input data files are fed to software to create synthetic population for city of Melbourne.

5.1.4 Results After generating both of the synthetic populations, which are expected to have micro-data of 100% population across considered control variables, results will be aggregated to check their match with the actual population. Total number of households and persons generated in micro-data of synthetic population are compared with actual numbers fed to the software by calculating percentage difference to check that no significant loss has occurred due to processing of the software. Also, the distribution of these aggregated numbers across categories of control variables is compared with actual distribution by

67 calculating change in percentage distribution for each category. The generated populations are validated by extracting aggregated cross-tables of any two household or person level control variables at a time from both synthetic and actual populations and comparing them. To check if results are consistent across all the geographical areas considered, distribution of number of SA1s for absolute percentage change in aggregated numbers generated are compared to actual population.

Table 5.1 presents overall comparison of synthesized and actual population of Greater Melbourne (for considered SA1s) by using PopGen and PopSynWin software.

Table 5.1: Overall comparison of synthesized and actual population

% % Actual PopSynWin PopGen Difference Difference

Households 1,430,581 1,430,387 -0.01 1,430,581 0.00

Persons 3,995,665 3,708,065 -7.20 3,763,258 -5.82

Tables 5.2 and 5.3 present the distribution of actual and synthesized population for Greater Melbourne by household and person level control variables respectively.

Table 5.2: Distribution of household level control variables

Actual PopSynWin Difference PopGen Difference

% % % point % % point

Dwelling Structure

Separate house 72.65 72.56 0.1 72.65 0

Semi-detached, 11.56 11.66 -0.1 11.56 0 row or terrace house, town house, etc.

Flat, unit or 15.31 15.37 -0.06 15.31 0 apartment

Other dwelling 0.43 0.39 0.04 0.43 0

Other groups 0.04 0.02 0.02 0.04 0

Number of Persons

1 23.3 23.83 -0.53 23.19 0.11

2 31.98 32.41 -0.43 32.02 -0.04

3 16.96 17.15 -0.19 16.98 -0.02

4 17.4 16.92 0.48 17.4 0

5 7.17 6.74 0.43 7.2 -0.02

6 or more 3.2 2.95 0.24 3.21 -0.01

Number of Motor Vehicles

1 35.03 34.98 0.05 34.96 0.06

2 36.93 36.34 0.59 36.91 0.02

3 10.79 10.76 0.03 10.79 0

4 or more 5.21 5.22 -0.02 5.22 -0.01

None 9.12 9.37 -0.24 9.18 -0.05

Other groups 2.92 3.33 -0.41 2.94 -0.02

Table 5.3: Distribution of person level control variables

Actual PopSynWin Difference PopGen Difference

% % % point % % point

Gender Male 49.14 49.24 -0.1 49.53 -0.39

Female 50.86 50.76 0.1 50.47 0.39

Age (years) 0-4 6.5 7.07 -0.57 6.39 0.11

5-9 6.05 6.6 -0.55 5.95 0.1

10-14 5.95 6.5 -0.55 5.82 0.13

15-19 6.32 5.95 0.37 6.31 0.01

20-24 7.47 7.62 -0.14 7.38 0.1

25-29 7.92 8.03 -0.11 7.92 0

30-34 7.51 7.53 -0.02 7.56 -0.05

35-39 7.51 7.64 -0.13 7.59 -0.08

40-44 7.51 7.68 -0.17 7.6 -0.09

45-49 6.95 7.17 -0.22 7.02 -0.07

50-54 6.44 6.52 -0.08 6.5 -0.06

55-59 5.63 5.51 0.12 5.71 -0.08

60-64 5.11 4.9 0.21 5.17 -0.06

65-69 3.86 3.53 0.33 3.9 -0.03

70-74 3.04 2.76 0.27 3.07 -0.03

75-79 2.42 2.18 0.24 2.42 0

80-84 1.97 1.57 0.4 1.94 0.03

85 and above 1.84 1.25 0.59 1.78 0.06

Labor Force Status Employed 48.22 51.98 -3.76 48.96 -0.73

Unemployed 2.79 1.46 1.33 2.78 0.01

Not in the labor force 26.18 24.59 1.59 25.94 0.24

Other groups 22.8 21.96 0.84 22.32 0.84

Table 5.4 presents weighted average of absolute percentage point difference in distribution.

Table 5.4: Weighted average of absolute percentage point difference in distribution

PopSynWin PopGen

Dwelling Structure 0.10 0.00

Number of Persons Usually Resident in Dwelling 0.42 0.04

Number of Motor Vehicles Owned by Household 0.27 0.04

Gender 0.10 0.39

Age 0.26 0.07

Labor Force Status 2.46 0.53

For further validation and comparison of synthetic populations, the TableBuilder tool (provided on the ABS website) is used to generate cross-tables of two control variables at a time for aggregated totals in actual population of entire Greater Melbourne (9658 SA1s), and compared with similar cross-tables from synthetic populations (9424 SA1s). Figure 5.1 shows the variation of ‘dwelling structure’ with ‘number of motor vehicles owned by household’, and their comparison for PopSynWin, PopGen, and actual population. Figure 5.2 shows the variation of ‘gender’ with ‘labor force status’, and their comparison for PopSynWin, PopGen, and actual population.

Figure 5.1: (Number of motor vehicles, Dwelling structure) variation.

Figure 5.2: (Labor force status, Gender) variation. Table 5.5 presents summation of squares of difference in square roots of numbers in cross- tables generated by software and in cross-tables of TableBuilder (=Delta1).

Table 5.5: Delta1 for all control variables

PopSynWin PopGen

Dwelling Structure - Number of Persons Usually 436 270 Resident in Dwelling

Number of Persons Usually Resident in Dwelling - 2056 889 Number of Motor Vehicles Owned by Household

Dwelling Structure - Number of Motor Vehicles Owned 8601 10802 by Household

Gender - Age 11712 4251

Gender - Labor Force Status 17944 3902

Age - Labor Force Status 49780 5735

Absolute percentage difference with respect to actual population in the number of households generated and the number of persons generated can be calculated at each SA1 level. Average of these percentage differences (=Delta2) is plotted against the number of SA1s in the range for both of the software in Figure 5.3.

2500

2000

1500

PopSynWin

Number of SA1 of Number 1000 PopGen

500

1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9

9-10

10-11 11-12 12-13 13-14 14-15 15-16 16-17 17-18 18-19 19-20 More than… More Less than Less 1 % Difference

Figure 5.3: Delta2 against number of SA1. To check the performance of both software across geographies, percentage difference in number of persons generated compared to actual population is plotted on maps for PopGen and PopSynWin. Numbers against SA1s are summed up to visualize them at SA3 and SA4 level. Firstly in Figure 5.4, a particular legend is used to plot maps at SA3 level and comparing them for PopSynWin and PopGen. Range of values in legend shows the percentage difference in number of persons generated compared to actual population (=Delta3).

Figure 5.4: Visualizing Delta3 at SA3 level for PopSynWin and PopGen.

PopSynWin PopGen

To better understand what characteristics of geography are affecting the performance of software, Delta3 is plotted on maps for PopSynWin and PopGen at SA4 level in Figure 5.5. Range of values in legend shows the percentage difference in number of persons generated compared to actual population (=Delta3).

Figure 5.5: Visualizing Delta3 at SA4 level for PopSynWin and PopGen.

PopSynWin PopGen

The percentage distribution of all of the control variables across their categories are calculated for each SA4 and subtracted from percentage distribution for entire study area to get percentage point difference. The absolute weighted average of percentage point difference across categories (=Delta4) is calculated for each control variable and presented in Table 5.6. This table shows that household and person characteristics are how much deviant for each SA4 compared to the entire study area.

Table 5.6: Delta4 for all control variables across all SA4

Melbou Melbou Melbou Melbour Melbour Melbou Melbourn Melbour Morningt rne - rne - rne - ne - ne - rne - e - Inner ne - on Inner North North Outer South Inner East West Peninsula South East West East East

Dwelling 30.78 2.17 8.82 10.36 10.90 11.92 9.35 10.11 10.23 Structure

Number of Motor Vehicles 10.96 1.17 1.72 2.15 2.07 4.22 2.65 0.91 1.35 Owned by Household

Number of Persons Usually 6.91 0.54 1.35 1.67 2.02 0.86 2.48 2.19 2.56 Resident in Dwelling

Labor Force 5.36 1.29 0.98 0.40 3.21 2.00 1.75 2.21 1.87 Status

Gender 0.24 1.04 0.85 0.08 0.25 0.09 0.67 0.68 0.49

Age 2.35 0.65 0.51 0.19 0.43 0.64 0.28 0.59 0.79

5.1.5 Interpretation of Results In Table 5.1, the number of households synthesized is 100% in case of PopGen and almost 100% in case of PopSynWin. But 100% persons are not synthesized in both cases. This is generally due to the inconsistency of household and person level variables interaction in micro-data and aggregate data and also due to the process of balancing. The difference can be reduced with more consistent data but is acceptable for our purpose of travel demand modeling. In Tables 5.2 and 5.3, the distribution across control variables in the synthetic population of both of the software match closely with the actual population but that is even better with PopGen which can be seen in Table 5.4.

Although the number of SA1s considered in generating cross-tables of control variables are lesser for synthetic populations than actual population, Figures 5.1 and 5.2 show that the distribution is matching closely with the actual population for both software. Table 5.5 further shows that these results are even better for PopGen. In Figure 5.3, the

76 distribution of Delta2 is positively skewed with a long tail for both software but PopGen results are more closely matching with the actual population even at each SA1 level.

In Figure 5.4, by visualizing percentage difference in the number of persons generated compared to the actual population at SA3 level, PopGen is performing better than PopSynWin across the study area. In Figure 5.5, it can be seen that both of the software are performing worse for some particular SA4s. For SA4, Melbourne – Inner (plotted in the darkest color), Delta3 is the highest for both PopGen and PopSynWin. This can be understood by looking at the distribution of categories across the control variables used for population synthesis. A more number of people live in flats than a separate house, more one-person households, and more people are employed in Melbourne – Inner compared to all other SA4s. In Table 5.6, the distribution across all control variables in Melbourne-Inner are more deviant compared to other SA4. Most of the distribution characteristics of all other SA4s are matching to each other, they look to be deviant from overall study area because of the effect of Melbourne-Inner SA4.

Synthetic population generated by PopGen is used for the purpose of travel diary assignment in the next section. For this purpose, a surveyed sample of individual travel diaries (containing travel and activity details of an individual on a particular day), for example from the Victorian Integrated Survey of Travel and Activity (VISTA), is used to assign travel diaries to the synthetic population.

5.2 Travel Diary Assignment Construction of individual-level activity schedules, also known as travel diaries, is an important component in urban transport models. These travel diaries can be aggregated to realistically represent population travel demand. Such travel diaries are generally comprised of the sequence of trips each individual in the population makes as well as trip attributes such as travel mode, trip purpose, and departure time. Such details of travel diary are generally available in travel surveys but only for a sample of a population. For predicting travel demand and pattern of an entire population of a region, these sample travel diaries can be assigned to a synthetic population. Various studies to date have used sets of decision making algorithms (Vaughn et al. 1999, Arentze and Timmermans 2004, Ma et al. 2009), such as determining activity patterns, travel time of day, activity durations, and travel mode choice for assigning travel details to individuals in a synthetic population. Each decision making step is modelled based on statistical models or decision trees, which has an associated error term (Huynh et al. 2013). As a result, the overall error

77 rate of an activity list assignment is compounded as the output of a particular decision making step is the input to the next step. Furthermore, these approaches consider individual level travel details assignment, ignoring the interdependencies that exist among individuals in a household. With a view to address these limitations, this study uses a single-step approach using household level semi-deterministic search method to assign travel diary to each individual in a synthetic population (Huynh et al. 2013).

In previous section, a synthetic population has been created for the Greater Melbourne. In this section, the focus is on one region of Greater Melbourne for travel diary assignment and further prediction of DRT susceptibility of its trips. Chapter 4 predicted certain SA3 regions having a higher susceptibility to DRT; Yarra Ranges was one of them. This region is also of interest for a wider research group, this study is part of. Hence, the synthetic population of Yarra Ranges SA3 generated by PopGen is assigned travel diaries of a particular day using sample VISTA travel diaries. Figure 5.6 presents the map of Yarra Ranges in Greater Melbourne.

Figure 5.6: Map of Yarra Ranges in Greater Melbourne. The synthetic population of Yarra Ranges has 136807 persons living in 48667 households. These persons are assigned travel diaries using VISTA sample of Yarra Ranges. Travel diaries are assigned to the synthetic population by matching household and person characteristics semi-deterministically. It involves selection of a synthetic person, followed by a search in the VISTA sample for a person which is having personal and household characteristics matching to the synthetic person as far as possible. This

78 search relaxes constrains of exact matching gradually. The criteria and order used for matching are presented in Figure 5.7. At the end, each person of SP has a travel diary consisting of an ordered set of trips, travel purpose, travel mode, departure time, and estimated trip time. This trip assignment resulted in 420136 trips by the synthetic population of Yarra Ranges.

Select a Synthetic Household (SH) for Travel Diary (TD) assignment

If SH has atleast one dependent child

YES (Find how many dependent children?)

Search for Vista Households having same or greater number of dependent children, same Motor Vehicle (MV) ownership, same number of employed persons, Found same number of females, same number of persons None found?

Search for Vista Households having same or greater number of dependent children, same MV ownership, same number of employed persons, Found same number of females None found?

Search for Vista Households having same or greater number of dependent children, same MV ownership, same number of employed persons Found None found?

Search for Vista Households having same or greater number of dependent children, same MV ownership None found? Found

Search for Vista Households having same or greater number of dependent children Found None found? Select one of them randomly Search for Vista Households having largest number of dependent children, select one of them randomly and duplicate dependent children in Vista household to match the number of dependent children in SH

Assign Person IDs of dependent children in selected Vista household to dependent children in considered SH, Assign Person IDs of student adults in selected Vista household to student adults in considered SH, Assign Person IDs of elderly people in selected Vista household to elderly people in considered SH, Assign Person IDs of other adults in selected Vista household to other adults in considered SH Randomly select Person IDs YES of student adults, elderly Any adult remains unassigned people and other adults from NO END any Vista households and assign to unassigned ones Search for Vista Households having same MV ownership, same number of employed persons, same number of females, same number of persons Found None found?

Search for Vista Households having same MV ownership, same number of employed persons, Found same number of females None found?

Search for Vista Households having same MV ownership, same number of employed persons Found None found?

Search for Vista Households having same MV ownership Select one of them randomly

Figure 5.7: Travel diary assignment to synthetic persons using VISTA sample.

The generated activity schedules are validated against Vista sample of Yarra Ranges. Figures 5.8, 5.9 and 5.10 compare the proportions by trip purpose, travel mode and the number of trips made daily in synthetic population travel diary with Vista sample respectively. These figures show that proportion of trips by various classifications is matching considerably in Vista sample and assigned trips. Hence, the generated travel diaries can be considered validated.

25 20 15 10

5 % of Total Trips Total of % 0 YR Synthetic Trips YR Vista Trips

Trip Purpose

Figure 5.8: Trip count proportions by trip purpose.

YP Synthetic Trips % of Total Trips Total of % 20 YR Vista Trips 10

Travel Mode

Figure 5.9: Trip count proportions by travel mode.

20 YR Synthetic Trips 10

% of Total Persons Total of % YR Vista Trips 0 1 2 3 4 5 6 7 8 or more Number of Trips in a Day

Figure 5.10: Population proportion by the number of trips in a day.

5.3 Exploring DRT Susceptibility of Trips in Yarra Ranges In Sections 5.1 and 5.2 a synthetic population has been created and trips of a particular day have been assigned to it using sample travel survey respectively. In this section, DRT susceptibility of those trips is checked. Following eleven parameters to affect the DRT susceptibility are identified in Chapter 4:

1. (15-24) and (55 or more) age groups

2. Trips of female

3. Not in workforce

4. No driving licence

5. Low household vehicle ownership (0 or 1)

6. Low household income

7. Single person household

8. Lack of train station proximity

9. Shopping and social trips

10. Higher trip waiting time

11. Higher trip walking time

Parameter 9 is related to a trip characteristic, i.e. trip purpose, which varies in different trips of a person on a day and mode choice of a person for an earlier trip affects the mode choice for a later trip and vice versa. For example, a person may leave home for a chain

82 of trip purposes (Social-Work-Shopping) and his mode choice cannot be predicted for a single trip. Similarly, Parameter 10 and 11 are in direct relation with the mode selected for a particular trip, for example, waiting/walking time is negligible in a personal transport trip. They were earlier considered to take into account the performance of public transport in different regions of Melbourne, but now the focus is on a particular study region. Hence, in this study only parameters 1 to 8 which are properties of a person and his household independent of the trip will be considered to test how they affect the susceptibility of a trip to convert to DRT.

Figure 5.11 presents the number of trips and the parameters they satisfy. Figure 5.12 presents the distribution of trips as per the number of parameters, they are satisfying. It can be seen that there are no trips satisfying all the 8 parameters and a few satisfy 7 or 6 parameters, but there are many trips satisfying 4 or 5 parameters, which can also be considered likely to convert to DRT. All of the considered parameters favor the use of DRT, and this research considered their effect with equal weight. It can be speculated that the trips which are satisfying more parameters are also more likely to convert to DRT. For example, if an elderly female has to make her travel mode decision for a particular trip, her decision may vary based on whether she owns a car or not. Based on our model, she is more likely to choose DRT if she doesn’t own a car, as car ownership is one of the parameter found to be affecting DRT susceptibility. For the purpose of visualization and to highlight some of the characteristics, the trips can be roughly classified in the category of likeliness. It can be considered that the trips satisfying five or more parameters are ‘highly likely’ to shift to DRT, trips satisfying three or four parameters are ‘likely’ and all other trips are ‘less likely’. Results based on this classification are presented in Figure 5.13. It can be seen that more than 12 % trips are highly likely to convert to DRT along with another 34.76 % trips which are also likely to convert. It should not be taken as all the highly likely or likely trips will convert to DRT definitely and none of the less likely trips will convert. It is a prediction based on the characteristics of the population and gives the likely travel demand to focus more for DRT.

250000

200000

150000

100000 Number of Trips of Number 50000

Number of trips satisfying a particular parameter

Parameter

Figure 5.11: Number of trips satisfying a particular parameter.

120000

100000 Number of trips 80000 satisfying as 60000 number of parameters

40000 Number of Trips of Number 20000

0 0 1 2 3 4 5 6 7 8 Number of Parameters

Figure 5.12: Number of trips satisfying as number of parameters.

250000 52.92 % 200000

34.76 % Likeliness of 150000 trips to convert to 100000 DRT

Number of Trips of Number 12.32 % 50000

0 Less Likely Likely Highly Likely

Figure 5.13: Likeliness of trips to convert to DRT. It is important to understand the current trip characteristics of the trips, which are in different categories of likeliness to convert to DRT. It will help to highlight what kind of trips are more or less likely to convert to DRT. It is to mention again that the trip related parameters are not considered while predicting the likeliness of the trips. Figure 5.14 presents the proportion of trip purposes in categories of likeliness.

35 30 25

20 % 15 10 Less Likely 5 Likely 0 Highly Likely

Trip Purpose

Figure 5.14: Proportion of trip purposes as per likeliness to convert to DRT. Figure 5.14 shows that the highest proportions of trips in highly likely category are for shopping, and social purposes. It is interesting to point out here that shopping and social

85 trips were also found as one of the dominant parameters to affect the susceptibility to DRT in the review of similar services. On the other hand, the work related trips are the least likely to convert to DRT. These findings are supported by what Lave and Mathias (2000) demonstrated as the traditional dial-a-ride market. But recently, Wang et al. (2015) highlighted the potential role of DRT in providing for demand from a different market segment, i.e. those in employment, although in rural areas. Cooper et al. (2008) also point out the potential contribution of DRT in providing employment access in rural areas.

In order to know about the current travel modes that will be more likely to convert to DRT, trips are classified as per their travel modes in different categories of likeliness. Figure 5.15 presents the proportion of travel modes in three categories of likeliness.

% 40 Less Likely 30 Likely 20 Highly Likely

0 Car Driver Car Passenger Walking Public Other Transport Travel Mode

Figure 5.15: Proportion of travel modes as per likeliness to convert to DRT. This figure shows that the proportion of public transport trips is almost similar in both highly likely and less likely trips which further verifies the hypothesis and finding in Chapter 4 that the existing public transport may not always be competing with DRT services. Also, it can be observed that share of walking trips is higher in highly likely trips. It has been a cause of concern advocated by health organizations that implementing convenient transport options such as DRT, which provide cheaper door-to-door transportation on demand may prevent people from walking to and from their homes or places of activities which involve short distance, or transit stops. But, at the same time, walking may indicate lack of proper transport accessibility especially for the people who

86 are transport disadvantaged. DRT is widely seen as tackling this problem of transport accessibility. It can also be seen that a significant proportion of trips highly likely to convert to DRT are currently undertaken as car passengers. This highlights the importance of DRT for people in certain age groups, who are dependent and also people who do not have access to private transport or driving licence and not using public transport either due to unavailability or unwillingness. They are likely to shift to DRT, if provided.

So far, the susceptibility of trips to convert to DRT on the basis of individual and household characteristics has been analyzed and trips have been classified into three categories of likeliness. Also, variations in current trip characteristics such as trip purpose and travel mode have been checked as per categories of likeliness. To see if there is any variation in likeliness of trips to convert to DRT based on what time of day they are undertaken, proportion of trips in three categories of likeliness is plotted in an hourly interval for a day. Figure 5.16 presents the findings. Observation from this figure is that there are no trips from 12 a.m. to 7 a.m. which are highly likely to convert to DRT. This is a representation of the age groups considered favorable to DRT and also lesser access to private transport, which prevents from traveling in odd hours when public transport is not available or is less frequent.

% 8

6 Less Likely Likely 4 Highly Likely 2

4 a.m. - 5 a.m. 5 - a.m. 4 a.m. 6 - a.m. 5 a.m. 7 - a.m. 6 a.m. 8 - a.m. 7 a.m. 9 - a.m. 8 a.m. 2 - a.m. 1 a.m. 3 - a.m. 2 a.m. 4 - a.m. 3

1 p.m. - 2 2 p.m. - p.m. 1 3 p.m. - p.m. 2 4 p.m. - p.m. 3 5 p.m. - p.m. 4 6 p.m. - p.m. 5 7 p.m. - p.m. 6 8 p.m. - p.m. 7 9 p.m. - p.m. 8

9 a.m. - 10 a.m. 10 - a.m. 9 a.m. 1 - a.m. 12

12 p.m. - 1 p.m. 1 - p.m. 12 p.m. 10 - p.m. 9

10 a.m. - 11 a.m. 11 - a.m. 10

11 a.m. - 12 p.m. 12 - a.m. 11 a.m. 12 - p.m. 11 10 p.m. - 11 p.m. 11 - p.m. 10 Time Period

Figure 5.16: Temporal distribution of trips across categories of likeliness. Out of total 136807 individuals of Yarra Ranges, 66872 (making 222323 trips) are less likely, 51053 (making 146029 trips) are likely and 18882 are highly likely to shift to DRT. To see if there is any variation in likeliness of trips to convert to DRT based on where the travelers reside within the Yarra Ranges, trips in three categories of likeliness are plotted on maps. Figures 5.17, 5.18 and 5.19 present the spatial distribution of household locations of individuals as per category of their likeliness.

Figure 5.17: Distribution of individuals who are less likely to convert to DRT in Yarra Ranges.

Figure 5.18: Distribution of individuals who are likely to convert to DRT in Yarra Ranges.

Figure 5.19: Distribution of individuals who are highly likely to convert to DRT in Yarra Ranges. Looking at these figures, it can be said that the distribution of individuals is in the same pattern across Yarra Ranges in all three categories of likeliness in the proportion of their share (52.92 % for less likely, 34.76 % for likely and 12.32 % for highly likely). This is probably the pattern of population density in the Yarra Ranges. In order to check, if there are some regions that are more susceptible to DRT than others within the Yarra Ranges, proportion of individuals highly likely to convert to DRT in each SA1 of Yarra Ranges is plotted in the map in Figure 5.20.

Figure 5.20: SA1 Regions in Yarra Ranges as per percentage of individuals who are highly likely to convert to DRT. It can be seen in Figure 5.20 that although, most of the SA1 regions have the proportion of highly likely individuals in the range of 10 – 20 %, there are some regions which are more likely to convert than others, but without any pattern.

5.4 Conclusions In this chapter, DRT travel demand has been explored for an entire population of a region. It involved the creation of a synthetic population using aggregate data from population census and sample micro-data, assignment of travel diary to the synthetic population using sample travel surveys and prediction of likeliness of generated trips to convert to DRT using favorable population parameters. This chapter presented synthetic reconstruction methods as valid tools for this purpose. Software tools based on both IPF and IPU are found to be capable enough in generating a representative population. Both of the tools, PopSynWin and PopGen have been developed in the US for generating the synthetic population of a particular region. They follow the terminology of US census but, were applied in this study for an Australian city with little difficulty. Hence, these

92 tools can be adapted to generate synthetic populations of different cities by matching terminology of the census. Both of them recognize the inconsistency in input data directly from census and require rearrangement and balancing of input datasets. PopGen (following IPU algorithm) is performing better in generating population at person level as expected. Generated results have been validated by checking their distribution across categories of all control variables with actual aggregate data. The difference in synthetic population compared to actual population is visualized spatially to check any significant change in performance of the tools across different geographies. It is found that performance varies according to variation in household and person characteristics of a particular geography compared to the overall study area.

Travel diaries from a sample travel survey such as VISTA in Melbourne have been used to estimate travel demand of an entire population for a particular day. The methodology used is based on a semi-deterministic search to find the best representative person of a synthetic person in the sample travel survey. This travel demand is checked for the likeliness of trips to convert to DRT using the parameters favorable to DRT identified from the review of various DRT services existing in the world. Current trip properties of the generated trips such as trip purpose and travel mode have been studied to find any variation on the basis of their predicted likeliness to convert to DRT. Also, temporal and spatial variations of the trips have been explored.

Chapter 6: Conclusions

This research is motivated to predict the susceptibility of various regions in a city to a proposed mode of transportation. Based on the review of literature, it is recognized that the service-specific preference surveys to know the willingness of a population to use a proposed mode of transportation, are not only expensive but ambiguous too. This argument is found to be more valid if the proposed services are novel in their character, which may be the case with the demand responsive forms of transport that are in the focus of this study. Also, importance of various population characteristics in affecting its travel decision making is clear from the literature review. Hence, this research developed a novel method to find spatial patterns of demand for DRT, or in principle, any particular transport service using the variation in socio-economic and demographic characteristics, and existing trip characteristics of the population. A review of similar transport services elsewhere is conducted to gain the parameters favoring use of DRT; the findings have been presented in the Chapter 3. These identified parameters are used in Chapter 4 to predict the susceptibility to use DRT. This research tested the hypothesis that the dominant parameters identified from multiple studies elsewhere can be used for the study area (here: Melbourne, but in principle the same method can be applied in any area). It attempted to verify this test by doing a similar prediction for an existing mode (here: public transport, but in principle any other mode could be used for validation as well) and comparing its results with already known ones. The predicted outcomes determined the regions that are more favorable to use DRT, and proved that existing public transport may not always compete with DRT.

The validation of the method shows that analysis of socio-economic, demographic, and trip characteristics of the population can be used to predict spatial demand pattern of a proposed transport service in a target city. Thus, the hypothesis can be considered to be independently proven, and the requirement of a service-specific preference/adaptation survey can be eliminated by gaining usage pattern insights from existing similar services elsewhere. This proves reliability of using experiences for designing new transport services rather than misleading preference surveys.

As the review of operating services presented in Chapter 3 is generic, it can be used in conjunction with respective travel surveys and population characteristics in different places. Also, the identified parameters favoring the use of DRT can be useful in developing mode choice models for the purpose of demand modeling and simulation. Similarly, a review can be done for different kind of proposed transport modes by identifying their own set of affecting parameters, and the provided methodology can be applied and validated for exploring demand patterns. Hence, it can be said that the provided methodology can be evaluated, and thus its predictive capacity is expected to be realistic and reliable, and is also it is generic and hence transferrable.

The size of census regions used in this research is relatively large to ensure availability of considerable data in travel survey for each region. Finer studies can be conducted by choosing regions of interest and finding susceptibility pattern by dividing into smaller statistical areas. Based on availability of reliable data, more parameters such as parking cost and availability or a public transport accessibility index with possible effects on usability of any transport mode can be considered in the model. Also, the effect of each individual parameter on the variation in susceptibility patterns can be analyzed in detail to make policy and investment decisions when targeting a particular segment. Different existing transport modes such as taxi or private car can be selected, both for further validation and also for studying the competing nature among existing modes and predicted services.

In this research, the parameters which are known to affect the value of our desired outcome, i.e. susceptibility to DRT are identified from the literature review. Actual values of these parameters for all the regions in the study area is available, which are standardized and added to find relatively which regions are more susceptible than others. The assumption here is that all the parameters are equally likely to affect the susceptibility

95 which isn’t true. The parameters can be given weights by ranking their importance in affecting the outcome. If the parameters can be prioritized, they can be weighted by using techniques such as ‘Rank Order Centroid (ROC)’ (Ahn 2011), or ‘Pairwise Comparison’ (Barzilai 1997). But, as the parameters come from the review of existing services, the ranking should also be based on the same. For example, if the trip and user details of existing DRT services is available, they can be analyzed to predict whether the purpose of the trip or the age of the user has affected the mode choice more. Similarly, weights can be given to all the parameters either directly on the basis of their importance ranking or better by actually quantifying the relative importance, using further insights from the existing DRT services.

It would also be important to assess the impact of different service levels on DRT demand. The importance of DRT systems’ level of service (LoS) on its uptake is mentioned in many studies and they have tried to find the optimized operational characteristics to keep a certain LoS but there is no research on how passengers would react to a DRT service with a certain LoS (Berbeglia et al. 2010, Cordeau and Laporte 2003, Dessouky et al. 2003). In other words, no study has considered and modelled people’s preferences to investigate directly how changing LoS would affect the demand for a certain DRT system. Since, investigating the effect of LoS on DRT uptake is beyond the scope of this work, it was assumed that the service is good enough that its LoS would not hinder anyone from using it, if other criteria are fulfilled. Hence, this study focused on the effects of population characteristics on DRT demand, effects of service levels and DRT service provision (i.e. supply side factors) are not dealt with.

In order to predict the travel demand at a finer level granularity, this research created travel diaries for a study area. It required the creation of a synthetic population, and assignment of sample travel diaries to the synthetic population, to predict the likeliness of travelers to shift to DRT. Findings are presented in Chapter 5. By validating the aggregated version of the synthetic population with the actual aggregate data, it can be concluded that the requirement of micro-data of population at household and person level, for example for travel demand modeling, can be met effectively by creating a synthetic population.

Trip count proportions by trip purpose, travel mode, and number of trips made daily in generated travel diaries for the synthetic population and original sample travel diaries are

96 found to be matching considerably. Hence, the generated trips can be considered as a realistic representation of travel demand for the concerned population. The generated travel demand can further be equipped by assigning location coordinates to the mentioned places of activities in the travel diaries. These travel diaries can act as direct input to agent based models or micro-simulation models for transport network of a region. This research applied the identified parameters favorable to the proposed transport mode to these travel diaries and predicted travel demand patterns. These identified patterns of travel demand for a proposed transport mode help in making policy decisions in terms of favorable regions for implementation and also to target particular sections of the population who are transport disadvantaged.

Appendix

Table A.1: Raw data from VISTA sample of 40 SA3 regions for DRT favorable parameters

Low (15-24) Lack Shopp househ and (55 of Low Higher Highe ing Trips No old Single or train househ trip r trip Not in and of driving vehicle person SA3 more) station old waitin walkin workfor social female licence owners househo age proxi income g time g time ce (%) trips (%) (%) hip (0 ld (%) groups mity (%) (sec) (sec) (%) or 1) (%) (%) (%)

Banyule 37.57 33.82 55.63 22.99 25.19 28.16 19.02 10.11 11.71 40.44 5.19

Bayside 33.86 34.73 57.23 21.12 27.11 6.32 13.70 12.43 11.61 41.65 8.36

Boroondara 35.21 35.00 55.23 22.61 31.29 22.55 13.18 8.64 7.61 37.37 6.64

Brimbank 34.69 30.87 51.06 28.02 32.87 13.88 34.90 14.26 12.65 47.20 6.44

Brunswick - 35.98 24.29 53.67 25.74 46.17 54.16 20.86 16.00 15.31 39.71 7.38 Coburg

Cardinia 31.58 26.72 52.61 28.43 21.68 16.69 27.25 8.99 13.50 46.38 5.97

Casey - 37.48 28.24 58.05 25.89 19.32 51.76 31.65 10.15 14.80 41.03 6.14 North

Casey - 32.17 24.50 54.84 31.86 32.10 54.66 31.36 12.48 9.70 49.33 4.02 South

Dandenong 34.80 33.35 46.85 24.63 30.42 48.64 31.05 10.49 13.16 42.88 6.18

Darebin - 39.39 36.37 53.19 24.44 41.06 25.48 26.61 9.77 13.77 40.61 5.44 North

Darebin - 39.70 28.25 55.85 26.58 45.83 45.99 25.48 6.83 12.33 41.32 11.23 South

Essendon 34.89 29.31 53.48 23.95 39.93 12.72 21.76 10.44 14.90 36.03 11.72

Frankston 34.28 34.09 56.33 21.81 28.21 51.66 29.10 14.87 13.87 42.21 10.10

Glen Eira 37.23 35.47 52.69 22.47 34.08 10.61 21.80 10.52 13.30 42.42 10.45

Hobsons Bay 33.72 30.33 51.91 24.66 32.91 34.12 24.36 10.70 11.93 39.37 9.19

Keilor 32.03 31.94 50.10 21.46 22.84 57.93 18.77 11.50 14.17 37.62 7.15

Kingston 36.49 35.09 53.85 20.97 33.09 24.03 20.65 10.54 13.51 38.07 9.86

Knox 34.50 34.40 55.41 21.35 16.66 51.54 22.09 9.99 12.52 39.20 5.81

Macedon 22.39 34.53 51.36 20.86 18.39 34.74 33.46 9.58 15.91 41.64 5.21 Ranges

Manningha 31.51 33.52 52.29 24.73 9.44 70.39 11.43 11.94 12.80 42.05 1.66 m - East

Manningha 37.99 37.59 53.97 20.73 19.18 83.06 20.04 8.62 12.31 43.86 3.92 m - West

Maribyrnon 36.92 27.02 52.41 25.39 34.38 16.36 27.91 12.31 13.15 35.39 9.10 g

Maroondah 35.14 36.12 55.75 23.99 24.40 39.94 23.27 11.44 13.99 41.10 6.64

Melbourne 34.15 29.25 49.68 11.91 47.63 20.40 17.12 10.19 12.28 19.86 15.84 City

Melton - Bacchus 26.25 23.36 49.70 31.86 18.53 49.18 25.18 11.64 15.83 49.36 4.50 Marsh

Monash 34.92 36.93 48.80 18.98 26.24 37.93 25.18 11.88 14.24 39.24 7.41

Moreland - 34.39 35.54 50.67 29.58 28.70 9.89 34.79 15.04 15.97 52.58 6.87 North

Mornington 37.71 43.97 54.83 19.41 24.60 59.20 29.06 9.06 12.81 47.42 7.94 Peninsula

Nillumbik - 32.29 27.79 54.77 25.96 11.14 45.28 14.44 11.49 12.61 40.21 3.01 Kinglake

Port Phillip 34.72 22.77 52.10 16.72 47.85 13.40 14.70 10.12 13.61 24.83 16.79

Stonnington 39.91 34.59 56.85 21.80 27.43 10.33 22.14 8.71 12.49 40.23 10.71 - East

Stonnington 43.13 30.27 53.40 15.05 42.58 8.73 19.89 10.27 16.10 27.86 16.86 - West

Sunbury 31.15 31.25 52.03 29.67 18.94 15.10 19.22 11.07 7.76 39.02 4.11

Tullamarine - 27.47 25.77 47.30 25.27 23.37 25.54 32.73 12.17 12.02 40.92 6.20 Broadmeado ws

Whitehorse - 38.02 35.07 51.65 20.57 24.97 31.94 21.15 10.78 14.50 40.07 5.71 East

Whitehorse - 35.09 36.85 54.75 23.13 31.43 33.69 23.92 9.58 12.63 40.99 7.68 West

Whittlesea - 34.59 29.07 53.00 26.52 20.49 59.50 22.70 11.04 13.95 42.44 2.52 Wallan

Wyndham 27.24 23.45 54.17 27.34 22.61 63.50 21.90 10.03 8.55 35.91 4.18

Yarra 38.85 26.23 51.22 18.38 54.75 18.01 17.93 11.02 14.18 26.73 12.03

Yarra 33.19 35.09 54.83 24.90 24.17 40.42 27.59 13.34 14.33 45.19 7.27 Ranges

Table A.2: Standardized data of 40 SA3 regions for DRT favorable parameters

(15- 24) Low Shopping Lack of Higher Higher and Trips No household Low Single and train trip trip Not in SA3 (55 or of driving vehicle household person social station waiting walking workforce more) female licence ownership income household trips proximity time time age (0 or 1) groups

Banyule 0.73 0.52 0.78 0.56 0.35 0.28 0.32 0.36 0.81 0.63 0.23

Bayside 0.55 0.56 0.93 0.46 0.39 0.00 0.10 0.61 0.98 0.67 0.44

Boroondara 0.62 0.58 0.75 0.54 0.48 0.21 0.07 0.20 0.69 0.54 0.33

Brimbank 0.59 0.38 0.38 0.81 0.52 0.10 1.00 0.81 0.59 0.84 0.31

100

Brunswick - Coburg 0.66 0.07 0.61 0.69 0.81 0.62 0.40 1.00 0.91 0.61 0.38

Cardinia 0.44 0.19 0.51 0.83 0.27 0.14 0.67 0.24 0.47 0.81 0.28

Casey - North 0.73 0.26 1.00 0.70 0.22 0.59 0.86 0.36 0.51 0.65 0.29

Casey - South 0.47 0.08 0.71 1.00 0.50 0.63 0.85 0.62 0.61 0.90 0.16

Dandenong 0.60 0.50 0.00 0.64 0.46 0.55 0.84 0.40 0.65 0.70 0.30

Darebin - North 0.82 0.64 0.57 0.63 0.70 0.25 0.65 0.32 0.85 0.63 0.25

Darebin - South 0.83 0.26 0.80 0.74 0.80 0.52 0.60 0.00 0.56 0.66 0.63

Essendon 0.60 0.31 0.59 0.60 0.67 0.08 0.44 0.39 0.78 0.49 0.66

Frankston 0.57 0.53 0.85 0.50 0.41 0.59 0.75 0.88 0.74 0.68 0.55

Glen Eira 0.72 0.60 0.52 0.53 0.54 0.06 0.44 0.40 0.77 0.69 0.58

Hobsons Bay 0.55 0.36 0.45 0.64 0.52 0.36 0.55 0.42 0.59 0.60 0.50

Keilor 0.46 0.43 0.29 0.48 0.30 0.67 0.31 0.51 0.67 0.54 0.36

Kingston 0.68 0.58 0.63 0.45 0.52 0.23 0.39 0.41 0.86 0.56 0.54

Knox 0.58 0.55 0.76 0.47 0.16 0.59 0.45 0.35 0.97 0.59 0.27

Macedon Ranges 0.00 0.55 0.40 0.45 0.20 0.37 0.94 0.30 0.00 0.67 0.23

Manningham - East 0.44 0.51 0.49 0.64 0.00 0.83 0.00 0.56 0.61 0.68 0.00

Manningham - West 0.75 0.70 0.64 0.44 0.22 1.00 0.37 0.20 0.55 0.73 0.15

Maribyrnong 0.70 0.20 0.50 0.68 0.55 0.13 0.70 0.60 1.00 0.47 0.49

Maroondah 0.61 0.63 0.80 0.61 0.33 0.44 0.50 0.50 0.65 0.65 0.33

Melbourne City 0.57 0.31 0.25 0.00 0.84 0.18 0.24 0.37 0.71 0.00 0.93

Melton - Bacchus Marsh 0.19 0.03 0.26 1.00 0.20 0.56 0.59 0.52 0.11 0.90 0.19

Monash 0.60 0.67 0.17 0.35 0.37 0.41 0.59 0.55 0.70 0.59 0.38

101

Moreland - North 0.58 0.60 0.34 0.89 0.42 0.05 1.00 0.90 0.98 1.00 0.34

Mornington Peninsula 0.74 1.00 0.71 0.38 0.33 0.69 0.75 0.24 0.25 0.84 0.41

Nillumbik - Kinglake 0.48 0.24 0.71 0.70 0.04 0.51 0.13 0.51 0.59 0.62 0.09

Port Phillip 0.59 0.00 0.47 0.24 0.85 0.09 0.14 0.36 0.75 0.15 1.00

Stonnington - East 0.84 0.56 0.89 0.50 0.40 0.05 0.46 0.21 0.58 0.62 0.60

Stonnington - West 1.00 0.35 0.59 0.16 0.73 0.03 0.36 0.38 0.73 0.24 1.00

Sunbury 0.42 0.40 0.46 0.89 0.21 0.11 0.33 0.46 0.02 0.59 0.16

Tullamarine - Broadmeadows 0.24 0.14 0.04 0.67 0.31 0.25 0.91 0.58 0.48 0.64 0.30

Whitehorse - East 0.75 0.58 0.43 0.43 0.34 0.33 0.41 0.43 0.52 0.62 0.27

Whitehorse - West 0.61 0.66 0.71 0.56 0.49 0.36 0.53 0.30 0.75 0.65 0.40

Whittlesea - Wallan 0.59 0.30 0.55 0.73 0.24 0.69 0.48 0.46 0.55 0.69 0.06

Wyndham 0.23 0.03 0.65 0.77 0.29 0.75 0.45 0.35 0.57 0.49 0.17

Yarra 0.79 0.16 0.39 0.32 1.00 0.15 0.28 0.46 0.77 0.21 0.68

Yarra Ranges 0.52 0.58 0.71 0.65 0.33 0.44 0.69 0.71 0.79 0.77 0.37

Table A.3: Raw data from VISTA sample of 40 SA3 regions for public transport favorable parameters

Low (11- household 30) Single vehicle Tertiary age Education person ownership student groups and work household (0 or 1) SA3 trips (%) (%) trips (%) (%) (%)

Banyule 4.95 29.43 28.04 5.19 25.19

Bayside 4.88 32.66 26.18 8.36 27.11

102

Boroondara 5.92 22.80 25.12 6.64 31.29

Brimbank 14.08 32.63 43.63 6.44 32.87

Brunswick - Coburg 8.79 27.17 28.56 7.38 46.17

Cardinia 6.62 20.13 22.05 5.97 21.68

Casey - North 10.78 25.59 27.14 6.14 19.32

Casey - South 7.97 26.80 28.71 4.02 32.10

Dandenong 10.85 19.61 22.92 6.18 30.42

Darebin - North 6.76 25.03 27.54 5.44 41.06

Darebin - South 2.57 24.93 27.85 11.23 45.83

Essendon 3.11 25.48 27.48 11.72 39.93

Frankston 7.31 22.64 27.55 10.10 28.21

Glen Eira 11.16 24.09 28.34 10.45 34.08

Hobsons Bay 10.96 26.15 25.74 9.19 32.91

Keilor 6.63 24.69 32.29 7.15 22.84

Kingston 5.27 25.77 30.90 9.86 33.09

Knox 6.13 25.35 33.35 5.81 16.66

Macedon Ranges 9.48 17.15 29.07 5.21 18.39

Manningham - East 8.99 22.08 23.47 1.66 9.44

Manningham - West 7.30 20.66 27.94 3.92 19.18

Maribyrnong 10.12 26.33 23.67 9.10 34.38

Maroondah 8.26 21.72 21.66 6.64 24.40

Melbourne City 8.67 24.83 31.90 15.84 47.63

Melton - Bacchus Marsh 4.65 24.74 32.25 4.50 18.53

Monash 8.00 22.62 27.02 7.41 26.24

Moreland - North 10.65 26.27 27.31 6.87 28.70

103

Mornington Peninsula 11.70 27.08 30.61 7.94 24.60

Nillumbik - Kinglake 8.04 24.88 26.30 3.01 11.14

Port Phillip 3.58 24.58 29.06 16.79 47.85

Stonnington - East 8.21 24.73 26.88 10.71 27.43

Stonnington - West 7.17 22.85 32.24 16.86 42.58

Sunbury 6.42 23.88 24.01 4.11 18.94

Tullamarine - Broadmeadows 6.29 21.72 23.07 6.20 23.37

Whitehorse - East 4.01 20.65 23.24 5.71 24.97

Whitehorse - West 7.04 23.12 23.85 7.68 31.43

Whittlesea - Wallan 6.83 22.55 20.89 2.52 20.49

Wyndham 6.58 22.88 23.86 4.18 22.61

Yarra 7.49 18.60 19.68 12.03 54.75

Yarra Ranges 4.09 22.53 26.51 7.27 24.17

Table A.4: Standardized data of 40 SA3 regions for public transport favorable parameters

Low (11- household Tertiary 30) Education Single vehicle student age and work person ownership SA3 trips groups trips household (0 or 1)

Banyule 0.39 0.38 0.17 0.23 0.35

Bayside 0.43 0.09 0.00 0.44 0.39

Boroondara 0.70 0.59 0.32 0.33 0.48

Brimbank 0.54 0.65 0.37 0.31 0.52

Brunswick - Coburg 0.56 0.32 0.16 0.38 0.81

Cardinia 0.00 0.50 0.34 0.28 0.27

104

Casey - North 0.47 0.62 0.38 0.29 0.22

Casey - South 0.21 0.79 0.35 0.16 0.50

Dandenong 0.35 0.49 0.53 0.30 0.46

Darebin - North 0.73 0.58 0.25 0.25 0.70

Darebin - South 0.72 0.16 0.14 0.63 0.80

Essendon 0.41 0.35 0.33 0.66 0.67

Frankston 0.36 0.51 0.33 0.55 0.41

Glen Eira 0.49 0.29 0.08 0.58 0.54

Hobsons Bay 0.41 0.23 0.34 0.50 0.52

Keilor 0.18 0.49 0.52 0.36 0.30

Kingston 0.47 0.35 0.31 0.54 0.52

Knox 0.49 0.49 0.30 0.27 0.16

Macedon Ranges 0.60 0.00 0.39 0.23 0.20

Manningham - East 0.13 0.35 0.29 0.00 0.00

Manningham - West 0.37 0.35 0.05 0.15 0.22

Maribyrnong 0.71 0.54 0.31 0.49 0.55

Maroondah 0.34 0.43 0.18 0.33 0.33

Melbourne City 1.00 1.00 1.00 0.93 0.84

Melton - Bacchus Marsh 0.05 0.54 0.33 0.19 0.20

Monash 0.53 0.49 0.51 0.38 0.37

Moreland - North 0.35 0.19 0.10 0.34 0.42

Mornington Peninsula 0.13 0.23 0.15 0.41 0.33

Nillumbik - Kinglake 0.35 0.37 0.17 0.09 0.04

Port Phillip 0.40 0.37 0.52 1.00 0.85

Stonnington - East 0.66 0.59 0.17 0.60 0.40

105

Stonnington - West 0.48 0.50 0.28 1.00 0.73

Sunbury 0.20 1.00 0.27 0.16 0.21

Tullamarine - Broadmeadows 0.31 0.53 0.57 0.30 0.31

Whitehorse - East 0.32 0.29 0.14 0.27 0.34

Whitehorse - West 0.75 0.45 0.36 0.40 0.49

Whittlesea - Wallan 0.23 0.56 0.47 0.06 0.24

Wyndham 0.09 0.48 0.39 0.17 0.29

Yarra 0.79 0.64 0.46 0.68 1.00

Yarra Ranges 0.29 0.36 0.23 0.37 0.33

106

Bibliography

ActiveAge (2008). "An introduction to Demand Responsive Transport as a Mobility Solution in an Ageing Society." http://www.activeage.org/publications/doc_download/13-An-introduction-to- demand-responsive-transport-DRT, last accessed 9, Nov 2015.

Ahn, B. S. (2011). "Compatible weighting method with rank order centroid: Maximum entropy ordered weighted averaging approach." European Journal of Operational Research, 212(3), 552-559.

Al-Ayyash, Z., M. Abou-Zeid and I. Kaysi (2015). Forecasting Students’ Demand for a Shared-Ride Taxi Service Using an Integrated Choice and Latent Variable Model 2. Transportation Research Board 94th Annual Meeting.

Ambrosino, G., J. Nelson and M. Romanazzo (2004). Demand responsive transport services: Towards the flexible mobility agency, ENEA, Italian National Agency for New Technologies, Energy and the Environment.

Andrejszki, T. and Á. Török (2014). "Pricing systems analysis of DRT systems." Acta Technica Jaurinensis 7(2): 123-129.

Anspacher, D., A. J. Khattak and Y. Yim (2004). "Demand-Responsive Transit Shuttles: Who Will Use Them?" California Partners for Advanced Transit and Highways (PATH).

107

Arampatzis, G., C. T. Kiranoudis, P. Scaloubacas and D. Assimacopoulos. (2004). "A GIS-based decision support system for planning urban transportation policies." European Journal of Operational Research, 152(2), 465-475.

Arentze, T. A. and H. J. Timmermans (2004). "A learning-based transportation oriented simulation system." Transportation Research Part B: Methodological 38(7): 613-633.

Arentze, T., H. Timmermans and F. Hofman (2014). "Population synthesis for microsimulating travel behavior." Transportation Research Record: 85-91.

Ashford, N. J. and W. G. Bell (EDs.) (1978). Mobility for the Elderly and the Handicapped. Proceedings of the International Conference on Transport for the Elderly and Handicapped at Cambridge, April 4-7, 1978. Loughborough University of Technology, Chicago.

Auld, J., A. Mohammadian and K. Wies (2008). Population synthesis with control category optimization. 10th International Conference on Application of Advanced Technologies in Transportation. Athens, Greece.

Axhausen, K. W. and T. Gärling (1992). "Activity‐based approaches to travel analysis: conceptual frameworks, models, and research problems." Transport Reviews 12(4): 323-341.

Bakker, P. (1999). Large scale demand responsive transit systems-a local suburban transport solution for the next millennium. Public Transport Planning and Management. Proceedings of seminar E held at the AET European Transport Conference, Robinson College, Cambridge, UK, 27-29 September 1999, Volume P433.

Balcombe, R., R. Mackett, N. Paulley, J. Preston, J. Shires, H. Titheridge, M. Wardman and P. White (2004). "The Demand for Public Transport, TRL Report 593." Transport Research Laboratory, Crowthorne, Berkshire.

Bar-Gera, H., K. C. Konduri, B. Sana, X. Ye and R. M. Pendyala (2009). Estimating survey weights with multiple constraints using entropy optimization methods. Transportation Research Board 88th Annual Meeting.

108

Barzilai, J. (1997). "Deriving weights from pairwise comparison matrices". Journal of the operational research society, 48(12), 1226-1232.

Basnal, P., V. Singh, A. Shukla, D. Kumar and A. Kadam (2015). "Demand Responsive Transport." IJISET - International Journal of Innovative Science, Engineering & Technology, Vol. 2 Issue 4, April 2015.

Bearse, P., S. Gurmu, C. Rapaport and S. Stern (2004). "Paratransit demand of disabled people." Transportation Research Part B: Methodological 38(9): 809-831.

Beckman, R. J., K. A. Baggerly and M. D. McKay (1996). "Creating synthetic baseline populations." Transportation Research Part A: Policy and Practice 30(6): 415-429.

Ben-Akiva, M. E. and S. R. Lerman (1985). Discrete choice analysis: theory and application to travel demand (Vol. 9). MIT press.

Ben-Akiva, M., J. L. Bowman and D. Gopinath (1996). "Travel demand model system for the information era." Transportation 23(3): 241-266.

Benjamin, J., S. Kurauchi, T. Morikawa, A. Polydoropoulou, K. Sasaki and M. Ben- Akiva (1998). "Forecasting paratransit ridership using discrete choice models with explicit consideration of availability." Transportation Research Record: Journal of the Transportation Research Board(1618): 60-65.

Berbeglia, G., J. F. Cordeau and G. Laporte (2010). "Dynamic pickup and delivery problems." European Journal of Operational Research. 202, 8–15. doi:http://dx.doi.org/10.1016/j.ejor.2009.04.024.

Berger, P. A. and S. Hradil (1990). "Lebenslagen, Lebensläufe." Lebensstile. Göttingen.

Best, H. and M. Lanzendorf (2005). "Division of labour and gender differences in metropolitan car use: an empirical study in Cologne, Germany." Journal of Transport Geography 13(2): 109-121.

Black, A. (1995). Urban mass transportation planning. New York: McGraw-Hill.

109

Boarnet, M. G. and S. Sarmiento (1998). "Can land-use policy really affect travel behaviour? A study of the link between non-work travel and land-use characteristics." Urban Studies 35(7): 1155-1169.

Bose, J. and J. Sharp (2005). Measurement of travel behavior in a trip-based survey versus a time use survey–A comparative analysis of travel estimates using the 2001 National Household Travel Survey and the 2003 American Time Use Survey. In American Time Use Survey Early Results Conference.

Bowman, J. L. (2009). "Population synthesizers." Traffic Engineering and Control 49(9): 342.

Bradley, M. and J. Bowman (2006). SACSIM/05 activity-based travel forecasting model for SACOG: Featuring DAYSIM—the person day activity and travel simulator, Technical memo.

Brake, J., J. D. Nelson and S. Wright (2004). "Demand responsive transport: towards the emergence of a new market segment." Journal of Transport Geography 12(4): 323- 337.

Braun, M. and S. Winter (2009). "Ad hoc solution of the multicommodity-flow-over- time problem." Intelligent Transportation Systems, IEEE Transactions on, 10(4), 658- 667.

Carson, R. T., J. J. Louviere, D. A. Anderson, P. Arabie, D. S. Bunch, D. A. Hensher, R. M. Johnson, W. F. Kuhfeld, D. Steinberg and J. Swait (1994). "Experimental analysis of choice." Marketing Letters 5(4): 351-367.

Cervero, R. (1997). Paratransit in America: Redefining mass transportation, Greenwood Publishing Group.

Chin, S.-F. and A. Harding (2006). Regional dimensions: creating synthetic small- area microdata and spatial microsimulation models, National Centre for Social and Economic Modelling.

Choi, J., W. Do Lee, W. H. Park, C. Kim, K. Choi and C.-H. Joh (2014). "Analyzing changes in travel behavior in time and space using household travel surveys in Seoul Metropolitan Area over eight years." Travel Behaviour and Society 1(1): 3-14.

110

Ciari, F. (2009). Modeling collective taxis in a multi-agent traffic simulation framework, 9th Swiss Transport Research Conference, Monte Verità, Ascona, September 9.-11. 2009.

Clifton, K. J. and S. L. Handy (2001). Qualitative methods in travel behaviour research, Institute of Transportation Studies, University of California, Davis.

Contrino, H. and N. McGuckin (2009). "Demographics Matter Travel Demand, Options, and Characteristics Among Minority Populations." Public Works Management & Policy 13(4): 361-368.

Cooper, U., J. D. Nelson, S. Wright and J. A. Cooper (2008). "Achieving access to work, improved social inclusion and reducing rural access costs." In: Scottish Transport Applications and Research (STAR) Conference, 24 April 2010, Glasgow.

Cordeau, J. F. and G. Laporte (2003). "The dial-a-ride problem (DARP): Variants, modeling issues and algorithms." Q. J. Belgian, French and Italian Operations Research Societies. 1(2), 89–101.

Cordeau, J. F. and G. Laporte (2007). "The dial-a-ride problem: models and algorithms." Annals of Operations Research 153(1): 29-46.

Curtis, C. and T. Perkins (2006). "Travel behaviour: A review of recent literature." In: Working Paper no. 3, Department of Urban and Regional Planning, Curtin University, Perth. http://urbanet.curtin.edu.au/local/pdf/ARC_TOD_ Working_Paper_3.pdf, last accessed 9, Nov 2015.

Dagsvik, J. K. (2000). "Multinomial choice and selectivity."In: Discussion papers 264, Research Department of Statistics, Norway.

Damm, D. (1980). Interdependencies in activity behavior. In Transportation Research Record 750, TRB, National Research Council, Washington, D.C., 1980. pp. 33-40.

Data Management Group (1996). Transportation Tomorrow Survey, Joint Program in Transportation, University of Toronto.

Data Management Group (2011). Transportation Tomorrow Survey, Joint Program in Transportation, University of Toronto.

111

Davison, L., M. Enoch, T. Ryley, M. Quddus and C. Wang (2014). "A survey of demand responsive transport in Great Britain." Transport Policy 31: 47-54.

Department of Transport, England (2014). Annual bus statistics of England 2013/14, Statistical Release.

Dessouky, M., M. Rahimi and M. Weidner (2003). "Jointly optimizing cost, service, and environmental performance in demand-responsive transit scheduling." Transportation Research Part D: Transport and Environment. 8(6), 433–465. doi: 10.1016/S1361-9209(03)00043-9.

Dieleman, F. M., M. Dijst and G. Burghouwt (2002). "Urban form and travel behaviour: micro-level household attributes and residential context." Urban Studies 39(3): 507-527.

Dix, M. and A. Layzell (1986). Behaviour changes in changing populations: problems and proposals for policy-directed longitudinal studies.

DRT for DRT (2013). "‘DRT for DRT’, Developing Relevant Tools for Demand Responsive Transport, http://www.drtfordrt.org.uk/downloads/130601_newsletter.pdf, last accessed 9, Nov 2015.

Enoch, M. P., S. G. Ison, R. Laws and L. Zhang (2006). Evaluation study of demand responsive transport services in Wiltshire. Final report. Loughborough Loughborough University.

Evans, J. D. (1996). Straightforward statistics for the behavioral sciences. Brooks/Cole Pub. Co.

Fyhri, A. and R. Hjorthol (2009). "Children’s independent mobility to school, friends and leisure activities." Journal of Transport Geography 17(5): 377-384.

Gärling, T., D. Ettema, R. Gillholm and M. Selart (1996). "Towards a theory of the intention-behavior relationship with implications for the prediction of travel behavior." In: Hensher D, King J & Oum T (eds) World Transport Research: Proceedings of the 7th World Conference on Transport Research: Vol. 1. Travel Behaviour (pp. 231-240). Oxford: Pergamon.

112

Gärling, T., R. Gillholm and A. Gärling (1998). "Reintroducing attitude theory in travel behavior research: The validity of an interactive interview procedure to predict car use." Transportation 25(2): 129-146.

Giannopoulos, G. A. (2004). "The application of information and communication technologies in transport." European Journal of Operational Research 152(2): 302- 320.

Golob, T. F., A. D. Horowitz and M. Wachs (1977). Attitude-behavior relationships in travel demand modelling. In D. A. Hensher, & P. R. Stopher (Eds.), Behavioural travel modelling (pp. 739–757). London: Croom Helm.

Goodwin, P., R. Kitamura and H. Meurs (1990). "Some principles of dynamic analysis of travel behaviour." P.M. Jones (Ed.), New Development in Dynamic and Activity-Based Approaches to Travel Study, Gower Publishing, Aldershot, England (1990), pp. 56–72.

Habib, K., J. Swait and S. Salem (2012). Investigating structural changes in commuting mode choice preferences with repeated cross-sectional travel survey data: the contexts of Greater Toronto and Hamilton (GTHA) area. 13th International Conference on Travel Behaviour Research, Toronto.

Häme, L. (2013). Demand-Responsive Transport: Models and Algorithms. Ph. D. thesis, School of Science, Aalto University, Espoo, Finland.

Hensher, D. A. (1994). "Stated preference analysis of travel choices: the state of practice." Transportation 21(2): 107-133.

Hensher, D. A., J. M. Rose and W. H. Greene (2008). "Combining RP and SP data: biases in using the nested logit ‘trick’–contrasts with flexible mixed logit incorporating panel and scale effects." Journal of Transport Geography 16(2): 126- 133.

Hirsh, M., J. N. Prashkea and M. Ben-Akiva (1986). "Dynamic model of weekly activity pattern." Transportation Science 20(1): 24-36.

Hirsh, M., J. Prashker and M. Ben-Akiva (1984). Day of Week Models of Activity Patterns. Publication No. 84‐049, Transportation Research Institute, Technion‐Israel Institute of Technology, Haifa.

113

Horn, M. E. (2002). "Fleet scheduling and dispatching for demand-responsive passenger services." Transportation Research Part C: Emerging Technologies 10(1): 35-63.

Horowitz, J. (1991). "Modeling the choice of choice set in discrete-choice random- utility models." Environment and Planning A 23(9): 1237-1246.

Huynh, N. N., N. Shukla, A. Munoz Aneiros, V. L. Cao and P. Perez (2013). "A semi- deterministic approach for modelling of urban travel demand." International Symposium for Next Generation Infrastructure 2013 (ISNGI), pp. 1-8.

Huynh, N. N., V. Cao, R. Wickramasuriya Denagamage, M. Berryman, P. Perez and J. Barthelemy (2014). "An agent based model for the simulation of road traffic and transport demand in a Sydney metropolitan area." In Proceedings of the Eighth International Workshop on Agents in Traffic and Transportation, Paris, France, 5–9 May 2014; pp. 1–7.

Huynh, N., M.-R. Namazi-Rad, P. Perez, M. J. Berryman, Q. Chen and J. Barthelemy (2013). "Generating a synthetic population in support of agent-based modeling of transportation in Sydney."20th International Congress on Modelling and Simulation, December 2013, Adelaide, Australia.

Joewono, T. B. (2009). "Exploring the willingness and ability to pay for paratransit in Bandung, Indonesia." Journal of Public Transportation 12(2): 5.

Jordan, L. (1988). "Conjoint analysis modelling of stated preferences." Journal of Transport Economics and Policy 22(1): 93-119.

Kaltheier, R. M. (2002). Urban transport and poverty in developing countries: Analysis and options for transport policy and planning. Division 44 Environmental Management, Water, Energy, Transport. Deutsche Gesellschaft für Technische Zusammenarbeit (GTZ) GmbH. Eschborn.

Kattiyapornpong, U. and K. E. Miller (2006). Understanding travel behavior using demographic and socioeconomic variables as travel constraints. ANZMAC 2006: Advancing theory, maintaining relevance: Proceedings of the 2006 Australian & New Zealand Marketing Academy Conference, [Queensland University of Technology, School of Advertising, Marketing and Public Relations].

114

Kitamura, R. and T. Lam (1984). A model of constrained binary choice. Papers presented during the Ninth International Symposium on Transportation and Traffic Theory held in Delft the Netherlands, 11-13 July 1984.

Koffman, D. (2004). Operational experiences with flexible transit services. A Synthesis of Transit Practice. Transportation Research Board of the National Academies, Washington, D.C., 2004.

Koppelman, F. S. and P. K. Lyon (1981). "Attitudinal analysis of work/school travel." Transportation Science 15(3): 233-254.

Lave, R. and R. Mathias (2000). "State of the Art of Paratransit." Transportation in the New Millennium. State of the Art and Future Directions (CD-ROM), TRB, National Research Council, Washington, D.C., 2000.

Laws, R. (2009). Evaluating Publicly-Funded DRT Schemes in England and Wales. PhD Thesis, Loughborough University, Loughborough.

Lerman, S. R., M. Flusberg, W. M. Pecknold, R. E. Nestle and N. H. Wilson (1980). "A model system for forecasting patronage on demand responsive transportation systems." Transportation Research Part A: General 14(1): 13-23.

Levin, I. P. and J. J. Louviere (1981). Psychological contributions to travel demand modeling. In: I Altman & J Wohlwill (eds) Human Behavior and Environment (Vol 5 pp 29-61). New York: Plenum Press.

Lim, P. P. and D. Gargett (2013). Population Synthesis for Travel Demand Forecasting. 36thAustralasian Transport Research Forum (ATRF), 2013, Brisbane, Queensland, Australia.

Litman, T. (2013). "Understanding transport demands and elasticities." How prices and other factors affect travel behavior.(Victoria Transport Policy Institute: Litman) Available at http://www.vtpi.org/elasticities.pdf, last accessed 27, April 2016.

Louviere, J. (1988). Analyzing Individual Decision Making: Metric Conjoint Analysis, Sage university series on quantitative applications in the social sciences No. 67. Newbury Park, CA: Sage Publications, Inc.

115

Louviere, J. J. and D. A. Hensher (1983). "Using discrete choice models with experimental design data to forecast consumer demand for a unique cultural event." Journal of Consumer research10(3): 348-361.

Louviere, J. J. and G. Woodworth (1983). "Design and analysis of simulated consumer choice or allocation experiments: an approach based on aggregate data." Journal of marketing research 20 (November):350-367.

Louviere, J. J., D. A. Hensher and J. D. Swait (2000). Stated choice methods: analysis and applications, Cambridge University Press, Cambridge.

Louviere, J. J., D. Street, L. Burgess, N. Wasi, T. Islam and A. A. Marley (2008). "Modeling the choices of individual decision-makers by combining efficient choice experiment designs with extra preference information." Journal of Choice Modelling 1(1): 128-164.

Lu, X.-Y. and G. D. Gosling (2007). System modeling and implementation of the intermodal airport ground access planning tool. 11th World Conference on Transport Research.

Ma, T.-Y., C. Raux, E. Cornelis and I. Joly (2009). "Multistate Nonhomogeneous Semi-Markov Model of Daily Activity Type, Timing, and Duration Sequence." Transportation Research Record: Journal of the Transportation Research Board(2134): 123-134.

Maddern, C. and D. Jenner (2007). Telebus mobility and accessibility benefits: final report. 12th International Conference on Mobility and Transport for Elderly and Disabled transport (TRANSED). Hong Kong.

Mageean, J. and J. D. Nelson (2003). "The evaluation of demand responsive transport services in Europe." Journal of Transport Geography 11(4): 255-270.

Mahmoud, M. S., P. Eng and A. Shalaby (2015). Survey of Cross-Regional Intermodal Passenger Travel (SCRIPT): A Joint RP/SP Survey Integrated with a Multimodal Trip Planner Tool 2. Transportation Research Board 94th Annual Meeting.

McFadden, D. (1974). "The measurement of urban travel demand." Journal of Public Economics 3(4): 303-328.

116

McGuckin, N. and Y. Nakamoto (2004). "Differences in trip chaining by men and women. Research on Woman's Issues in Transportation: Report of a Conference, Vol. 2: Technical Papers." Transportation Research Board.

Moeckel, R., K. Spiekermann and M. Wegener (2003). Creating a synthetic population. Proceedings of the 8th International Conference on Computers in Urban Planning and Urban Management (CUPUM).

Müller, K., K. W. Axhausen, K. W. Axhausen and K. W. Axhausen (2010). Population synthesis for microsimulation: State of the art, ETH Zürich, Institut für Verkehrsplanung, Transporttechnik, Strassen-und Eisenbahnbau (IVT).

Nelson, J. D. and T. Phonphitakchai (2012). "An evaluation of the user characteristics of an open access DRT service." Research in Transportation Economics 34(1): 54-65.

NZ Transport Agency (2008). Planning, programming and funding manual (PPFM). NZTA. 728pp.

Olaru, D., N. Smith and J. Peachman (2005). Whereabouts from Monday to Sunday? Australasian Transport Research Forum (ATRF), 28th, 2005, Sydney, New South Wales, Australia.

Osman, A. O., K. M. N. Habib and A. Shalaby (2012). Modal Shift Forecasting Model for Transit Service Planning: Survey Instrument Design. In Proc., 12th International Conference on Advanced Systems for Public Transport.

Pas, E. (1990). "Is travel demand analysis and modelling in the doldrums." Developments in dynamic and activity-based approaches to travel analysis: 3-27.

Pas, E. I. (1985). "State of the art and research opportunities in travel demand: another perspective." Transportation Research Part A: General 19(5): 460-464.

Piatkowski, D. P. and W. E. Marshall (2015). "Not all prospective bicyclists are created equal: The role of attitudes, socio-demographics, and the built environment in bicycle commuting." Travel Behaviour and Society 2(3): 166-173.

Pinjari, A. R., N. Eluru, R. B. Copperman, I. N. Sener, J. Y. Guo, S. Srinivasan and C. R. Bhat (2006). Activity-based travel-demand analysis for metropolitan areas in Texas: CEMDAP Models, Framework, Software Architecture and Application Results.

117

Polk, M. (2003). "Are women potentially more accommodating than men to a sustainable transportation system in Sweden?" Transportation Research Part D: Transport and Environment 8(2): 75-95.

Polk, M. (2004). "The influence of gender on daily car use and on willingness to reduce car use in Sweden." Journal of Transport Geography 12(3): 185-195.

Poulenez-Donovan, C. J. and C. Ulberg (1994). "Seeing the trees and missing the forest: qualitative versus quantitative research findings in a model transportation demand management program evaluation." Transportation Research Record(1459).

Prud’homme, J., D. Josselin and J. Aryal (2011). Quantitative analysis of pollutant emissions in the context of demand responsive transport. Computational Science and Its Applications-ICCSA 2011, Springer: 439-453.

Public Transport Management Act (2008). The New Zealand Public Transport Management Act (PTMA 2008).

Rasouli, S. and H. Timmermans (2014). "Applications of theories and models of choice and decision-making under conditions of uncertainty in travel behavior research." Travel Behaviour and Society 1(3): 79-90.

Rayle, L., S. Shaheen, N. Chan, D. Dai and R. Cervero (2014). App-Based, On- Demand Ride Services: Comparing Taxi and Ridesourcing Trips and User Characteristics in San Francisco University of California Transportation Center (UCTC), UCTC-FR-2014-08.

Rodier, C. J., R. A. Johnston and D. R. Shabazian (1998). "Evaluation of advanced transit alternatives using consumer welfare." Transportation Research Part C: Emerging Technologies 6(1): 141-156.

Ronald, N., R. Thompson, J. Haasz and S. Winter (2013). Determining the Viability of a Demand-Responsive Transport System under Varying Demand Scenarios. Proceedings of the Sixth ACM SIGSPATIAL International Workshop on Computational Transportation Science. Orlando, FL, USA, ACM: 7-12.

Rosenbloom, S. (2004). Understanding Women’s and Men’s Travels Patters. Research on Woman’s Issues in Transportation. Report of a Conference. Volume 1.

118

Washington D.C.: Transportation Research Board of the National Academies. http://onlinepubs.trb.org/onlinepubs/conf/CP35v1.pdf, last accessed 27, April 2016.

Rosenbloom, S. and G. J. Fielding (1998). Transit markets of the future: the challenge of change. Transit Cooperative Research Program. National Research Council, Washington, DC.

Ryan, J., H. Maoh and P. Kanaroglou (2009). "Population synthesis: Comparing the major techniques using a small, complete population of firms." Geographical Analysis 41(2): 181-203.

Ryley, T. J. (2005). A study of individual travel behaviour in Edinburgh, to assess the propensity to use non-motorised modes, PhD thesis, Edinburgh Napier University.

Ryley, T. J. E., M. P. Quddus, M. A. Stanley, P. Zanni and A. M. (2013). Developing Relevant Tools for Demand Responsive Transport (DRT). ATCO Conference. Liverpool.

Salvini, P. and E. J. Miller (2005). "ILUTE: An operational prototype of a comprehensive microsimulation model of urban systems." Networks and Spatial Economics 5(2): 217-234.

Scheiner, J. and C. Holz-Rau (2007). "Travel mode choice: affected by objective or subjective determinants?" Transportation 34(4): 487-511.

Scott, R. and Booz & Company (2010). Demand responsive passenger transport in low-demand situations. Research Report, 425, NZ Transport Agency, Wellington.

SG Associates, Inc., L. Scott, Cleary, Inc., and C.M. Research, Inc. (1995). Workbook for Estimating Demand for Rural Passenger Transportation. Transit Cooperative Research Program (TCRP) Report 3, published by Transportation Research Board, Washington.

Shaheen, S. and C. Rodier (2004). Carsharing and Carfree Housing: predicted travel, emission, and economic benefits, paper presented at the the 83th Annual Meeting of the Transportation Research Board, Washington, DC, January.

Sihvola, T., L. Häme and R. Sulonen (2010). Passenger-pooling and tripcombining potential of high-density demand-responsive transport. TRB 89th Annual Meeting.

119

Silva, P., F. Antunes, R. Gomes and C. Bento (2015). Demand Modelling for Responsive Transport Systems Using Digital Footprints. Progress in Artificial Intelligence, Springer: 181-186.

Simma, A. and K. W. Axhausen (2001). "Structures of commitment in mode use: a comparison of Switzerland, Germany and Great Britain." Transport Policy 8(4): 279- 288.

Spielberg, F. and R. H. Pratt (2004). "Demand-Responsive/ADA-Traveler Response to Transportation System Changes." Chap. 6, Report 95, TCRP, Federal Transit Administration, Washington, DC (2004).

Srinivasan, S. and L. Ma (2009). Synthetic population generation: A heuristic data- fitting approach and validations. 12th International Conference on Travel Behaviour Research (IATBR), Jaipur.

Su, F. and M. G. Bell (2009). "Transport for older people: Characteristics and solutions." Research in Transportation Economics 25(1): 46-55.

Swait, J. and W. Adamowicz (2001). "Choice environment, market complexity, and consumer behavior: a theoretical and empirical approach for incorporating decision complexity into models of consumer choice." Organizational Behavior and Human Decision Processes 86(2): 141-167.

Takeuchi, R., I. Okura, F. Nakamura and H. Hiraishi (2003). Feasibility Study on Demand Responsive Transport systems (DRTS). 5th Eastern Asia Society for Transportation Studies Conference.

TCRP (1995). Workbook for Estimating Demand for Rural Passenger Transportation, Synthesis Report 3, Transit Cooperative Research Program (TCRP), Federal Transit Administration, Washington, DC.

TCRP (2004). Demand Responsive/ADA Traveller Response to Transportation System Changes, Synthesis Report 95, Transit Cooperative Research Program (TCRP), Federal Transit Administration, Washington, DC.

The Transport Data Centre (TDC), of Transport NSW (2010). 2008/09 Household Travel Survey Summary Report.

120

The Victorian Department of Transport (2011). Victorian Integrated Survey of Travel & Activity 2009-10, Survey Procedures and Documentation. Final Data Release v1.0.

Timmermans, H. and R. G. Golledge (1990). "Applications of behavioural research on spatial problems II: preference and choice." Progress in Human Geography 14(3): 311-354.

Train, K. E. (1979). "A comparison of the predictive ability of mode choice models with various levels of complexity." Transportation Research Part A: General 13(1): 11-16.

Tranter, P. and J. Whitelegg (1994). "Children's travel behaviours in Canberra: car- dependent lifestyles in a low-density city." Journal of Transport Geography 2(4): 265- 273. van Bladel, K., T. Bellemans, D. Janssens, G. Wets, L. Nijland, T. Arentze and H. TIMMERMANS (2008). Design of Stated Adaptation Experiments: Discussion of Some Issues and Experiences. Paper presented at the International Conference on Survey Methods in Transport: Harmonization and Data Comparability. Annecy, France, 25-31 May 2008.

Vance, C., S. Buchheim and E. Brockfeld (2005). Gender as a determinant of car use: evidence from Germany. In Research on Women's Issues in Transportation Conference Proceedings, Volume 2, The National Academies Press, 2005, pp. 59-67.

Vaughn, K. M., P. Speckman and D. Sun (1999). "Identifying relevant socio- demographics for distinguishing household activity-travel patterns: A multivariate regression tree approach." National Institute of Statistical Science (NISS) report, Research Triangle Park, North Carolina.

Wang, C., M. Quddus, M. Enoch, T. Ryley and L. Davison (2014). "Multilevel modelling of Demand Responsive Transport (DRT) trips in Greater Manchester based on area-wide socio-economic data." Transportation 41(3): 589-610.

Wang, C., M. Quddus, M. Enoch, T. Ryley and L. Davison (2015). "Exploring the propensity to travel by demand responsive transport in the rural area of Lincolnshire in England." Case Studies on Transport Policy 3(2): 129-136.

121

Wang, H. and S. Winter (2010). Utilizing taxi empty cruise time to solve the short distance trip problem. In: ITS World Congress, Busan, Korea.

Wardman, M. (1988). "A comparison of revealed preference and stated preference models of travel behaviour." Journal of Transport Economics and Policy XXII(1): 71- 92.

Wardman, M. (1991). "Stated preference methods and travel demand forecasting: an examination of the scale factor problem." Transportation Research Part A: General 25(2): 79-89.

Williamson, P., M. Birkin and P. H. Rees (1998). "The estimation of population microdata by using data from small area statistics and samples of anonymised records." Environment and Planning A 30(5): 785-816.

Wilson, A. G. and C. E. Pownall (1976). "A new representation of the urban system for modelling and for the study of micro-level interdependence." Area: 246-254.

Winter, S. and S. Nittel (2006). "Ad hoc shared‐ride trip planning by mobile geosensor networks." International Journal of Geographical Information Science 20(8): 899-916.

Yang, L., C. Choudhury, M. Ben-Akiva, J. Abreu e Silva and D. Carvalho (2009). Stated preference survey for new smart transport modes and services: Design, pilot study and new revision, MIT Portugal Transportation Systems, Working Paper Series, ITS-SCUSSE.

Ye, X., K. Konduri, R. M. Pendyala, B. Sana and P. Waddell (2009). A methodology to match distributions of both household and person attributes in the generation of synthetic populations. 88th Annual Meeting of the Transportation Research Board, Washington, DC.

Zografos, K. G., K. N. Androutsopoulos and T. Sihvola (2008). "A methodological approach for developing and assessing business models for flexible transport systems." Transportation 35(6): 777-795.

122

Minerva Access is the Institutional Repository of The University of Melbourne

Author/s: Jain, Shubham

Title: Exploring susceptibility to use demand responsive transport (DRT)

Date: 2016

Persistent Link: http://hdl.handle.net/11343/112495

File Description: Exploring Susceptibility to Use Demand Responsive Transport (DRT)