Journal of Transport Geography 76 (2019) 221–231

Contents lists available at ScienceDirect

Journal of Transport Geography

journal homepage: www.elsevier.com/locate/jtrangeo

Using walkability measures to identify train stations with the potential to T become transit oriented developments located in walkable neighbourhoods ⁎ Dana Jeffreya, Claire Boulangéb, Billie Giles-Cortib, Simon Washingtonc, Lucy Gunnb, a School of Design, The , Melbourne, 3010, VIC, b Centre for Urban Research, RMIT University, Melbourne, 3000, VIC, Australia c School of Civil Engineering, The University of Queensland, , 4072, QLD, Australia

ABSTRACT

Classifying train stations into typologies is a useful way to simplify their complex characteristics to assess their potential to become Transit Oriented Developments (TODs). However, researchers are yet to fully explore the walkability of train station neighbourhoods. Walkable areas have many features but typically include high residential density, greater street connectivity, and mixed land uses. Such characteristics facilitate access to the train station and can indicate both the effectiveness and potential of a train station to function as a TOD. This research explores the walkability of 230 train stations in metropolitan Melbourne, Australia using 14 different walkability measures. A two-stage cluster analysis was employed to group the train stations to determine their degree of walkability. Thetrainstation typology was validated using train station patronage data by different transport modes. Three clusters were found: cluster 1 train stations were more walkableand generally located in inner city areas, whilst those in cluster 2 were the least walkable and were generally located in middle and outer suburban areas. Cluster 3 train stations had the most potential for development as a TOD, having similar walkability features to those in cluster 1 but more car parking facilities and local living destinations when compared with the least walkable cluster 2 train stations. These findings suggest potential for cluster 3 train stations to become TODs, particularly if residential densities were increased in the surrounding neighbourhood. The patronage data validated the cluster findings in that the most walkable cluster 1train stations had the highest percentage of pedestrian entries. TODs offer a way for planners to increase public transit use by co-locating a variety of services, destinations, residences and places of employment. Our findings provide a typology useful for exploring development strategies for developing train stations into TODasameans of managing population growth and creating healthy, liveable and sustainable cities.

1. Introduction identify those train stations ripe for development and revitalization (Center for Transit-Oriented Development, 2010; Lyu et al., 2016; Transit Oriented Developments (TODs) co-locate high density re- Reusser et al., 2008; Zemp et al., 2011). To develop typologies, TODs sidential housing with public transit and a variety of destinations such are typically assessed using node and place measures using Bertolini's as retail shops, supermarkets, employment, health and community node-place model (Bertolini, 1999). This typology provides a frame- services, and are intended to encourage walking and produce commu- work for assessing the transit stop (‘the node’), the surrounding area nity benefits (Higgins & Kanaroglou, 2016; Kamruzzaman et al., 2014; (‘the place’), and the relationship between the two (Bertolini, 1999). Vale, 2015). TODs facilitate local living by allowing people to live, Measures such as transit frequencies and diversity of transit services work and play close to home with access to other destinations via measure the accessibility of the node, whilst built environment measures transit (, 2015; Commission on the Social such as street connectivity and employment density are used to assess Determinants of Health, 2008; Newman et al., 2015). TODs contribute the quality of the place (i.e., the neighbourhood). Under the node-place to and promote a range of co-benefits including reduced single occu- model, transit stops that balance node and place values are ‘balanced’ pant vehicle use, road congestion, and transportation costs (Renne, and have development potential because they match transit accessi- 2009; Stutzer & Frey, 2008), as well as improved air quality (Woodcock bility with demand, whilst unbalanced nodes require further interven- et al., 2009), enhanced social cohesion, (Giles-Corti et al., 2015) and tion to optimise TOD outcomes (Bertolini, 1999). active lifestyles conducive of improved health and well-being (Higgins Whilst the node-place model provides an understanding of the & Kanaroglou, 2016). transit stop itself, it overlooks the importance of walking accessibility Typologies for transit stops and the surrounding area are commonly between the node and the place (Lyu et al., 2016; Schlossberg & Brown, used to assist policymakers, planners and urban designers understand 2004). Without this additional information, a future development may which TODs have similar characteristics, enabling further analysis to fail to achieve its full potential as a TOD. This shortcoming is

⁎ Corresponding author. E-mail address: [email protected] (L. Gunn). https://doi.org/10.1016/j.jtrangeo.2019.03.009 Received 27 September 2018; Received in revised form 13 February 2019; Accepted 19 March 2019 0966-6923/ © 2019 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/BY-NC-ND/4.0/). D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231 epitomized in what is referred to as Transit Adjacent Developments et al. (2007), also incorporated the proportion of land use mix relating (TADs) where there is poor access between the transit stop and a to residential, recreational, commercial or industrial land in their study compact, mixed land use local neighbourhood (Kamruzzaman et al., using built environment features. Whilst Adams et al. (2011) used 2014; Renne, 2009). To overcome this limitation, emphasis on walking perceived built environment measures relating to residential density, accessibility or the walkability of the transit neighbourhood has been land use mix, street connectivity, walking and cycling facilities, aes- made (Schlossberg & Brown, 2004; Vale, 2015). thetics, safety, transit stops and proximity to parks and recreational The walkability of the neighbourhood surrounding a transit stop is a facilities from the Neighbourhood Environment Walkability Scale key aspect of its success, as it facilitates convenient, safe and direct (NEWS) in a latent class profile analysis of individuals and their access to and from the transit node (Jacobson & Forsyth, 2008; Vale, neighbourhood built environments. 2015), encouraging greater use of the transit stop and ensuring that Research typically finds that more walkable neighbourhoods and TOD residents have access to more local amenities to meet their local areas are associated with more walking (Cerin et al., 2007; Gunn et al., needs. Successful TODs maximise access by active modes. A walkable 2017b; McCormack et al., 2012). Furthermore, a common finding to neighbourhood can help achieve this, because it is characterised by date, is that more walkable areas are located in older more established well-connected streets, high residential densities, and proximate access areas, often located closer to the inner-city or downtown areas which to jobs, services and public transport (Giles-Corti et al., 2015). How- typically have more commercial zoning, retail destinations and higher ever, one key difference between a TOD and a walkable neighbourhood residential densities (Gunn et al., 2017b; McCormack et al., 2012). is that, although a walkable neighbourhood has access to transit, it is not necessarily centred on a transit stop. These urban environment 2.2. Typology literature relating to TODs and commonly used measures for features are often presented in the literature through Ewing and Cer- node and place vero's ‘5Ds’ framework which includes: Density, Diversity, Design, Destination accessibility and Distance to transit (Ewing & Cervero, Understanding the node and place features of transit nodes can re- 2010; Ewing et al., 2009; Giles-Corti et al., 2016). In combination, these veal their context-specific structure leading to improved selection, de- features promote walking and cycling and lead to health and commu- sign and development of TODs (Vale, 2015). TOD typologies derived nity benefits (Frank, 2005; Giles-Corti et al., 2015; Saelens et al., 2003). from cluster or latent class analysis assist in this process, as they sim- Despite the importance of neighbourhood walkability to TOD de- plify large datasets (based on individual transit nodes and their area- velopment, only a few researchers have explored how the walkability of based features) into a small set of clusters based on similar features. neighbourhoods surrounding transit stops could inform future TOD This distillation helps policymakers and planners understand the cur- development (Schlossberg & Brown, 2004; Vale, 2015). This lack of rent state of TODs relative to planning expectations, and can assist in attention may arise because a detailed examination and calculation of identifying transit stops and their surrounding areas that require con- complex built environment features surrounding transit nodes goes text-specific development (Higgins & Kanaroglou, 2016). beyond the broad definition of a TOD. A variety of node-place measures that capture the features of transit Existing TOD typologies to date have not fully accounted for this stops are typically used in the TOD literature. Kamruzzaman et al. complexity. Hence, in this paper we propose and describe a new ty- (2014) calculated six built environment measures for administrative pology that better accounts for the complexity of walkability of the areas based on 1734 Census Collector Districts (CCDs) in Brisbane, TOD's neighbourhood and its role in the TOD's success. This research Australia. These included net employment density, net residential seeks to develop and explore a train station typology based on walk- density, land use diversity, intersection density, cul-de-sac density and ability measures with the aim of providing refined and improved me- transport accessibility. Cluster analysis was then used to create a ty- trics that can contribute to identifying potential successful TOD de- pology resulting in four clusters. However, by using CCDs, this research velopments. We develop and present a train station typology analysis did not explicitly focus on train station precincts. using 14 objective walkability features measured within close proximity Other researchers have used similar measures for characterizing the to 230 train stations in metropolitan Melbourne, Australia. area surrounding train stations. Using data from Switzerland, Reusser et al. (2008) used population, the number of primary and tertiary sector 2. Literature review jobs, and land-use diversity within 700 m of 1684 train stations, and from questionnaires they identified number of full time jobs in educa- 2.1. Typology literature relating to walkability tion, distance from town centre and the presence of grocery, restau- rants, pharmacies and florists as being important measures of place. Walkability within active living research primarily focuses on the Similarly, Zemp et al. (2011) used population and number of jobs area surrounding people's homes (Leal & Chaix, 2011) with only a few within 700 m of 1700 train stations. Atkinson-Palombo & Kuby (2011) researchers considering other important locations such as workplaces, used measures based on parcels of residential, vacant, TOD-compatible neighbourhood/retail activity-centres (Gunn et al., 2017b) or transit and TOD-incompatible land and measures relating to people which stops (Schlossberg & Brown, 2004; Vale, 2015). A walkable environ- included jobs, population density, household income, percentage of ment increases access to these high use locations without the need for a owner-occupied housing units, and percentage of people with a ba- motor vehicle (Giles-Corti et al., 2016; Higgins & Kanaroglou, 2016; chelor degree for their study of TOD potential conducted in Phoenix, Sallis et al., 2016) leading to numerous health, environmental and Arizona. These authors also included various node measures for transit economic co-benefits. frequency and accessibility, which were combined using factor analyses In seeking to understand the features in such environments, active prior to undertaking cluster analysis. Higgins & Kanaroglou (2016) used living researchers have used cluster and latent profile analysis to derive density, street connectivity, land use mix, development mix and a typologies. Both methods create homogenous groups (or clusters) based gravity measure of station interaction-potential based on population, on a set of built environment measures that help characterize the areas employment and station-to-station transit time for exploring TOD of interest. Generally these studies use objective measures such as ped- typologies on 372 train stations in Toronto, Canada. They measured sheds ratios, population density, land use mix, intersection density, cul- connectivity using ped-shed ratios based on 10 min of walking. Instead de-sac density, proximity to services, ratio of commercial area to total of using factor analysis and prior to undertaking the latent profile area, and various other density measures including density of busi- analysis, they used a latent class structure based on the two most cor- nesses, bus stops, park types, recreational destinations, sidewalks, and related input variables of density and interaction potential. Despite cycle-ways (Charreire et al., 2012; Higgins & Kanaroglou, 2016; arguing for the incorporation of measures relating to the walkability of McCormack et al., 2012; McNally & Kulkarni, 1997). The work of Cerin an area, in a study in Lisbon, Portugal, Vale (2015) only included a ped-

222 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231 shed ratio, measured as the area in a Euclidean buffer to that from a mode choice behaviour according to their four cluster TOD solution. street network buffer, as an add-on to another 13 measures used inhis Using a multinomial regression with mode choice as an outcome node-place assessment and cluster analysis of train and ferry stations. measure and the cluster solution as the exposure, they found that In contrast to these studies using cluster or latent class analysis, people living in non-TOD areas were less likely to use active modes of Schlossberg & Brown (2004) focused explicitly on the walkability of 11 transport. train stations in Portland, Oregon using six place-based measures di- In this study, the focus was to explore the typology of metropolitan chotomized into high and low categories measured at a quarter and half Melbourne's train stations based on measures of walkability. We built mile (400 m and 800 m) distance from each train station in their on previous research, and used a comprehensive and established set of comparison of TOD sites. These included: quantity of accessible paths walkability measures based on Cervero and Ewing's 5Ds framework measured as the length of minor roads, quantity of impedance paths (Cervero & Kockelman, 1997; Ewing & Cervero, 2001) that have been measured as the length of major roads, pedestrian catchment areas (i.e., used widely in public health literature (Saelens & Handy, 2008). The ped-sheds), impedance pedestrian catchment areas (i.e., ped-shed based specific measures chosen for this study have been found to beasso- on area accessible once high volume, high speed roads are removed), ciated with walking and wellbeing outcomes (Gunn et al., 2017a; Gunn and density of dead ends. Their analysis ranked the train stations on et al., 2017b; King et al., 2015, Davern et al. 2018). In validating our each of these measures and grouped the three best and worst train cluster solution and typology, we followed the approach of Higgins & stations. The remainder of their analysis explores the walkability of two Kanaroglou (2016) and measured the performance of the train stations station areas as case studies using maps of the built environment in metropolitan Melbourne separately to the derivation of TOD typol- variables. A major finding was that evaluating and understanding the ogies, using train station patronage data. We used the percentage of node-place structure and how it relates to the transit-node, is key for train station entries made by transport mode (Higgins & Kanaroglou, facilitating the location of TODs. 2016). Lyu et al. (2016) included a third aspect to the node-place model relating to orientation, which was used to explore the interrelation 3. Data and methods between the node and place. They explained that the area surrounding transit stations may not have adequate street connectivity to support 3.1. Study area walking or cycling or may have densities situated adjacent to the station instead of toward it, resulting in a TAD. Lyu et al. (2016) referred to The metropolitan train network in Melbourne, Australia was the these area-based aspects surrounding the transit node as “functional case study for this research. and morphological interrelations,” which have the potential to facilitate Melbourne is a large sprawling city of approximately 10,000 km2, transit use but also increase the utility and value of the area sur- with a hub and spoke train network, and one of the largest tram net- rounding the train station leading to improved transit services ac- works in the world. Nevertheless, access to transit is limited because cording to the land use feedback cycle (Wegener, 2004; Wegener & density overall is relatively low at 13 dwellings per hectare (Arundel Fuerst, 1999). Measures included in the ‘orientation’ aspect include et al., 2017). Melbourne's population is projected to reach 8 million by average block size, average distance from station to jobs, average dis- 2050. This rapid growth will require an additional 1.6 million re- tance from station to residents, length of paved footpath per acre, in- sidential dwellings to be built in the next 32 years (State Government of tersection density, and the walk score – features which are also related , 2017a; State Government of Victoria, 2017b). Until recently, to the walkability of an area. Although using different terminology, Melbourne, like most Australian cities, has favoured urban sprawl as a Vale (2015) and Lyu et al. (2016) have essentially articulated the need development model (Newman & Kenworthy, 1999), resulting in high to include the walkability of the TOD's neighbourhood into the node- rates of car dependency, poor health outcomes and social inequities place model (Bertolini, 1999; Vale, 2015). (Dodson & Sipe, 2008). However, recent state government planning has recognised the need to accommodate future population growth in es- 2.3. TOD performance and validation tablished areas including the development of TODs (State Government of Victoria, 2017a; State Government of Victoria, 2017b). Node-place measures can be categorized as input and outcome Melbourne's train network currently consists of 17 metropolitan measures. Input measures relate to the node and place explicitly whilst train lines (with 230 train stations), which converge on the central outcome measures relate to TOD use. TOD performance relates to business district located at the centre of the city via an underground whether a TOD is doing well on the outcome measures (Renne, 2009). loop that supports three underground stations. The hub and spoke In the literature, there is some confusion on whether outcome measures network form means each of the train lines converge in the city. To should be included in node-place indexes or within the creation of TOD date, there are no train lines that traverse the city and distances be- typologies. Indeed, including outcome measures in typology analyses tween train lines increase in the middle and outer city areas making may strengthen the distinction between well performing and balanced access to train lines difficult. A new metro connecting the wes- TODs compared to those that are not. tern and south-eastern sections of the city is currently under construc- For example, Reusser et al. (2008) used passenger frequency and tion and a has recently been proposed by the current vehicle miles travelled in their node-place index. However, Renne state government (State Government of Victoria, 2018). The tunnel will (2009) and Zemp et al. (2011) argued that such measures are actually facilitate an uplift of services across the network by 2025. Until re- outcome measures that arise from the station inputs. The distinction cently, Melbourne's train network has been predominantly above- between these measures is important as it is the outcome measures that ground, inter-connecting with the road network across the city. This indicate whether a TOD is performing well as a transit stop. network configuration decreases efficient travel for road and rail users. Instead of directly incorporating outcome measures, several authors To mitigate and alleviate road congestion, in 2017–2018, 50 rail use them to validate their TOD typologies. Renne (2009), for example, crossings were removed to streamline rail and road transit (State argued that TOD performance can be validated in various ways ac- Government of Victoria, 2017a; State Government of Victoria, 2017b), cording to a framework based on: (1) travel behaviour; (2) the with sky-rails being constructed on one of the most heavily used lines economy; (3) the natural environment; (4) the built environment; (5) (i.e., Cranbourne-Pakenham). the social environment; (6) the policy context; and by comparing (a) The analysis presented here serves as a baseline measure for the TODs to other TODs; (b) TODs with non-TODs; or (c) TODs against walkability of the train stations on Melbourne's rail network. It offers a regional averages. Kamruzzaman et al. (2014) used this approach when novel and unique opportunity to explore both the walkability of areas they validated their typology from a travel behaviour perspective using surrounding train stations in a generally low density sprawling city and

223 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231 the implications this might have for future train station planning, de- 3.4. Statistical analysis velopment and policymaking. Given the timing of data collection, it does not take into account if, and how, removal of level crossings or Summary statistics for the 14 walkability measures were computed sky-rail sections might have affected the walkability of the relevant and a two-stage cluster analysis of Melbourne's train stations using train stations neighbourhoods. standardized built environment variables performed. Cluster analysis was chosen ahead of latent profile analysis because it is an intuitive and well-established method for processing large datasets quickly. Pearson 3.2. Data and methods correlations were calculated prior to conducting the cluster analysis to check for multicollinearity amongst the walkability measures. Two types of data were used in this study: (1) spatial data on which First, hierarchical cluster analysis was employed to help determine walkability measures were derived using Geographic Information the number of clusters. Hierarchical clustering typically starts with Systems (GIS) analysis; and (2) train station patronage data by mode single objects which are combined to create clusters based on similarity. representing annual counts of train station entries. Clusters are then combined until one single cluster is achieved. The The spatial data were acquired from a variety of government and appropriate number of clusters is determined using the distance mea- non-government agencies including: sure between cluster formations from the agglomeration schedule where a large jump in the agglomeration schedule is indicative of the • Public Transport Victoria: public transport data (open-access data- appropriate number of clusters. In undertaking hierarchical cluster sets)(Public Transport Victoria, 2014) analysis several linking algorithms exist and were used here including • VicMap: street network, parcel boundaries and zoning data from Ward's, Complete and Average Methods. Each method links objects in (open-access datasets) (State Government of Victoria, 2012a; State different ways and the three algorithms were used to mitigate linking Government of Victoria, 2012b; State Government of Victoria, problems that can occur because of each linking method. Nevertheless, 2012c) Complete and Average linkage methods in particular have been cited as • Pitney Bowes: business points data (commercial dataset) (Pitney being appropriate for many styles of data (Hair Jr et al., 2014). Bowes Ltd, 2014) In the second stage, the K-means clustering algorithm was used as it is more robust to outliers and irrelevant clustering variables (Hair Jr GIS-techniques in ArcGIS and analytical tools in ArcMap were used et al., 2014). To validate the number of clusters, the Calinksi-Harabasz to derive the walkability measures. A geodatabase was deployed in pseudo F statistic was computed. Following the cluster analysis, Tukey ArcMap to store and organise all spatial layers and tables of data col- post-hoc tests were performed to determine statistically significant lected for this study. differences between built environment measures by cluster. The train station patronage data was sourced from the Public Transport Victoria (PTV) open dataset “Station patronage data for 3.5. Validating the cluster solution Melbourne's rail network”. Each station was surveyed by PTV in 2012 for one weekday (excluding Friday) to represent an average day across The final cluster solution was mapped using GIS and a desk-top the network. The patronage data provides the count and proportion of audit using Google maps was undertaken to validate the cluster analysis train station entries made by mode (Public Transport Victoria, 2015). solution to local knowledge of the characteristics of Melbourne's train stations and urban form. A final validation of the cluster solution was undertaken using pa- 3.3. Walkability measures tronage data by mode and includes both descriptive and visual analyses presented as the percentage of train station entries made by walking Fourteen walkability measures were derived and categorized ac- overlaid onto the GIS map of the final cluster solution. cording to the 5Ds framework for each buffer and train station Statistical analyses were performed in R and final GIS maps were (Table 1). As discussed earlier most design, diversity and density related produced using the R Leaflet package. GIS built environment data measures have been used in previous typology research. However, some manipulation was undertaken in ESRI ArcGIS Desktop version 10.3.1. measures, including car parking and bike facilities, whilst considered important in the literature, are often not included owing to data 4. Results availability issues (Chorus & Bertolini, 2011; Griffiths & Curtis, 2017). Inclusion of these variables is important because it maximises the po- 4.1. Cluster analysis tential of using transit for those living further away. In this study, we also included walkability measures relating to destination accessibility Table 2 presents summary statistics and the three cluster solution that have been found to be associated with walking (Gunn et al., 2017a; for the built environment measures surrounding Melbourne's train Gunn et al., 2017b; King et al., 2015) and wellbeing outcomes (Davern, stations chosen based on Tukey post-hoc tests, the Calinksi-Harabasz Gunn et al. 2018). These measures are related to the concept of ‘local pseudo F statistic, and the overall interpretability of the results. Pearson living’ which ensures access to essential destinations allowing people to correlations revealed no issues with multicollinearity. live within their immediate neighbourhood (Badland et al., 2017). Overall, three different clusters of train stations were identified. Living close to essential destinations and amenity is important in Compared with other clusters, cluster 1 train stations had the highest shifting travel behaviour toward more active modes and for alleviating values for gross dwelling density, community resources, educational road congestion. Locating TODS in walkable neighbourhoods would facilities, convenience stores, entertainment venues, destination di- maximise use of the TOD and its amenities by local residents; and versity, and number of tram stops. These high values suggest that the provides TOD residents with access to more local amenities required for area surrounding these train stations is highly walkable. Cluster 1 train daily living. stations had similar values to cluster 3 train stations for the connected Prior to analysis all freeways were removed from the road network node and ped-shed ratios, and access to sporting facilitates. Whilst there spatial layer, as they are inaccessible to pedestrians. Areas around train are a high number of bus stops in the area surrounding cluster 1 train stations were defined using an 800 m road network buffer. This distance stations, there were fewer bus stops than the areas surrounding cluster was chosen as it has been associated with transport walking measures 3 train stations. The number of bike facilities were also low, although such as walking trips and sufficient physical activity and has been used similar in value to cluster 2 train stations. Finally, cluster 1 train sta- in other typology analyses (Gunn et al., 2017b; Vale, 2015). tions had the lowest amount of car parking in the area surrounding the

224 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231

Table 1 Built environment measures calculated within 800 m of a train station.

Measures Definition Data source GIS methods

Design Connected node The connected node ratio was calculated using the following Intersection data were derived from the The join command was used to count the ratio formula: VicMap Transport road centrelines (State number of intersection points in each buffer. x Government of Victoria, 2012a). y Where x is the number of three or more ways intersection and y is the total number of intersections.

Ped-shed ratio Ratio of area within 800 m street network buffer to the area Street network data were sourced from The ArcGIS Network Analyst extension within 800 m Euclidean buffer. the VicMap Transport road centrelines was used to define a 1600 m road network (State Government of Victoria, 2012a). buffer (i.e., areas that encompass all accessible streets within specified cut-off distances) created around each train station.

Diversity Land use mix Calculated as an entropy measure based: A customised land use layer was created The land use mix measure was calculated n LUM = 1( i=1 piln( p i ))/ln( n ) where parcel boundaries (State using the entropy formula, adapted from Where i is an index for land use category, n is the total number of Government of Victoria, 2012b), zoning Frank et al. (Frank et al., 2005)(Frank et al., land uses included, and p is the area of each land use i. Five land data (State Government of Victoria, 2005) and is the same as the one used by use categories were included in the calculation: residential, 2012c) and geocoded business points data Christian and colleagues (Christian et al., retail, commercial, industrial and “other”. (Pitney Bowes Ltd, 2014) were combined. 2011). The parcel boundaries data and zoning The intersect command was used to data were readily available nationally calculate the area of each land use type while the business points were available within each buffer. on a fee-for-service basis.

Car parking Land area dedicated to car parking (m2). Car parking data were sourced from The join command was used to measure Public Transport Victoria data (Public the land area dedicated to car parking in Transport Victoria, 2012). each buffer.

Density Gross dwelling Number of commercial dwellings + number of residential Dwelling count data were sourced from The join command was used to sum the density dwellings ÷ commercial and residential area the meshblock-level census data number of dwellings in each buffer. The dwelling density measure is the same as used by Badland (Australian Bureau of Statistics, 2011). et al. (2017).

Destination accessibility Community Count of all the libraries and post offices. Destinations data were sourced from The join command was used to retrieve the resources geocoded business points data (Pitney destination points within the boundaries of Education Count of all the primary and secondary schools, childcare, Bowes Ltd, 2014). each buffer. kindergartens and creches. Summary statistics were used to compute Convenience Count of all the supermarkets, specialty food stores, the number of destinations in each buffer. stores fishmongers, butchers, poultry stores, fruit and vegetable shops, milk bars, petrol stations and news-agencies. Entertainment Count of all the cinemas, theatres, art galleries, gaming venues. venues Sporting facilities Count of all the tennis courts, swimming pools, gyms, ovals and fields. Destination Sum of the presence/absence of each of the following 10 diversity destination categories in a buffer: 1. Childcare: Public primary and secondary schools, kindergartens and crèches 2. Specialty food stores: Fruit and vegetable grocers, fish stores, poultry stores and butchers 3. Supermarket 4. Convenience store: Convenience stores, petrol stations and newsagents 5. Transport: Bus and Tram stops 6. Pharmacy 7. Medical: General practitioners, maternal health clinics, community health centres 8. Dentist 9. Library 10. Post Office

(continued on next page)

225 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231

Table 1 (continued)

Measures Definition Data source GIS methods

Transit accessibilitya,b Bus stops Count of all the bus stops present in a buffer. Bus stops data were sourced from Public The join command was used to retrieve Transport Victoria data (Public Transport the number of bus stop, tram stop and bike Victoria, 2014). infrastructure points within the boundaries of each buffer.

Tram stops Count of all the tram stops present in a buffer. Tram stops data were sourced from Public Summary statistics were used to compute Transport Victoria data (Public Transport the number of services in each buffer. Victoria, 2014).

Bike facilities Sum of the presence/absence of bike related infrastructure Bike infrastructure data were sourced including bike hoops, lockers, sheds and parkiteersb. from Public Transport Victoria data (Public Transport Victoria, 2014).

a Transit accessibility is used in place of distance to transit. b Bike parkiteers are large sheds secured through access cards capable of storing many bikes. train station, which is commensurate with the high walkability of the clusters for the land use mix entropy measure, with similar values found surrounding area. for each cluster of train stations. Land use mix is generally found to be Cluster 2 train stations had the lowest values of all clusters for the associated with walking behaviour (Christian et al., 2011). However, in connected node and ped-shed ratios, gross dwelling density, sporting the context of train station neighbourhoods there did not appear to be facilities, destination diversity and bus stops. The values for community sufficient variation in land use mix across any of the train station resources, education, convenience stores, entertainment venues, tram neighbourhoods. All three clusters differed statistically in their amount stops and bike facilities were also lower than those found for cluster 1 of car parking. train stations, but were not statistically different to those for cluster 3 train stations. The amount of car parking for cluster 2 train stations was between that of the other clusters. These findings suggest cluster 2 train 4.2. Location of train station clusters stations are located in low walkable neighbourhoods. The walkability attribute values of cluster 3 train stations were low Fig. 1 shows the location of the different train stations clusters. This and similar to cluster 2 stations for community resources, education, map illustrates that cluster 1 train stations - typically located in high convenience stores, entertainment venues and tram stops. However, walkable neighbourhoods - are predominantly located in the inner and cluster 3 stations had the highest values for bike facilities and bus stops middle suburbs. Cluster 3 train stations - found in neighbourhoods with and had the highest values for car parking. The gross dwelling density some but not all walkable features - are typically located in the middle and destination diversity values for cluster 3 train stations were be- suburbs. Whilst cluster 2 train stations are found in low walkable tween the other clusters. neighbourhoods and are typically located in the outer suburbs and re- Notably, there were no statistical differences between train station gional areas of metropolitan Melbourne.

Table 2 Descriptive statistics for the three cluster solution for built environment measures around train stations.

All train stations Cluster 1 train stations Cluster 2 train stations Cluster 3 train stations

n = 230 n = 91 n = 53 n = 86

MEAN SD MEAN SD MEAN SD MEAN SD

Design Connected node ratio 0.76 0.16 0.79a 0.10 0.61a,c 0.22 0.81c 0.10 Ped-shed ratio 0.48 0.12 0.52a 0.09 0.32a,c 0.10 0.52c 0.08

Diversity Land use mix 0.39 0.15 0.38 0.16 0.37 0.13 0.42 0.14 Car parking 8250.98 7798.46 4429.06a,b,c 4857.08 7358.12a,b,c 7733.10 12,845.36a,b,c 8049.25

Density Gross dwelling density 17.69 9.34 23.23a,b,c 10.96 10.61a,b,c 5.28 16.19a,b,c 4.96

Destination accessibility Community resources 1.64 1.54 2.74a,b 1.45 0.81a 0.98 0.99b 1.19 Education 3.47 2.85 5.23a,b 3.12 2.34a 1.88 2.29b 1.98 Convenience stores 7.14 8.23 12.23a,b 10.23 3.45a 4.44 4.04b 3.66 Entertainment venues 0.44 1.76 0.88a,b 2.65 0.04a 0.19 0.22b 0.71 Sporting facilities 2.27 2.60 2.82a 2.45 1.02a,c 2.04 2.44c 2.82 Destination diversity 6.97 2.73 9.03a,b,c 1.05 4.81a,b,c 2.86 6.12a,b,c 2.38

Transit accessibility1 Bus stops 12.50 8.56 12.51a,b,c 8.19 5.45a,b,c 4.84 16.83a,b,c 7.88 Tram stops 3.53 7.63 7.50a,b 10.39 0.59a 2.48 1.15b 3.29 Bike facilities 0.97 0.77 0.75b 0.71 0.55c 0.57 1.4b,c 0.66

SD = Standard Deviation. Superscript letters indicate significant differences between clusters at a 5% significance level. Specifically:a shows significant differences between clusters 2 and 1; b shows significant differences between clusters 3 and 1; c shows significant differences between clusters3and2. 1 Transit accessibility is used in place of distance to transit.

226 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231

Fig. 1. Map of metropolitan Melbourne train stations displayed by cluster type.

4.3. Train station entry patterns – validation of the cluster solution cluster 2, and 43% for cluster 3). Table 3 also shows that train station entries to cluster 1 train sta- Table 3 shows the percentage of train station entries by transport tions made by either train and tram were higher at 16% and 8% re- mode and cluster. Notably, 82% of train station entries were made by spectively than the other two clusters where the percentages were re- active transport modes: 56% by walking, 25% by public transport; and latively small being less than 5%. Only 3% of train station entries to 0.68% by cycling. cluster 1 train stations were made by bus. The higher percentages of Cluster 1 train stations located in highly walkable neighbourhoods train and tram entries compared to bus entries for cluster 1 train sta- were accessed by the highest percentage of pedestrians (64%), whilst tions is unsurprising as the tram network is predominantly based in the the remaining clusters were accessed by fewer but also high percen- inner-city, and train lines become more disperse with limited service tages of pedestrians relative to other active transport modes (41% for variety in middle and outer areas negating this mode as a connecting service and entry point. Supporting this finding is the jump in train station entries made by bus for cluster 2 and cluster 3 train stations at Table 3 11% and 16% respectively. Notably, the percentage of train station Percentage of train station entries made by transport mode by cluster. entries made by cycling was relatively small for all clusters, however Cluster Transport mode the percentages were slightly higher for cluster 2 and 3 train stations at 0.8% and 1.0% respectively when compared with those made to cluster Walk Cycle Bus Train Tram Active Private Motor 1 train stations (0.5%). Transport Vehicle Overall, 18% of all train station entries were made by private motor 1 63.94 0.54 3.19 16.09 8.43 28.25 7.58 vehicle. However only 8% of drivers accessed cluster 1 train stations, 2 40.53 0.77 10.58 1.90 0.39 13.64 45.47 with the largest percentage made to cluster 2 train stations (45%) fol- 3 42.90 0.96 15.80 3.16 0.96 20.95 35.45 lowed by cluster 3 train stations (35%). Total 56.29 0.68 7.44 11.40 5.73 81.56 18.06 Fig. 2 Panels (a) – (d) overlay the cluster solution with the

227 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231

Fig. 2. Map of metropolitan Melbourne train stations displayed by cluster type with patronage data over-layed. percentage of train station entries made by walking, cycling, bus and by with the 25 dwellings per hectare recommended to support local private motor vehicle respectively. Similar to the location of cluster 1 walking (Boulangé et al., 2017). As a result, these stations were located and cluster 3 train stations, the proportion of train station entries made in neighbourhoods that encouraged patrons to walk which was re- by walking shown in panel (a) is generally highest for train stations in flected in the patronage by mode results that showed most peopleac- the inner and middle suburbs, and declines by distance from the centre cessing cluster 1 train stations by active transport, and in particular of Melbourne. The walkability measures for Cluster 1 train stations walking (64%). indicate that these train stations were located in highly walkable Panel (b) shows the proportion of train station entries made by neighbourhoods. Indeed, neighbourhoods surrounding cluster 1 train cycling, although there is no clear pattern, the map shows that some stations had an average of 23 dwellings per hectare, which is consistent train stations located in the middle and outer areas of the city have a

228 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231 high proportion of train station entries by cycling. Panel (c) shows low bike facilities and a low proportion of train station entries made by proportions of train station entries made by bus, although there appear cycling. Indeed, better integration of active transport networks and to be some stations in the middle and outer suburbs in the south east of provision of storage and changing facilities within the train station the city that have slightly higher proportions of train station entries could improve access by active transport and in particular entries by made by bus. Panel (d) shows train station entries made by private cyclists. motor vehicle, which are generally low in the inner-city areas, and high For cluster 3 train stations investment in on-road cycling infra- for middle and outer suburban train stations. This reflects the walk- structure could also potentially reduce the need for nearby car parking, ability findings for cluster 2 train stations, which were generally located the space for which could be re-purposed into mixed land use devel- in low walkable neighbourhoods with poor pedestrian access. This was opments that increase gross dwelling density, the number of destina- evidenced by the poor connected node and ped-shed ratios, the lowest tions and employment density of these precincts, further increasing gross dwelling density of all clusters (10 dwellings per hectare), low their walkability and potential to become fully fledged TODs. However, destination counts and diversity, and the lowest quantity of bus stops. for cluster 2 train stations a different strategy may be required since Most entries in cluster 2 train stations were made by private motor they were located in the least walkable neighbourhoods and did not vehicle (45%), even though there was less car parking available at these appear to be designed to maximise accessibility. They were poorly stations than around cluster 3 train stations. served by public transport (i.e., bus and tram stops); and had less car Cluster 3 train stations were located in moderately walkable parking available than cluster 3 train stations. For these train stations, neighbourhoods. These train stations had some neighbourhood walk- providing a greater variety and frequency of inter-connected public ability attributes similar to those found in cluster 1, in particular the transport services may improve the accessibility and patronage of connected node and walkable ped-shed ratios. However, cluster 3 train cluster 2 train stations, even if they are not accessed by active modes of station neighbourhoods differed significantly in the quantity and walking and cycling. Similarly, for those living further away, cluster 2 variety of local destinations available. This finding is not surprising, train stations may benefit from having larger car parking facilities or because the average gross dwelling density of neighbourhoods sur- park and ride facilities available to maximise the number of people who rounding cluster 3 stations was considerably lower at 16.2 dwellings could make use of train travel (Olaru et al., 2014). per hectare. Of the three clusters, cluster 3 train stations had the most Research of this type is ideally positioned to inform the identifica- car parking and bike facilities. The patronage results indicated that to tion of context-specific strategies for enhancing the accessibility ofin- get to these train stations 43% of people walked, 35% used a private dividual train stations. Moreover, the results could be used to guide motor vehicle, 16% used buses, and only 1% cycled. train station development in established and new areas. However, some built environment features are more difficult to upgrade than others, 5. Discussion and are best planned in the site design phase, to ensure that train sta- tion areas are located in walkable neighbourhoods from the outset. This This research derived a typology for evaluating access to me- would make it easier for subsequent development to convert train sta- tropolitan Melbourne's train stations based upon a cluster analysis tions into TOD at a later stage (Messenger & Ewing, 1996). methodology using 14 walkability measures. In deriving this typology, In this study, we found that train stations in the middle and outer we sought to understand what constitutes a walkable train station areas may benefit from the provision of on-road and end-of-trip cycling neighbourhood and to identify possible interventions and redevelop- facilities that would facilitate access to train stations from distances ment that could create more walkable neighbourhoods surrounding further away. However, we identified fewer options for supporting established and new train stations with the aim of identifying effective access by walking for train stations built in low walkable areas, and TOD location. We identified and defined three types of train stations, without massive redevelopment and investment, it is unlikely that these which we validated using Google maps and with train station patronage train stations could ever become well-functioning TODs (Vale, 2015). data by transport mode. With few exceptions, the three clusters were To increase the patronage of these stations, different strategies are re- generally located in inner, middle and outer areas of metropolitan quired including enhancing the cycling facilities, providing more inter- Melbourne. The degree of walkability declined with increasing distance connected transit services as well as providing more car parking facil- from the Central Business District. ities. However, preferably, future train stations will not be built in low Generally, cluster 1 train stations were located in the most walkable walkable neighbourhoods. neighbourhoods and as a consequence, were accessed by approximately Our findings, can be used to mitigate low walkability by using de- 30% more pedestrians. Access by cyclists was the lowest of all the velopment strategies informed by the train station typology and based clusters, and notably these stations had few bike facilities. For Cluster 1 on the features of the most walkable train station neighbourhoods. This train stations located in inner cities areas, the on-road cycling infra- research supports a revision to the old adage, ‘if we build it, they will structure may have already existed, and for people who cycle, the come’ to perhaps — “if we build it with the right features, they will distances to key destinations are close enough to avoid transit use al- come’. Train stations designed to encourage walking and cycling, re- together. However, for other cluster 1 train stations located in suburban quire a supportive and broader land use strategy. This is important as areas, greater investment in safe on-road cycling infrastructure may cities grow because it maximises public transport use, provides more increase both the number of train patrons overall, and those accessing sustainable transport options and supports the health and well-being of the train station by cycling (Pucher & Buehler, 2009). nearby residents (Watts et al., 2018). Overall, train stations with the most potential for redevelopment to become TODs were the cluster 3 stations. These were located in the next 6. Strengths and limitations most walkable neighbourhoods, although the levels of gross density and access to destinations were considerably lower than cluster 1 train The primary strength and contribution of this research is it devel- stations. Cluster 3 train stations had twice the availability of bike fa- oped a rigorous, evidence-driven typology derived using 14 measures of cilities compared to the remaining cluster 1 and 2 train stations, how- walkability of the train station neighbourhood that included measures ever only 1% of patrons cycled. More cycling and use of existing bike for car parking and bike facilities. Many researchers advocate for the facilities could be encouraged if they were complemented by on-road use of car parking and bike facility measures in analyses of train sta- cycling infrastructure provided within 5 km. Such infrastructure has tions and TODs but lack the data to do so (Griffiths & Curtis, 2017; been shown to connect large tracts of land across metropolitan Renne, 2009). In this study the use of these two measures in addition to Melbourne (Giles-Corti et al., 2014). This approach could also be ap- the remaining walkability measures revealed useful and important plied to cluster 2 train stations which also had both a low number of findings that could be used by policymakers and practitioners for

229 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231 improving station access and train patronage by both of these transport Funding declaration modes. This typology can be used to identify train stations ripe for devel- CB and LDG are support by the NHMRC Centre for Research opment, and for understanding which features could be changed and by Excellence in Healthy Liveable Communities Grant (#1061404). BGC is how much. It can also be used by policymakers, planners and urban supported by an NHMRC Senior Principal Research Fellowship designers to derive a normative or standardized TOD typology (Higgins (#1107672) and Clean Air and Urban Landscapes Hub – National & Kanaroglou, 2016). Such a typology could guide future train station Environmental Sciences Programme. CB and BGC are also supported by design, improvements to existing train stations with differing levels of the NHMRC Australian Prevention Partnership Centre (# 9100001). walkability features and inform neighbourhood design with both medium and long-term development and policy targets. The train sta- Competing interests tion typology developed in this work provides validated evidence on the types of train station features that could be incorporated to improve The authors declare they have no actual or potential competing train stations with low walkability. Train stations located in neigh- interests. bourhoods with low walkability could be improved using the higher standards set by those in the remaining clusters in terms of accessibility, Authors' contribution design, and performance (Gunn et al., 2017b). Hence, using a com- prehensive set of walkability measures is a major strength of this paper DJ, LG, CB, BGC, and SW conceived the research. DJ, LG and CB as it facilitates the distinction between the walkability of neighbour- conducted the analyses. LG, DJ, BGC, and CB drafted the manuscript. hoods surrounding train stations revealing information on their po- All authors contributed to data interpretation, added intellectual con- tential to become a TOD (Vale, 2015). tent, commented on drafts and approved the final version. All authors Another strength of this paper is the use of 800 m street network read and approved the final manuscript. buffers for calculating train station features. The 800 m buffer isevi- dence-based: i.e., in Melbourne the policy is that residents live within References 800 m of train stations. However, the network buffers also better re- present people's walking experience as they are based on connected Adams, M.A., Sallis, J.F., Kerr, J., Conway, T.L., Saelens, B.E., Frank, L.D., Norman, G.J., streets rather than Euclidean or as the crow flies approaches used in Cain, K.L., 2011. Neighborhood environment profiles related to physical activity and weight status: a latent profile analysis. Prev. Med. 52, 326–331. other studies. Much past research draws upon Euclidean buffers Arundel, J., Lowe, M., Hooper, P., Roberts, R., Rozek, J., Higgs, C., Giles-Coti, B., 2017. (Kamruzzaman et al., 2014; Lyu et al., 2016), although research in- Creating Liveable Cities in Australia: Mapping Urban Policy Implementation and corporating ped-shed ratios necessarily uses data calculated via both Evidence-Based National Liveability Indicators. Centre for Urban Research RMIT University, Melbourne. methods (Higgins & Kanaroglou, 2016; Vale, 2015). Although many Atkinson-Palombo, C., Kuby, M.J., 2011. The geography of advance transit-oriented de- papers use a distance of either 700 m or 800 m, we recommend that velopment in metropolitan Phoenix, Arizona, 2000–2007. J. Transp. Geogr. 19 (2), 800 m be used as a standard for future work as there is now supporting 189–199. literature on 800 m being a walkable distance linked to food related and Australian Bureau of Statistics, 2011. 2074.0 – Census of Population and Housing: Mesh Block Counts. A. B. S. Statistics. local living destinations (Gunn et al., 2017a; Gunn et al., 2017b; King Badland, Mavoa, Boulangé, Eagleson, Gunn, Stewart, David, S., Giles-Corti, B., 2017. et al., 2015). However, a limitation of our approach was that our street Identifying, creating, and testing urban planning measures for transport walking: network layer used road centre-line data, which may not accurately findings from the Australian national liveability study. J. Transp. Health 5, 151–162. Bertolini, L., 1999. Spatial development patterns and public transport: the application of represent the full extent of the pedestrian network potentially under- an analytical model in the Netherlands. Plan. Pract. Res. 14 (2), 199–210. estimating some of the walkability measures. There was also a differ- Boulangé, C., Gunn, L., Giles-Corti, B., Mavoa, S., Pettit, C., Badland, H., 2017. Examining ence in the timing of the GIS data, which is based on 2011 spatial layers associations between urban design attributes and transport mode choice for walking, cycling, public transport and private motor vehicle trips. J. Transp. Health 6, and the train station patronage data which was collated in 2014. Whilst 155–166. a temporal mismatch such as this is common practice given the avail- Center for Transit-Oriented Development (2010). Performance-Based Transit-Oriented ability and complexity of creating GIS data, it is not clear how this Development Typology Guidebook. www.ctod. Cerin, E., Leslie, E., Toit, L.d., Owen, N., Frank, L.D., 2007. Destinations that matter: affects the analysis or the results. associations with walking for transport. Health Place 13 (3), 713–724. Future research could include additional walkability measures such Cervero, R., Kockelman, K., 1997. Travel demand and the 3Ds: density, diversity, and as perceptions of safety, the presence of street lighting and aesthetic design. Transp. Res. D 2 (3), 199–219. Charreire, H., Weber, C., Chaix, B., Salze, P., Casey, R., Banos, A., Badariotti, D., Kesse- aspects of the built environment such as street trees. We opted to use Guyot, E., Hercberg, S., Simon, C., Oppert, J.-M., 2012. Identifying built environ- cluster analysis ahead of other methods for combining variables be- mental patterns using cluster analysis and GIS: relationships with walking, cycling cause it is an intuitive, simple and well-established method used in and body mass index in French adults. Int. J. Behav. Nutr. Phys. Act. 9 (1), 59. previous built environment studies. Other methods such as latent class Chorus, P., Bertolini, L., 2011. An application of the node place model to explore the spatial development dynamics of station areas in Tokyo. J. Transp. Land Use 4 (1), analysis may have been preferable, and these could be explored in fu- 45–58. ture studies. Christian, H.E., Bull, F.C., Middleton, N.J., Knuiman, M.W., Divitini, M.L., Hooper, P., Our validation of the clusters by transport mode is a final strength of Amarasinghe, A., Giles-Corti, B., 2011. How important is the land use mix measure in understanding walking behaviour? Results from the RESIDE study. Int. J. Behav. this paper. However, we recommend that future work explicitly model Nutr. Phys. Act. 8, 55–66. walking and active transport behaviours in-conjunction with built en- City of Melbourne, 2015. Places for People. Melbourne, Australia. vironment features using other methods including the use of Poisson or Commission on the Social Determinants of Health, 2008. Final Report—Closing the Gap in a Generation. Geneva, World Health Organization, Commission on Social Binomial regression models. Such approaches are now being used to Determinants of Health. measure the health benefits of good built environment design andare Davern, M., Gunn, L., Whitzman, C., Higgs, C., Giles-Corti, B., Simons, K., Villenaueva, K., being used in economic evaluation frameworks (Boulangé et al., 2017; Mavoa, S., Roberts, S., Badland, H., 2018. Using spatial measures to test a conceptual model of social infrastructure that supports health and wellbeing. Cities & Health 1 Veerman et al., 2016), which are now being used for advocating for (2), 194–209 (Published online: 12 March 2018). healthy, liveable and sustainable city-wide development. Dodson, J., Sipe, N., 2008. Shocking the suburbs: urban location, homeownership and oil vulnerability in the Australian City. Hous. Stud. 23 (3), 377–401. Ewing, R., Cervero, R., 2001. Travel and the built environment: a synthesis. Transp. Res. Acknowledgments Rec. 1780, 87–114. Ewing, R., Cervero, R., 2010. Travel and the built environment: a meta-analysis. J. Am. This manuscript is based on the research of Dana Jeffrey as part of Plan. Assoc. 76 (3), 265–294. her Masters of Urban Planning undertaken at The University of Ewing, R., Greenwald, M., Walters, J., Feldman, M., Cervero, R., Thomas, J., 2009. Measuring the Impact of Urban Form and Transit Access on Mixed Use Site Trip Melbourne.

230 D. Jeffrey, et al. Journal of Transport Geography 76 (2019) 221–231

Generation Rates—Portland Pilot Study. Washington, D.C, U.S. Environmental Pitney Bowes Ltd, 2014. Axiom Business Points. Pitney Bowes Ltd, Sydney. Protection Agency. Public Transport Victoria, 2012. Database. Frank, L., 2005. Linking objectively measured physical activity with objectively measured Public Transport Victoria, 2014. Public Transport a Collection of PTV Datasets. urban form: findings from SMARTRAQ. Am. J. Prev. Med. 28 (supplement), 117–125. Public Transport Victoria, 2015. Passenger Activity by Metropolitan Station 2008–09 to Frank, L.D., Schmid, T.L., Sallis, J.F., Chapman, J., Saelens, B.E., 2005. Linking objec- 2013–14. tively measured physical activity with objectively measured urban form: findings Pucher, J., Buehler, R., 2009. Integrating bicycling and public transport in North America. from SMARTRAQ. Am. J. Prev. Med. 28 (2), 117–125. J. Public Transp. 12 (3), 79–104. Giles-Corti, B., Eagleson, S., Lowe, M., 2014. Securing Australia's Future – Sustainable Renne, J.L., 2009. In: Curtis, C., Renne, J.L., Bertolini, L. (Eds.), Measuring the Success of Urban Mobility: The Public Health Impact of Transportation Decisions. (Australian Transit-oriented Development. Transit-oriented Development: Making It Happen, pp. Council of Learned Academies). 241–255 Ashgate, Burlington, VT. Giles-Corti, B., Foster, S., Koohsari, M., Francis, J., Hooper, P., 2015. The influence of Reusser, D.E., Loukopoulos, P., Stauffacher, M., Scholz, R.W., 2008. Classifying railway urban design and planning on physical activity. In: Barton, H.E., Thompson, S.E., stations for sustainable transitions – balancing node and place functions. J. Transp. Grant, M.E., Burgess, S.E. (Eds.), The Routledge Handbook of Planning for Health and Geogr. 16 (3), 191–202. Well-Being: Shaping a Sustainable and Healthy Future. Taylor and Francis, Hoboken. Saelens, B.E., Handy, S.L., 2008. Built environment correlates of walking: a review. Med. Abingdon, Oxon New York, NY. Sci. Sports Exerc. 40 (7 Suppl 1), S550–S566. Giles-Corti, B., Vernez-Moudon, A., Reis, R., Turrell, G., Dannenberg, A.L., Badland, H., Saelens, B., Sallis, J., Frank, L., 2003. Environmental correlates of walking and cycling: Foster, S., Lowe, M., Sallis, J.F., Stevenson, M., Owen, N., 2016. City planning and findings from the transportation, urban design, and planning literatures. Ann. Behav. population health: a global challenge. Lancet 388 (10062), 2912–2924. Med. 25 (2), 80–91. Griffiths, B., Curtis, C., 2017. Effectiveness of transit oriented development inreducing Sallis, J.F., Bull, F., Burdett, R., Frank, L.D., Griffiths, P., Giles-Corti, B., Stevenson, M., Car use: case study of Subiaco, Western Australia. Urban Policy Res. 35 (4), 391–408. 2016. Use of science to guide city planning policy and practice: how to achieve Gunn, L.D., King, T.L., Mavoa, S., Lamb, K., Giles-Corti, B., Kavanagh, A., 2017a. healthy and sustainable future cities. Lancet 388 (10062), 2936–2947. Identifying destination distances that support walking trips in local neighborhoods. J. Schlossberg, M., Brown, N., 2004. Comparing transit-oriented development sites by Transp. Health 5, 133–141. walkability indicators. Transp. Res. Rec. 1887 (1887), 34–42. Gunn, L.D., Mavoa, S., Boulangé, C., Hooper, P., Kavanagh, A., Giles-Corti, B., 2017b. State Government of Victoria, 2012a. Vicmap Transport. Victoria, Department of Designing healthy communities: creating evidence on metrics for built environment Sustainability and Environment. features associated with walkable neighbourhood activity centres. Int. J. Behav. Nutr. State Government of Victoria, 2012b. Vicmap Property. Victoria, Department of Phys. Act. 14 (1), 164. Sustainability and Environment. Hair Jr., J.F., Black, W.C., Babin, B.J., Anderson, R.E., 2014. Multivariate Data Analysis, State Government of Victoria, 2012c. Vicmap Planning. Victoria, Department of Harlow Pearson Education Limited, 7th ed. Pearson new international edition. Sustainability and Environment. Higgins, C.D., Kanaroglou, P.S., 2016. A latent class method for classifying and evaluating State Government of Victoria, 2017a. Level Crossing Removal Project: PROGRAM the performance of station area transit-oriented development in the Toronto region. BUSINESS CASE EXECUTIVE SUMMARY. J. Transp. Geogr. 52 (C), 61–72. State Government of Victoria, 2017b. Plan Melbourne – Metropolitan Planning Strategy. Jacobson, J., Forsyth, A., 2008. Seven American TODs good practices for urban design in Victoria, Department of Transport. transit-oriented development projects. J. Transp. Land Use 1 (2), 51–88. State Government of Victoria, 2018. https://www.premier.vic.gov.au/underground- Kamruzzaman, M., Baker, D., Washington, S., Turrell, G., 2014. Advance transit oriented suburban-rail-loop-to-connect-victoria/ Accessed January 19, 2019. development typology: case study in Brisbane, Australia. J. Transp. Geogr. 34 (C), Stutzer, A., Frey, B.S., 2008. Stress that Doesn't pay: the commuting paradox* Scand. J. 54–70. Econ. 110 (2), 339–366. King, T., Bentley, R., Thorton, L., Kavanagh, A., 2015. Does the presence and mix of Vale, D.S., 2015. Transit-oriented development, integration of land use and transport, and destinations influence walking and physical activity? A multilevel study. Int.J. pedestrian accessibility: combining node-place model with pedestrian shed ratio to Behav. Nutr. Phys. Act. 12, 115. evaluate and classify station areas in Lisbon. J. Transp. Geogr. 45 (C), 70–80. Leal, C., Chaix, B., 2011. Influence of geographic life environments on cardiometabolic Veerman, J.L., Zapata-Diomedi, B., Gunn, L., McCormack, G.R., Cobiac, L.J., Mantilla risk factors: a systematic review, a methodological assessment and a research agenda. Herrera, A.M., Giles-Corti, B., Shiell, A., 2016. Cost-effectiveness of investing in Obes. Rev. 12 (3), 217–230. sidewalks as a means of increasing physical activity: a RESIDE modelling study. BMJ Lyu, G., Bertolini, L., Pfeffer, K., 2016. Developing a TOD typology for Beijing metro Open 6 (9), e011617. station areas. J. Transp. Geogr. 55, 40–50. Watts, N., Amann, M., Arnell, N., Ayeb-Karlsson, S., Belesova, K., Berry, H., ... Costello, McCormack, G.R., Friedenreich, C., Sandalack, B.A., Giles-Corti, B., Doyle-Baker, P.K., A., 2018. The 2018 report of the lancet countdown on health and climate change: Shiell, A., 2012. The relationship between cluster-analysis derived walkability and shaping the health of nations for centuries to come. Lancet 392 (10163), 2479–2514. local recreational and transportation walking among Canadian adults. Health Place Wegener, M., 2004. Overview of land-use transport models. In: Hensher, D.A., Button, K. 18, 1079–1087. (Eds.), Transport Geography and Spatial Systems. Pergamon/Elsevier Science, McNally, M., Kulkarni, A., 1997. Assessment of influence of land use-transportation Kidlington, UK. system on travel behavior. Transp. Res. Rec. 1607 (1607), 105–115. Wegener, M., Fuerst, F., 1999. Land-use transport interaction: State of the art. Deliverable Messenger, T., Ewing, R., 1996. Transit-oriented development in the Sun Belt. Transp. 46. Institut für Raumplanung, Dortmund. Res. Rec. 1552 (1552), 145–153. Woodcock, J., Edwards, P., Tonne, C., Armstrong, B.G., Ashiru, O., Banister, D., Beevers, Newman, P., Kenworthy, J., 1999. Sustainability and Cities: Overcoming Automobile S., Chalabi, Z., Chowdhury, Z., Cohen, A., Franco, O.H., Haines, A., Hickman, R., Dependence. Island Press, Washington, DC. Lindsay, G., Mittal, I., Mohan, D., Tiwari, G., Woodward, A., Roberts, I., 2009. Public Newman, L., Baum, F., Javanparast, S., O'Rourke, K., Carlon, L., 2015. Addressing social health benefits of strategies to reduce greenhouse-gas emissions: urban land trans- determinants of health inequities through settings: a rapid review. Health Promot. port. Lancet 374 (9705), 1930–1943. Int. 30 (Suppl. 2), ii126–ii143. Zemp, S., Stauffacher, M., Lang, D.J., Scholz, R.W., 2011. Classifying railway stations for Olaru, D., Smith, B., Xia, J., Lin, T., 2014. Travellers' attitudes Towards Park-and-ride strategic transport and land use planning: context matters!. J. Transp. Geogr. 19 (4), (PnR) and choice of PnR Station: evidence from Perth, Western Australia. Procedia 670–679. Soc. Behav. Sci. 162 (C), 101–110.

231