Measuring Transit-Oriented

Development Network Synergy

Based on Node Typology

RUNJIE HUANG
March, 2017

SUPERVISORS:
Dr. A.B. Grigolon
Dr. M. Madureira

ADVISOR:
ir. M.J.G. Brussel
MEASURING TRANSIT-ORIENTED
DEVELOPMENT NETWORK
SYNERGY BASED ON NODE
TYPOLOGY

RUNJIE HUANG
Enschede, The Netherlands [March, 2016]

Thesis submitted to the Faculty of Geo-Information Science and Earth

Observation of the University of Twente in partial fulfilment of the
requirements for the degree of Master of Science in Geo-Information
Science and Earth Observation.

Specialization: Urban Planning and Management

SUPERVISORS:
Dr. A.B. Grigolon
Dr. M. Madureira

ADVISOR:
ir. M.J.G. Brussel

THESIS ASSESSMENT BOARD:

Prof. dr. ir. M.F.A.M. van Maarseveen (Chair)
Dr. Lissy La Paix Puello (External Examiner, University of Twente)
 MEASURING TRANSIT-ORIENTED DEVELOPMENT NETWORK SYNERGY BASED ON NODE TYPOLOGY

ABSTRACT:
Recent research has focused on assessing the Transit-Oriented Development (TOD) level of individual
transit nodes within a TOD network. However, this thesis argues that having such a TOD level value is not
sufficient for understanding the role each particular transit node plays within a TOD network. In other
words, a transit node may have a low performance when evaluating its individual TOD level, but it may
serve an important role within the TOD network as a whole, for example as a feeder node.

In this thesis, a TOD typology is developed based on five built-form indicators to identify their roles in the
transit network, and to discuss how their effects contribute to the large-scale TOD. Based on the TOD
typology, a correspondence analysis is conducted to measure the potential synergy effect of the TOD
network system in the Arnhem-Nijmegen city region, in the Netherlands.

In the results, the urban core central nodes reach high TOD scores in the 5D built-form aspects as well as
the aspect of ridership, whereas, in the aspect of ridership, some low TOD level suburban stations
outperform some urban stations with higher TOD level values. This outperformance indicates the
importance of these low TOD-ness suburban stations. A correspondence analysis is conducted based on
the developed typology. The results suggest that there are differentiations among the TOD nodes, in terms
of residential housing price and building use. Based on thesis result, this thesis argues that these
differentiations would lead to the complementarity among the stations. Thus, a station should not be seen
as an isolated node. In other word, TOD should be assessed and planned in a network system view.

Keywords:
Transit-oriented development, TOD Typology, TOD Network, Network Synergy, Complementarity

i
 MEASURING TRANSIT-ORIENTED DEVELOPMENT NETWORK SYNERGY BASED ON NODE TYPOLOGY

ACKNOWLEDGEMENTS
This thesis is the culmination of being a student for 18 month at the Faculty of Geo-Information Science
and Earth Observation (ITC) of the University of Twente. It was submitted to World Symposium on
Transport and Land Use Research (WSTLUR) 2017 in Brisbane, Australia, and accepted for presentation at
WSTLUR 2017.

I would like to sincerely thank my supervisors Dr. A.B. Grigolon and Dr. M. Madureira, my advisor ir.
M.J.G. Brussel for their enlightening guidance and invaluable advices. I deeply appreciate them for
promoting and giving hopes to an amateur researcher. I would also like to thank Dr. Y.J. Singh for providing
some of the crucial data.
Doing research is always like standing on the shoulders of the giants, therefore, I would like to thank ITC
for providing us the essential facilities and resources so that we can have a good environment to study and
read abundant literatures to broaden our minds. Through this one and a half year in ITC, my world view
was greatly widened. The international cultures converging here in ITC give me a profound experience
which will definitely benefit the rest of my life. The international friendships are with no doubt indelible. I
will be missing my friends from the five continents.

Finally, I would like to express my love and gratitude to my parents. I am more than grateful to their
enormous love and unyielding support.

ii
 MEASURING TRANSIT-ORIENTED DEVELOPMENT NETWORK SYNERGY BASED ON NODE TYPOLOGY

TABLE OF CONTENTS
INTRODUCTION ..............................................................................................................................................1

1.1. Background and Justification ....................................................................................................................................1

1.2. Research Problem........................................................................................................................................................1

1.3. Research Objectives and Questions .........................................................................................................................2

1.4. Research Questions .....................................................................................................................................................2

1.5. Thesis Structure and the Conceptual Framework .................................................................................................2

LITERATURE REVIEW...................................................................................................................................5
2.1. Concept of TOD .........................................................................................................................................................5

2.2. TOD Measurement .....................................................................................................................................................5

2.3. TOD Typology ............................................................................................................................................................6

2.4. Synergy Effect of Network ........................................................................................................................................8

2.5. Measuring Network Synergy .....................................................................................................................................9

2.6. Conclusion ................................................................................................................................................................. 10

METHODOLOGY .......................................................................................................................................... 11
3.1. Study Area.................................................................................................................................................................. 11

3.2. Unit of Analysis ........................................................................................................................................................ 12

3.3. Operationalization of the 5D Indicators .............................................................................................................. 13

3.4. Latent Class Cluster Model (LCCM)..................................................................................................................... 19

3.5. Measuring the synergy in the TOD network ....................................................................................................... 20

3.6. Correspondence Analysis ........................................................................................................................................ 22

RESULTS ........................................................................................................................................................... 27

4.1. Variables Correlation ............................................................................................................................................... 27

4.2. TOD Typology ......................................................................................................................................................... 27

4.3. Complementarity in TOD Network ..................................................................................................................... 32

Discussion and conclusion ............................................................................................................................... 37

5.1. Discussion .................................................................................................................................................................. 37

5.2. Limitation ................................................................................................................................................................... 38

5.3. Future study ............................................................................................................................................................... 38

iii
 MEASURING TRANSIT-ORIENTED DEVELOPMENT NETWORK SYNERGY BASED ON NODE TYPOLOGY

LIST OF FIGURES
Figure 1 Conceptual Framework of the thesis ......................................................................................................... 3

Figure 2, typologies of unsupervised cluster analysis methods .............................................................................. 7

Figure 3 Arnhem-Nijmegen region is located in the Netherlands. ..................................................................... 11

Figure 4 The TOD network of the region, with its 20 train stations.................................................................. 11
Figure 5 Railway Service in the Region of Arnhem-Nijmegen ............................................................................ 12
Figure 6 Buffer areas of 800 meters from the three train stations ...................................................................... 13

Figure 7 population distribution (source: CBS)...................................................................................................... 14

Figure 8 The network distance to transit calculated by Network Analysis in ArcGIS .................................... 15
Figure 9 The urban fabric of the area around Station Arnhem Centraal, Arnhem Velperpoort, Arnhem
Presikhaaf and Velp. ................................................................................................................................................... 15
Figure 10 The ‘1’ and ‘0’ points captured by the fishnets ..................................................................................... 16

Figure 11 Buildings of residential and non-residential .......................................................................................... 16

Figure 12The raster image of residential buildings’ density.................................................................................. 16
Figure 13, The raster image of non-residential buildings’ density ....................................................................... 17

Figure 14, result of the focal statistics for residential buildings ........................................................................... 17
Figure 15, result of the focal statistics for non-residential buildings .................................................................. 17
Figure 16, the point containing the focal statistics information of the residential and non-residential ........ 18
Figure 17 merits and demerits of latent class model-based clustering method ................................................. 19
Figure 18 Illustrations of the club type (a) and the web type (b) TOD networks ............................................ 20
Figure 19 Conceptual framework of identification and measurement of complementarity in TOD network.
........................................................................................................................................................................................ 22
Figure 20 Mathematical abstract of the contingency table ................................................................................... 23
Figure 21, a probability table ..................................................................................................................................... 24
Figure 22, Typology of the TOD network of region Arnhem-Nijmegen.......................................................... 29
Figure 23, TOD characteristics of Station 4. For illustration, the indicator values were normalized ........... 30
Figure 24, TOD characteristics of Station 21 (indicator values normalized) .................................................... 30
Figure 25, TOD characteristics of Station 1 (indicator values normalized)....................................................... 31
Figure 26, Location of stations 6 and 19 within the urban fabric. ...................................................................... 31

Figure 27, TOD node typologies associated with Residential Housing Price ................................................... 33
Figure 28, TOD node typologies associated with building use ........................................................................... 35

iv
 MEASURING TRANSIT-ORIENTED DEVELOPMENT NETWORK SYNERGY BASED ON NODE TYPOLOGY

LIST OF TABLES
Table 1, Method adopted by Higgins & Kanaroglou (2016) ..................................................................................6

Table 2, Method adopted by Atkinson-Palombo & Kuby (2011) .........................................................................7

Table 3,Method adopted by Key literature 3: Kamruzzaman et al. (2014) ..........................................................7

Table 4, Summery of the club- and web-type network characteristics from Meijers (2005) .............................8
Table 5, 5D Indicators and Measurement Variables............................................................................................. 13
Table 6 Latent Class model fit statistics .................................................................................................................. 20

Table 7, sample data of CA ....................................................................................................................................... 23

Table 8, probability table for the two variables ..................................................................................................... 24
Table 9 Correlation matrix of the TOD variables ................................................................................................. 27
Table 10 Latent Class Cluster Model of TOD nodes ........................................................................................... 28
Table 11 Validation of the Typology with Passenger Count Information ....................................................... 32
Table 12, inertia value for the CA of TOD node typologies vs. residential housing price ............................ 33

Table 13, inertia value for the CA of TOD node typologies vs. building uses ................................................ 34

v
 INTRODUCTION
1.1. Background and Justification

With increasing urbanization rates worldwide, cities are being challenged by urban problems such as traffic
congestion, sprawl, environmental degradation and spatial segregation. The automobile-oriented, road-
based transport systems are criticized for contributing to these challenges. While the demands on higher
mobility and land are still increasing, the resources of land and fuel are limited. Moreover, the overuse of
private automobiles is one of the main contributors to greenhouse gas emissions (U.N. Habitat, 2009). Thus,
it is an urgent and significant issue to develop a sustainable transport system that can improve efficient use
of the scarce resources.
Seen as a planning tool that integrates transport and land use, transit-oriented development (TOD) is a
promising approach to promote transit patronage and dissuade citizens from the usage of the private
automobile. It is typically understood as the development of transit stations around high-density (density),
land use mixed (diversity) and, walkable and cycling-friendly environments (design). Density, diversity and
design are disseminated as the three of the most important dimensions of TOD (Cervero & Kockelman,
1997). It is such a development located at a rapid transit system that can compete with automobiles in long
trips and attract more people to walk or cycle in short trips. This can lead to lower usage of the automobile
and a denser urban fabric, consequently, less consumption of fossil fuel and concentrated opportunities and
services (Calthorpe, 1993; Cervero and Kockelman, 1997; Cervero, Murphy, Feerell, Goguts & Tsai, 2004).
‘TOD-ness’, a term first developed by Evans, Stryker, Kuzmyak & Pratt, (2007) was defined as “potential
device for considering the degree to which a particular project is intrinsically oriented towards transit”. TOD
has been championed as one of the most effective solutions for maximizing the potential return on
investment for transit projects (Higgins & Kanaroglou, 2016), but due to the starting conditions of different
stations, the actual implementation may be different. Assessing the existing TOD conditions and
understanding the heterogeneity of the built environment before implementing a TOD project become vital
to enhance the success rate of the project.

1.2. Research Problem

Recent research has been focusing on assessing the performance of individual transit nodes within a TOD
network, using methods such as Spatial Multiple Criteria Analysis (SMCA) (Singh, 2015), in order to arrive
at a TOD-ness value for different transit nodes. Such measurement is important to evaluate the extent to
which an area is oriented towards transit, but usually insufficient for the understanding of the heterogeneous
built environment characteristics of area around train stations. This is because the TOD characteristics at
the node level (in this thesis, defined as TOD nodes) tend to be lost in the process of indicator aggregation.
Studies have suggested that the heterogeneous built environments critically affect the implementation of
TOD, which means TOD might have different typologies according to its built environment. This implies
that general TOD-ness values might not be able to reflect the performance across different types of nodes
in the transit network because the nodes might play different roles in the network, and as a network system,
there might be synergy effects between nodes of the TOD network.

1
1.3. Research Objectives and Questions

General Objective

This thesis aims to develop a TOD typology, which will serve to identify the characteristics of different
nodes in the transit network. Based on this typology, to further analyse synergy effects between TOD nodes
of the TOD network, understanding how the individual nodes interact and contribute to the large-scale
TOD network, potentially to network synergy.

Sub-objectives

In order to achieve the research objective, several sub-objectives are derived:

i. To identify the reasons of the variation of TOD at node scale

ii. To identify the dimensions, and the indicators to enable the TOD typology development
iii. To develop the TOD typology
iv. To identify the roles of the TOD nodes in a TOD network
v. To measure the synergy in a TOD network

1.4. Research Questions

The research questions related to the sub-objectives are:

(i.) To identify the factors causing the variation of TOD at node scale
(1) What are the factors that causes the difference in TOD implementation?
(ii.) To identify the dimensions, and the indicators to enable the TOD typology development
(2) What are the main dimensions of TOD?
(3) What indicators of the dimensions should be used to develop a TOD typology.
(iii.) To develop the TOD typology
(4) What methods can be used to develop a TOD typology?

(iv.) To identify the roles of the TOD nodes in the TOD network
(5) What kind of interaction might exist between the nodes?
(v.) To measure the synergy in a TOD network

(6) What is synergy effect?

(7) What method should be used to measure the network synergy?
(8) In which of the 5D dimension of a TOD are network synergy likely to play a role?

1.5. Thesis Structure and the Conceptual Framework

The thesis is structured as follows: in chapter 1, the background, research problem, research objectives and
problem, and the conceptual framework of the thesis are presented; chapter 2 contains the discussions of
the literatures about the concept of TOD, its measurement, existing studies which have also developed

2
TOD typologies, the general economic theory of network synergy and the transferring of network synergy
theory to the urban issue; chapter 3 explains the methodology, presenting the study area, illustration of the
unit of analysis, the geo-processing of 5D indicators, the development of TOD typology and the network
synergy measurement; chapter 4 discusses the results of the two main analyses: latent class analysis and
correspondence analysis; finally, conclusions are drawn and plans for future research are discussed in
chapter.

Transit-Oriented
Development
Enable to better
understand the
roles of the nodes
Measuring
The 5 Ds of Built
Existing TOD
Inform planning

Environment
Level

Latent Class
TOD Index TOD Typology
Analysis

Complementarity Correspondence
between Transit Nodes Analysis

Figure 1 Conceptual Framework of the thesis

Figure 1 illustrates the conceptual framework of this thesis. As the essence of TOD, the 5Ds dimension
(density, diversity, design, distance to transit and destination accessibility) are adopted to first measure the existing
TOD level (which is referred as TOD-ness in this thesis), and based on the measurement, the TOD node
typology will be developed with the latent class cluster method. Later on, based on the developed TOD
node typology, correspondence analysis will be conducted to explore the complementarity between the
transit nodes.

3
4
 LITERATURE REVIEW
2.1. Concept of TOD

Concisely, the main characteristics of a TOD development are commonly referred to as the 3D’s of the
built environment: density, diversity and design (Cervero & Kockelman, 1997). Later on, these were expanded
into 5Ds, adding another two criteria of the built environment: destination accessibility and distance to transit
(Ewing & Cervero, 2001). Cervero and Kockelman (1997) argue that compact neighbourhoods (density) can
dissuade vehicle trips and encourage non-motorized travel by bringing origins and destinations closer
together to create an opportunities-concentrated (diversity) zone within walking or cycling distance (design).
While destination accessibility refers to the ease of access to trip attractions, distance to transit is the factor
measured as an average of the shortest street routes from the residences or workplaces to the nearest transit
node. Therefore, distance to transit delineates the impedance to take transit. These criteria are interrelated (e.g.
Ewing (2008) argues that since density and diversity co-exist, many of their benefits are inextricable) and
even overlap (e.g., to some extent, diversity and destination accessibility are related to trip attraction (Ewing
& Cervero, 2010). Yet, the interrelation shows that the criteria of TOD are not isolated, which also indicates
that the implementation of TOD is multi-dimensional.
When implemented correctly, the potential benefits of TOD are abundant. Cervero and Murakami (2009)
found that TOD is well suited for financing railway infrastructures, especially in the cases characterized by
high densities. They take Hong Kong as example, and demonstrate how the investment in TOD might be
paid off with the increase in transit patronage and the gaining from higher real estate prices. In the cases of
suburbanization, TOD can also be adopted to address the problem of declining accessibility. Cervero and
Day (2008) found that relocation to the suburban areas near a transit station could moderate the decline in
job accessibility, encouraging more people to take transit. Studies also find that TOD has positive effects in
creating vibrant, rich and liveable urban places, increasing physical activities and thus enhancing the public
health and quality of life (Frank, 2000; Dittmar & Poticha, 2004). From these points of views, it can be
concluded that TOD is multi-functional, as its core assumptions are manifold, such as to reduce sprawl,
increase transit ridership, reduce car use, etc.

2.2. TOD Measurement

Depending on the conditions around the different stations, the implementation of a TOD may be different.
Assessing the existing TOD conditions and understanding the heterogeneity of the built environment before
implementing a TOD project become vital to enhance the success rate of the project. A thoughtful analysis
of the existing built environment can ease facilitation of a future TOD (Kamruzzaman, Bakker, Washington
and Turrell, 2014). (Singh, 2015) shares a similar opinion and she argues that a scientific analysis on
measuring existing TOD levels is a prerequisite that can help to uncover to what extent an area is transit-
oriented, and because of what reasons. Therefore, she carried out her research in the TOD network of the
city region of Arnhem and Nijmegen, in the Netherlands, by aggregating multiple spatial indicators under a
SMCA framework to arrive at a general TOD level value.

However, for measuring the existing levels of TOD, having such a general value is essential but it is
insufficient for the understanding of the heterogeneous built environment because the TOD characteristics
are lost in the process of indicator aggregation. Understanding the heterogeneous built environment of
TOD is important since it help to understand why TOD unfolds differently across the system (Atkinson-

5
palombo & Kuby, 2011). Given the fact that TOD is multi-dimensional and multi-functional, with different
purposes and different aims, TOD implementation can take a variety of forms and the individual node can
have different roles within the network system, complementing each other (Atkinson-Palombo & Kuby,
2011). Furthermore, there is no “one-size-fits-all” TOD mode (Austin et al., 2010). Based on these
arguments, it can be indicated that a well-performing TOD transit node does not necessarily have to be ideal
in all dimensions of the conventional TOD criteria and that TOD planning rarely starts from zero.
Therefore, the various characteristics of the existing built environment could be the determinant factors
forming the heterogeneous typologies of TOD. Studying the typology of TOD is, therefore, in a prominent
position to address the problem.

2.3. TOD Typology

Interest in developing a TOD typology is emerging which regards TOD typology as a tool for informing
policy prescription and evaluation (Higgins & Kanaroglou, 2016). According to (Kamruzzaman al., 2014),
“developing a typology is a way to group together areas that have a common set of characteristics”. The
authors also indicate the benefit of developing a TOD typology which are as following: (1) bring
convenience to urban planning, design, and operation since within the same TOD typology they can share
a same set of strategies; (2) identify the general development potentials so that once a TOD is classified, the
optimal figuration can be deducted (e.g. the desirable density, mixed use); (3) reduce the complexity of
managing the infrastructure by enabling a common standard within the same TOD typology (e.g.
operations); (4) enable comparison within the same typology, identifying the benchmark station of a certain
type which can help to assess the other stations of the same type.
Research regarding the definition of TOD typologies has been carried out since the birth of TOD. Calthorpe
(1993) made a dichotomy between residential TOD and job-generating TOD. Dittmar and Poticha (2004)
expanded these dichotomous typologies with the distinction between urban and suburban areas. In addition,
Hancock et al., (2014) identified 5 types of TOD for strategic planning. These typologies are regarded by
Higgins and Kanaroglou (2016) as the “normative TOD typologies” since it only generally “outlines the
characteristics of what different TOD contexts should look like in terms of factors such as densities, housing
types, and transit service”. Aside from these normative typologies, research has been conducted to achieve
a more rigorous assessment by using the “positive approaches”, which aims to reduce the complexity created
by the heterogeneity in node context. Table 1 to 3 are the gist of relevant researches conducted with the
“positive approaches” and figure 2 is the typologies of unsupervised cluster analysis methods used by
different literatures.
Table 1, Method adopted by Higgins & Kanaroglou (2016)

Method Latent Class Analysis

Objectives To develop a TOD typology with latent class analysis to reduce the
complexity creating by the heterogeneity in node contexts.

Conclusion Latent Class Analysis is a good method for deriving TOD typologies that
are statistically rigorous, geographically rich, easily interpretable, and
readily transferable.

6
 Table 2, Method adopted by Atkinson-Palombo & Kuby (2011)

Methods Hierarchical Cluster Analysis, Factor Analysis, ANOVA Tests

To evaluate the relationship between geography and the amount and type
of advance TOD that takes place across a light-rail transit system, and to
Objectives
provide a baseline for future studies that will investigate how TOD
unfolds over time.

Overlay zoning was used most in areas of urban poverty and least in node
types with the most single-family housing. Advance TOD coincided
Conclusion
strongly with overlay zoning in areas of urban poverty and least in
employment and amenity centres.
Table 3,Method adopted by Key literature 3: Kamruzzaman et al. (2014)

Methods Two-step Cluster Analysis, Multinomial Logistic Regression Model

Objectives  To develop a typology for existing neighbourhoods in order to

understand the potential for different types of TODs in Brisbane,
Australia;
 To validate the typologies with performance indicators;
 To support the planning of advanced TOD typologies based on
readily available policy indicators.

Conclusion  Two types of TODs may be suitable for classification and effect mode
choice in Brisbane;
 TOD typology should be developed based on their TOD profile and
performance matrices;
 Both bus stop and train station based TODs are suitable for
development in Brisbane.

There is a variety of methods to develop a TOD typology, through the comparison, LCCM has more
advantages as: (1) it uses a probabilistic approach (in case of categorical variables) or means/rates (in case
of continuous or count variables) to clustering methods, (2) accommodates unscaled/unstandardized
variables, allowing the model outputs to be analysed in their own units, (3) accommodates different data
formats (nominal, ordinal, continuous) (Higgins and Kanaroglou, 2016).

Clustering
(unsupervised)

Partitional
Hierarchical
(nonhierarchical)
Represented by

Agglomerative Divisive K-means

Figure 2, typologies of unsupervised cluster analysis methods

7
2.4. Synergy Effect of Network
As one of the objectives of this research is to understand the synergy effects among different TOD nodes
within a TOD network system, a discussion of the literatures about synergy effects and network system are
undoubtedly important.

Described by Meijers (2005), linguistically, the word synergy refers to “a situation in which the effect of two
or more co-operating or combined bodies or functions is larger than the sum of the effects each body or
function alone can achieve”. The concept of network system is highly associated with synergy effects. A
network is basically consisted of the four elements: nodes, linkages between the nodes, flows and meshes
(Meijers, 2005). In the sense of urban transport and land use, nodes can be understood as cities, stations etc.
and while linkages can be seen as the roads or railways that connect the nodes, flows can be referred to
people with travel demands or the goods to be transferred, and meshes, the undeveloped or preserved land
(or hinterland) compassed by the nodes and linkages. When these elements function as a network, the nodes
cease to be the isolated nodes. With such a network, the interactions between the nodes create synergy
effects.
However, the meaning of synergy is too fuzzy to be a concept to understand the mechanism of network
synergy (Meijers, 2005). Extensive researches have been done in explaining the concept of synergy and
network synergy. Capello & Rietveld (1998) distil the essence of the synergy concept from economic
theories, arriving at co-operation, complementarity and externalities, by which they reckon synergy is
manifold and it is the expression of beneficial results of co-operation and complementarity.
Linking synergy to network, Capineri & Kamann (1998) categorize networks in two types: ‘club’ type
networks and ‘web’ type networks. In the club networks, actors (the nodes, the linkages or the flows) ‘share
a common objective, activity or service, while also having parallel interests and transaction chains’. This is
also known as ‘horizontal synergy’ that co-operating with each other, actors contribute the common goals,
leading to economies of scale, thus an averaging lower fixed cost and positive externalities. Meijers (2005)
exemplifies the club networks with the memberships system of a tennis club: individuals cannot afford the
facilities but coalesced, they can and the more members the lower membership fees or the longer opening
hours. However, if the members all want to play at the same time, it causes negative externalities. In web
networks, actors usually have different characteristics (the differentiation). By complementing each other,
they reach a positive outcome. This is also known as ‘vertical synergy’ that ‘results from a specialisation
process, redistributing resources and activities among the participating actors according to their
competence’. The example given by Meijers (2005) is the chain of enterprises. The highly-specialized
enterprises focus the effect on a particular stage of the production to achieve elaboration. Together, they
form a more exquisite industrial chain. To summarize, table 4 shows the revised types of network and the
characteristics.
Table 4, Summary of the club- and web-type network characteristics from Meijers (2005)

Types of Types of Types of Interaction Actor Behaviour Results

Network Synergy
Club network Horizontal Co-operation Common objective or Agglomeration
synergy interest among actors leading to
economies of scale
Web network Vertical Complementarity Differentiation in the Actors being
synergy roles specialized in their
fields

8
2.5. Measuring Network Synergy
Transferring the concept of synergy to the urban issues, Meijers (2005) reckon multitude of networks can
be found. For example, on a micro level, the urban network nodes can be households, firms, individuals,
organisations. On a macro level, it can be the network of cities, in which the nodes are the cities. To build
the conceptual framework to measure the network synergy, the urban network synergy needs to be defined.
To structure the framework, Meijers (2005) put up two questions: (1) whether a polycentric urban region is
a club- or a web- type; (2) how co-operation and complementarity should be interpreted in the context of
polycentric urban regions to provide a meaningful framework for analysis.
For the first question, Capineri & Kamann (1998) reckon (as cited by Meijers (2005)), in real life, networks
will have both club-type aspects and web-type dimensions, which means co-operation and complementarity
can coexist. This coexistence can be understood as: within a network, nodes can co-operate with certain
nodes, at the same time, complement each other with other nodes.

The second question has two pillars: for co-operation and for complementarity. In term of co-operation,
Meijers & Romein (2003) argue what is needed is the regional organising capacity, which helps to co-
ordinate the relationship between the members and generate the network cohesion. Such a regional
organising capacity can be achieved through an institutionalised framework of co-operation, debate,
negotiation and decision-making. As for complementarity, according to Meijers (2005), in urban regions,
complementarity exists between relatively similar ‘activities’ or ‘places’. ‘Activities’ include economic, social
and environmental activities, e.g. commercial services, education or medical care. On the other hand, ‘place’
can be referred to business milieus or residential milieus.

Moreover, two important preconditions related to supply and demand must be satisfied: (a) “there must
be differentiation in the supply of activities or places”, otherwise, the activities or places will be in a
relationship of competition (since they are more or less similar and mutually replaceable) instead of
complementarity; (b) “the geographical markets of demand for these activities or places must at least
partly overlap”, which means if they are not in the same markets, the supplies cannot meet the demands
even if the supplies can cater the demands. Therefore, no overlapping geographical markets will result in no
supply-demand interaction, thus no complementarity occurrence. Example given by Musterd & van Zelm
(2001) is that if the provisions of different types of residential milieus matches the need of different
preferences of a regional population, these provisions are complementary. Additionally, Meijers (2005) states
“complementarity often leads to spatial interaction”, however, on the other way around, it can be argued
that without spatial interaction, the synergy from complementarity can never be achieved. For example, the
economies of two cities are complementary, however, if the infrastructure between the two cities (the
network linkage) is inefficient and ineffective to enable the spatial interaction (or most extremely, they are
completely separated from each other), there will be no synergy generated from the ‘complementary
economies’ of these two cities.
Ullman (1956) (as quoted by Meijers (2005)) points out the importance of intervening opportunities and
the role of transferability (the costs of interaction) in the occurrence of spatial interactions. As for
intervening opportunities, the term inventor Stouffer (1940) states “The number of persons going to a given
distance is directly proportional to the number of opportunities at that distance and inversely proportional
to the number of intervening opportunities”, which can be understand, in the network view, as that the
more opportunities provided by one node the stronger linkage can this node build with the others; the
intervening opportunities offered by the in-between nodes can weaken the linkage between the two end
nodes. On the other hand, transferability (the costs of interact) emphasizes on the capacity of the linkage
between the nodes.

9
Interpretation of how intervening opportunities and transferability can be measured in the TOD context is
elaborated further in sections 3.5 and 4.3.

2.6. Conclusion
TOD, in simple terms, aims to encourage people to drive less, walk or cycle more in short distance and take
public transit in the long trips. In this sense, whenever the trip-makers feel it is easy, comfortable, efficient
or effective to travel or commute by walking, cycling or taking public transit, this type of planning can be
regarded as successful TOD planning. Therefore, to understand the current TOD level, which is esteemed
to be a prerequisite of successful TOD planning, should not be constrained too much by theories or
empirical rules. Thus, this thesis argues individual nodes should not be seen in an isolated way but in a
systematic view that nodes may play different roles in the system. Based on what has been done by
researchers, developing a typology for the transit nodes can facilitate the understanding of the system.
Moreover, as a network system, a TOD network in nature may generate network synergy which is usually
associated with network systems. Based on existing studies that measure network synergy in the polycentric
urban region, this thesis will explore synergy effects in the TOD network. Further explanation and analysis
of the TOD network synergy are in section 4.3.
The bundled concepts: places and activities, supplies and demands can help to understand why there are
complementarity in the network. As places make up cities, and most activities take place within cities
(Meijers, 2005), the concepts of ‘places’ and ‘activities’ explain where do people stay/depart to, and why do
they stay/depart to. With the rules of ‘overlapped market for demands and supplies’ and ‘differentiation in
places and activities’, complementarity can be stimulated. In this sense, the supplies and demands of places
and activities offer a way to understand the interaction between land use and transport: activities stimulate
travel demands, related to transport; places offer the site to hold the activities, related to land use.
Transferring this thinking to a TOD network, the TOD nodes are the places, whilst activities around the
nodes characterize the nodes (e.g. different land uses invite different activities) and become the reason why
people travel to these nodes (because activities have attractiveness). Based on this conceptual transformation,
the complementarity within a TOD network can be discussed. The further discussion is in section 3.5, and
application in section 4.3.

10
 METHODOLOGY
In this section, the unit of analysis, the measurement of the indicators and the Latent Class Cluster Method
(LCCM) will be described. The TOD typologies will be discussed based on the results of the LCCM and
the TOD-ness ranking of the nodes calculated, and validated with passenger-count data at each transit
station. To further explore the roles of the nodes, the framework of TOD network synergy is established
and followed by the measurement with correspondence analysis.

3.1. Study Area

Figure 3 Arnhem-Nijmegen region is Figure 4 The TOD network of the region, with its 20 train
located in the Netherlands. stations.

The Arnhem Nijmegen City Region is located at the heart of a vast metropolitan area in the east of the
Netherlands. Consisting of twenty municipalities, the region aims to promote regional cooperation. The
region is flanked by the Randstad conglomeration in the west, the Flemish Diamond to the south and the
Ruhr to the east. With the ambition to become the second largest economic area in the Netherlands after
Randstad by 2020, the city region aims to promote regional co-operation (“Arnhem-Nijmegen Metropolitan
area,” 2013). As it is esteemed to be one of the beacons of TOD in the Netherlands (Pojani & Stead, 2014),
the city region of Arnhem and Nijmegen was considered a suitable case study for this thesis. Moreover, the
city region of Arnhem and Nijmegen is a public body and inter-municipal institution, which accords with
the requirement of “regional organizing capacity” as discussed in Section 2.5.
The city region covers more than 1000 km2 with a population over 750,000 inhabitants. However,
individually, Arnhem and Nijmegen are relatively small-size cities, with the respective population
about154,000 and 170,000, comparing to Amsterdam (845,000) and Rotterdam (620,000). Arnhem-
Nijmegen city region can be subdivided into seven regions (Geography of the Arnhem-Nijmegen region,
2015), served by a rail-based national as well as regional transit system composed of 22 train stations, as

11
shown in figures 3 and figure 4. The railway connection in the region is shown in figure 5. Notably, the rail
service shown in figure 5 is the partial railway service of the large railway network in the Netherlands. There
are 8 lines serving through the region. The two major stations of the region are Arnhem Centraal Station
and Nijmegen Station. Several companies operate the train system in the Netherlands among which NS,
ARRIVA and VEOLIA are the three railway companies providing train service in the region. Unlike the
usual metro system in other metropolis, only the Sprinter trains (local service) stop at the small stations
(such as the stations with relatively low ridership) while the Intercity trains (express service) only stop at the
central station of the cities. Nevertheless, the railway service provided in the region is still fast and frequent
enough to form the ‘rapid transit corridor’ to enable TOD. According to the time tables published by NS
(“Nederlandse Spoorwegen”, the Dutch Railway), in 2016, the daily train frequency in Arnhem Central
Station is 597 times which is the highest among the 22 stations in the region.

Legend: Train Line

46 3
14 24
53 10
50 23

Figure 5 Railway Service in the Region of Arnhem-Nijmegen

3.2. Unit of Analysis

Based on the study of Guerra, Cervero, & Tischler (2013), we take the 800-metre buffer around the transit
station as the unit of analysis. The size of the buffer can simulate the catchment of people walking to/from
the station along the pedestrian network within 10 minutes, which is a fundamental assumption of this
thesis. Figure 6 illustrates the buffers around 3 train stations in Arnhem and the distribution of different
uses of buildings. If the Euclidean distance between two station is less that 1600m, the solution in this thesis
was to divide the overlapped part into two equal parts and assign them to the each corresponding station,
as show in figure 6. Since the value of indicators will be finally be averaged by the buffer size and this is the
only case in the region, the overlapping procedure will not cause problems.

12
 Figure 6 Buffer areas of 800 meters from the three train stations

3.3. Operationalization of the 5D Indicators

The indicators selected for this thesis are quantitative, and derived from the 5D of the built environment
indicators (Ewing & Cervero, 2001, 2010). The measurement variables for the indicators and the
corresponding data source are presented in Table 5.
Table 5, 5D Indicators and Measurement Variables
Indicator Measurement variables Data source
Population Density Dutch Statistics Office (CBS）
Provincial Employment Survey,
Job density*
2015 (Province of Gelderland)
Density
Dutch Cadastral Office -
Business Density Basisregistratie Adressen en Gebouwen
(BAG)
Dutch Cadastral Office -
Distance to transit Walking Impedance Basisregistratie Topografie
(TOP10NL)
Diversity Land Use Diversity (BAG)
Destination
Mixed-ness of Land Uses (BAG)
Accessibility
Intersection Density (BAG)
Design Length of Bicycle Network
Dutch Cyclist Union (Fietsersbond)
Length of Pedestrian Network

*Data not available for stations 5 and 16.

13
 Density

Density was expressed using three measurement variables: population, job and business density, as
residential population density is a primary element to promote ridership; job density indicates the
attractiveness of a node as trip destination; and business density measures the small business for example
the small pedestrian-friendly shop that can maximize the possibility for people to access by walking. All
variables were calculated for the buffer area (800 meters), and reflect the density per Km2, i.e., population
density includes the number of residents living within 800-meter buffer area, around each of the 20 train
stations. Job density includes the number of jobs in 21 different sectors, in the buffer area, and was used
to reflect the intensity of employment that brings commuters to the buffer area. Business density includes
industrial, shops and offices facilities, and it is used to reflect the trip attraction of the buffer area,
including the services and the opportunities provided within the buffer.

Figure 7 population distribution (source: CBS)

The population data used in this thesis is derived from CBS (Statistics Netherlands) in the form of 100m x
100m grid shape file. Each grid contains the demographic information of corresponding geographic area.
Information was extracted by using the 800m buffers of the stations (as shown in figure 7).

Distance to transit

Distance to transit was calculated as the network distance to the transit station within the buffer area. The
“closest facility” function of Network Analysis in ArcGIS was used to calculate the distance from the
centroid of each cell (CBS data) to the station (illustrated in figure 8), through the pedestrian network. The
distances were averaged and calculated for each station. Since the patterns of road network around the
stations are different, high values on this variable mean high impedance walking from the resident to the
station, i.e., the street is less direct; therefore, it is more detouring to reach the station by walking.

14
 Figure 8 The network distance to transit calculated by Network Analysis in ArcGIS

Destination accessibility

The local destination accessibility was measured with the land use mixed-ness index, which indicates
the easiness to access resources within short trips. It is calculated with a formula adapted from (Zhang
& Guindon, 2006):
∑𝑗 𝐿𝑜
MI(𝑖) = ∑
𝑗(𝐿𝑟 +𝐿𝑜 )

Equation 1, the mixed-ness formula

where MI(𝑖) is the Mixed-ness Index of buffer area 𝑖, 𝐿𝑟 and 𝐿𝑜 are respectively residential land uses
and non-residential land uses. The proportion of non-residential land uses within a small calculation area
is calculated for each residential point 𝑗. The value of MI(𝑖) shows a balanced mixed-ness when it is
0.5. The index is closer to 0 or 1, the more biased to the respective land use.
The geoprocessing was conducted in ArcGIS with the function of ModelBuilder. Since the proportions
of residential land uses and non-residential land uses are the key element, their calculation within a
certain area is the core. The building use data is derived from BAG. Followed is the procession.

Figure 9 The urban fabric of the area around Station Arnhem Centraal, Arnhem Velperpoort, Arnhem
Presikhaaf and Velp.

First, enlarged buffers (r=1000m) are created to capture the mixed-ness of margin. The uses of the building
were divided into to dichotomy: residential and non-residential, representing by 1 and 0. In figure 9, the blue
pointsare the redidential building (marked as 1),the bright orange are the non-residential.

15
 Figure 10 The ‘1’ and ‘0’ points captured by the fishnets

Then, As shown in figure 10, fishnets (the size is 10m x 10m, simulating the average size of the buildings)
are created to capture the numbers of ‘1’ and ‘0’ points. They are then translated in two raster images with
the cell value representing the number of the respective type of building. (figure 11)

Figure 11 Buildings of residential and non-residential

Figure 12The raster image of residential buildings’ density

16
 Figure 13, The raster image of non-residential buildings’ density

As the next step, the function of “focal statistics” is performed to calculate the number of the building of
the same type. The values of the cells within the 100m buffers (simulating the proximity of the buildings)
of each cell were summed up, producing a ‘hot spot’ map, with the red spots indicating high
residential/none-residential density within the 100m buffers, and low residential/none-residential density in
blue (figure 14 show the focal statistics for residential buildings, figure 15 for none-residential)

Figure 14, result of the focal statistics for residential buildings

Figure 15, result of the focal statistics for non-residential buildings

Then, the two images are transferred into vector point and aggregated into one layer which contains the
focal information of the two types of building use. (figure 16)

17
 Figure 16, the point containing the focal statistics information of the residential and non-residential

Diversity

Land use Diversity, referring to the number of different land uses in a given area and the degree to which
they are represented in land area, is measured with the widely used ‘entropy’ concept. High land use diversity
is indicated by a high value of entropy, thus, higher levels of TOD-ness. This indicator was applied to the
Dutch spatial planning context by Ritsema van Eck and Koomen (2008) and adapted by Singh (2015)as
follows:
− ∑𝑖 𝑄𝑙𝑢𝑖 ×ln(𝑈𝑙𝑢𝑖 )
LU𝑑 (𝑖) =
ln(𝑛)
where,
𝑆𝑙𝑢𝑖
𝑄𝑙𝑢𝑖 =
𝑆𝑖
LU𝑖 = Land use class within the buffer area i
𝑄𝑙𝑢𝑖 = The share of specific land use within the buffer area i

𝑆𝑙𝑢𝑖 =Total area of the specific land use within the buffer area i
𝑆𝑖 = Total area of buffer i
Equation 2, the entropy formula

Adopted from Cervero and Kockelman (1997), the mean entropy for land use categories is
calculated within a half-mile (804.5m) radius which is almost the same as the buffer size of this
thesis. To conduct this calculation, information of the different land use points within the 800m-
radius buffer around the stations extracted to Excel table, using the above formula.

Design

As for the Design aspect, the total length of pedestrian and bicycle lanes was taken as a measure that
reflects the length of accessible roads for non-motorized transport, key to the success of a TOD. The total
length of pedestrian and bicycle lanes of each node was calculated by summarizing the sub-network
clipped with the buffer.
Besides that, intersection density was calculated to measure the street connectivity of the road network of
each node. The higher the density, the better for cyclists and pedestrians as the routes may become shorter
in terms of travel distance although not necessarily in terms of travel times.

18
A straight but artful way to calculate this indicator was as follows: (1) to transform the network (which is
jointed polylines) into points, since one line segment has two end point, the “intersections” are then the
points containing more than two overlapping points which have the same x-y coordinates; (2) to calculate
the geometry adding the coordinates to the points, then after giving the point the station ID, extract the
data to Excel file; (3) to create an ID for each point attribute using the X-Y coordinates in Excel, then, use
the formula “COUNTIF” to count the repeating time of the unique X-Y coordinates, after that use the
advance filter to keep one record for one unique X-Y coordinates, finally use pivot table to summarize the
numbers of records for each of the stations, which are the numbers of intersections with the buffer. (4) to
calculate the density, dividing the numbers of intersections by the buffer area.

3.4. Latent Class Cluster Model (LCCM)

This thesis adopts the positive perspective by applying a Latent Class Cluster Method (LCCM) as a tool
for developing a TOD typology for the Arnhem-Nijmegen region. LCCM was chosen for this thesis
because (1) it uses a probabilistic approach (in case of categorical variables) or means/rates (in case of
continuous or count variables) to clustering methods, (2) accommodates unscaled/unstandardized
variables, allowing the model outputs to be analyzed in their own units, (3) accommodates different data
formats (nominal, ordinal, continuous) (also illustrated in figure 17) (Higgins and Kanaroglou, 2016).

Intensive
Computation

Probabilistic Latent Class model-based Clustering Flexible

Approach (Supervised) Structure

Formal More type of

decisions Data

Figure 17 merits and demerits of latent class model-based clustering method

The quantified 5D indicators were input to the software LatentGold (version 5.1) where the LCCM was
estimated. The class selection choice was done by analysing two output statistics: the BIC value and
entropy value. The BIC statistic is a complement to model fit, and takes into account the parsimony
(degrees of freedom or number of parameters) of the model. As for entropy, it indicates how well the
model predicts class memberships. The closer these values are to 1 the better the prediction.
In table 6, the model fit statistics of the LCCM are shown. To demonstrate the class selection, the
distribution of BIC and Entropy are presented across different classes. The model was estimated from 2
to 7 classes. The selected number of classes was the one with the highest Entropy value, followed by the
lowest BIC value. The model with 3 classes was selected as a better solution for the class division.

19
 1880 0.7

1860 0.6

0.5
1840
0.4

Entropy
BIC

1820
0.3
1800
0.2

1780 0.1
BIC
Entropy
1760 0
2 3 4 5 6 7
Classes

Table 6 Latent Class model fit statistics

3.5. Measuring the synergy in the TOD network

3.5.1 Co-operation and complementarity in TOD network

As a network system, a TOD network also consists of nodes, linkages, flows and meshes, where nodes are
the nodes with TOD planning; linkages can mean the pedestrian paths, bicycle lanes and transit services;
flows are the trip-makers, and meshes, the none-TOD areas (e.g. farmlands, wetlands, forest). In this
sense, a well-coordinated TOD network could also generate network synergy. The TOD network is
usually composed by club-type aspects and the web-type dimensions.

(a) (b)
Figure 18 Illustrations of the club type (a) and the web type (b) TOD networks

In terms of club-type network, TOD nodes of the same type may achieve horizontal synergy by jointly
contributing to the ridership of the shared transit line. Similar to the tennis club example in section 2.4, it is
not economical to build a railway line or a station for just one small, stand-alone municipality, considering
the potentially low ridership. But if a railway line will traverse and connect several relatively small
municipalities, together, the relatively small nodes of the municipalities can contribute a considerably high
level of ridership. As illustrated in figure18 (a), if all the small nodes on the line are residential nodes (thus,
they may have lower land use diversity) and the big one is a node with high attractiveness (e.g. high land use
diversity and provide high density of jobs), the small nodes might form a row of feeder nodes with a
common objective to access the resources in the big node and together generating stable ridership to the
line, which is corresponding to the intervening opportunities theory, as discussed in section 2.6, that the

20
number of people (in this case, ridership) going to a place (the big node) is in positive correlation to the
attractiveness of that place. And since the small nodes in figure 18 (a) are of the same type, the intervening
opportunities offered by the in-between nodes will be considerably limited. Thus, in the club type network,
the co-operation among the similar nodes can agglomerate the individual demands into a non-negligible
one, inviting services. As ridership is a key performance indicator of the railway service, this kind of co-
operation among the small nodes could be valuable to railway operation.

As for the web-type network aspects, a TOD network can generate the vertical synergy by the
complementation between two different types of TOD nodes in different aspects. For example, the
differentiation of different type to the nodes in terms of density can provide different groups of people their
preferable choices.

However, in this thesis, the emphasis of measuring the synergy in TOD network will be focus on
complementarity, the vertical synergy from web-type network, since (1) the co-operation between the nodes
is not the main interaction between the nodes in the network, considering such an interaction requires all
the mentioned conditions at the same time to stimulate travel demand; (2) such co-operation is hard to
quantify, thus, there is no suitable statistical method to measure and analyse co-operation among these
nodes. Nevertheless, measuring only complementarity based on the TOD typologies is still significant to
identify the TOD network synergy.
The TOD network in Arnhem and Nijmegen City Region fits well into the requirements of network synergy:
(1) for the requirement of ‘regional organising capacity’, the city region of Arnhem and Nijmegen is a
public body and inter-municipal institution; (2) as for ‘the overlapped geographical markets of demand’,
the city region has the ambition to become the second largest economic area in the Netherlands, aiming to
promote regional co-operation within the region; (3) in term of ‘spatial interaction’, the railway services
can be assured, as discussed in section 3.1. What remains to be measured and testified is the ‘differentiation
in the supply of activities or places’.
To summarize, in this thesis, measuring TOD network synergy will be carried out as measuring the
differentiation in the supply of activities or places.
3.5.2 Indicators to measure TOD network complementarity

Ideally, the 5D dimensions, which reflex the TOD characteristics, would structure a coherent framework
corresponding to the typology framework. However, not every aspect of TOD can be complementary in
terms of demands and supplies of the places or activities. Some of the aspects appear to be unshareable to
the other nodes, breaking the rule of ‘overlapped geographical markets’. For example, the dimension of
‘distance to transit’ measures the catchment area of a station, which cannot fit the rule of complementarity that
the demand market should be geographically overlapped. As for the ‘design’ dimension, it seems to be of
little significance to discuss how the street connectivity or the friendly walking environment of a particular
node complement those of the other nodes, because these built-environments tend to be exclusive. Only
the dimensions of density and diversity can represent the demands and supplies of activities or places and
at the same time meet the two fundamental rules of complementarity, which are ‘the overlapped
geographical markets of demand’ and ‘the differentiation in the supply of activities or places’. However, due
to the limitation of the data availability, the collected data of density (job density, population density) cannot
reflect well the characteristics of TOD network synergy, because these data are just the sum within the
buffer.
The Residential Housing Price (RHP) is, instead, the selected indicator to measure people’s housing
choice. The reasons are: (1) RHP is related to several variables, for example land price, the size of the house,
convenience degrees to different activities, etc., these variables lead to a variety of RHP, providing different
residential choice and catering different residential preference (as discussed in section 2.6, an important
aspects of complementarity is to provide different choices to different preference); (2) residential housing
fits the ‘places’ and ‘activities’ concepts as discuss in section 2.7, since the residential milieus are the ‘places’
providing residential function (activities) and usually the start point or the end point of a trip, thus RHP, a

21
reveal of the relationship between supplies and demands, can be considered an indicator to measure
complementarity among the nodes in the TOD network.
The data used in the analysis is from Dutch Statistics Office (CBS), it is a geographical dataset providing
information at the community level. Since the analysing unit in this thesis is the 800-metre buffer around
the station, the communities that are spatially intersected with a station buffer were assigned to the buffer.
The indicator of Building Use (BU)of the nodes is also adopted, because the building use distribution
indicate the association between node types and activity types. After analysing the correlation between the
node typology and the building uses, and the potential differentiation in activities can be identified.
The data source of building use is Basisregistratie Adressen en Gebouwen (BAG). The building within a station
buffer is assigned to the buffer. After that, in ArcGIS, the data can be summed up in a table, which is the
input in the following analysis.
It should be also indicated that complementarity is multi-dimensional (considering the varieties of ‘places’
and ‘activities’), the two indicators of RHP and BU are just two (albeit important) ones to demonstrate the
potential complementarity of a TOD network. Assumed that the ‘null hypothesis’ is ‘there is no
complementarity among the nodes in a TOD network’, if only an indicator could prove complementarity,
it is good enough to demonstrate the idea. Therefore, the point of the following exploratory analysis is not
to be exhaustive, but to test, since not all the dimensions of the nodes can be complementary. The
framework in figure 19 illustrates the method to identify and measure the complementarity in TOD network.

Residential Housing
Nodes Typology Building Use
Price

Dimension 1 Dimension 2

Correspondence
Analysis

Differentiation

Web Type Network

Complementarity of
TOD Network

Figure 19 Conceptual framework of identification and measurement of complementarity in TOD network.

3.6. Correspondence Analysis

Correspondence analysis (CA) in this thesis is used as a tool to measure the differentiation among the
TOD typology. The analysis was conducted with IBM SPSS Statistics (version 24). It is an analysis that aims
to visualize the nature of correlations between the column and the row of the contingency table. The final
visualization result of CA is a map with two axes, in which positions of the points (representing the
categories) reveal the correlations of the points. Followed are the basic steps to conduct CA, as well as its
fundamental formulas and the interpretation. The following kernel of CA is derived from Husson, Lê, &
Pagès (2010).

22
i. Data
The data required is a contingency table where the row and column respectively represent two qualitative
variables. Table 7 shows a sample data of CA, where variable 1 is the station types and variable 2, the types
of building uses, and value in the table means the number of building. The total sum of the buildings is 3519.
As we want to inspect the correlation between the two qualitative variables (in the case of table 7, how
different station type are associated with the building uses), the mathematical language is needed to explore
the potential correlation.
If we abstract the contingency table in a mathematical way, it could be shown in the way as figure 20, where
𝑥𝑖𝑗 is value of row i and column j, with the maximum row/column number as I/J. Therefore, in a table like
figure 20, the number of 𝑥𝑖𝑗 individuals are I × J (the table cells, in the case of table 7, it is 3×4=12).
Variable 2
Building Use
Social Health Care Industrial Office Sum Mean
Station Type
Suburban Residential Station 141 91 233 231 696 174
Variable 1 Urban Residential Station 250 115 149 393 907 227
Urban Mixed Core Station 572 42 459 843 1916 479
Sum 963 248 841 1467 3519 \
Mean 321 83 280 489 \ \
Table 7, sample data of CA

Variable V2

1 j J Sum

1
Variable V1

i Xij Xi

I XI

Sum Xj XJ N

Figure 20 Mathematical abstract of the contingency table

ii. Independence model

Thinking in a probabilistic way, total sum of the building can be seen as the sample size, which is 3519, then
in the probability of a building being in social category is the sum of social use buildings divided by the total
sum: 963/3519=0.274. In this way, the probability table can be calculated (see tables 8).

23
 Variable V2
Social Health Care Industrial Office Sum
Suburban Residential 0.040 0.026 0.066 0.066 0.198
Variable V1 Urban Residential 0.071 0.033 0.042 0.112 0.258
Urban Mixed Core 0.163 0.012 0.130 0.240 0.544
Sum 0.274 0.070 0.239 0.417 1
Table 8, probability table for the two variables

As a generally acknowledged rule of probability theory, if event A and event B are independent, the
probability of ‘A and B happen at the same time’ is equal to the probability of A times the probability of B,
which can be formulated as: 𝑃(𝐴|𝐵) = 𝑃(𝐴)×𝑃(𝐵). Again, take the social use buildings near the suburban
station for example (see tables 8), the probability for a building falling into social use type is 0.274 while the
probability for a building being a suburban residential type station is 0.198. According to probability theory,
if being a social use building is independent (set as event A) from being a building near suburban residential
station (set as event B), the probability of a building being social use building near suburban residential
station is 𝑃(𝐴|𝐵) = 𝑃(𝐴)×𝑃(𝐵) = 0.274×0.198 = 0.054 , which however, is not equal to the
corresponding value in table 8, which means the category of ‘social’ is not independent from the category
of ‘suburban residential’. The difference indicates correlation. To further analyse this indicated correlation,
the concept of chi-square distance is introduced.
Before looking into the chi-square distance, some mathematics derivation is needed. Since the probability
of an incident can be understand as the proportion of the sample, figure 20 can also be transfer into figure
21, wherein 𝑓𝑖. represents the probability of being in variable i, 𝑓𝑖. is the probability of falling into variable j,
and 𝑓𝑖𝑗 is the probability of being in row i, at the same time, in column j.

Variable V2

1 j J Sum

1
Variable V1

i fij fi.

Sum f.j 1

Figure 21, a probability table

iii. Chi-square distance

In the theory of statistics, the general idea of Chi-square test is to determine the difference between the
observed number and the theoretical number. In the case of correspondence analysis, this idea of Chi-square
is adapted to measure the deviation of the observed probability from the theoretical probability (as shown

24
in the below formula reasoning). In this light, the chi-square distance can be understood a distance
measuring the deviation from independence of the two variables, since the theoretical probabilities is based
on the hypothesis that the two variables are independent, so deviating from the theoretical probability means
deviating from independence between the two variables. The higher chi-square distance is, the further the
observed number is from the theoretical number, thus, the two variables from the corresponding row and
column are more independent (less correlated). In a nutshell, CA works with the table of probabilities and
aims to visualize the nature of correlation between the two variables.
𝐼 𝐽
2
(𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 𝑛𝑢𝑚𝑏𝑒𝑟 − 𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟)2
𝜒𝑜𝑏𝑠 =∑ ∑
𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎 𝑛𝑢𝑚𝑏𝑒𝑟
𝑖=1 𝑗=1

𝐼 𝐽 2
(𝑛𝑓𝑖𝑗 − 𝑛𝑓𝑖 ⋅ 𝑓𝑗 )
=∑ ∑
𝑛𝑓𝑖 ⋅ 𝑓𝑗
𝑖=1 𝑗=1

𝐼 𝐽
(𝑜𝑏𝑠𝑒𝑟𝑣𝑒𝑑 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦 − 𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎𝑙 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦)2
=∑ ∑ 𝑛
𝑡ℎ𝑒𝑜𝑟𝑒𝑡𝑖𝑐𝑎 𝑝𝑟𝑜𝑏𝑎𝑏𝑖𝑙𝑖𝑡𝑦
𝑖=1 𝑗=1

iv. Inertia
𝜒2
Statistically, inertia can be defined as 𝑛 , a measure without regard to the total number. It describes the
independence between the row categories and the column categories. The value of inertia is between 0 and
1, and it is usually far lower than 1. It would have the value of 1 of there would be an utterly exclusive
association between the row categories and the column categories. For example, if all the incidents of social
building uses belong to the urban mixed core nodes, at the same time the urban mixed core nodes do not
have the other building uses, the two categories have the ‘utterly exclusive association’, by which the inertia
= 1. In reality, situations are far away from such an extreme hypothesis. The land uses are, to some extent,
spread out well (albeit some may lean towards certain types) to the nodes, thus the value of inertia is far
lower than 1. On the contrary, the inertia value would be 0, if the observed probability is equal to the
theoretical probability, which mean the row categories and the column categories are completely
independent. For example, if the proportion of the different building uses are completely the same across
the three types of nodes, the inertia value will be 0.
The total inertia of the rows or the columns can be seen as a value measuring the general differentiation
among the categories in the rows or the columns. It would be interesting to compare the temporal data to
track the path of differentiation changes in time.
v. Graphical interpretation

As described by Meijers (2005):’Correspondence analysis is a technique to analyse the association between

rows and columns of a table or matrix by representing the rows and columns as points in a low-dimensional
Euclidean space (in practice, often a two-dimensional plot). Categories with similar distributions will be
represented as points that are close in space and categories that have very dissimilar distributions will be
positioned far apart.’
Till here, the introduction of correspondence analysis is enough to understand the result interpretation. For
further discussion of correspondence analysis, please see Jolliffe (2002)

25
26
 RESULTS
4.1. Variables Correlation

All the measurement variables are quantified by using ArcGIS. All the indicators are continuous variables.
Correlations between the variables were examined (Pearson’s correlation coefficient), and results do show
high and significant correlation between the variables (Table 9). But according to the fundamental
assumption of latent class analysis: the “local independence”, input indicators can be highly correlated while
output classes are conditionally independent of each other, which means they are not correlated within the
class (Masyn, 2013).
Table 9 Correlation matrix of the TOD variables

Impedance
Population

Land Uses
Intersectio

Pedestrian
Land Use

Bicycle +
n Density

Diversity
Business

Network
Walking
Density

Density

Density

Mixed-

Length
ness of
Job

Population
1 .562** .874** .748** -0,126 -0,077 -0,160 .797**
Density

Job
.562** 1 .681** 0,436 0,382 0,213 0,147 .518*
Density

Business
.874** .681** 1 .529* 0,073 0,196 0,117 .587**
Density

Intersection
.748** 0,436 .529* 1 -0,109 -0,333 -0,403 .876**
Density

Walking
-0,126 0,382 0,073 -0,109 1 0,220 0,174 0,114
Impedance

Land Use
-0,077 0,213 0,196 -0,333 0,220 1 .961** -0,244
Diversity

Mixed-ness
of Land -0,160 0,147 0,117 -0,403 0,174 .961** 1 -0,312
Uses

Length
Bicycle +
.797** .518* .587** .876** 0,114 -0,244 -0,312 1
Pedestrian
Network
*; ** Correlation is significant at the 0,05; 0,01 level (2-tailed)

4.2. TOD Typology

Table 10 presents the full model results. Model coefficients indicate each class’s mean value for each TOD
indicator, and associated statistical test. The percentages in parentheses correspond to each cluster’s
deviation in percentage terms from the sample mean. It is important to say that LCCM’s do not require
variable transformation, therefore the model outputs are interpreted based on the original units of each
variable.

27
 Table 10 Latent Class Cluster Model of TOD nodes

1. 2. 3. p-value R2
Measurement
Indicator Suburban Urban Urban Mixed
variables
Residential Residential Core

Population -2614,01 -582,40 3196,41

0,00 0,70
Density (-131%) (-14%) (41%)
-2466,79 -612,30 3079,09
Density Job Density 0,00 0,58
(-248%) (-21%) (47%)

Business -352,28 -238,99 591,27

0,00 0,86
Density (-195%) (-81%) (53%)

Distance to Walking -7,20 -15,81 23,01

0,79 0,04
transit Impedance (-1%) (-2%) (3%)

Land Use 0,003 -0,06 0,05

Diversity 0,06 0,18
Diversity (1%) (-28%) (17%)
Destinatio
n Mixed-ness of 0,005 -0,04 0,03
0,08 0,16
accessibilit Land Uses (3%) (-37%) (19%)
y

Intersection -45,96 23,84 22,12

0,00 0,66
Density (-46%) (14%) (13%)
Design
Length Bicycle -14853,73 5456,63 9397,10
+ Pedestrian 0,00 0,71
Network (-42%) (10%) (16%)

Percentage between parentheses corresponds to each class deviation from the sample mean.

28
 Figure 22, Typology of the TOD network of region Arnhem-Nijmegen

The LCCM clustered 20 train stations of the Arnhem-Nijmegen region into three TOD typologies. Base on
the characteristics of the typologies, names are given to enable the recognition and discussion: Suburban
Residential Station, Urban Residential Station and Urban Mixed Core Station (figure 22).
The Suburban Residential class contains 10 stations, illustrated in yellow in Figure 23. It is characterized by low
population density (131% smaller than the sample average) and low job and employment density
(respectively 248% and 195% smaller than the sample average). These stations also feature lower-than-
average intersection density (a measure for street connectivity), length of bicycle and pedestrian networks.
These features are predominantly residential, and located farther away from employment centres. They
might be, however, important as trip origins. For instance, station 4 (Zevenaar) ranks low in TOD-ness and
the absolute population is relatively low. However, it has a relatively high transit ridership, which accounts
for its important role within the TOD network as a feeder station. This means that the TOD-ness value
does not necessarily capture the significance of this station for the TOD network.

29
 TOD characteristics of Station 4
Population Density
1.000
Pedestrian Bicycle 0.800
Jobs Density
Path Length 0.600
0.400
0.200
Mixed-ness Index 0.000 Business Density

Intersection
Land Use Diversity
Density
Distance to Transit

Figure 23, TOD characteristics of Station 4. For illustration, the indicator values were normalized

The second typology was labelled as Urban Residential and it is characterized by lower destination accessibility
(37% smaller than the average) and lower diversity of land uses (28% smaller than the average) (table 10,
figure 24). However, it has more job opportunities when compared to the suburban residential class. The
design indicator scores higher for this class, meaning that the street design around the station is suitable for
cyclists and pedestrians. Seven stations were classified in this typology and Station 21 is taken as illustration.

TOD Characteristics of Station 21

Population Density
1.000
Pedestrian Bicycle 0.800
Jobs Density
Path Length 0.600
0.400
0.200
Mixed-ness Index 0.000 Business Density

Intersection
Land Use Diversity
Density
Distance to Transit

Figure 24, TOD characteristics of Station 21 (indicator values normalized)

The Urban Mixed Core class contains the three most urban stations, with all estimated indicators higher than
the sample mean (figure 25). These features are characteristic of central business districts and indeed this is
where these three stations are located. Station 1 is used to exemplify this station type.

30
 TOD Characteristics of Station 1

Population Density
1.000
Pedestrian Bicycle0.800
Jobs Density
Path Length 0.600
0.400
0.200
Mixed-ness Index 0.000 Business Density

Land Use Diversity Intersection Density

Distance to Transit

Figure 25, TOD characteristics of Station 1 (indicator values normalized)

Some unexpected classifications were encountered for some stations. For instance, station 6 was
hypothesized as a suburban residential instead of urban residential. This can be explained by the location of
the station within the urban fabric. Figure 26 shows that station 6 is located right in the middle of the town,
while station 19 is located at the edge of the town.

Figure 26, Location of stations 6 and 19 within the urban fabric.

In an attempt to validate the TOD typology, Table 11 discusses the relationship between the ridership of
each station (passenger count) with its TOD-ness value (TOD-ness). The assumption is that high ridership
would positively correlate with TOD-ness because one of the objectives of TOD is to dissuade travelers
from taking the automobile. However, by looking at the difference between passenger count rank and TOD-
ness rank, we can explain characteristics of the different stations by highlighting variables other than trip

31
generators, namely job density, diversity of usage, etc. The passenger count is a good predictor for TOD-
ness, but it is not the only, therefore it was used in this analysis as a control variable. For example, station 8
ranks 8th in relation to passenger count, but ranks 2nd is TOD-ness. It is located in the inner city, with the
university and the hospital in its surroundings, generating a lot of employment opportunities. The difference
in rank may be explained by the power of attraction of this station as an important business center, whereas
passenger counts are associated with origins of trips. Also, since this station is near to Arnhem central, which
has a higher train frequency, a lot of people cycle directly to this station, thus causing lower passenger counts
in station 8. The opposite is observed when analysing the differences in rank of station 4 (3rd in passenger
count and 16th in TOD-ness). This station is characterized mainly as a feeder station, performing low within
the built-form indicators.
Table 11 Validation of the Typology with Passenger Count Information

Passenger Passenger TOD-ness Difference in

Typology Station TOD-ness
Count count rank rank rank
3 1 43149 1 0,695 1 0
3 7 38442 2 0,656 3 1
3 8 3672 8 0,668 2 -6
2 2 3863 6 0,375 10 4
2 6 3865 5 0,392 8 3
2 9 3162 10 0,404 7 -3
2 15 2790 11 0,334 11 0
2 17 3287 9 0,426 5 -4
2 18 4214 4 0,458 4 0
2 21 2151 13 0,379 9 -4
1 3 732 18 0,304 14 -4
1 4 4652 3 0,280 16 13
1 10 482 20 0,226 19 -1
1 11 1625 14 0,405 6 -8
1 12 542 19 0,243 17 -2
1 13 907 17 0,228 18 1
1 14 3848 7 0,306 13 6
1 19 2250 12 0,282 15 3
1 20 1000 15 0,319 12 -3
1 22 925 16 0,140 20 4
Passenger count data was supplied from the Dutch train service provider (Nederlandse Spoorwegen – [Link]).

4.3. Complementarity in TOD Network

The method of correspondence analysis was discussed in section 3.5. In section 4.2, the typology of the
nodes in the Arnhem-Nijmegen TOD network is identified. Base on this typology, this section presents the
results of the identified complementarities of the TOD network in Arnhem-Nijmegen City Region by using
the data of ‘TOD node typologies vs. residential housing price’ and ‘TOD node typologies vs. building uses’
in correspondence analysis, IBM SPSS Statistics (version 24). Many other relations could be explored, for
example, job type or house type differentiation among the nodes. However, only Residential Housing Price
(RHP) and Building Use (BU) were selected based on data availability.
4.3.1 TOD node typologies vs. residential housing price

32
As shown in table 12, the inertia values are far lower than 1, indicated no utterly exclusive association (as
discussed in section 3.6) between the TOD node typologies and the RHP. However, the inertia of Urban
Mixed Core (0.056) is significantly higher than the other two (0.022 and 0.01), indicating a higher diversity
of the residential housing price in the Urban Mixed Core nodes. It means, within the nodes, Urban Mixed
Core nodes provide more choices in terms of residential housing price.
Table 12, inertia value for the CA of TOD node typologies vs. residential housing price

Station Type Inertia

Suburban Residential 0.022
Urban Residential 0.01
Urban Mixed Core 0.056
Total 0.087

Figure 27, TOD node typologies associated with Residential Housing Price

As an advantage of correspondence analysis, the result of CA is revealed not only in tables but also
graphically visualized. Figure 27 suggests the association between different type of TOD nodes and different
levels of RHP. In the graph, the origin (0,0) is the mean profile of the three TOD node typologies (as
discussed in section 3.6, the origin can also be understood as the theoretical probability wherein the node
typologies are independent from levels of RHP). If two node typologies lie close together, then their RHP
profile are more homogeneous, whilst the further they stay away, the more heterogeneous. The same
condition applies to the RHP: the points of RHP levels lie closer, the more similarly the levels of RHP

33
distributed between the node typologies. In this light, the point of medium RHP lying very close to the
point of the suburban residential node typology indicates the suburban residential nodes is characterized by
the medium-priced residential housing. And the urban mixed core typology is more or less closer to the
low-price residential housing, whereas, the three node typologies are all not highly associated with high-price
residential housing. These can be understood as: (1) since it is the housing price not the land price, the
average house size in the urban mixed core nodes might be smaller than the other two typologies, given the
prominently higher population density within the urban mixed core nodes (as shown in section 4.2, table
10), an assumption could be that in the urban mixed core nodes, there more multi-story apartments, lowering
the average housing price in the communities. With proper data (e.g. the housing type), such an assumption
can be tested; (2) the urban/ suburban residential nodes consist of less low-priced house and more medium-
priced house. (1) and (2) indicate the differentiation in the provision of different level of residential housing
among the three node typologies, therefore, indicate the complementarity among the three node typologies
in terms of residential housing price and that people with different preference in residential housing can
have more choices within the network than stay in one municipality.

4.3.2 TOD node typologies vs. building uses

Comparing 4.3.1, the inertia of TOD node typologies vs. building uses is relatively lower which indicate the
building uses among the three node typologies are relatively closer to the mean profile, whereas, the urban
mixed core nodes still have the highest inertia indicating the highest building use diversity, which is
consistent with what was shown on table 10.
Table 13, inertia value for the CA of TOD node typologies vs. building uses

Station Type Inertia

Suburban Residential 0.011
Urban Residential 0.007
Urban Mixed Core 0.017
Total 0.034

34
 Figure 28, TOD node typologies associated with building use

The visualization of TOD node typologies associated with building use in CA is shown in figure 28 and
some patterns are observed. First, the spread-out points can be explicitly grouped into 3 categories according
to the node typologies, indicating strong building use differentiation among the three typologies. Compared
to the other two, urban mixed core is much closer to the building use points of office, social, commercial,
logistic and industrial, which indicates, as expected, a relatively dominant position of this typology in these
business and commercial activities. Second, the residential use is the closest to the origin (0,0), meaning the
three node typologies have more or less similar proportion of residential building, whilst the residential point
is relative closer to urban residential, confirming that urban residential have higher residential proportion.
Third, sport uses are closer to suburban residential typology, which implies that suburban nodes have larger
tract of land to provide sporting ground. Additionally, health care use should also be noticed, that it is
relatively further away from the urban mixed cored nodes in the result graph, indicating more health care
facilities are located around residential type nodes.
The distinct differentiation of building uses among the nodes strongly indicates and confirms different roles
of the different TOD node types in terms of the provision of activities. Therefore, within the TOD network,
the complementarity of activities aroused. Travelling within the TOD network in the City Region of
Arnhem-Nijmegen, passengers can access more activities according to their needs.

35
36
 DISCUSSION AND CONCLUSION
5.1. Discussion
In this thesis, a TOD typology was developed to identify the characteristics of different nodes in the transit
network. Based on the developed typology, synergy effects between TOD nodes were analysed to
understand how their effects contribute to the large-scale TOD network. Three main questions were
addressed: (1) whether the TOD-ness values could properly reflect the performance across different nodes;
(2) what are the roles of the nodes within the TOD network; and (3) whether there are synergy effects
between TOD nodes within a TOD network.

The measurement of TOD-ness in nodes (at the nodes) in the Arnhem-Nijmegen City Region was based
on the 5D built-environment dimensions. The results of the analysis indicated variation in TOD-ness among
the nodes, as it is similarly demonstrated by Singh (2015). However, when TOD-ness was compared to
station ridership, some stations show contradictions in the rankings, as stations that rank low in TOD-ness
can still have good performance in term of ridership. This phenomenal ranking difference supports the first
claim of the research that a general TOD-ness value might not be able reflect the performance every node.
It also indicates that the nodes might not need to be perfect in all the dimensions of TOD.
The above indication leads to second research problem addressed in this study, which was to investigate the
roles of the nodes within the TOD network. Latent Class Clustering Method (LCCM) was conducted to
develop typologies for TOD nodes. Three typologies were identifiable, and the reason might be that the
network is relatively small and dominated by two very strong core stations, which influence the
characteristics of the remaining stations in the network. The distribution of the three types of nodes suggests
a hierarchical structure of the stations, where the typologies of urban residential and suburban residential
serves around the urban mixed core type. The LCCM applied in this thesis allowed us to reduce the
complexity inherent to each station environment, by using a means approach to analyse the clustering
outputs.
Developing the typology was the first step, to further understand the roles of the nodes within the TOD
network; the next step was to identify the interaction between the nodes. Transferring the concept of
network synergy from economic field, the interaction between the nodes can be co-operation and
complementarity, accordingly, the sub-network are the club type network and web type network. Both
interactions between the nodes can contribute to relative network synergy. However, due to the reason that
the horizontal synergy is hard to quantify and conditional, this thesis focuses on the vertical synergy of
complementarity.

According to Meijers (2005), to achieve complementarity, two fundamental rules should be followed: (1)
differentiation in the supplies of activities or places; (2) geographically overlapping market of demands. To
measure complementarity between the TOD nodes, the 5D concepts were first considered as a framework
to keep consistency with the TOD-ness measurement. However, the rules of complementarity constrain the
dimension of design, distant to transit and destination accessibility. As complementarity between the nodes
is the complementation of supplies and demands of activities and places, the indicators of residential housing
price and building use are selected. And given that Arnhem-Nijmegen City Region runs as a planning body
and that the TOD network enables the spatial interaction between the nodes, this thesis argues that
measuring complementarity between the nodes can be done by measuring its differentiation.
Correspondence analysis was considered to be the right tool to analyse differentiation since it is based on a

37
probabilistic method to depict the association between two variables of a contingency table. And the result
of the analysis of TOD typology versus housing prices (RHP) suggested the urban mixed core nodes offer
more choices in lower-price residential housing, which is inconsistent with the usual perception. Since the
data is based on the house unit, factors like housing condition or house size are not considered. A good
replacement of this data would be the housing type. But due to the data limitation, RHP was adopted to
reveal somehow the differentiation in housing choices and trigger discussions. On the other hand, the result
of ‘TOD typology versus Building Use’ suggests strong differentiation between the node types in terms of
building uses. As the nodes are connected by the rapid transit service, the differentiation in building use
between the nodes add to the total building use diversity of the nodes, which fit well the complementation
of diversity. Undeniably, selecting more factors to measure complementarity is necessary to fully
demonstrate how the TOD nodes could complement each other between different typologies with the
possible spatial interaction enabled by the rapid transit service.

As a result, the nodes interaction based on the network synergy theory indicates the roles of the nodes within
a TOD network, that they can be co-operative and contribute to the ridership, they can be complementary
in terms of supplies and demands of places and activities. Therefore, with the complementarity from the
other nodes, one node does not need to be ideal in some of the dimensions. Outside of the network synergy
theory, complementarity between the nodes can also be generally seen as providing services that the other
nodes do not have. For example, at the urban mixed core station, there are more trains service, especially
the Intercity train travelling outside the region. The residential typologies nodes are complemented as
passengers from the residential type station transferring at the urban mixed core station to access the
Intercity service. Consequently, the analysis suggested that there are synergy effects among the nodes in
Arnhem-Nijmegen TOD network.

5.2. Limitation
To some extent, the results of these analyses are also limited due to the selection of the spatial unit of 800m-
radius buffer. This area refers to a 10-minute walking distance. However, the catchment area of a station
can be even larger, considering cycling as a feeder mode to transit services. Moreover, since cycling lanes or
bus lines are other form of other types of network, the overlay of these networks smooths the rigidness of
a single network. Especially in the case of this thesis, bicycle shares 35% of the modal split and the number
of bicycles per inhabitant reaches 1.1 bikes per inhabitant, which means averagely every person has more
than one bike (Statistics Netherlands, 2015). The reason to conduct the analysis of single railway network is
twofold: (1) Considering the relatively higher speed of bicycle, the cycling network can enlarge the buffers
and even make the buffers of near stations overlapped. The overlapping will lead to competition of stations
in the access trips, which is much influenced by the traveler perception; (2) simplifying the complexity in
reality can enable a better understanding of how a regional railway system can promote TOD and achieve
network synergy by setting the roles of the nodes.
In this thesis, due to the data unavailability, only the data about passenger count was used to validate the
typology, however other measures can be used, namely commute mode share and travel characteristics.
Ideally, information about inbound and outbound stations, i.e., the station origin-destination information
(the O-D pair of the train trip), would allow the identification of station correlations. However, the Dutch
train service provider does not make such information readily available.

5.3. Future study

For future studies, extending the station buffer to simulate the catchment area for bicycle will be
recommended. Such an extension will lead to the overlapping of the buffers, which will then lead to the
analysis of competition between stations in the access trips. It is vital to understand different relationship
between the stations (for example competition or subordination). As the buffers are extended, the unit to
calculate built-environment indicator will also be enlarged, including more areas which might contain

38
different characteristics from the kernel buffers. The increased complexity need to be treated carefully as
traveller perception or station level might play a role in assigning the overlapped area to the nodes.

39
LIST OF REFERENCES
Arnhem-Nijmegen Metropolitan area. (2013). Retrieved from [Link]
nijmegen-city-region
Atkinson-palombo, C., & Kuby, M. J. (2011). The geography of advance transit-oriented development in
metropolitan. Journal of Transport Geography, 19(2), 189–199. [Link]
Austin, M., Belzer, D., Benedict, A., Esling, P., Haas, P., Miknaitis, G.,Wampler, E., Wood, J., Young, L. &
Zimbabwe, S. (2010). Performance-based transit-oriented development typology guidebook. Center for Transit-Oriented
Development, Oakland, CA.
Baker, D., Washington, S., & Turrell, G. (2014). Advance transit oriented development typology : case
study in Brisbane , Australia. JOURNAL OF TRANSPORT OF GEOGRRAPHY, 34, 54–70.
[Link]
Calthorpe, P. (1993). The next American metropolis: Ecology, community, and the American dream. Princeton
architectural press. New York.
Capello, R., & Rietveld, P. (1998). The Concept of Network Synergies in Economic Theory: Policy
Implications. Transport Networks in Europe: Concepts, Analysis, and Policies.
Capineri, & Kamann. (1998). Synergy in Networks: Concepts. In Transport Networks in Europe:
Concepts, Analysis, and Policies (pp. 35–56).
Cervero, R., & Day, J. (2008). Suburbanization and transit-oriented development in China. Transport Policy,
15 (5), 315–323.
Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: density, diversity, and design.
Transportation Research Part D: Transport and Environment, 2 (3), 199–219.
Cervero, R., & Murakami, J. (2009). Rail and property development in Hong Kong: experiences and
extensions. Urban Studies, 46, 2019–2043.
Cervero, R., Murphy, S., Feerell, C., Goguts, N., & Tsai, Y.-H. (2004). Transit-oriented development in the United
States: Experiences, challenges, and prospects. Transit Cooperative Research Program (TCRP) Report 102.
Transportation Research Board.
Dittmar, H., & Poticha, S. (2004). Defining transit-oriented development: the new regional building block.
In: The New Transit Town: Best Practices in Transit-Oriented Development, 20–55.
Evans, J. E., Stryker, A, Kuzmyak, R. & Pratt, R. H. (2007). Transit oriented development: traveler response to
transportation system changes. Transit Cooperative Research Program (TCRP) Report 95. Transportation
Research Board.
Ewing, R., & Cervero, R. (2001). Travel and the built environment – a synthesis. Transportation Research Record:
Journal of the Transportation Research Board, 1780, 87-114.
Ewing, R., & Cervero, R. (2010). Travel and the built environment: A Meta-Analysis. Journal of the American
Planning Association, 73 (6) 265–294.
Ewing, R. H. (2008). Characteristics, causes, and effects of sprawl: A literature review. In: Urban Ecology: an
international perspective on the interaction between humans and nature, 21 (2), 519–535.
Frank D, L. (2000). Land use and transportation interaction: implications on public health and quality of
life. Journal of Planning Education and Research, 20 (1), 6–22.
Geography of the Arnhem-Nijmegen region. (2015). Retrieved from
[Link]
region.

40
Guerra, E., Cervero, R., & Tischler, D. (2013). Half-mile circle: does it best represent transit station
catchments? Transportation Research Record: Journal of the Transportation Research Board, 2276, 101–109.
Hancock, M. M. B., Shepherd, D. S., Faatz, D. J., Lopez, D. P. D., Lehmann, D. P., Burns, A., … Duffany,
B. (2014). Transit Oriented Denver, Tod Strategic Plan 2014. Denver.
Higgins, C. D., & Kanaroglou, P. S. (2016). A latent class method for classifying and evaluating the
performance of node transit-oriented development in the Toronto region. Journal of Transport Geography, 52,
61–72.
Higgins, C. D., & Kanaroglou, P. S. (2016). A latent class method for classifying and evaluating the
performance of station area transit-oriented development in the Toronto region. JTRG, 52, 61–72.
[Link]
Husson, F., Lê, S., & Pagès, J. (2010). Exploratory multivariate analysis by example using R. CRC press.
Jolliffe, I. T. (2002). Principal Component Analysis, Second Edition. Encyclopedia of Statistics in
Behavioral Science, 30(3), 487. [Link]
Kamruzzaman, M., Baker, D., Washington, S., & Turrell, G. (2014). Advance transit oriented development
typology: Case study in Brisbane, Australia. Journal of Transport Geography, 34, 54–70.
Masyn, K. E. (2013). Latent Class analysis and finite mixture modeling. The Oxford Handbook of Quantitative
Methods, 2, 551–611.
Meijers, E. (2005). Polycentric urban regions and the quest for synergy: Is a network of cities more than
the sum of the parts? Urban Studies, 42(4), 765–781. [Link]
Meijers, E., & Romein, A. (2003). Realizing potential: Building regional organizing capacity in polycentric
urban regions. European Urban and Regional Studies, 10(2), 173–186.
[Link]
Musterd, S., & van Zelm, I. (2001). Polycentricity, Households and the Identity of Places. Urban Studies
(Routledge), 38(4), 679–696. [Link]
Pojani, D., & Stead, D. (2014). Dutch planning policy: The resurgence of TOD. Land Use Policy, 41, 357–
367.
Ritsema van Eck, J., & Koomen, E. (2008). Characterising urban concentration and land-use diversity in
simulations of future land use. Annals of Regional Science, 42 (1), 123–140.
Singh, Y. J. (2015). Measuring transit-oriented development ( TOD ) at regional and local scales – a planning support tool.
PhD dissertation, University of Twente, Enschede.
Statistics Netherlands. (2015). Transport and Mobility 2015. Statistics Netherlands.
Stouffer, S. A. (1940). Intervening opportunities: a theory relating mobility and distance. American
Sociological Review, 5(6), 845–867.
Ullman, E. L. (1956). The role of transportation and the bases for interaction. In W. L. Thomas (Ed.),
Man’s Role in Changing the Face of Earth (pp. 862–880). Chicago, IL: The University of Chicago Prss.
U.N. Habitat (2009). Global report on human settlements: planning sustainable cities. Lodon, Earthscan.
Zhang, Y., & Guindon, B. (2006). Using satellite remote sensing to survey transport-related urban
sustainability. Part 1: Methodologies for indicator quantification. International Journal of Applied Earth
Observation and Geoinformation, 8 (3), 149-164.

41
42