Visualisation of trip chaining behaviour and mode choice using household travel survey data

Abstract

Planning for transport infrastructure requires forecasting of future travel demand. Various factors such as future population, employment, and the travel behaviour of the residents drive travel demand. In order to better understand human travel behaviour, household travel surveys—which require participants to record all their trips made during a single day or over a whole week—are conducted. However, the daily travel routines of people can be very complex, including routes with multiple stops and/or different purposes and often may involve different modes of transport. Visualisations that are currently employed in transport planning are, however, limited for the analysis of complex trip chains and multi-modal travel. In this paper, we introduce a unique visualisation approach which simultaneously represents several important factors involved in analysing trip chaining: number and type of stops, the quantity of traffic between them, and the utilised modes of transport. Moreover, our proposed technique facilitates the inspection of the sequential relation between incoming and outgoing traffic at stops. Using data from the South-East Queensland Travel Survey 2009, we put our developed algorithm into practice and visualise the journey-to-work travel behaviour of the residents of inner Brisbane, Australia. Our visualisation technique can assist transport planners to better understand the characteristics of the trip data and, in turn, inform subsequent statistical analysis and the development of travel demand models.

Full text

1 23
Public Transport
Planning and Operations
ISSN 1866-749X
Public Transp
DOI 10.1007/s12469-018-0183-5
Visualisation of trip chaining behaviour
and mode choice using household travel
survey data
Günter Wallner, Simone Kriglstein,
Edward Chung & Syeed Anta Kashfi

1 23
Your article is protected by copyright and
all rights are held exclusively by Springer-
Verlag GmbH Germany, part of Springer
Nature. This e-offprint is for personal use only
and shall not be self-archived in electronic
repositories. If you wish to self-archive your
article, please use the accepted manuscript
version for posting on your own website. You
may further deposit the accepted manuscript
version in any repository, provided it is only
made publicly available 12 months after
official publication or later and provided
acknowledgement is given to the original
source of publication and a link is inserted
to the published article on Springer's
website. The link must be accompanied by
the following text: "The final publication is
available at link.springer.com”.

Vol.:(0123456789)
Public Transport
https://doi.org/10.1007/s12469-018-0183-5
1 3
ORIGINAL PAPER
Visualisation oftrip chaining behaviour andmode choice
using household travel survey data
GünterWallner1· SimoneKriglstein1,2· EdwardChung3,4 · SyeedAntaKash4
Accepted: 31 July 2018
© Springer-Verlag GmbH Germany, part of Springer Nature 2018
Abstract
Planning for transport infrastructure requires forecasting of future travel demand.
Various factors such as future population, employment, and the travel behaviour
of the residents drive travel demand. In order to better understand human travel
behaviour, household travel surveys—which require participants to record all their
trips made during a single day or over a whole week—are conducted. However, the
daily travel routines of people can be very complex, including routes with multiple
stops and/or different purposes and often may involve different modes of transport.
Visualisations that are currently employed in transport planning are, however, lim-
ited for the analysis of complex trip chains and multi-modal travel. In this paper,
we introduce a unique visualisation approach which simultaneously represents sev-
eral important factors involved in analysing trip chaining: number and type of stops,
the quantity of traffic between them, and the utilised modes of transport. Moreover,
our proposed technique facilitates the inspection of the sequential relation between
incoming and outgoing traffic at stops. Using data from the South-East Queensland
Travel Survey 2009, we put our developed algorithm into practice and visualise the
journey-to-work travel behaviour of the residents of inner Brisbane, Australia. Our
visualisation technique can assist transport planners to better understand the charac-
teristics of the trip data and, in turn, inform subsequent statistical analysis and the
development of travel demand models.
Keywords Trip chain· Multi-modal travel· Trip scheduling· Human travel
behaviour· Visualisation· Household travel survey
* Edward Chung
edward.cs.chung@polyu.edu.hk
1 Institute forDesign andAssessment ofTechnology, Vienna University ofTechnology,
Argentinierstrasse 8, 1040Vienna, Austria
2 Center forTechnology Experience, AIT Austrian Institute ofTechnology GmbH, Giefinggasse
2, 1210Vienna, Austria
3 Department ofElectrical Engineering, Faculty ofEngineering, Hong Kong Polytechnic
University, Hung Hom, Kowloon, HongKong
4 Queensland University ofTechnology, 2 George Street, BrisbaneQLD4000, Australia
Author's personal copy

G.Wallner et al.
1 3
JEL Classication R29· R41· R42
1 Introduction
Effective transport planning is crucial for building modern, productive, and environ-
mentally sustainable cities and regions. Therefore, understanding the travel activities
of the population is of particular importance as it helps to guide the planning of
transport infrastructure and services in order to make our communities better places
to live. Visualisation techniques used in the field of transportation planning aim to
represent the complex nature of human travel activities. They are a means to facili-
tate and clarify the understanding of multifaceted information in relation to trans-
portation. Transportation visualisation, according to Manore (2007), can be defined
as “any progressive visual means of representing static or temporal spatial and geo-
metric information”. In the context of transport planning, visualisation techniques
are an important aspect not only for analysing data (e.g., by means of Geographic
Information Systems—GIS) but also for clearly conveying the outcomes of trans-
portation planning to policy makers or the general public. Lately, visualisations in
transportation are also increasingly employed in interactive public transport journey
planners such as the journey planner of London (Transport for London), the Public
Transport Victoria journey planner (Public Transport Victoria), the Rail travel plan-
ner Europe (Rail Europe—Rail travel planner Europe), or the TransLink Journey
planner Brisbane (TransLink Journey Planner) to name but a few examples.
While individual trips have long served as the basic unit for analysing travel
behaviour, there has been a shift of focus to journeys—so-called trip chains—
as those are deemed to better reflect the actual travel demands (Ma etal. 2014).
Trip chaining arises for several reasons including the desire of people to search for
ways to perform multiple activities in a single journey within less time and travel
distance (Shiftan 1998; Hensher and Reyes 2000). Trip chaining behaviour consti-
tutes a complex phenomenon influenced by a variety of variables (McGuckin and
Murakami 1999; Ma etal. 2014) such as household characteristics. This is further
complicated by the fact that the individual trips themselves can be complex and
often involve different or multiple modes of transport. Islam and Habib (2012) state
that trip chaining and mode choice are two critical factors influencing various pat-
terns of urban travel demand. In many instances increasing the complexity of trip
chains leads to higher auto dependency (Strathman etal. 1994; Wallace etal. 2000;
Ye etal. 2007) and to spreading the urban peak periods (Ye etal. 2007; Habib etal.
2009). Generally, trip chaining is prevalent among workers in urban areas. Anal-
ysis of trip chains so far has mainly concentrated on the home-to-work commute
(McGuckin and Murakami 1999; Primerano etal. 2008). The mandatory work jour-
ney serves as an important anchor in many households around which the daily travel
activity is scheduled (Hensher and Reyes 2000; Xianyu 2013). Most of the previous
Journey-to-Work (JTW) studies applied statistical analysis to understand the pro-
portions and types of trip chains (e.g., McGuckin and Murakami 1999; Primerano
etal. 2008; Islam and Habib 2012) or to estimate the role of various demographic
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
or socioeconomic characteristics on the daily travel behaviour (e.g., Strathman etal.
1994; Hensher and Reyes 2000; Ma etal. 2014).
Understanding such multi-modal trip chaining through an appropriate visualisa-
tion technique has become an increasingly important aspect in transport planning
to optimise public transport and to contribute to sustainable urban transporta-
tion (Fiorenzo-Catalano et al. 2004). However, visualisation techniques currently
employed in transport planning mostly focus on depicting the quantity of trips
between zones, for example, by means of choropleth or pie chart maps (Corcoran
etal. 2009; Xu etal. 2011) or by using desire lines (Xu and Milthorpe 2010). Other
visualisations, in turn, focus on the travel between different stations but often restrict
themselves to single modes of transport, such as rail or bus travel (Xu etal. 2011;
Nagel etal. 2014). Thus, these visualisations are limited for analysing series of trips
or multi-modal trips.
The visualisation technique introduced in this paper aims to provide an overview
of human travel behaviour and to facilitate the analysis of such multi-modal travel
and trip chaining. Its aim is to convey how different activities using different modes
of transport are linked together during a trip (or trip chain) in order to reach a des-
tination in a single picture. As a use case we visualise the JTW travel behaviour of
residents of inner Brisbane based on Household Travel Survey (HTS) data of South
East Queensland (SEQ) from 2009.
The remainder of this paper is structured as follows. After highlighting our
research objectives and significance, we provide a detailed literature review of
existing studies on trip chaining and multi-modal human travel behaviour and the
employed visualisation approaches. Section3 gives a detailed description of the
algorithm and the graphical representation. Section 4 provides a brief description
of the study area and the employed data. In Sect.4 we also apply the visualisation
approach to data from the SEQ Travel Survey 2009 (SEQTS09) and interpret the
results. Before the paper is concluded in Sect.6, Sect. 5 discusses implications of
the developed visualisation approach in theoretical and practical settings as well as
the runtime of the algorithm. Section5 also explores avenues for future research.
1.1 Research objectives
Our literature review presented in Sect.2 suggests that there has been little research
on a systematic way of visualising HTS data to provide an overview of the com-
plexity of trip making such as trip chaining and travel mode choice (single or
multi-modal). This paper contributes to filling this gap by proposing a visualisation
technique that can be used to analyse trip chains, multi-modal travel, or both in com-
bination. To achieve this goal this paper makes the following contributions:
• It provides a review of visualisation techniques currently employed in studies
dealing with trip chaining and multi-modal travel behaviour.
• It proposes a new visualisation technique for analysing complex human travel
behaviour, specifically trip chaining and multi-modal travel. For this purpose, the
Author's personal copy

G.Wallner et al.
1 3
visualisation shows several important variables such as frequency of trips, mode
choice, as well as the number and succession of stops in a single drawing.
• It demonstrates the applicability of our approach by applying it to real-world
HTS data from Brisbane, Australia from 2009 to highlight how transport plan-
ners and agencies can benefit from it.
1.2 Research signicance
This study aims to make a contribution to the complex task of modelling trip behav-
iour. Our proposed visualisation can help to unlock stories behind the trip data or
present information in an easy-to-digest manner and, in turn, can aid transport plan-
ners in forming hypotheses for subsequent statistical analysis. It can also inform the
creation of transport models used in urban studies to forecast travel demand and
travel behaviour (Bhat and Koppelman 1999). Commonly used models like agent-
based modelling (Monteiro et al. 2014) or tour-based travel demand modelling
(Omer etal. 2010) rely on personal trip data. Understanding how trips are performed
is therefore essential for building such models.
2 Literature review
This section demonstrates various visualisation methods that were adopted in pre-
vious studies concerned with trip chaining and/or travel mode choice (single or
multi-modal).
Using 2011 census data on modes of travel to work in England and Wales, Leve-
son (2013) used bar charts, stacked bar charts, and pie charts to compare different
modes of travel and their changing patterns over the years. Adopting a different
approach, Xu and Milthorpe (2010) analysed JTW travel patterns in Sydney using
data from 1981 to 2006. Besides using bar charts to compare the share of differ-
ent modes of travel and line charts to depict the relation between trip length and
distance between home and workplace, desire lines were used to visualise the traf-
fic volumes between origin–destination pairs. Taken together, these desire lines can
be viewed as a (directed or undirected) weighted graph with the weights represent-
ing the traffic volume. Xu etal. (2011), in turn, focused specifically on the analy-
sis of travel patterns of rail users based on data from the Sydney Household Travel
Survey. Again, stacked bar and line charts were used for analysing trip lengths. In
addition, pie chart maps—pie charts superimposed over a map—were created to
visualise the modes of transport used to access the stations (e.g., by bus or on foot).
Using Nationwide Personal Transportation Survey data, McGuckin and Murakami
(1999) investigated the types of trip chains made by adult men and women. This
study made use of bar charts to compare the trip chaining behaviour between men
and women on their home-to-work or work-to-home journey with respect to the
number and purpose of stops on their trips. Unlike the above studies, which made
use of static charts, Nagel etal. (2014) proposed a multi-touch tabletop application
to analyse specific aspects pertaining to bus travel in Singapore’s bus network. For
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
example, a map representation was used to show the number of alighting and board-
ing passengers at the different bus stops by means of two concentric circles. In addi-
tion, arc diagrams were used to depict the number of passengers travelling between
stops to assess passenger flow and connectivity of bus stops. Recently, Sun etal.
(2016) discussed visualisation methods for representing aggregated passenger flow
characteristics between stations using the Shanghai Metro as a use case. A radial
node-link diagram was used to convey the amount of passenger flow between sta-
tions. However, unlike our visualisation intermediate stops are not discernable. In
general, the studies mentioned above either focus only on a single mode of travel or
on the utilization of different transport modes whereas our concern is to also facili-
tate the analysis of multi-modal travel. It shows how different modes of transport are
linked together during a trip (or trip chain) to reach a destination.
While above-mentioned studies explored either trip chaining or multi-modal
travel behaviour, other studies analysed both (McGuckin et al. 2005; Walle and
Steenberghen 2006; Golob and Hensher 2007; Hensher 2007; Ye etal. 2007; Currie
and Delbosc 2011; Islam and Habib 2012; Harney and Rajoo 2015). For example,
McGuckin etal. (2005) used bar and stacked bar charts to examine the percentage
of stops during commutes to comprehend trip chaining trends in the United States.
Harney and Rajoo (2015) analysed how intermediate stops, mode choice, and tim-
ing of trips differ depending on the tours undertaken in the South East Queensland
and Cairns regions. 3D surface plots were used to depict the temporal distributions
of the tours. Golob and Hensher (2007) as well as Hensher (2007) used three-year
Sydney travel survey data to investigate senior citizens’ trip chaining travel activity
(either work or non-work-centric trips). Line graphs were employed to illustrate the
average home-based trip chains based on different demographic attributes such as
age, gender, living circumstance, trip purpose, complexity of trips, and mode. Simi-
larly, Ye etal. (2007) examined the relation between mode choice and trip chaining
behaviour in the context of multi-stop (complex) vs. single-stop (simple) trips. Their
results suggest that the complexity of the trip chaining patterns drives mode choice.
Table1 offers an overview of studies including the various visualisation techniques
used to represent data.
As Table 1 shows, most of the studies used either descriptive statistics or
advanced statistical techniques to perform their analysis. However, mainly bar
charts, pie charts, scatter plots, line graphs, or simple tables were used to visualise
data such as the percentage of used travel modes and the stops/number of legs per
trip chain, sometimes in combination with the activities at each stop. It also shows
that it is uncommon to find visualisation techniques specifically intended for vis-
ualising trip-chains (single or complex), activities, and/or mode choice (single or
multi) in transportation research. Furthermore, it is rare to find studies that focus on
trip chaining and multi-modal travel behaviour simultaneously. Hence, our goal is to
facilitate the analysis of multi-modal travel, that is, how different modes of transport
are linked together during a trip (or trip chain) in order to reach a destination by
means of a visualisation technique specifically designed for this purpose.
Apart from the scientific literature reviewed above there are also several tools
for visualizing trips. For example, TransCAD (Caliper Corporation) is a transpor-
tation planning software, which provides different visualisation methods (e.g., pie
Author's personal copy

G.Wallner et al.
1 3
Table 1 Previous studies related to trip chaining or multi-modal travel behaviour or both
Author and year Study location Data used Subject Analysis method Visualisation technique(s)
Leveson (2013) England and Wales Census Data, 2011 Method of travel to work Descriptive statistical
analysis
Bar charts, stacked bar
charts, pie charts,
choropleth maps
Xu etal. (2011) Sydney, AU Sydney Household Travel
Survey
Mode share (single), access
mode to rail station, trip
length
Descriptive statistical
analysis
Bar charts, stacked bar
charts, line graphs, scatter
plots, tables, pie chart
maps
Xu and Milthorpe (2010) Sydney, AU JTW Data (1981–2006) Mode share (single or mul-
tiple) and trip length
Descriptive statistical
analysis
Bar charts, stacked bar
charts, pie charts, line
graphs, pie chart maps,
desire lines
McGuckin etal. (2005) US Nationwide Personal Trans-
portation Survey 1995 &
National HTS 2001
Trip chaining patterns and
travel patterns of com-
muters
Descriptive statistical
analysis
Bar charts, line graphs,
tables
Walle and Steenberghen
(2006)
Belgium Belgian Mobility Survey,
1998–1999
Trip chaining and mode
choice
Regression model Bar charts, choropleth maps
McGuckin and Murakami
(1999)
US Nationwide Personal Trans-
portation Survey, 1995
Trip chaining behaviour
between men and women
Descriptive statistical
analysis
Bar charts
Currie and Delbosc (2011) Melbourne, AU Melbourne Household
Travel Survey Data,
1994–1999
Complexity of trip chaining
behaviour
Statistical significance
analysis using t test
Bar charts
Harney and Rajoo (2015) SEQ and Cairns, AU SEQ and Cairns Travel
Survey
Mode choice, trip timing,
length and purpose of
stops
Descriptive statistical
analysis
Stacked bar charts, 3D sur-
face plots, tables
Hensher (2007) Sydney, AU Sydney Household Travel
Survey, 2002–2004
Individual trip chaining
travel activity
Statistical analysis, nested
logit model
Line graphs
Golob and Hensher (2007) Sydney, AU Sydney Household Travel
Survey, 2002–2004
Trip chaining travel activity Multiple correspondence
analysis
Line graphs
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
Table 1 (continued)
Author and year Study location Data used Subject Analysis method Visualisation technique(s)
Alsnih and Hensher (2005) Sydney, AU Sydney Household Travel
Survey, 2002
Mode of travel, trip chain-
ing patterns of aging
people
Descriptive statistical
analysis
Tables
Islam and Habib (2012) Switzerland Six-Week Travel Diary
Data, 2003
Hierarchical trip chaining
and mode choice
Structural equation model-
ling (SEM)
Tables
Primerano etal. (2008) Adelaide, AU Household Travel Surveys
Metropolitan Adelaide,
1999
Trip chaining behaviour of
households
Descriptive statistical
analysis
Tables
Xianyu (2013) Beijing, CN Household Travel Surveys,
2005
Trip chaining and mode
choice of the home-based
work tour
Co-evolutionary approach
combining two multi-
nominal logit models
Tables
Strathman etal. (1994) Portland, OR, US Household Travel Survey,
1985
Trip chaining Logit model Tables
Ye etal. (2007) Switzerland Swiss Microcensus Travel
Survey, 2000
Mode choice and com-
plexity of trip chaining
patterns
Tour-based and activity-
based travel demand
modelling systems
Tables
Philip etal. (2013) Kerala, IN Interviews based on ques-
tionnaire
Mode choice behaviour Multinominal logit model Pie charts
Ma etal. (2014) Beijing, CN Beijing Activity Diary
Survey 2007 and Land
Use Data
Trip chaining Multinominal logit models Scatter plots, spider graphs,
choropleth maps, tables
Author's personal copy

G.Wallner et al.
1 3
chart maps or weighted node-link diagrams superimposed over maps). These can
be used to analyse, for instance, trip generation and distribution. The JTW Visu-
aliser (Bureau of Transport Statistics, New South Wales) is an interactive tool for
visualising trip flows. In contrast to TransCAD it does not use map-based repre-
sentations but makes use of a radial graph layout to show the outgoing or incom-
ing traffic between a certain region of interest and other suburbs. When select-
ing an edge, the utilized modes of travel are depicted next to the graph using
bar charts. However, as the bar charts are only displayed for the current selec-
tion it is not easy to compare the modes of travel for different origin–destination
pairs. While the JTW Visualiser shares the abstract representation of trips with
our approach, its focus on binary links between suburbs makes it not well suited
for analysing trip chains. Another publicly available interactive data visualisa-
tion facility is offered by the Victorian Integrated Survey of Travel and Activ-
ity (Department of Economic Development, Jobs, Transport and Resources 2016)
which allows users to explore Melbourne’s JTW information by region (inner,
middle, and outer) and mode of transport. JTW data is represented by interactive
bar and pie charts. While this visualisation method allows users to compare mode
choice for JTW by region it only shows summary statistics. It does not provide
information on the complexity of trip chains and about which modes are used for
which legs of the JTW. Both aspects are addressed by our proposed visualisation.
From a more general perspective, this work shares certain similarities with
flow maps. These flow maps have long been used in cartography to represent
movements of objects from one location to another. A well-known example of
such a flow map was drawn in 1861 by Charles Joseph Minard to visualise Napo-
leon’s Russian campaign of 1812 (Tufte 2001). In recent years, different flow map
layout approaches (Phan etal. 2005; Verbeek etal. 2011) have been developed.
However, flow maps show the flow between a single source and several destina-
tions and consequently resemble a tree structure, where flow can split into dis-
tinct branches but not rejoin. As such flow maps are limited for the purpose of
analysing trip chaining as passengers can arrive at a certain location from dif-
ferent points of origins. Alternatively, Sankey diagrams (Riehmann etal. 2005)
are another way to visualise quantities of flow. Sankey diagrams have tradition-
ally been used to visualise energy, gas, heat, or water distribution and flow or
cost transfers between processes. Recently they have also been adopted to other
domains and applications (Rosvall and Bergstrom 2010; Wongsuphasawat and
Gotz 2012; Perer and Wang 2014). In contrast to flow maps, Sankey diagrams not
only visualise the splitting but also the merging of flows and thus can be used to
represent directed weighted graphs. Thus they would be more appropriate for our
purposes. However, like flow maps, Sankey diagrams do not represent the rela-
tion between incoming and outgoing flow at a single node, an issue we address
with our proposed visualisation.
Perhaps the approach most similar to ours is the use of droplet maps as discussed
in Andrienko etal. (2013). Droplet maps show the different locations along paral-
lel vertical axes connected by lines that convey the magnitude of flow between the
places. These maps allow the analysis of drivers’ trips in order to see temporal or
ordering relations between different places (e.g., home, work, or shopping places).
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
However, it does not distinguish between different modes of travel and similar to
Sankey diagrams the relation between incoming and outgoing flow is not deducible.
Table2 visually compares commonly used representation techniques in the con-
text of trip chaining and multi-modal travel analysis. In summary, we can say that
many of these visualisations focus on representing summary statistics or concentrate
on depicting the overall traffic between areas. In general, these visualisation tech-
niques are thus not well suited for representing trip chaining and travel mode choice
simultaneously.
In contrast to existing techniques, this paper presents a visualisation technique
aimed at analysing trip chaining data which overcomes shortcomings of existing
approaches when it comes to displaying complex multi-modal trips (or trip chains).
It makes use of node-link diagrams to graphically show how stops within a trip are
connected and which modes of transport are used for the various legs of a jour-
ney. Such graphs can, for example, be of use for city planners as they show detailed
information about travel data in a single drawing.
3 Visualisation approach
Our visualisation technique provides an aggregated view of the trip scheduling
behaviour of people by constructing a directed weighted graph from HTS data and
then visualising this graph using node-link diagrams to show how people concat-
enate various activities (or modes of transport) on their journey toward a destination
(e.g., JTW). Each node corresponds to a stop that people make, either to change
the mode of transport or to engage in an activity (e.g., shopping). It is important to
note that nodes do not correspond to geographical locations or physical facilities. In
this paper the term ‘node’ refers particularly to the types of facilities provided. For
example, all different shops visited during the first stop on a trip will be represented
by the same node, although in reality they might be different shops at different loca-
tions. Nodes with multiple incoming and outgoing edges are split into sub-nodes
to convey the sequential relation between the arriving and departing traffic flows.
Edges show the means of transport people use to travel from one stop to another.
The volume of traffic (weights) is conveyed through the width of the edge and the
size of the nodes.
3.1 Algorithm
Figure1 provides an overview of the principal steps involved in creating an aggre-
gated trip-scheduling graph. In brief, HTS data is first filtered according to user-
definable criteria (e.g., weekday, origin) and converted into a graph structure which
is then laid out in a left-to-right fashion. Subsequently, nodes with more than one
incoming and more than one outgoing edge are split and edges are merged to reduce
visual clutter later on. In the last step, the graph is visualised. In the following the
different stages are discussed in detail.
Author's personal copy

G.Wallner et al.
1 3
Table 2 Comparison between commonly used representations for depicting information about trips or
mode choice
Visualisaon technique Descripon Used by, e.g.,
Tables
Chain type frequency%
Home-Work2,120 22
Home-School 1,590 19
Home-School-Work 378 3.7
Home-Park-Work 201 2
Home-Shop-Work1,189 12
Tabular representaons to,
e.g., list frequencies of
different trip chains
Strathman et al. (1994) and Primerano
et al. (2008) to depict the frequency of
various trip chains
Choropleth maps
Choropleth maps shade areas
in proporon to a certain
variable, e.g., public transport
use
Leveson (2013) to visualisecensus data
of England and Walesfrom 2011
Desire lines
Desire lines indicate the
magnitude of traffic between
regions, superimposed over a
map
Xu and Milthorpe (2010) to represent
JTW data from Sydney, AU
from 1981 to 2006
Radial graph layout
Radial graph layout showing
the outgoing or incoming traffic
between a certain region of
interest and other regions
the JTW Visualiser of the Bureau of
Transport Stascs of New South
Walesfrom 2011 to show JTW flows
between districts (accessible online:
Bureau of Transport Stascs) and Sun
et al. (2016) to visualise passenger flow
between subway staons
Charts and diagrams
Various kinds of charts to show
frequencies, percentage
distribuons, etc.
the Victorian Integrated Survey of
Travel &Acvity to represent weekday
JTW data from 2012 and 2013 in
Melbourne, AU in an interacve
fashion (accessible online: Department
of Economic Development, Jobs,
Transport and Resources)
Droplet maps
Droplet maps show the
ordering relaons between
different places
Andrienko et al. (2013) to analyse
personal driving data collected through
GPS
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
For the purpose of exposition we focus on the JTW. As mentioned earlier, how-
ever, the algorithm can be applied to other types of journeys such as journey-to-
school or journey-to-home, too. In the following, we assume that we have an input
database (e.g., HTS dataset) that contains a record for each trip taken by a person
in a household, where a trip can contain multiple stops. In this discussion a stop is
defined as a change in mode or purpose.
Step 1 The process starts by extracting all stops from the input database which
belong to a trip (or trip chain) originating at the respondent’s home and ending at the
respondent’s workplace and which match certain criteria (e.g., only trips conducted
on a weekday or only trips originating in a certain neighbourhood). For each trip
segment, the type of the origin (e.g., accommodation, transport place), the type of
the destination, the stop number, the mode of travel (e.g., walking, public bus), and
the ID of the trip the segment belongs to are retrieved from the database. For all
except the last trip segment of a trip, the destination of one trip segment is the origin
of the next segment.
Step 2 The information gathered during Step 1 is used to derive a directed acyclic
multigraph, formally G =(N, E) where N is a set of weighted nodes and E is a multi-
set of directed edges, i.e., multiple edges between the same two nodes are permitted.
Nodes represent the various types of places visited during a trip. To be more precise,
N is partitioned into n + 1 subsets
with n being equal to the number of trip segments of the longest trip. The nodes in
each subset Ni, except N0, represent the types of the destination places of the i-th trip
segment. Note, that N0 as well as Nn contain only a single node, as all trips have the
same origin and destination type in common, i.e., the respondents home and work-
place. A specific type of place can only appear at most once in each Ni. The weight
of a node is given by the number of people who visited the associated type of stop at
some point during a trip. For example, if three people stopped at a bus station after
the second trip segment, then the node “bus station” in N2 has a weight of three.
Edges
depict the trip segments, pointing from the origin u to the destination v of the trip
segment. Each edge is further associated with the trip ID the segment belongs to,
the stop number, and the travel mode. Figure2a shows an example how a graph may
look like after this step.
(1)
N=N
0
∪N
1
∪
⋯
∪Nn,Ni∩Nj=�,i≠j
(2)
E
→N×N=
{
(u,v)
|
u∈N
i
,v∈N
j
,j=i+1
}
Fig. 1 Overview of the principal steps of the visualisation algorithm
Author's personal copy

G.Wallner et al.
1 3
Step 3 In this step, a reduced graph G′ =(N, E′) is derived from G by replacing
parallel edges (i.e., edges between the same two nodes) with a single weighted edge,
with the edge’s weight being equal to the number of parallel edges (Fig.2b). Each
weighted edge e in E′ also maintains a list of the individual edges which are aggre-
gated by e. This information will be used for colouring the edges later in the process
(see Sect.3.2). G′ will be used for visualisation purposes, whereas calculations on
the graph structure itself are performed on the more detailed graph G.
Step 4 Once G′ has been derived we obtain a two-dimensional embedding of G′
by using the layout algorithm of Sugiyama (Sugiyama etal. 1981)—a graph draw-
ing algorithm which is well suited for directed acyclic graphs, as it is the case in our
setting. Moreover, Sugiyama’s heuristic addresses several aesthetic criteria for draw-
ing graphs, among others, minimization of edge crossings, uniform distribution of
nodes, and uniform direction of edges (Healy 2013) and thus produces graph layouts
which enhance readability. The Sugiyama heuristic arranges nodes in layers, gener-
ally along horizontal lines, with edges going from top to bottom but for our purposes
we arrange nodes in columns with edges proceeding from left to right. Secondly,
as nodes represent the stops made during a trip we want to ensure that all stops at a
particular stage of a trip are not scattered over different layers but are instead located
in the same layer to increase the clarity of the graph. Therefore, nodes are preas-
signed to layers, that is, all nodes of a subset Ni are confined to the i-th layer.
Step 5 After the layout of G′ has been determined some ambiguities need to be
resolved which arise due to people arriving at a stop from different locations and
leaving for different places. Consider, for example, the case depicted in Fig.2a. Peo-
ple are arriving at the train station (at the centre of the graph) either by train from
another train station or by walking from a car park and then travel onwards with
another train or on foot. From this representation, however, it is not clear how the
individual stages are chained together as it is not possible to discern how the incom-
ing and outgoing edges are related to each other. For example, one could get the
impression that somebody was walking to the train station and then continued walk-
ing although this was not really the case. To prevent such misinterpretations, nodes
Fig. 2 Splitting of nodes to convey the sequential relation of incoming and outgoing traffic flow. a Initial
graph derived by extracting trip segments from a household travel survey, b reduced graph obtained by
replacing parallel edges with a single weighted edge, c, d graph and reduced graph after node-splitting
to counteract ambiguities which arise due to ‘junction nodes’, e, f graph and reduced graph after merging
edges that lead to the same node
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
with more than one incoming and more than one outgoing edge in G′ (called junc-
tion nodes henceforth) are split into several nodes and incident edges are rerouted
as follows. Let
n∈N
be a junction node and let m be the number of incoming
edges of n in G′ then n is split into m sub-nodes
ns1,ns2,…,nsm
. Let I(n) further be
the set of incoming edges connecting nodes
ni1,ni2,…,nim
with n in G and O(n)
be the set of outgoing edges of n in G. At this point it should be emphasized that
the splitting itself is performed in the underlying graph G as the rerouting of an
outgoing edge requires knowledge of its preceding edge leading to n. An incom-
ing edge
e=(nik,n)∈I(n)
,
1≤k≤m
is then rerouted to
(nik,nsk)
and an outgoing
edge
e=(n,no)∈O(n)
with preceding edge
f=(nik,n)
, that is fstopNr= estopNr−1 and
ftripID= etripID, is rerouted to
(nsk,no)
. The weights of the newly created sub-nodes are
set to the number of incoming edges after rerouting. The resulting graph is depicted
in Fig.2c. Note that the splitting is performed after layouting to ensure that the sub-
nodes of a junction node are placed close to each other (the Sugiyama heuristic,
however, attempts to distribute nodes uniformly and may even place associated sub-
nodes apart from each other). Hence, sub-nodes are vertically centred around the
position of their respective junction node in G′ with the vertical order of the sub-
nodes chosen in such a way to minimise edge crossings of incident edges (Fig.2d).
Step 6 In this step, rerouted outgoing edges starting at different sub-nodes but
leading to the same end node are bundled at a merge node as shown in Fig.2e. This
is done to avoid cluttering the resulting visualisation with multiple edges running
alongside each other, especially if the end node is located far away (cf. Pupyrev
etal. 2011). More precisely, let
̃
E
be the set of outgoing edges which connect dif-
ferent sub-nodes ns of a junction node with a specific adjacent node no then each
̃e
=(n
sk
,n
o
)∈
̃
E,1 ≤k≤
m
is rerouted to
(nsk,nM)
, where nM is the new merge
node. Furthermore, a new edge between nM and no inheriting all attributes of
̃e
(e.g.,
mode of transport) is created. Finally, edges in G′ are updated accordingly to reflect
the changes made to G (Fig.2f).
Step 7 The graph resulting from Step 6 is visualised as described in the next
section.
3.2 Visual representation
Figure3 shows how the developed visualisation technique represents multi-modal
travel behaviour. The figure is based on 478 trips. The resulting graphs are to be read
from left to right with the node in the first column representing the origin of the trip
(or trip chain), the nodes in the second column representing all different types of
stops people encountered at their first stop toward their destination, the nodes in the
third column showing all different types of places when stopping for a second time
and so forth.
Each node has a label which shows the type of the place and the number of peo-
ple who stopped there. The size and thickness of nodes and edges correspond to the
number of people. Sub-nodes due to the splitting of a junction node are surrounded
by a grey border to visually indicate their grouping (see Fig.3, label A). Edges orig-
inating at the same ‘junction node’ and sharing the same destination are merged to
Author's personal copy

G.Wallner et al.
1 3
reduce edge clutter (Fig.3, label B). Merge nodes are rendered using a black vertical
line to highlight that edges are merged at that point into a single edge and are not
running alongside each other.
The algorithm can be easily adapted to handle different shapes for nodes such as
circles or rectangles. However, depending on the shape the edge routing may need
to be slightly adjusted to reduce overlappings in the proximity of a node. In this
paper, we will restrict the discussion to rectangles. The height of the rectangle is
proportional to the weight of the node (i.e., number of people passing through a
place). Edges are drawn using piecewise cubic Bezier curves (edges returned by the
Sugiyama heuristic can consist of multiple segments to prevent edge-node overlaps)
with the width of the curve being linearly proportional to the edge weight. In case
of rectangular nodes, the control points of a curve are set in such a way that the
curve is entering and leaving a node horizontally. The vertical order of the outgoing
edges corresponds to the vertical order of the positions of their respective end nodes
to circumvent edge crossings. Incoming edges at the target are ordered in a similar
fashion as are edges converging at a merge node. As the outgoing flow matches the
incoming flow at each node (the exception being the start and target location) the
edges sum to the same height.
As all edges are pointing from the left to the right, we follow the recommendation
of Holten and van Wijk (2009b) and omit arrowheads as an indicator of direction.
Their user study has shown that arrowheads should be avoided whenever possible as
user performance is quite low with them, probably due to occlusion and visual clut-
ter caused by the arrowheads. Edges are rendered in order of decreasing width, i.e.,
thinner edges are rendered on top of thicker edges such that they lie over the thicker
edges where they cross. This way occlusions of thinner edges by thicker edges
are reduced. We also make use of alpha blending as suggested by Holten and van
Wijk (2009a) to emphasise individual edges in areas with a high density of edges.
Fig. 3 Graphical elements of the visualisation. The size and thickness of nodes and edges reflect the
number of people. A, a junction node split into several sub-nodes to visualise how the previous stop
affects the onward journey; B, edges originating at the same ‘junction node’ and sharing the same desti-
nation are merged to reduce edge clutter; C, the colouring of edges reflects the percentage share of dif-
ferent modes of transport; D, edges whose size have been limited are visually differentiated with a stripe
pattern. The label of the edge provides an indicator of the actual width
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
In addition, edges are colour-coded to reflect the travel mode. For that purpose, we
used an established colour scheme for qualitative data from ColorBrewer (Harrower
and Brewer 2003). If different modes of transport have been used to reach a des-
tination from the same origin, the curve is divided into sections with each section
representing one mode of transport. The width of a section is proportional to the
percentage share of the transport mode. For example, the stretch labelled C in Fig.3
has been covered in approximately equal shares by driving, as passenger in another
vehicle, or by walking.
As nodes and edges are scaled in relation to the number of people, popular routes
are visually accentuated. Nevertheless, the scaling of edges and nodes can be chal-
lenging due to possible large variations in traffic volumes, for example, when a few
trips (or parts thereof) are being taken by a disproportionately large number of peo-
ple (such as the direct home-to-work trip). Using linear scaling with a scaling factor
that scales down the width of such dominant edges to a reasonable size would cause
the other less popular edges to be too thin to remain readable. Using a larger scaling
factor, on the other hand, would lead to some very thick edges which would occlude
large parts of the graph. On this account, we have chosen to impose an upper limit
on the edge widths. Edges that are affected by this limit are visually distinguished
with a stripe pattern as illustrated in Fig.3 (label D). The multiplier on the edge is
displayed to provide an indicator of the actual width. Logarithmic scaling would
have been another alternative to compress the large range of values. However, it may
make it more difficult to visually assess the differences in traffic volume. Moreover,
if logarithmic scaling is used, the sum of the edge widths of the incoming edges
will, in general, not be equal to the sum of the edge widths of the outgoing edges at
a node. This can give the observer the false impression that the number of people
changes at a node or merge node.
4 Case study Brisbane
The city of Brisbane, Queensland’s capital, serves as a case study to illustrate our
approach. Brisbane is located within the SEQ region of the state. Brisbane, com-
prising 189 suburbs, has a population of about 1.1 million as of 2009 (Queensland
Government 2010). The TransLink Division of Queensland Department of Trans-
port and Main Roads delivers transit throughout SEQ, including Brisbane City, via
operator contracts. The SEQ region includes 23 transit zones of which the City of
Brisbane encompasses the five innermost zones (Fig.4). This study focuses on the
travel behaviour of residents living in inner Brisbane, which comprises TransLink’s
travel zones 1–3.
4.1 Study area anddata
The SEQTS09 data (Department of Transport and Main Roads 2009) are utilised to
demonstrate the applicability of our approach for trip chain and multi-modal travel
analysis. The SEQTS09 database contains information about the day-to-day travel
Author's personal copy

G.Wallner et al.
1 3
behaviour of persons living in approximately 10,000 randomly sampled private
residential households within the Brisbane Statistical Division, the Gold Coast City
Council, and the Sunshine Coast Regional Council area. The data were collected
during a ten-week period from late April to late June in 2009 using self-completion
questionnaires. Each household was assigned a specific day for which to fill out the
questionnaire.
The dataset contains detailed information for each trip taken by one person of
a household. A trip, as stored in the database, represents a one-way travel from an
origin to a destination for a single purpose but may involve several modes of trans-
port (stages). For each stage, several features such as start-time, type of the origin
and destination, and mode of transport are recorded. All trips conducted by a sin-
gle person over a day are numbered in chronological order. Tables3 and 4 show
small excerpts from the dataset for two people. For example, the person with the ID
Y09H040119P01 undertakes four trips with a car (TIMEMODE = 1) on a single day
with each trip consisting only of one stage (no change of transport). However, the
person makes two stops at educational institutions (504 and 508) on the way from
home (201) to work (301). In this case, the JTW thus consists of three trips. In con-
trast, person Y09H020138P02 does not stop on its way to work to pursue some other
purpose but needs to make use of three modes of transport (TRIPSTAGES = 3).
First, the person drives (MAINMODE = 1) to a train station (103) then takes the
train to another train station before walking (MAINMODE = 4) the rest of the way to
the workplace. After work, the person returns back home using the same way.
Based on household trip data we derived trip chains from home to work. In addi-
tion, we assigned each household its related public transport zone (1–3) based on
Fig. 4 Left: TransLink’s 23 travel zones of the SEQ area. Right: inner areas of Brisbane City (zones
1–3). Source: https ://trans link.com.au
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
Table 3 Excerpt from the trip table of the SEQTS09 database for two persons
ORIGPLACE2 and DESTPLACE2 represent the type of place of the origin and destination. TIMEMODE indicates the mode with the longest travel time for a trip. Places
as well as transport modes are number-coded in the database
PERSID TRIPID TRIPNR TRIPSTAGES ORIGPLACE2 DESTPLACE2 TIMEMODE …
Y09H040119P01 Y09H040119P01T01 1 1 201 504 1 …
Y09H040119P01 Y09H040119P01T02 2 1 504 508 1 …
Y09H040119P01 Y09H040119P01T03 3 1 508 301 1 …
Y09H040119P01 Y09H040119P01T04 4 1 301 201 1 …
Y09H020138P02 Y09H020138P02T01 1 3 201 301 7 …
Y09H020138P02 Y09H020138P02T02 2 3 301 201 7 …
Y09H020138P02 Y09H020138P02T03 3 1 201 103 1 …
Y09H020138P02 Y09H020138P02T04 4 1 103 201 1 …
… … … … … … … …
Author's personal copy

G.Wallner et al.
1 3
its geographic area, indicated by Statistical Area Level 1 (SA1) codes, stored in the
SEQTS09 database. Some SA1 zones are, however, relatively large and may inter-
sect more than one public transport zone. In that case, the public transport zone cov-
ering most of the area of that SA1 zone was used as the public transport zone for the
entire SA1 area.
4.2 Results
The results presented in this section make use of the SEQTS09 dataset. To begin
with, Fig.5 provides an overall view of trip chains from residents living in inner
Brisbane (zones 1–3) during weekdays on their JTW. Each intermediate node
reflects a category of location (indicated by the label below each node) at which
Table 4 Excerpt from the stops table of the SEQTS09 database showing the stages for the two 3-stage
trips from Table3
MAINMODE indicates the mode of transport
TRIPID STOPID STOPNR ORIGPLACE2 DESTPLACE2 MAINMODE …
Y09H020138P02T01 Y09H020138P02S01 1 201 103 1 …
Y09H020138P02T01 Y09H020138P02S02 2 103 103 7 …
Y09H020138P02T01 Y09H020138P02S03 3 103 301 4 …
Y09H020138P02T02 Y09H020138P02S04 4 301 103 4 …
Y09H020138P02T02 Y09H020138P02S05 5 103 103 7 …
Y09H020138P02T02 Y09H020138P02S06 6 103 201 1 …
… … … … … … …
Fig. 5 JTW trip chains undertaken by residents in inner Brisbane (zones 1–3, based on 978 journeys)
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
people stopped during their work commute. In this example we used the course
classification provided by the SEQTS09 dataset which distinguishes between ten
different types of locations (e.g., educational institution, shop, accommodation).
The numbers in brackets indicate how many people make the stop. Please note
that each trip (= edge between two stops) represented in this graph may itself con-
sist of multiple stages travelled by different modes of transport, i.e. be a multi-
modal trip (see also Figs. 6, 7). For that reason, we adopted the longest time-
mode. This means that the mode used for the longest time of the trip contributes
to the colouring of the aggregated edges.
It is evident from Fig.5 that most of the people (~ 82%) did not make any addi-
tional stop on their way to work. About 50% of the people who went directly to
work used their own vehicle for their journey, followed by about 25% of the peo-
ple who used public transport (train and bus combined). Other transport modes
such as walking or bicycle are used far less often to go directly to work. If peo-
ple decide to make additional stops on their way to work then they mostly used
their own vehicle. The convenience of a private vehicle possibly accounts for the
relatively low use of public transport. This finding also aligns with previous stud-
ies (Strathman etal. 1994; Bhat 1997; Hensher and Reyes 2000; Ye etal. 2007).
However, none of the trip chains contains more than three intermediate stops. The
second most frequent class of trip chain is that with a single intermediate stop.
Of these, the most common trip chain is Home–Educational Institution–Work-
place with people stopping at an educational institution, for instance, to drop off
their children at school. The second most frequent trip chain with a single stop,
Home–Shop–Workplace, includes a visit to some sort of shop, for example, to
buy something to eat before going to work. This is followed by trip chains where
people make a stop at some accommodation or transport feature (e.g., to pick-up a
co-worker) or to visit a social place. Interestingly, in the latter case (Home–Social
Place–Home), walking is a popular alternative to the private vehicle. Trip chains
with two intermediate stops are few in number with people stopping at two
Fig. 6 Multi-modal journeys of people going directly from their homes to work (zones 1–3)
Author's personal copy

G.Wallner et al.
1 3
Fig. 7 Integrated view of all multi-modal JTW trip chains conducted by residents of inner Brisbane on weekdays
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
educational institutions constituting the majority of cases. Trip chains with three
stops are very uncommon.
As mentioned earlier, Fig. 5 only visualises the trip chains without showing
details of the intermediate stages of the trips. Figure 6 illustrates to what degree
residents of inner Brisbane engage in multi-modal travel while commuting to work.
It provides a detailed breakdown of multi-modal direct Home–Workplace trips (the
most common trip chain according to Fig.5). Intermediate nodes in Fig. 6 depict
transfer locations. Labels again show the type of location but this time the more
detailed classification provided by the SEQTS09 dataset was adopted to distinguish
between different types of transport features (e.g., bus stop, train station). According
to Fig.6, most of the surveyed households which do not pursue any activities dur-
ing their JTW do not engage in multi-modal travel at all. They normally use a single
mode of transport to go to work. Again, in this case residents prefer their own vehi-
cle but cycling, walking, or travelling as vehicle passenger are also used, although
to a much lesser extent. If public transport is part of the work commute then mostly
only a single public transport service is used. Typically mainly either a bus or a train
with ferries only playing a marginal role. When the main leg is made of a single
bus or train ride, people usually walk to and from the stops. It will be noted that the
proportion of commuters who use their own vehicles to go to a railway station is
greater than that for those who go to bus stations. This may be due to the fact that
train stations are located farther apart from each other than bus stations. A large
proportion of people takes a single train to the destination station, while others take
up two trains. Everyone completes the last leg of the journey from the station to the
workplace on foot. If people go to a train station with a public bus then they do so
to catch a train for their next part of the journey before walking the rest of the way.
While Fig.6 focuses on multi-modal direct home-to-work trips, Fig.7 offers an
integrated view of both trip chaining behaviour and mode choice. People are making
at most six intermediate stops although work commutes with more than three inter-
mediate stops are exceptional. Two intermediate stops, on the other hand, are quite
common. These trip chains mainly arise from people who rely on public transport
and who walk from and to the bus and train stations. In terms of modes of transport,
mainly car followed by bus and train are the preferred modes. Public ferries, for
instance, only play a very minor role for the daily work commute.
5 Discussion
Our proposed visualisation method simultaneously represents several vari-
ables (e.g., number and types of stops) which contribute to the complexity of
trip chains as well as the frequency of occurrence of trip chains, the concatena-
tion of stops made during a trip chain, and people’s preferred modes of travel.
Its purpose is making the interrelations between mode choice and trip chaining
behaviour more tangible. It may help transport planners to better appreciate and
identify travel demands, for example, to ensure that public transport meets the
demands of the residents. Moreover, our visualisation method can clearly indi-
cate the number of people visiting specific locations on their way to work (or
Author's personal copy

G.Wallner et al.
1 3
between any other two locations). The complexity of trip chaining often leads
to higher car dependency. Our proposed visualisation technique can, for exam-
ple, assist town planners or the city council authority in planning land-use around
transportation hubs. For instance, mixed-use land development and multi-purpose
activity centres allow people to fulfil a variety of activities at a single location.
This may encourage travellers to fulfil their desire of activities at a single location
and discourage them to undertake additional trips. This, in turn, can contribute to
reducing car dependency and traffic congestions. The versatility of our approach
also allows for easy inclusion of additional data such as SA1 level codes. Hence,
it equips researchers with the ability to zoom into the travel specifics of a certain
area or to easily identify travel mode choice of a subset of the population and the
dynamics behind the mode choice. Consequently, transport planners can use this
information to evaluate the availability of public transport of an area along with
land-use development to encourage people to reduce car travel and instead rely on
other sustainable means of transportation. Besides being useful for analysing trip
chaining and mode choice we can also see potential of our approach for commu-
nicating the data to policy-makers. At this point it is worth noting that although
we used HTS data as input, other data sources can be used as well as long as they
can be mapped to the graph structure described in Sect.3.1.
For example, smart card data or data collected by electronic fare payment sys-
tems could be used to visualise passenger flows in public transportation systems.
In terms of runtime, the proposed algorithm is mainly dominated by the com-
plexity of the Sugiyama layout algorithm when the time taken to import the data
(e.g., from a database or file) is excluded. The runtime of the Sugiyama layout
itself depends on the running times of the algorithms used for its individual phases.
Assuming that the fastest available algorithms for each phase are used, the algorithm
has a time complexity of
O(|N||E�|log|E�|)
(see Eiglsperger etal. (2005) for an in-
depth discussion). However, as the layouting is performed on the reduced graph
G′
which has considerably fewer edges than G and because the number of nodes of
G′
is typically small this should not be problematic. Nevertheless, if necessary, more
efficient implementations such as the one proposed by Eiglsperger etal. (2005) are
available and could be used instead as well.
The second major factor influencing the overall runtime of our approach is the
time complexity of the crossing reduction step which takes place after the splitting
of junction nodes. This reduction is currently performed using the greedy-switch
heuristic proposed by Eades and Kelly (1986), which has a time complexity which
is quadratic in the number of nodes to be switched (Bastert and Matuszewski 2001).
Besides, as the reduction is carried out for the sub-nodes ns of each junction node
separately, with
|ns|
usually being small, this should not be a seriously limiting fac-
tor. Moreover, if runtime is a concern then other heuristics with near-linear time
complexity can be considered (see Bastert and Matuszewski (2001) for a discussion
of different heuristics and their runtime and quality trade-offs). All other steps can
be implemented in linear time. As such the algorithm is applicable to larger datasets
such as the one used in this paper, too, especially because the runtime is not directly
dependent on the number of individual trips to be processed but rather on the num-
ber of edges in the reduced graph, whereby usually
|E′|
≪
|E|
.
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
To give some understanding of the actual timings achieved with our prototype
implementation: The graph shown in Fig.5 (
|N|=
21,
|E|=
1195,
|E�|=45
)1 took
about 344.38ms to calculate of which 270.21ms are spend on the Sugiyama lay-
out and 4.55ms on the node-splitting including crossing reduction. The remaining
time is spread across the other stages and for setting up the necessary data structures
for rendering. The timings for the graph in Fig.6
(|N|=
18,
|E|=
1381,
|E�|=33
)1
are quite similar, with 336.32ms in total of which the Sugiyama layout consumes
279.70 ms and the node-splitting 4.43 ms. For comparison, a larger graph with
|N| = 77, |E| = 9405, and |E′| = 273 takes about 944.74ms to calculate with 488.78ms
and 117.33 ms for layouting and node splitting, respectively. All given values are
averaged over 10 runs of the algorithm and were measured on an Intel Core i5-6400
2.7 GHz Quad-Core CPU with 8 GB RAM and are exclusive the time for data
import.
In terms of space requirements, we maintain two graphs—G and
G′
—in parallel,
requiring
O(|N|+|E|)+O(|N|+|E�|)
space. Please note that the number of nodes
and thus the visual complexity of the resulting graph is partly influenced by the
number of types of places which are distinguished. In Fig.5, as pointed out above,
we directly used the course classification of the SEQTS09 dataset which differenti-
ates between ten types of places. Some of these places are, however, rarely visited
during the work commute (e.g., recreational place or natural feature). If desired by
the analyst such seldom visited places may be merged together. By doing so, edges
connecting these types of places also become thicker which might make it also eas-
ier to see the modal differentiation.
One challenge in visualising HTS data is the complexity of factors influencing
the travel choices of people. A multitude of variables has been recognized to influ-
ence the travel behaviour, including individual or household characteristics (e.g.,
age, gender, marital status, presence of children, or number of vehicles per house-
hold) and geographical area (Hensher and Reyes 2000; Ma etal. 2014; McGuckin
and Murakami 1999). While we focused on analysing JTW patterns of residents liv-
ing within certain public transport zones in the above case study, these other char-
acteristics can be used as well to generate individual graphs for certain segments
of the public. For example, consider separate JTW graphs for males and females or
individual graphs for each district in a city. For comparison purposes, these graphs
could then be visualised side-by-side by using small multiples.
Similarly, there exist a number of attributes which can be used for describing and
understanding trip chains and multimodal travel. Our visualisation currently repre-
sents an important subset of these attributes, such as frequency of trips, mode choice,
and number of stops. However, there exist other attributes (e.g., purpose, travel time,
or distance travelled) which should also be considered for inclusion. Small multi-
ples could, for example, also be used to generate graphs based on travel time or dis-
tance. Currently it is also not apparent from the visualisation which activities people
1 At the time of layouting, the final number of nodes and edges is slightly higher due to the introduction
of sub-nodes during the node-splitting stage and of merge nodes and additional edges during the merge
edges stage of the algorithm.
Author's personal copy

G.Wallner et al.
1 3
pursue at the different locations. Activities, however, play an important role in activ-
ity-based approaches to travel behaviour analysis (Xianyu 2013). For example, a
commuter may not necessarily stop at a train station to catch a train but to withdraw
money from an ATM. One possibility to show such information could be by colour-
ing the nodes in proportion to the purpose of the stop. However, we refrained from
doing so mainly for two reasons. (1) As we are concerned with nominal data it is
important that the colours are reliably distinguishable by the human observer (cf.
Silva etal. 2011). Research suggests that humans can rapidly perceive about five to
ten different colours (see Ware 2004). The SEQTS09 dataset we used, however, dif-
ferentiates between at least ten purposes which would have approximately doubled
the amount of colours or one and the same colour would have to be used for different
categories. (2) As the nodes can be rather small the percentage share of the different
activities may be hard to assess visually. Such information could, for instance, be
displayed by tooltips. Tooltips may also be used to display more detailed informa-
tion on demand (e.g., exact percentage shares for the different modes of transport).
As part of future work we are thus considering to develop the visualisation into an
interactive system which allows users to access detailed information on demand and
offers them the possibility to compare multiple trip scheduling graphs using small
multiples. We are also considering ways to incorporate temporal attributes into the
visualisation. This would offer new opportunities for analysis such as investigating
trip chains or mode choice with respect to commute time.
6 Conclusions
Research on the complex phenomena of trip chaining and travel mode choice can
benefit from visualisations which complement existing tabular and statistical meth-
ods. However, our literature review has shown that visualisations currently used for
analysing trip chaining and multi-modal travel are limited in the amount of simulta-
neously displayed variables. Most visualisations focus on representing aggregated
statistics and the amount of traffic or passenger flow between pairs of regions or
stations (i.e. single trips without intermediate stops). In contrast, the visualisation
introduced in this paper provides an aggregated view of trip chains and mode choice
at the same time by means of node-link diagrams. Several variables, including the
number and type of stops (e.g., bus station, shopping mall) within trip chains, the
quantity of traffic at and between stops, as well as the different modes of transport
used to reach a location/destination are displayed. Moreover, our proposed technique
allows to inspect the sequential relation between incoming and outgoing traffic at
stops. We achieve this by splitting nodes where people arrive from more than one
location and continue to different destinations into sub-nodes. This way we can
resolve ambiguities which would arise in related approaches like flow maps (Phan
et al. 2005) or Sankey diagrams (Riehmann etal. 2005) (see Sect. 2 for details),
where it is not clear which part of the incoming flow continues to which subsequent
node. Using data from the South East Queensland 2009 household travel survey
we demonstrated how the visualisation can be used to reveal patterns in the travel
behaviour of residents. While we focused on the JTW in this paper we would like
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
to reemphasize that the algorithm is general enough to be applied to other types of
journeys as well. Future work will focus on the inclusion of further variables (e.g.,
activities performed at stops) and on integrating the visualisation into an interactive
system which allows users to compare different graphs and to access detailed infor-
mation on demand.
Acknowledgements We would like to thank the Queensland Department of Transport and Main Roads
for providing the HTS data.
References
Alsnih R, Hensher DA (2005) The travel behaviour of seniors in an ageing population in Sydney: an
exploratory study of trip chains and modal preferences in the greater metropolitan area of Sydney.
Road Transp Res 14(4):60–76
Andrienko G, Andrienko N, Bak P, Keim D, Wrobel S (2013) Visual analytics focusing on space. In:
Visual analytics of movement. Springer, Berlin, Heidelberg, pp 253–305. ISBN:978-3-642-37583-5
Bastert O, Matuszewski C (2001) Layered drawings of digraphs. Drawing graphs: methods and models.
Springer, Berlin, pp 87–120
Bhat CR (1997) Work travel mode choice and number of non-work commute stops. Transp Res Part B
31(1):41–54
Bhat CR, Koppelman FS (1999) Activity-based modeling of travel demand. In: Hall RW (ed)
Handbook of transportation science, vol 23. Springer, New York, pp 35–61. https ://doi.
org/10.1007/978-1-4615-5203-1_3
Bureau of Transport Statistics: JTW visualiser. http://visua l.bts.nsw.gov.au/jtwdy namic /. Accessed July
2017)
Capiler Corporation. TransCAD transportation planning software. http://www.calip er.com/TCTra velDe
mand.htm. Accessed July 2017
Corcoran J, Chhetri P, Stimson R (2009) Using circular statistics to explore the geography of the journey
to work. Pap Reg Sci 88(1):119–132. https ://doi.org/10.1111/j.1435-5957.2008.00164 .x
Currie G, Delbosc A (2011) Exploring the trip chaining behavior of public transport users in Melbourne.
Transp Policy 18:201–210
Department of Economic Development, Jobs, Transport and Resources (2016) VISTA 2012–13 - Journey
to work - victorian integrated survey of travel & activity (VISTA). https ://trans port.vic.gov.au/data-
and-resea rch/vista /vista -data-and-publi catio ns/
Department of Transport and Main Roads (2009) South East Queensland Travel Survey 2009. https ://
data.qld.gov.au/datas et/2009-south -east-queen sland -house hold-trave l-surve y
Eades P, Kelly D (1986) Heuristics for reducing crossings in 2-layered networks. ARS Combin 21:89–98
Eiglsperger M, Siebenhaller M, Kaufmann M (2005) An efficient implementation of Sugiyama’s algo-
rithm for layered graph drawing. J Graph Algorithms Appl 9(3):305–325
Fiorenzo-Catalano S, Van Nes R, Bovy PH (2004) Choice set generation for multi-modal travel analysis.
Eur J Transp Infrastruct Res 4(2):195–209
Golob TF, Hensher DA (2007) The trip chaining activity of Sydney residents: a cross-section assessment
by age group with a focus on seniors. J Transp Geogr 15:298–312. https ://doi.org/10.1016/j.jtran
geo.2006.09.005
Habib KMN, Day N, Miller EJ (2009) An investigation of commuting trip timing and mode choice in
the Greater Toronto Area: application of a joint discrete-continuous model. Transp Res Part A
43(7):639–653. https ://doi.org/10.1016/j.tra.2009.05.001
Harney D, Rajoo PD (2015) Moving to tour based models in QLD, is it time? In: Australian Institute of
Traffic Planning and Management (AITPM) national conference, 28–31 July 2015, Brisbane
Harrower M, Brewer C (2003) ColorBrewer.org: an online tool for selecting colour schemes for maps.
Cartogr J 40(1):27–37. https ://doi.org/10.1179/00087 04032 35002 042
Healy N (2013) Hierarchical drawing algorithms. In: Tamassia R (ed) Handbook of graph drawing and
visualization. Chapman and Hall/CRC, Boca Raton
Author's personal copy

G.Wallner et al.
1 3
Hensher DA (2007) Some insights into the key influences on trip chaining activity and public transport
use of seniors and the elderly. Int J Sustain Transp 1(1):53–68. https ://doi.org/10.1080/15568 31060
10970 04
Hensher DA, Reyes AJ (2000) Trip chaining as a barrier to the propensity to use public transport. Trans-
portation 27(4):341–361. https ://doi.org/10.1023/A:10052 46916 731
Holten D, Van Wijk JJ (2009a) Force-directed edge bundling for graph visualization. In: Proceedings
of the 11th Eurographics/IEEE—VGTC conference on visualization (Aire-la-Ville, Switzer-
land, Switzerland), EuroVis’09. Euro graphics Association, pp 983–998. https ://doi.org/10.111
1/j.1467-8659.2009.01450 .x
Holten D, Van Wijk JJ (2009b) A user study on visualizing directed edges in graphs. In: Proceedings
of the SIGCHI conference on human factors in computing systems, CHI’09. ACM, New York, pp
2299–2308. https ://doi.org/10.1145/15187 01.15190 54
Islam MT, Habib KN (2012) Unraveling the relationship between trip chaining and mode choice: evi-
dence from a multi-week travel diary. Transp Plan Technol 35(4):409–426
Leveson GT (2013) 2011 Census analysis—method of travel to work in England and Wales Report. Tech.
Rep., Office for National Statistics
Ma J, Mitchell G, Heppenstall A (2014) Daily travel behaviour in Beijing, China: an analysis of workers’
trip chains, and the role of socio-demographics and urban form. Habitat Int 43:263–273
Manore MA (2007) Visualization in transportation 101. In: TR news, visualization in transportation:
empowering innovation, transportation research board of the national academes, vol 252, Septem-
ber-October, p 3. http://onlin epubs .trb.org/onlin epubs /trnew s/trnew s252.pdf
McGuckin N, Murakami E (1999) Examining trip-chaining behavior. A comparison of travel by men
and women. Transp Res Rec 1963:79–85. http://nhts.ornl.gov/1995/Doc/Chain 2.pdf, https ://doi.
org/10.3141/1693-12
McGuckin N, Zmud J, Nakamoto Y (2005) Trip chaining trends in the U.S.—understanding travel behav-
iour for policy making. Transp Res Rec 1917:199–204. https ://doi.org/10.3141/1917-22
Monteiro N, Rossetti R, Campos P, Kokkinogenis Z (2014) A framework for a multimodal transportation
network: an agent-based model approach. Transp Res Procedia 4:213–227. https ://doi.org/10.1016/j.
trpro .2014.11.017
Nagel T, Maitan M, Duval E, Moere AV, Klerkx K, Ratti C (2014) Touching transport—a case study on
visualizing metropolitan public transit on interactive tabletops. In: Proceedings of the 2014 interna-
tional working conference on advanced visual interfaces, AVI’14. ACM, New York, pp 281–288.
https ://doi.org/10.1145/25981 53.25981 80
Omer M, Kim H, Sasaki K, Nishii K (2010) Tour based travel demand model using person trip data
and its application to advanced policies. KSCE J Civil Eng 14(2):221–230. https ://doi.org/10.1007/
s1220 5-010-0221-6
Perer A, Wang F (2014) Frequence: interactive mining and visualization of temporal frequent event
sequences. In: Proceedings of the 19th international conference on intelligent user interfaces,
IUI’14. ACM, pp 153–162. https ://doi.org/10.1145/25575 00.25575 08
Phan D, Xiao L, Hanrahan P, Wino-Grad T (2005) Flow map layout. In: Proceedings of the 2005 IEEE
symposium on information visualization. IEEE Computer Society, pp 219–224
Philip M, Sreelatha T, George S (2013) Activity based travel behavioural study and mode choice model-
ling. Int J Innov Res Sci Eng Technol 2(1):181–190
Primerano F, Michael APT, Pitaksringkarn L, Tisato P (2008) Defining and understanding trip chaining
behavior. Transportation 35(1):55–72. https ://doi.org/10.1007/s1111 6-007-9134-8
Public Transport Victoria—Journey Planner. https ://www.ptv.vic.gov.au/journ ey. Accessed July 2017
Pupyrev S, Nachmanson L, Kaufmann M (2011) Improving layered graph layouts with edge bundling.
In: Brandes U, Cornelsen S (eds) Graph drawing, vol 6502. Lecture Notes in Computer Science.
Springer, Berlin, pp 329–340. https ://doi.org/10.1007/978-3-642-18469 -7_30
Queensland Government—Office of Economic and Statistical Research (2010) Information brief—
Regional Population Growth: 2008–09. http://www.qgso.qld.gov.au/produ cts/repor ts/pop-growt
h-reg-qld/reg-pop-growt h-2008-09.pdf. Accessed August 2017
Rail Europe—Rail travel planner Europe. http://www.raile urope -world .com. Accessed July 2017
Riehmann P, Hanfler M, Froehlich B (2005) Interactive Sankey diagrams. In: Proceedings of the IEEE
symposium on information visualization, INFOVIS, pp 233–240. https ://doi.org/10.1109/infvi
s.2005.15321 52
Rosvall M, Bergstrom T (2010) Mapping change in large networks. PLoS One 5:1. https ://doi.
org/10.1371/journ al.pone.00086 94
Author's personal copy

1 3
Visualisation of trip chaining behaviour and mode choice
Shiftan Y (1998) Practical approach to model trip chaining. Transp Res Rec 1645:17–23
Silva S, Santos BS, Madeira J (2011) Using color in visualization: a survey. Comput Graph35(2):320–333
Strathman JG, Dueker KJ, Davis JS (1994) Effects of household structure and selected travel characteris-
tics on trip chaining. Transportation 21(1):23–45. https ://doi.org/10.1007/BF011 19633
Sugiyama K, Tagawa S, Toda M (1981) Methods for visual understanding of hierarchical system struc-
tures. IEEE Trans Syst Man Cybern 11(2):109–125. https ://doi.org/10.1109/TSMC.1981.43086 36
Sun Y, Shi J, Schonfeld PM (2016) Identifying passenger flow characteristics and evaluating travel time
reliability by visualizing AFC data: a case study of Shanghai Metro. Public Transp 8(3):341–363
TransLink—Journey Planner. http://jp.trans link.com.au. Accessed July 2017
Tufte E (2001) The visual display of quantitative information. Graphics Press, Cheshire
Verbeek K, Buchin K, Speckmann B (2011) Flow map layout via spiral trees. IEEE Trans Vis Comput
Graph17(12):2536–2544
Wallace B, Barnes J, Rutherford GS (2000) Evaluating the effect of traveler and trip characteristics on
trip chaining, with implications for transportation demand management strategies. Transp Res Rec
1718:97–106
Walle SV, Steenberghen T (2006) Space and time related determinants of public transport use in trip
chains. Transp Res Part A 40:151–162
Ware C (2004) Information visualization: perception for design, 2nd edn. Morgan Kaufmann, Burlington
Wongsuphasawat K, Gotz D (2012) Exploring flow, factors, and outcomes of temporal event sequences
with the outflow visualization. IEEE Trans Vis Comput Graph 18(12):2659–2668. https ://doi.
org/10.1109/TVCG.2012.225
Xianyu J (2013) An exploration of the interdependencies between trip chaining behavior and travel mode
choice. Procedia Soc Behav Sci 96:1967–1975. https ://doi.org/10.1016/j.sbspr o.2013.08.222
Xu B, Milthorpe F (2010) Analysis of journey to work travel patterns in Sydney. In: Proceedings of the
Australasian transport research forum.http://atrf.info/paper s/2010/2010_Xu_Milth orpe.pdf
Xu M, Milthorpe F, Tsang K (2011) Detailed analysis of the travel patterns of rail users in Sydney. In:
Proceedings of the Australasian transport research forum. http://atrf.info/paper s/2011/2011_Xu_
Milth orpe_Tsang .pdf
Ye X, Pendyala RM, Gottardi G (2007) An exploration of the relationship between mode choice and
complexity of trip chaining patterns. Transp Res Part B 41:96–113. https ://doi.org/10.1016/j.
trb.2006.03.004
Author's personal copy