Understanding neighbourhood design impact on travel behaviour: An application of structural equations model to a British metropolitan data

Transportation Research Part A 46 (2012) 22–32

Contents lists available at SciVerse ScienceDirect

Transportation Research Part A

journal homepage: www.elsevier .com/locate / t ra

Understanding neighbourhood design impact on travel behaviour: Anapplication of structural equations model to a British metropolitan data

Paulus Teguh Aditjandra a,⇑, Xinyu (Jason) Cao b, Corinne Mulley c

a NewRail – Centre for Railway Research, School of Mechanical and Systems Engineering, Newcastle University, Stephenson Building, Claremont Road,Newcastle upon-Tyne NE1 7RU, UKb Humphrey School of Public Affairs, University of Minnesota, 301 19 Ave. S., Minneapolis, MN 55455, USAc Institute of Transport and Logistics Studies (ITLS), University of Sydney Business School, Newtown Campus, 144 Burren Street, C37,The University of Sydney, NSW 2006, Australia

a r t i c l e i n f o

Article history:Received 14 February 2011Received in revised form 8 July 2011Accepted 7 September 2011

Keywords:Longitudinal analysisNeighbourhood characteristicsResidential self-selection

0965-8564/$ - see front matter � 2011 Elsevier Ltddoi:10.1016/j.tra.2011.09.001

⇑ Corresponding author. Tel.: +44 191 222 5997.E-mail address: [email protected] (P.T

a b s t r a c t

The objective of this study is to explore whether changes in neighbourhood characteristicsbring about changes in travel choice. Residential self-selection is a concern in the connec-tions between land-use and travel behaviour. The recent literature suggests that a longitu-dinal structural equations modelling (SEM) approach can be a powerful tool to assess theimportance of neighbourhood characteristics on travel behaviour as opposed to the atti-tude-induced residential self-selection. However, the evidence to date is limited to partic-ular geographical areas and evidence from one country might not be transferrable toanother because of differences in land-use patterns and land-use policies. The paper is toaddress the gap by extending the evidence using British data. The case study is based onthe metropolitan area of Tyne and Wear, North East of England, UK. A SEM is applied to219 respondents who reported residential relocation. The results identify that neighbour-hood characteristics do influence travel behaviour after controlling for self-selection. Forinstance, the more people are exposed to public transport access, the more likely they driveless. Neighbourhood characteristics also impact through their influence on car ownership.A social environment with vitality also reduces the amount of private car travel. These find-ings suggest that land-use policies at neighbourhood level can play an important role inreducing driving.

� 2011 Elsevier Ltd. All rights reserved.

1. Introduction

Urban sprawl has been widely criticised for its contribution to the car-dependent lifestyle of many societies. For manyresearchers this has been the motivation for investigating the relationships between urban form and travel behaviour.The unknown answer and the intriguing question for the studies is ‘‘If we develop metropolitan areas in an alternativeway, will people reduce their driving and increase their use of public transport and non-motorised transportation?’’ Thatis, is there a form of neighbourhood development that makes urban development more sustainable than sprawldevelopment?

During the past two decades, the literature has shown that urban form characteristics, such as density, settlement size,land-use mix, accessibility and local street layout, are cumulatively affecting travel behaviour alongside socio-economiccharacteristics and planning strategies such as jobs-housing balance, location and regional structure (CfIT, 2009). Plan-ning-based studies have found that there are significant associations between urban form characteristics and travel patterns

. All rights reserved.

. Aditjandra).

P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32 23

irrespective of whether travel behaviour is measured as travel mode choice, travel distance, travel frequency, travel purposeor travel time. Nevertheless, despite identifying these links quantitative and qualitatively, there have not been many studies,especially in the UK, which have developed a robust design to address causal connections between the built environmentand travel behaviour, taking account of the fact that individuals may self select a residential location with specific neighbour-hood characteristics. Moreover, even when the study design is to take account of self-selection, the temporal impact has notbeen accounted for. This study offers a quasi longitudinal design to investigate the causal relationship between the builtenvironment and travel behaviour.

Using data from Tyne and Wear metropolitan area, England, this study aims to contribute to the understanding of theimpact of neighbourhood characteristics on travel behaviour, so as to provide evidence to support the implementation ofland-use policies that aim to encourage alternative development (such as the Compact City in Europe and the Smart Growthand New Urbanism in the US) and hence reduce driving and car dependence. These policies are predicated on the idea that bysituating residential, employment and service locations closer to each other, trip lengths will become shorter, and individ-uals will drive less and/or are more likely to travel on foot, by bicycle, and by public transport. However, it should beacknowledged that Compact City is more than just travel-related issues and is extended to other aspects including the con-servation of the countryside, efficient utility and infrastructure provision, and the revitalisation and regeneration of innerurban areas (Howley et al., 2009). It is also important to note that ‘‘new urbanist’’ inspired developments, such as the‘‘Eco-Town’’ in the UK, are to a large extent designed to be located outside of urban centres, and will depend on privatecar to reach all but local destinations. The purpose and argument of this paper is not to exclude car in the design of neigh-bourhoods but instead to understand better how neighbourhoods can be designed in terms of their characteristics to accom-modate other modes of travel to give the same opportunities as the car offers but to reduce driving without compromisingthe residents’ daily needs.

This study is one of few applications of dynamic (quasi-longitudinal) structural equation model (SEM) in the field of land-use and travel behaviour. Using US data, Cao et al. (2007) demonstrated the methodology and provided results on neighbour-hood design and travel decisions in a US context. This study, with its similar survey design to and with a methodology mod-elled to a large extent on Cao et al. (2007), contributes in three further significant ways. First, through the use of a newdataset, this study is able to offer confirmation or otherwise of connections between neighbourhood design and travel deci-sions and therefore to point to potential generalisation of results. Second, this paper is using one of few disaggregate studiesusing British data in the literature. It produces important implications for planning policy and practice in the British context.Third, although there are differences, the study enables us to compare and contrast the results between different geograph-ical and planning contexts, which offer a unique opportunity to discuss transferability of results.

The paper is structured as follows: the next two sections briefly outline the context of the study related to the connectionsbetween the built environment and travel behaviour and the potential role of land-use planning, before turning to discussionof recent SEM applications in travel behaviour research. Having established the rationale for the paper, described the appli-cability of the methodology, this paper then focuses on the data and variables followed by analysis of results. The last sectionsummarises key findings and discusses the contribution of this study to the debate on the role that the built environment(and therefore planning) can play in creating sustainable mobility in the future where sustainable mobility is defined as tra-vel using less carbon-based fuel (Banister, 2008).

2. Context of study

Many studies investigating the relationship between urban form and travel behaviour have been criticised for their failureto take account of the issue of residential self-selection. The argument, as comprehensively revisited since Handy et al.(2005), is that if particular characteristics of a residential neighbourhood area are associated with particular travel behav-iour, the direction of causality is not defined. Do urban form characteristics influence individuals’ travel behaviour? Do indi-viduals’ travel behaviour preferences lead individuals to select their residential neighbourhood conducive to particular travelpatterns? The latest evidence from the literature suggests that the impact of urban form characteristics on travel behaviourmay result from the two confounding sources of attitudes and demographics (Cao et al., 2009). Furthermore, Bohte et al.(2009) highlights that the impact of attitudinal attributes are as important as socio-economic characteristics in the determi-nation of travel behaviour.

A better understanding of the role of residential self-selection can support a policy of sustainable spatial planning, there-by also addressing the issue of housing supply. Naess (2009) argues that if households are able to self-select their residentialneighbourhood, this does not mean the urban structure does not influence travel behaviour, but that the urban structureactually enables households to self-select. Naess argues that it is possible to persuade car-oriented households to use walk-ing and cycling (Næss, 2005, 2006). Activity participation, location of activities, choice of travel mode and route choice con-tribute to a higher amount of motorised travel among outer-area residents than among inner-city dwellers, regardless of anyself-selection of residents to particular types of neighbourhoods (Naess, 2009; Cao et al. 2010). This is consistent with anactivity-based theory of urban travel demand rather than those which are reliant only on utility based principles. Axhausenand Gärling (1992) observe that travel behaviour models solely based on the utility maximisation principle are not sufficientto understand how people make decisions. Travel demand is generally a derived demand – derived from the desire to reachplaces, whether work places, parks, shopping centres, town centres or just local amenities – although travel also carries a

24 P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32

positive utility; and under certain circumstances, travel is desired for its own sake (Mokhtarian and Salomon, 2001). Despitethe growing interest in activity-based modelling within transportation research, a better understanding of the direction ofcausality for how urban form influences travel behaviour is a crucial link between sustainable land-use and transport stra-tegic decision making.

In the current debate of the development of the Eco-Town in Britain that has been envisaged to encourage low carbon-based energy use including travel, proponents argue that the Eco-Town could open new land for living and relieve some ofthe housing market pressure. Eco-Town promotes compact mixed-use development geared with green technologies to meetzero carbon housing. In terms of transport, the design has been envisaged to encourage sustainable mobility. A number ofEco-Town projects are currently under discussion between the UK government and local authorities but it is still unclearhow the planned goals will be realised in these cases. Although land-use changes in Britain have not been particularly envi-ronmentally unfriendly (Bibby, 2009), it is nevertheless true that a better understanding of the impact of the built environ-ment on travel behaviour would provide a sounder basis to know how to meet future sustainable mobility. It is in theinterest of the British public to understand how people who would live in Eco-Town differ from those in conventional sub-urbs, especially in terms of travel behaviour; and this paper directly addresses this issue.

3. Methodological framework: The appropriateness of structural equation modelling

The use of SEM in travel behaviour research has a long track record that dates back to the 1980s (Golob, 2003). An SEMcan include measurement models, which identify latent constructs underlying a group of observed variables, and/or struc-tural equations, which depict the directional relationships among latent and observed variables. Unlike regression, the SEMis estimated through covariance analysis. SEM can illustrate direct effects between variables and indirect effects throughmediating variables, such as the influence of attitudes on travel behaviour through the residential choice. SEM enablesthe estimation of bi-directional relationship (feedback loops) between variables whereas regression allows only unidirec-tional relationship. Thus, the SEM is superior to linear regression.

Application of SEM in the urban form impact on travel behaviour can be traced back from the work of Golob (2000). Usingdata from Portland in the US, he incorporated a residential accessibility index (as an exogenous variable) to explain time useand trip generation. The work of Simma and Axhausen (2003), based on Austria, incorporated measures of residential acces-sibility and local land-use (as exogenous factors) to explain travel distance differences and personal household characteris-tics (endogenous variables). In the cross-sectional data, they concluded that local accessibility measures were moreinfluential than regional measures developed from gravity models and land-use characteristics.

Recently, de Abreu e Silva et al. (2006) applied SEM to model land-use characteristics for the work and residential locationto predict commuting distance and other travel variables. They concluded that land-use and urban design strongly influencecar ownership and mode choice after controlling for socio-economic and demographic characteristics. Using Flemish (Bel-gian) regional travel survey, van Acker et al. (2007) found that socio-economic characteristics played a greater role thanland-use characteristics in predicting trip frequency, distance and time. However, Maat and Timmermans (2009) using activ-ity-based theory concluded that indirect effects can lead to a different total effect with the apparent effects of one variable onanother variable giving rise to a trade-off of opposite effects. For example, the effects of residential density on travel distancesuggest that people in dense residential environment travel shorter distances, although this effect is partly cancelled out byextra trip activities. Additionally, workplace density/mix increases total daily travel distances but decreases distances by car(Maat and Timmermans, 2009). The arguments used to support activity-based theory are that travel distance and the urbanform relationship are a statistical association, as distances are not travel choices in itself but the consequence of other deci-sions (Maat and Timmermans, 2009). Nevertheless, there is a significant literature from studies based on the traditional util-ity-maximisation theory and these are discussed next.

Using cross-sectional data from San Francisco Bay Area, California, US, Bagley and Mokhtarian (2002) developed the firstSEM in addressing residential self-selection resulting from attitudes. The SEM includes urban structure characteristics suchas traditional and suburban neighbourhood and various travel attitudes as endogenous variables – this is in contrast with theaforementioned studies that treated urban structure as exogenous. Demographic, lifestyle and other attitudinal factors wereincluded as exogenous variables in this study. They concluded that residential location type had few separate impacts ontravel behaviour; attitudes and lifestyles were the most important predictors of travel behaviour.

Scheiner and Holz-Rau (2007) used cross-sectional data collected in Cologne, Germany to analyse the relationships be-tween life situation (socio-economic and demographic characteristics), lifestyle (preferences and location attitudes mea-sures), choice of residential location (density of supply, quality of public transport and mixed land-use) and travel mode.Their findings show that life situation influences mode choice more than lifestyle. But lifestyle plays an important role byaffecting location attitudes and residential location type that in turn influence mode choice. The effect of location attitudeson travel behaviour (travel mode choice and distance) are found to be equal or even stronger than the effects of residentiallocation attributes on travel behaviour, thus indicating the importance of residential self-selection issue.

Conventionally, travel time has been treated as total disutility on the assumption that demand for travel exists only be-cause travellers want to reach their destination. Whether some travellers in some circumstances actually derive some utilityfrom travel itself is a growing literature, building on the contribution of Mokhtarian and Salomon (2001). It is clear that sometravellers experience a positive utility of travel. Whilst not directly relevant to this study, the exploration of this literature

P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32 25

can offer an additional reason for the self selection of neighbourhoods by individuals as well as explain differences betweenurban and suburban dwellers. Ory and Mokhtarian (2009) used more than 1300 commuters in San Francisco Bay Area col-lected in 1998 to understand the connections among perceptions, affections and desires of short-distance travel. Though theoverarching aim of the study was to bring evidence in support of a positive utility theory of travel demand, it offers evidencerelevant to the research question posed in this paper. Their model includes residential location as a dummy variable of urbanform characteristic. They found that the shorter distances travelled by urban dwellers is relatively stable (and less influencedby enjoyment) regardless of the perceptions and affections compared to their suburban counterparts. Further, suburban res-idents who travel a great deal in their car do not have a strong desire to reduce that travel, which suggests the existence ofresidential self-selection.

A comprehensive review on the methodologies used to understand the impact of residential self-selection on travelbehaviour concludes the superiority of longitudinal structural equations modelling. Longitudinal analysis with the collectionof travel-related attitudinal data before and after a residential move provides the best way to investigate the causal relation-ship between the built environment and travel behaviour (Mokhtarian and Cao, 2008). The most recent study that uses suchapproach (or at least the closest approach) is Cao et al. (2007). They conducted a quasi-longitudinal SEM analysis using themovers (people who relocate house within the last 1 year) from a survey of eight neighbourhoods in Northern California.They found that neighbourhood preferences and travel related attitudes indirectly influence travel behaviour through resi-dential choice and directly influence car ownership and driving behaviour and walking behaviour to a lesser extent. Theaccessibility factor (that has high associations with access to shopping mall and town centre) is the most influencing factorin explaining changes in driving behaviour.

Aditjandra (2008), using a similar survey employing quasi-longitudinal design and analysis to Handy et al. (2005), inves-tigated the relationships of urban form and travel behaviour in British context and showed significant differences betweenUS and British data. More factors of neighbourhood characteristics and travel attitudes are found in the British study than theUS study. In particular, the British study found three neighbourhood accessibility factors including public transport, shop-ping and open space access whilst the US study found only two. Furthermore, British study has shown that (public transport)travel accessibility – which did not appear in the US data – is associated with changes in walking and public transport use.This suggests that residents of British neighbourhoods are more aware of public transport service than their US counterparts(Aditjandra et al., 2009).

One rationale underlying this study is that planning activities are heavily influenced by past experience. Whilst the prob-lem of car dependency through the urban sprawl effect is well known worldwide, the implications of land-use planning andpolicies to different countries may be different. This means that evidence from one country might not be transferable to an-other. In this study, we develop an SEM using the British case to evaluate the transferability of model results between dif-ferent planning contexts – British and US.

4. Methodology

This section describes the collection and manipulation of the data to construct relevant variables for SEM. Since the objec-tive of the study is to examine a British case, the selection of neighbourhoods to represent the typical British residentialneighbourhoods was important. Ten neighbourhoods were selected to represent five Districts of Tyne and Wear metropol-itan area in the North East of England. The neighbourhoods were selected to vary systematically by neighbourhood type, theDistricts of the metropolitan conurbation and size of neighbourhoods. Neighbourhood types were characterised by variousstreet pattern layouts based on typo-morphology classification advocated by Marshall (2005) (Fig. 1).

The neighbourhood unit was captured by reference to the lowest administration area used in the latest available BritishCensus data (2001), the Lower Layer Super Output Area (LSOA). Tyne and Wear metropolitan area contains 719 differentLSOAs in total and on average, a LSOA consists of 1500 household with 7500 individual persons. The potential neighbour-hoods were screened District by District to ensure that income and other characteristics were above the average for the areausing Index of Multiple Deprivation (IMD) 2004, weighted by seven aspects: income, employment, health, education, barri-ers to housing and services, crime and living environment and a UK measure of the deprivation of an area, to control for thesecharacteristics. The purpose of this screening was to find neighbourhoods where people would choose to live rather than

Fig. 1. ABCD Typology as transect. Source: Marshall, 2005.

26 P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32

areas where housing might be allocated on the basis of need or affordability as it is preferences in the choice in the builtenvironment that is being considered.

To combine the census screening and neighbourhood design screening, Google Earth™ was used to capture the aerialview of a shortlist of potential neighbourhoods as well as to identify the homogeneity of street lay out within the LSOA.A total of 190 LSOAs from the 38 highest IMD of each district were image-captured and analysed in this way. After filteringthe potential neighbourhoods through controlling for income (higher IMD relative to the average of the district) and sustain-able mobility (the percentage of walking, cycling and public transport use), the most representative residential neighbour-hoods according to ABCD typology street layout were selected as the areas for the case-study.

Table 1 shows that how the chosen areas are classified according to the ABCD typology, as well as the characteristics ofhigh vs. low percentages of (sustainable) travel to work derived from the 2001 British Census data. It is noticeable that a clearcut example of A type is missing as it is unavailable within Tyne and Wear. The thresholds of high and low percentages ofsustainable travel (walking, cycling, and public transport use) for commuting relate to the predominant travel behaviour ofthe neighbourhood. ‘‘High’’ refers to a higher percentage of commuting by metro/bus/walking/cycling selected and rangesbetween 26% and 34% of travel to work by these modes. The reverse is true for ‘‘low’’ which refers to a lower percentageof commuting by metro/bus/walking/cycling with a range between 13% and 21% of commuting by these modes.

The eight-page survey was administered using a delivered-out, mail-back approached out in Spring 2007. It was person-ally addressed using names and addresses from the electoral register and delivered to households in each of the 10 neigh-bourhoods. A sample of approximately 220 households in each neighbourhood was selected to meet the number of theneighbourhood catchment represented by the LSOA unit. A pre-paid self-addressed envelope was enclosed inside each ques-tionnaire delivered. One week later, a reminder postcard with individual names stated on the postcard was delivered to therespondents. In total 2157 questionnaires were delivered. The number of returned questionnaires totalled 716 giving a re-sponse rate of 33% of which 32% provided valid data for the analysis.

This paper studied 219 respondents who reported they had moved to their current residence within the last 8 years,which had been identified as the upper average duration of respondents identified that they could recall their travel behav-iour change in the pilot prior to the main survey (Aditjandra, 2007). The survey captured respondents’ socio-demographicvariables including gender, age, economic status, educational background, household income, household size and numberof children, as well as changes in household income, household size and number of children before and after household relo-cation. Table 2 presents sample characteristics of these movers.

When working with samples, it is desirable to identify whether the sample represents the population to understandwhether results might be generalised to that population. Population data for residents who moved within the last 8 yearsare not available. However, since the focus of our study is on explaining the relationships of other variables to travel behav-iour rather than on describing travel behaviour per se, these differences are not expected to materially affect the results(Babbie, 1998).

The questionnaire was designed to capture changes in travel behaviour that result from different neighbourhood charac-teristics. This was planned by asking respondents who had moved to their current address to indicate how they drive now ascompared to before they moved, on a five-point scale from ‘‘a lot less’’ to ‘‘a lot more’’. This variable indicates respondents’changes in driving behaviour in the SEM.

Neighbourhood characteristics and neighbourhood preferences were measured using 27 statements which were dividedinto six aspects of neighbourhood design. The questionnaire design was loosely based on Handy et al. (2005) and there werea number of differences. In this study the preference statements were grouped under different sub-headings of neighbour-hood design aspects: the motivation for this was to make it easier for the respondents to become familiar with the questionsasked and their context.

These statements were measured using a four-point scale from ‘‘not at all true’’ to ‘‘entirely true’’ to obtain respondents’opinions on the perceived neighbourhood characteristics. Respondents were asked to rate the characteristics of their current

Table 1Case-study areas classified by ABCD typology, British Census 2001 percentage of sustainable travel to work and neighbourhood housing types.

ABCD typology sorting % Sustainable travel to work (walk, cycle, metro and bus)

High Low

B prone to C type South Shields, South Tyneside (terraced)Low Fell, Gateshead (terraced)

C type Lemington, Newcastle (semi-detached) Cleadon Park, South Tyneside (semi detached and detached)Fulwell, Sunderland (terraced and semi-detached) Tynemouth, North Tyneside (semi detached and detached)

D type Pelaw–Wardley, Gateshead (detached) Chapel Park, Newcastle (semi detached and detached)Preston Grange, North Tyneside (detached)Washington, Sunderland (detached)

Note: ‘‘Terraced housing’’ is a style of medium-density housing that originated in Europe in the late 17th century, where a row of identical or mirror-imagehouses share side walls. ‘‘Semi-detached’’ housing consists of pairs of houses built side by side as units sharing a party wall and usually in such a way thateach house’s layout is a mirror image of its twin.

Table 2Sample characteristics. Source: this study.

Housing types Terraced to semi Semi to detached Detached

SouthShields

LowFell

Fulwell Lemington Tynemouth CleadonPark

ChapelPark

PrestonGrange

Pelaw–Wardley

Washington

Number 27 30 14 20 19 18 23 28 23 17Percent Female 58 57 58 37 35 39 59 58 52 56Average car ownership 0.93 1.21 1.00 1.20 1.28 1.56 1.48 1.57 1.39 1.47Average age 40.0 40.3 49.3 39.5 48.9 47.1 43.6 49.3 36.5 48.8Average Household

size1.52 1.97 2.36 2.45 2.42 2.78 2.87 2.71 3.00 2.82

Percent Householdwith children

15 27 36 35 37 28 43 36 65 53

Percent home owners 74 87 86 100 74 89 87 96 100 100Mean Household

income (£k)21.7 28.1 29.0 29.7 29.7 32.1 31.4 31.9 34.0 45.8

Median Householdincome (£k)

30 30 30 40 40 40 40 40 40 40

P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32 27

neighbourhood as well as the neighbourhood in which they previously resided. In identifying the residents’ preference forthe same neighbourhood characteristics in selecting their residence, a four-point scale from ‘‘not at all important’’ to ‘‘extre-mely important’’ was used. Travel attitudes/preferences were measured using a series of 28 statements on a five-point Likertscale from ‘‘strongly disagree’’ to ‘‘strongly agree’’.

This paper uses the SEM methodology to investigate links between neighbourhood design and travel behaviour and fol-lows the two step estimation approach recommended by Anderson and Gerbing (1988). The first step is to perform a factoranalysis to provide latent constructs which are subsequently used as continuous variables in the SEM. The advantage of usingfactor analysis in this way is that it has been shown to reduce complexity in the development of the SEM in the second step.Thus, common Factor Analysis (CFA, called principle axis factoring in SPSS) with oblique rotation was used as a first step toidentify the latent constructs underlying the 27 statements on neighbourhood characteristics and the 28 statements on tra-vel attitudes/preferences. The CFA is more appropriate than principle component analysis when the purpose of the proce-dure is to identify latent constructs (Widaman, 1993). Oblique (Oblimin) rather than orthogonal rotation was usedbecause, in theory, the latent factors of neighbourhood design perceptions and preferences and travel attitudes might cor-relate with each other and would not be statistically independent (Field, 2005). The criterion ‘‘Eigenvalue > 1’’ was used todetermine the number of factors. Through this analysis, perceived and preferred neighbourhood characteristics were ex-tracted into seven factors: safety, travel accessibility, residential spaciousness, social factors, shopping/facilities accessibility,outdoor space accessibility and neighbourhood unattractiveness. The travel attitudes were reduced to eight factors includingpro-public transport use, travel minimising awareness, dislike-cycling, positive utility of travel, safety of car, pro-walking,dislike-travel and car dependent. Factor loadings are shown in Table 3. Accordingly, these factors (latent constructs) are trea-ted as observed continuous variables when we develop the SEM.

5. Model construction and results

5.1. Conceptual model and model estimation

In this study, changes in driving behaviour, changes in built environment and changes in car ownership were initially se-lected as endogenous variables (Fig. 2). Changes in driving behaviour were captured from the quasi-longitudinal survey thatasked respondents to recall their changes in driving before and after relocation. Changes in built environment factors werecomputed by taking the difference between factor scores for current neighbourhoods and the corresponding scores for pre-vious neighbourhoods. In a longitudinal analysis, the directions of the hypothesised effects are particularly important due tothe temporal sequences of events. Previous research has well documented that residential choice is a long-term choice, carownership is a medium term decision and travel behaviour is conditional on both residential choice and car ownershipchoice (Ben-Akiva and Atherton, 1977). Therefore, we assumed that changes in the built environment affect changes incar ownership and driving behaviour with changes in car ownership in turn impacting on changes in driving behaviour. Itwas hypothesised that endogenous variables are also affected by a number of exogenous variables: demographic character-istics and their changes, current neighbourhood characteristics, and current travel attitudinal factors.

The maximum likelihood estimation (MLE), as is common practice, is used to develop the SEM. Because the data containmissing values, we use the option of ‘‘estimate means and intercepts’’. Since few variables were significantly associated withchanges in built environment factors, a parsimonious model structure, shown in Fig. 3, was proposed in which changes in thebuilt environment were treated as exogenous variables.

Table 3Factors of neighbourhood characteristics and travel attitudes.

Neighbourhood characteristicsfactors

Statements Loadings

Safety Safe neighbourhood for walking 0.829Low crime rate within neighbourhood 0.777Safe neighbourhood for children to play outdoor 0.686Low level of car traffic on neighbourhood streets 0.673Quiet Neighbourhood 0.603Good street lighting 0.364

Travel accessibility Easy access to a good public transport service (bus/metro/rail) 0.877Good public transport service (bus/metro/rail) 0.804Easy access to highway network (main road) 0.417Pavements – easy walking routes throughout the neighbourhood 0.394Local shops within walking distance 0.353

Residential spaciousness Adequate space of garden at the front 0.919Adequate space of garden at the back 0.857Adequate off-street parking (garages or driveways) 0.560

Social factors Lots of people out and about within the neighbourhood 0.787Lots of interaction among neighbours 0.665Diverse neighbours in terms of ethnicity, race and age 0.465Economic situation of neighbours similar to my level 0.386

Shopping/facilities accessibility Easy access to a district shopping centre (Tesco, ASDA, etc.) 0.913Easy access to town centre 0.713Other amenities/facilities such as a community/leisure centre or facilities for childrenavailable nearby

0.468

Local shops within walking distance 0.316

Outdoor space accessibility Parks and open spaces nearby 0.586Extension of cycle routes beyond the neighbourhood 0.576Other amenities/facilities such as a community/leisure centre or facilities for childrenavailable nearby

0.309

Neighbourhood unattractiveness Attractive appearance of neighbourhood �0.771High level of neighbourhood’s upkeep (well maintained) within the neighbourhood �0.723Variety in housing style �0.440

Travel attitude factors Statements Loadings

Pro-public transport use I like travelling by public transport 0.876I prefer to take public transport than drive whenever possible 0.870Public transport can sometime be easier for me than driving 0.743

Travel minimising awareness I prefer to organise errands so that I make as few trips as possible 0.634Fuel efficiency is an important factor for me in choosing a vehicle 0.617I try to limit my driving to help improve air quality 0.598The price of fuel effects the choices I make about my daily travel 0.570I often use the telephone/internet to avoid having to travel somewhere 0.399When I need to buy something, I usually prefer to get it at the closet store possible 0.393Vehicle should be taxed on the basis of the amount of pollution they produce 0.368

Dislike-cycling I prefer to cycle rather than drive whenever possible �0.930I like riding a bicycle �0.782Cycling can sometimes be easier for me than driving �0.751

Positive utility of travel Travel time is generally wasted time �0.643The only good thing about travelling is arriving at your destination �0.618

Safety of car Travelling by car is safer overall than taking public transport 0.801Travelling by car is safer overall than walking 0.775Travelling by car is safer overall than riding a bicycle 0.488

Pro-walking I like walking 0.730I prefer to walk rather than drive whenever possible 0.728Walking can sometimes be easier than driving 0.582

Dislike-travel The trip to/from work is useful break between home and work (the importance of yourjourney to work)

�0.720

I use my time to/from work productively �0.618

Car dependent I need a car to do many things I like to do 0.632Getting to work without car is a hassle 0.551

Extraction method: Principal Axis Factoring.Rotation method: Oblimin with Kaiser Normalisation.

28 P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32

Changes in the built environment

Changes in car ownership

Changes in driving behaviour

Demographics and their changes

Neighbourhood characteristics

Current attitudes and preferences

Changes in the built environment

Changes in car ownership

Changes in driving behaviour

Demographics and their changes

Neighbourhood characteristics

Current attitudes and preferences

Fig. 2. Conceptual model.

Changes in car ownership

Demographics and their changes

Neighbourhood attributes and their changes

Current attitudes and preferences

Changes in driving behaviour

Fig. 3. Parsimonious conceptual model.

Table 4Model Goodness-of-Fit (GOF).

Degrees of freedom 65v2: measures discrepancy between the sample and model-implied covariance matrices; the smaller the bettera 103.8v2/d.f.: a ‘‘relative chi-square value’’ corrected for degrees of freedom; values of 3 or less indicate a good fit and values as high as 5 represent

an adequate fit1.60

Hoelter Critical N: A parameter to judge if sample size is adequate. A critical N of 200 or better indicates a satisfactory fit and a value under 75is unacceptable

196

Root Mean Square Error of Approximation (RMSEA): measures the estimated discrepancy between the model-implied and true populationcovariance matrix, corrected for degrees of freedom; values less than 0.05 indicate a good fit, and values as high as 0.08 represent areasonable fit

0.053

a The chi-squared statistic increases with the sample size and so it is not a good GOF measure. However as the basis for other GOF measures, it is alwaysreported anyway (Byrne, 2001).

P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32 29

The parsimonious model has an acceptable fit as shown in Table 4. We would prefer to evaluate whether the data meetthe required multivariate normality assumption of the data but AMOS does not produce statistics regarding normalityassumption when the option of ‘‘estimate means and intercepts’’ is selected. If cases with missing values are removed,the effective sample size is reduced by about 25% (from 219 to 169). As the sample size is already relatively small, it is morevaluable to retain observations than confirm normality.

5.2. Model discussion

Table 5 presents the matrix of standardised direct effects and total effects of the parsimonious model. In terms of endog-enous variables, changes in car ownership were found to have a positive relationship with changes in driving behaviour. Forexogenous variables, an increase in household income and/or an increase in household size are associated with an increase incar ownership but the direct influences of the two variables on changes in driving behaviour are insignificant. Current mea-sures of travel attitudes and residential preferences influence changes in car ownership and driving behaviour, particularlythe latter. Individuals who favour public transport and walking are more likely to reduce their driving. Interestingly, the fac-tor ‘‘dislike cycling’’ is negatively associated with changes in driving. That is, those who prefer cycling are more likely to in-crease their driving. In the data, people who live in Pelaw–Wardley and Cleadon Park tend to prefer cycling. These suburbanneighbourhoods are perhaps places where residents used cycling recreationally or these neighbourhoods are simply locatedtoo far from shopping facilities for cycling to be an option. People who value positive utility of travel time are likely to have alower car ownership and drive less. Those who prefer a high access to shopping facilities tend to increase their driving.

Table 5Standardized direct and total effects for the Parsimonious model (N = 219).

Variables Changes in car ownership Changes in driving behaviour

Endogenous variablesChanges in car ownership 0.173 (0.173)Changes in driving behaviour –

Exogenous variablesSocio-demographics

Changes in income 0.262 (.262) 0 (0.045)Changes in household size 0.406 (.406) 0 (0.070)

Travel attitudesPro-public transport �0.173 (�0.173)Pro-walking �0.117 (�0.117)Dislike-cycling �0.116 (�0.116)Positive utility of travel �.144 (�.144) �0.124 (�0.149)

Residential preferencesShopping accessibility 0.124 (0.124)

Neighbourhood characteristicsChanges in safety factors 0.119 (119) 0 (0.021)Changes in shopping accessibility �0.123 (�0.123) 0 (�0.021)Changes in travel accessibility �0.157 (�0.157)Change in social factors �0.145 (�0.145)

Squared multiple correlations 0.248 0.169

Notes: The numbers in brackets are total effects. A blank cell indicates that this variable was found to be insignificant at the 0.1 level in the model thereforeestimated as a zero coefficient.

30 P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32

After controlling for demographics and attitudinal factors, four built environment variables are significantly associatedwith changes in car ownership and driving behaviour. First, an increase in the safety aspect of the residential environmenttends to increase car ownership. However, this safety variable may act as a proxy for the suburbaness of the neighbourhoodbecause the perception of safety tends to be positively associated with suburban neighbourhoods and suburbanites tend tohave a high level of car ownership. Therefore, a substantial increase in neighbourhood safety (such as a move from towncentre to a suburban community) may require the moving household to acquire an additional vehicle. An individual whomoves to an area with easy access to a shopping centre and town centre (high shopping accessibility) is more likely to sheda private car on moving. People who experienced an increase in access to the transportation system (especially public trans-port) on moving tend to reduce their driving. Interestingly, moving to a more vibrant social area is associated with less pri-vate car driving, which provides support for the development of ‘‘café’’ style areas in some neighbourhoods within the surveyarea. Singleton and Straits (2005) suggested three prerequisites for a causal inference: time precedence, non-spuriousness,and association. As modelling change variables in the longitudinal analysis addresses the time precedence of an associationand the SEM controls for confounding factors and interactions among variables, the significant relationships found in thisstudy provide a robust inference for the causal influences of the built environment on driving behaviour.

A comparison of the standardised total effects shows that the size of neighbourhood characteristics’ influence on drivingbehaviour is similar to that of other variables. For a resident moving from a suburban to an urban neighbourhood, one mightexpect safety within the neighbourhood environment to decrease whereas shopping accessibility, travel accessibility and so-cial factors to increase. If the change was one standard deviation for each variable, this model suggests that on average, driv-ing behaviour overall would be reduced by 0.344 standard deviations {=0.021 + 0.021 + 0.157 + 0.145 where these values arethe total effects of changes in each of neighbourhood characteristics on changes in driving, identified in Table 5}. Roughlyspeaking, this indicates the overall marginal effects of neighbourhood characteristics on driving behaviour; and from thisstudy, the magnitude of change on driving from these variables is similar to the finding in Cao et al. (2007).

6. Conclusions

This study applies an SEM in a British quasi-longitudinal dataset to understand the relationships between neighbourhooddesign and travel behaviour. It has a few limitations: firstly, sample size is related to the power of a test. Because the samplesize for the SEM is not large, some neighbourhood characteristics may be found statistically insignificant even if they do im-pact travel behaviour with the implication is that these characteristics may be overlooked. However, sample size limitationsdo not affect identifying as important those neighbourhood characteristics variables that are significant in the SEM. Sec-ondly, an ideal measurement of change variables should be based on longitudinal design: neighbourhood characteristics, tra-vel behaviour, attitudes and demographics are measured before and after residential relocation. Given limited resources, atrue longitudinal analysis was not able to be conducted. However this study is one of few applications of the SEM on quasi-

P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32 31

longitudinal data. It offers insights on the connections between neighbourhood design and travel choices and transferabilityof the connections between different geographical boundaries and planning contexts.

The study shows that in terms of standardised coefficients, changes in socio-demographic characteristics are the maincontributors to changes in car ownership but changes in neighbourhood characteristics, particularly safety factors and shop-ping accessibility, have important influences. The neighbourhood variables also affect changes in driving behaviour indi-rectly through their influences on changes in car ownership. Furthermore, changes in neighbourhood characteristics, suchas travel accessibility and social factors, also tend to bring about changes in driving behaviour and the size of the effect iscomparable to the size of the effect generated by attitudes and changes in car ownership. These findings are similar toCao et al. (2007), suggesting that neighbourhood design impacts on travel behaviour, after controlling for residential self-selection, may be similar in different geographical settings despite different planning contexts. Moreover, driving behaviourwas influenced by similar dimensions of neighbourhood characteristics (such as travel accessibility and social aspect ofneighbourhoods) in both cases. In contrast, the specific neighbourhood characteristics that influence car ownership are dif-ferent between the cases. In the US, it is yard size and off-street parking that are associated with car ownership whereas inthe British case, the influential factors are shopping/facility accessibility and safety aspects of residential neighbourhoods.Overall, the similarity shows that neighbourhood design does matter to travel behaviour and the size of the effect is consis-tent despite different political and planning contexts; whereas the differences suggest that the design of land-use policiesshould be tailored to specific contexts.

The results lay the foundations for a number of policy directions to increase sustainable mobility of residents. The safetyand shopping accessibility factors have a strong contribution to car ownership. An inference is that a good shopping acces-sibility tends to create an environment less conducive to driving. This finding provides evidence for the recommendations ofthe Barker review (2006): it is important to develop accessible supermarkets/shops to meet local residential market andlarge supermarket chains should expand in terms of sites rather than concentration on few large stores. However it shouldbe noted that accessibility in this context refers to all modes and not just private car. This suggests that how British policiesguide commercial development, for example, building a new supermarket such as Tesco (the biggest supermarket chain inBritain which usually located out of town with huge parking area) versus smaller Tesco Metro (local supermarket size), maysignificantly influence sustainable mobility of residents.

Safety plays an important role in sustainable mobility and higher safety is associated with suburban neighbourhoods. Re-sources to make traditional neighbourhoods as safe as suburban ones would be an effective policy to increase sustainablemobility.

Changes in travel accessibility (highly associated with public transport access) and changes in social factors (highly asso-ciated with interactions between neighbours) significantly reduce driving. Planning or developing a new town (such as anEco-town) or a new neighbourhood should give extra attention to the provision of (public transport) accessibility and thelayout of housing to promote social interactions among neighbours.

In addition, this study confirms the role of residential self-selection in changing travel behaviour. Individuals who favoursustainable modes of transport (public transport and walking) are more likely to reduce their driving. This evidence supportsthe development of more compact city type of neighbourhoods which are self supporting in terms of facilities, and meetingthe housing market that fits residents with less car dependent orientation. This is consistent with the recommendations ofother studies: shaping more balanced, smarter infrastructure growth, mixed-use patterns in urban development towardsmeeting the low carbon future (Falk, 2009; Scheiner, 2010).

After all, the implication made to the British land-use context above can be seen as more than just addressing the sus-tainable environmental aspect as suggested earlier in the paper. Promoting local (supermarket) shops as well as creatinga social conducive environment with safety and easy access to public transport are in fact touching the issues of economicand social sustainability.

References

Aditjandra, P.T., 2007. Relationships between built environment and travel behaviour: initial evidence in North Tyneside. In: Proceedings of the UniversitiesTransport Studies Group (UTSG) 39th Annual Conference, Harrogate. Leeds, UK.

Aditjandra, P.T., 2008. The relationship between urban form and travel behaviour: a micro-analysis in Tyne and Wear. PhD Thesis. November 2008. School ofCivil Engineering and Geosciences, Newcastle University, UK.

Aditjandra, P.T., Mulley, C.A., Nelson, J.D., 2009. Neighbourhood design impact on travel behaviour: a comparison of US and UK experience. Projections 9,28–56.

Anderson, J.C., Gerbing, D.W., 1988. Structural equation modeling in practice. a review and recommended two-step approach. Psychological Bulletin 103,411–423.

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

Babbie, Earl., 1998. The Practice of Social Research, eighth ed. Wadsworth Publishing Company, Belmont, CA.Bagley, M.N., Mokhtarian, P., 2002. The impact of residential neighbourhood type on travel behaviour: a structural equation modelling approach. Annals of

Regional Science 36, 279–297.Banister, D., 2008. The sustainable mobility paradigm. Transport Policy 15 (2), 73–80.Barker, K., 2006. Barker Review of Land Use Planning Final Report – Recommendations. HMSO, Norwich. TSO. <http://www.hm-treasury.gov.uk/media/3/A/

barker_finalreport051206.pdf>.Ben-Akiva, M., Atherton, T.J., 1977. Methodology for short-range travel demand predictions: analysis of carpooling incentives. Journal of Transport

Economics and Policy 11, 224–261.Bibby, P., 2009. Land use change in Britain. Land Use Policy 26S, S2–S13.

32 P.T. Aditjandra et al. / Transportation Research Part A 46 (2012) 22–32

Bohte, W., Maat, K., Wee, B.v., 2009. Measuring attitudes in research on residential self-selection and travel behaviour: a review of theories and empiricalresearch. Transport Reviews 29 (3), 325–357.

Byrne, B.M., 2001. Structural Equation Modelling with AMOS: Basic Concepts, Applications, and Programming. Lawrence Erlbaum Associates, Inc., Mahwah,New Jersey.

Cao, X., Mokhtarian, P.L., Handy, S.L., 2007. Do changes in neighbourhood characteristics lead to changes in travel behaviour? A structural equationsmodelling approach. Transportation 34 (5), 535–556.

Cao, X., Mokhtarian, P.L., Handy, S.L., 2009. Examining the impacts of residential self-selection on travel behaviour: a focus on empirical findings. TransportReviews 29 (3), 359–395.

Cao, X., Xu, Z., Fan, Y., 2010. Exploring the connections among residential location, self-selection, and driving: propensity score matching with multipletreatments. Transportation Research Part A 44 (10), 797–805.

CfIT – Commission for Integrated Transport, 2009. Land Use and Transport: Settlement Patterns and the Demand for Travel. Halcrow Group Ltd. inassociation with P. Headicar, D. Banister and T. Pharoah. Background technical report. October 2009. <http://cfit.independent.gov.uk/pubs/2009/sustainable/technical/pdf/pst-technical.pdf>.

de Abreu eSilva, J., Golob, T.F., Goulias, K.G., 2006. Effects of land use characteristics on residence and employment location and travel behavior of urbanadult workers. Transportation Research Record 1977, 121–131.

Falk, N., 2009. Briefing: Learning from Europe. Proceedings of the Institution of Civil Engineers: Urban Design and Planning 162, 53–58.Field, A., 2005. Discovering Statistics Using SPSS, second ed. SAGE Publications Ltd., London.Golob, T., 2000. A simultaneous model of household activity participation and trip chain generation. Transportation Research Part B 34, 355–376.Golob, T., 2003. Structural equation modelling for travel behaviour research. Transportation Research Part B 37, 1–25.Handy, S., Cao, X., Mokhtarian, P.L., 2005. Correlation or causality between the built environment and travel behaviour? Evidence from Northern California.

Transportation Research Part D 10, 427–444.Howley, P., Scott, M., Redmond, D., 2009. Sustainability versus liveability: an investigation of neighbourhood satisfaction. Journal of Environmental Planning

and Management 52 (6), 847–864.Maat, K., Timmermans, H.J.P., 2009. A causal model relating urban form with daily travel distance trough activity/travel decisions. Transportation Planning

and Technology 32 (2), 115–134.Marshall, S., 2005. Streets and Patterns. Spon Press, London.Mokhtarian, P.L., Cao, X., 2008. Examining the impacts of residential self-selection on travel behaviour: a focus on methodologies. Transportation Research

Part B 42, 204–228.Mokhtarian, P.L., Salomon, I., 2001. How derived is the demand for travel? Some conceptual and measurement considerations. Transportation Research Part

A 35, 695–719.Næss, P., 2005. Residential location affects travel behaviour – but how and why? The case of Copenhagen Metropolitan Area. Progress in Planning 63, 167–

257.Næss, P., 2006. Urban Structure Matters. Residential Location, Car Dependence and Travel Behavior. Routledge, London/New York.Naess, P., 2009. Residential self-selection and appropriate control variables in land-use-travel studies. Transport Reviews 29 (3), 293–324.Ory, D.T., Mokhtarian, P.L., 2009. Modeling the structural relationships among short-distance travel amounts, perceptions, affections, and desires.

Transportation Research Part A 43, 26–43.Scheiner, J., 2010. Far, far away – trip distances and mode choice in the context of residential self-selection and the built environment. In: Geller, P. (Ed.),

Built Environment: Design, Management and Applications. Nova Publishers, Hauppauge, NY, pp. 215–237.Scheiner, J., Holz-Rau, C., 2007. Travel mode choice: affected by objective or subjective determinants? Transportation 34 (4), 487–511.Simma, A., Axhausen, K.W., 2003. Interactions between travel behaviour, accessibility and personal characteristics: the case of upper Austria. European

Journal on Transport Infrastructure and Research 3 (2), 179–197.Singleton Jr., R.A., Straits, B.C., 2005. Approaches to Social Research, 4th ed. Oxford University Press, New York.van Acker, V., Witlox, F., Wee, B.v., 2007. The effects of the land use system on travel behavior: a structural equation modeling approach. Transportation

Planning and Technology 30 (4), 331–353.Widaman, K.F., 1993. Common factor analysis versus principal component analysis: differential bias in representing model parameters? Multivariate

Behavioural Research 28 (3), 263–311.