Content uploaded by Doina Olaru
Author content
All content in this area was uploaded by Doina Olaru on Feb 12, 2022
Content may be subject to copyright.
Environment and Planning A 2013, volume 45, pages 2495 – 2514
doi:10.1068/a45490
Lifecycle stages and residential location choice in the
presence of latent preference heterogeneity
Brett Smith, Doina Olaru
UWA Business School, The University of Western Australia, 35 Stirling Highway, Crawley,
WA, 6009, Australia; e-mail: brett.smith@uwa.edu.au, doina.olaru@uwa.edu.au
Received 6 September 2012; in revised form 29 November 2012
Abstract. The choice of residential locations is affected by both dwelling and location
characteristics. Preferences for these characteristics vary with each household’s
requirements, traditionally attributed to the household’s lifecycle stage. With a cross-
sectional study that identifies lifecycle stages according to household structure, this
paper offers an investigation of residential location and shows that not all components
of preference heterogeneity can be accounted for by household structure. Latent class
choice models examine household segments according to lifestyle preferences. The results
reveal the degree of association between identified household lifecycle segments and
estimated lifestyle latent classes. The composition of the latent structure differs for each
lifecycle segment; income and the age of the head of household strongly affect housing
preferences, but do not lead to the same latent class structure for households at different
lifecycle stages.
Keywords: lifecycle stage, lifestyle, residential location, latent class
1 Introduction
Location factors and individual needs and preferences clearly are associated (Li and Wu,
2004; Rabe and Taylor, 2010; Rossi, 1955; Scheiner and Kasper, 2003), and mounting
evidence indicates that households self-select residential location on the basis of their
lifestyle preferences (Mokhtarian and Cao, 2008; Van Wee, 2009). Prior research suggests
that household context is a strong determinant of relocation (Clark and Dieleman, 1996;
Clark and Onaka, 1983; Rossi, 1955), but residence in a particular neighborhood also drives
the adoption of a particular lifestyle, values, and attitudes that then influence decision
making and behavior. To explore the mechanisms by which residential change depends on
lifecycle and lifestyle orientations, we focus on the valuation that households express for
various dwelling and neighborhood characteristics, along a new railway corridor. We seek
to identify latent classes, or lifestyle segments, and investigate their association with the
household’s structure, which we call lifecycle.
1.1 Lifecycle stages and residential location
According to Rossi (1955), residential moves are a function of push and pull factors. Push
factors reflect mainly changes in household structure as people move through different
lifecycle stages (Clark and Dieleman, 1996; Clark and Onaka, 1983; Mulder and Lauster,
2010), which affect consumption patterns for housing and nearby facilities (eg, Clark et al,
2006). These lifecycle stages are the demographic configurations of the household, measured
by marital status, the relationships among household members, the presence of children, and
the ages of the household head and the youngest child (ABS, 2006).
Lifecycle housing adjustments generally relate to marriages or divorces (Aassve et al,
2007; Feijten, 2005), the birth of a child (Clark and Onaka, 1983; Ström, 2010), job retirement
(Ermisch and Jenkins, 1999), or losing a partner (Clark and Deurloo, 2006). Ström (2010)
notes a positive association of homeownership and the number of rooms with the birth of the
2496 B Smith, D Olaru
first child, though this factor shows no relationship with the type of dwelling. Mulder and
Lauster (2010) speculate that the choice of dwelling type depends instead on the difference
in household valuations of comfortable and high-status housing. Clark and Deurloo (2006)
also question a common theory that growing families who purchase larger houses later
downsize when the children leave the nest. Instead, they suggest that housing stock upgrades
are due primarily to increased wealth and find little evidence that older couples move back
into housing of medium to high density. Other factors inducing residential moves include:
a change in the workplace or employment/career (Bailey and Livingston, 2008; Dieleman,
2001; Wadell et al, 2007), migration (Dieleman, 2001), and a change to the household’s social
status (Myers and Gearin, 2001). In some cases, unsatisfactory quality (Clark and Onaka,
1983; Rabe and Taylor, 2010) or altered preferences due to changes in the housing market
[eg, low interest rates (Clark and Onaka, 1983; De Groot et al, 2011)] may trigger a move
out of a neighborhood. Finally, the desire for homeownership is a significant motivation for
moving (Blaauboer, 2010; Clark and Dieleman, 1996), which may be connected to family
formation (Mulder, 2006). The affordability of dwellings or limited opportunities in the rental
sector further affect residential moves (De Groot et al, 2011).
In contrast, pull elements include the attractiveness of the area, quality of the dwelling,
proximity to jobs and to other locations of interest, and evaluations of the available alternatives
in the housing market. Housing-market characteristics can function as both push and pull
determinants (ie, constraints or enticements) and affect lifestyle preferences at various
lifecycle stages (Clark and Dieleman, 1996). Intergenerational housing preferences tend to
persist, as reflected in the quality of housing rather than tenure in one location (Blaauboer,
2010; Mulder and Lauster, 2010).
Because housing consumption is a function of so many elements, as well as of paramount
importance for societal well-being (Dieleman, 2001; Gerstorf et al, 2010), substantial
research has explored housing choices. However, these investigations appear incomplete
without incorporating lifestyle (Æro, 2006).
1.2 Lifestyle and residential location
Research that accounts for lifestyle can address the gap between aspirations/values and
the resources needed to achieve those goals when relocating. Veal (2001) indicates that the
concept of lifestyle has ultimately emerged as an amorphous concept that provides a general
framework for describing clusters of household choices (eg, relocation, labor activity,
car ownership). In marketing literature, lifestyle offers a segmentation option to describe
consumption behavior (Cahill, 2006). Travel behavior research treats lifestyles as exogenous,
reflecting a repertoire of cultural values and preferences that affect the way households
move (Æro, 2006) and choose daily activities or travel (Veal, 2001; Walker and Li, 2007).
Æro (2006) ascribes the choice of residence to disposition (personal tradition). According
to Schwanen and Mokhtarian (2005), individuals and households adopt and adapt their
lifestyles and behaviors in response to environmental conditions, in their attempt to achieve
consonance between their values and their surroundings.
In a relocation setting, households assess combined housing and neighborhood features
subjectively (Galster, 2001), so their predisposition toward a certain type of lifestyle may
dominate their choice of a new residential location (Cao et al, 2009). For example, households
disposed to drive less and be more physically active look for places that are conducive to
greener and healthier lifestyles (Cao et al, 2009; Lund, 2006; Schwanen and Mokhtarian, 2005).
Households that want to upgrade their dwelling status are likely to move to more desirable
locations (Myers and Gearin, 2001), whereas other households (eg, older or single persons)
may choose to move to ‘better’ neighborhoods without changing the quality of their specific
dwelling (Clark et al, 2006). Omitting such self-selection would bias any behavioral analyses.
Lifecycle stages and residential location choice 2497
Regardless of whether housing or neighborhood elements dominate, lifecycle stages lead
to changes in lifestyles (Blaauboer, 2010; Veal, 2001). With latent class choice models, Walker
and Li (2007) and Olaru et al (2011) show that attitudes help define preference heterogeneity
in residential location decisions. But both studies find a correlation between latent (lifestyle)
classes and household characteristics. For example, larger families with more vehicles and
higher incomes constitute a latent class that reports higher preference for dwelling quality
(Olaru et al, 2011). Household structure covaries weakly with household preferences, but
income and the age of the head of the household (HOH) are discriminant characteristics
(Walker and Li, 2007). Other research notes significant associations between lifecycles and
lifestyles (Dieleman, 2001) that in turn influence relocation and travel: education or changes
in employment that lead to higher status and enable households to express their desired
lifestyles; union and children, with concomitant needs for the stability of homeownership;
changing cultural backgrounds due to marriage that shift location preferences to other ethnic
communities; long-distance migration that requires adjustments to new sociocultural settings
and mobility styles; or retirement, which provides more discretionary time and preference for
picturesque environments.
1.3 Residential location choices and affordability
Homes are typically the most valuable asset that people own, and they consume a substantial
proportion of most households’ income (Rahman, 2010). On another level, housing and
urban planning represent a significant government investment. Yet, planning decisions often
get undertaken without a strong base of evidence about what households value most or what
kinds of housing they are likely to choose.
In their discussion of choice with regard to English housing policy, Brown and King
(2005) suggest the need to connect effective choice (ie, the capability to select from any
alternatives—even if they are less than perfect solutions) with access to resources, because the
“impetus for choice is coming from government regulatory and policy agencies” (page 60),
which can enhance choice. In this sense, some reforms simply shape the achievement of
choice, such that customers take a more passive role, waiting for housing supply opportunities
to arise that enable them to exercise their choice. Similarly, Evans and Unsworth (2012) show
that planning policies and a set of financial market circumstances (rather than consumer
preferences) drove people in England to move to higher density areas after 2000. Although
Australia seems to have gone further toward systems accounting for demand (Brown and
King, 2005), the question of choice persists.
In addition, Australian house prices have risen significantly in the past three decades—
faster than average household incomes, construction costs, or rents (ABS, 2010; Richards,
2008). Prevailing housing-market conditions have created unequal distributional effects,
such that some population groups suffer impaired enjoyment of their housing. For some
households, housing costs exceed 50% of their earned income [9% of renters, 8% of mortgage
payers (Rahman, 2010)]. Of those with paid or mortgaged houses, 78.92% are couples
(Hendershott et al, 2009); the minority of single-person households reflects affordability
limitations. However, the spatial variation for housing and the policies aimed at promoting
mixed-income neighbourhoods indicate that, in Perth at least, households actually have a
range of options available. Yates (2001) argues that two camps dominate Australian debates
about urban consolidation, and both accept that planning (for smart growth or transit-oriented
developments or embedding the new urbanist principles) increase choice. One school of
thought advocates consolidation as a means to improve affordability, whereas the other
challenges this effect. Because housing choices are constrained by what the households
can afford and the housing stock available, adding lower value housing stock (eg, higher
density, smaller properties) and creating diverse housing bundles in many areas that also
2498 B Smith, D Olaru
offer urban facilities, may provide more previously unavailable options. The key question to
be investigated is how well the supply of housing matches population needs, resources, and
preferences.
Despite considerable research into how lifecycle conditions affect relocation decisions,
much remains to be learned about heterogeneity within classes of households that appear to
embody similar circumstances. We adopt an alternative perspective by considering lifecycle
and lifestyle jointly. In so doing, we recognize explicitly that residential choices reflect the
need to accommodate changing functional needs throughout the household’s lifecycle, but
they also respond to households’ preferences and idiosyncrasies and are limited by housing
affordability.
2 Housing choices on a new rail link
This research was conducted on a railway corridor opened in December 2007 in Western
Australia. The state government invested more than AUD$1.6 billion to create a new
strategic railway corridor through metropolitan Perth’s southern suburbs, adding 72 km to
a radial network of 173 km with five lines and sixty-nine railway stations. Perth has a low
population density (308 inhabitants/km2 or 1.7 million spread along an Indian Ocean coast
line of 150 km) and high car dependence (0.72 vehicles/person).
Along the new corridor there were opportunities to (re)develop urban areas near the
station and plan for new transit-oriented developments (TODs) at suitable locations. The rail
route travels along the center of an existing freeway reserve and through existing suburban
development, then along its own dedicated reserve through greenfield land. The application
of TOD principles aimed to integrate transport and land-use activity. The stations we study
represent different TOD models, such that two of the three stations in this study appear in
existing low-density suburbs, and the third was built in a greenfield. In figure 1 we display
the boundaries of the metropolitan area, the railway network, and the precincts, in ascending
order of distance from Perth.
The Wellard station precinct provides the most developed example of integration
focusing on walking. Bull Creek represents redevelopment in a well-established area,
with a greater focus on transit interchange, rather than direct integration with land use.
The Cockburn Central precinct contains features of both Wellard and Bull Creek, as an
example of a new TOD with a 100 m freeway reserve bisecting it. Their histories and
locations give these three precincts different socioeconomic characteristics. For example,
with its location near the Canning River, Bull Creek has the highest density, the highest
level of education among residents, and the highest income and real-estate values; it enjoys
good public transport, high-quality cycling and walking networks, and parks and natural
reserves. Cockburn Central features several new estates in which reasonably sized blocks
have allowed large families with children to find good-quality suburban living. It also has
the highest employment rate, second highest income, a mix of features (eg, multifunctional
town centre), and a large park-and-ride facility. Finally, Wellard reveals the smallest
households in the least dense precinct and with the highest car dependence, along with the
lowest employment rates, incomes, car ownership, and housing prices. In Wellard, the main
street combined with residential land use is not fully implemented; hence it has the lowest
walking and cycling mode shares.
Households within a five-minute-drive catchment area of each precinct were surveyed
(by random selection, using a framework from a utility company) during November and
December 2006, using computer-assisted personal interviews. The good response rates
of 42–55% reflected our efforts, including an introductory letter, three consecutive visits
on various days and times of day, and two reminders. Statistical tests confirmed the
Lifecycle stages and residential location choice 2499
representativeness of the sample in terms of demographics (gender, age, household size,
education), using 2001 Census data. The surveys included sections on:
●House: size and type of dwelling; type of tenancy and duration of occupancy; number of
parking bays, vehicles, and bicycles.
●Household structure: age, gender, level of education, place of work/education, and
number of weekly hours involved in paid and voluntary work for each family member;
relationships among household members; household income.
●Travel diary for each independent traveler in the household: origin, destination, departure
and arrival time, purpose/activity at destination, transport mode(s) and connections, out-
of-pocket costs.
●Previous location: size, type of dwelling; type of tenancy; number of vehicles and
bicycles.
●Push and pull reasons for moving from the previous residence and choosing current
location: attitudinal questions covering the importance of having access to built and
natural facilities when selecting the residential location.
●Stated importance: accessibility by car, walking, cycling, and public transport.
Figure 1. [In color online.] Metropolitan area, railway network, and the three precincts.
Railway network
100 20 km
Perth Metro area
2500 B Smith, D Olaru
●Stated choice scenarios. This unlabeled location experiment was administered only to
households that moved between 2002 and 2006. This five-year period helped us avoid
rationalization of the location decision and account for information disseminated in the
community about the railway line.
Further details on the geographical setting and the daily travel patterns appear in Olaru
et al (2011). This paper focuses on the responses to the stated preference survey.
2.1 Stated choice experiment
The stated choice experiment included nine attributes grouped in three categories: dwelling
attributes (block size and number of floors), neighborhood attributes (proximity to schools,
shops, medical facilities, and railway station, and quality of the amenity), and transport factors
(changes in daily travel costs and time for commuting). The households chose their preferred
combination of attributes from two hypothetical alternatives; each household answered eight
scenarios (figure 2). The nature of the dwelling was either a single-floor or multiple-floor
detached house, with block sizes ranging from 400 to 600 m2. The purchase price had three
levels, pivoted around the median house value in that precinct. For local accessibility we used
the time needed to walk to the closest shop, school, and railway station (5 to 15 minutes).
Three photographs illustrated the amenity of the streetscape, such as a main road with heavy
traffic or a ‘greened’ street with a park. To explore the influence of proximity to employment,
we included increases or decreases in daily commuting time and cost.
The experimental design was based on prior parameter estimates obtained from a pilot
study; it applied the D-optimum criterion, which minimized the determinant of the asymptotic
matrix of variances–covariances (Rose and Bliemer, 2009). It was optimized using a genetic
algorithm as the search mechanism.
Figure 2. [In color online.] Example scenario stated choice experiment (unlabeled experiment with
two alternatives).
Lifecycle stages and residential location choice 2501
2.2 Modeling strategy
Discrete choice modeling is based on random utility theory, which posits that rational
decision makers, in a homogeneous market segment and with perfect information, choose
the most preferred alternative from the available set. Their utility functions are a composite
of all characteristics of the alternatives. Accordingly, we incorporated two sources of lifestyle
preference data in the model. First, attitudinal responses to the importance of access to local
facilities by mode offered a direct approach to inferring lifestyle or attitudes to urban locality
and housing. Second, the observed choices in the experiment provided an indirect measure.
A latent class structure can capture the discrete variations in tastes, because “people with
different lifestyles will exhibit different location choice behavior” (Walker and Li, 2007,
page 82).
As we show in figure 3, the lifecycle variables (household characteristics S) may
moderate housing preferences in two ways: directly as taste moderators, in which housing
attributes interact with household characteristics (eg, residential lot price divided by income)
or as demographic translations, such that the parameter estimate includes a population mean
and component specific to a household type. We also could include lifecycle stages in the
utility expression by estimating choice models based on a priori segmentation. Because we
investigate preference differences for households in different stages of their lifecycle, we
adopt the segmentation strategy.
However, estimating separate choice models for different household categories could
lead us to overlook different household types within these categories, based on their lifestyle
preferences. In response, we could include household characteristics as correlates in the class
membership model. Different household characteristics might be of more importance for
households in specific stages of their lifecycle, such that the statistical association between
housing choices and household characteristics is confounded by the dissimilar correlation
across lifecycle groups. To investigate housing choice variability exhibited by lifecycle groups,
Figure 3. Modeling approach. Solid arrows distinguish structural relations in the discrete choice
model; dashed arrows indicate the structural part of the latent model and measurement relationships.
2502 B Smith, D Olaru
we thus examine the relationships between lifecycle and estimated latent classes. We begin
by exploring a descriptive association between latent classes and a priori lifecycle groups,
then investigate the specific latent class structure of each lifecycle segment. This strategy
reveals similarities in lifestyle classes across lifecycle segments. To analyze housing choices
for both the pooled and the segmented data, we use a latent class choice model.
2.3 Latent class choice model
The choice probabilities comprise two stages (figure 3). The class-specific (conditional)
choice probability can be estimated with a multinomial logit model (Swait, 1994), where
the parameter estimates c
b are class dependent. The utility derived from alternative i for
respondent r is a function of the housing and neighborhood variables Xirt in choice situation t:
.UX
irtcir
ti
rt
bf
=+
(1)
and the class-specific choice probabilities are
()
()
.
exp
exp
Pc X
X
{, }
irt
cjpt
j
cipt
12
;=
!
b
b
/ (2)
Class membership is unknown and we estimate a prior probability using observable household
characteristics (ie, current tenancy and reported household income), as well as the latent
variables formed using twenty-four attitudinal questions. The probability that respondent r
belongs to class c derives from multiple observations for each respondent, which offers
a better estimate of the utility function in equation (1). The class membership is also a
multinomial logit choice function, with the form:
(
(,1,2,...,,
0,
exp
exp
Pz
zcC
rc
cr
cC
cr
C
i
ii
==
=
!
/ (3)
where zr are characteristics or attitudes expressed by the responding household, and c
i
are
parameter estimates for membership. One parameter vector c
i
is set to 0 for identification.
The likelihood that a household makes t observed choices in the stated preference survey is
given as the expectation over C possible classes:
.PP
Pc
rr
cir
C
;
=
/
(4)
Finally, we used a sequential approach to include all the attitudinal responses. We form
latent constructs with confirmatory factor analysis, then use the constructs in the choice
models as covariates with the class membership component of the choice model.
3 Profile of sampled households
Our sample consists of 1034 households for which members provided all requested housing
and travel information. Only 471 households provided location preference information and
responded to the eight scenarios (ie, the last two sections of the survey).
We applied cluster analysis to the sociodemographic and economic characteristics of the
households; a six-cluster solution emerged as appropriate, according to its interpretability
and group differentiation. The clusters represent various lifecycle stages—single occupant,
sole parent, couple only, nucleus family, pensioner couple, extended or shared households
(see table 1). To gain additional insights, we split several categories according to the age
of the household head (HOH): young couples with children (HOH < 35) or couples with
children (HOH 35+) and extended/shared household with HOH < 35 or extended/shared
household with HOH 35+.
Lifecycle stages and residential location choice 2503
Senior couples and single individuals have lower incomes and are likely to live in smaller
houses. Sole parents and young couples display the lowest proportion of house ownership
Nucleus and extended families have the highest income, the largest homes, and most cars,
reflecting the multiple needs of their household members.
To capture preferences for the neighborhood—a bundle of spatially based attributes,
including structural, sociodemographic, and environmental characteristics (Galster, 2001)—
we asked the respondents to perform qualitative rankings that indicated the importance of
transport access, accessibility by car and active modes (cycle and walk), and attitudes toward
housing, local urban features, and social-interactive characteristics to them. These items were
factor analyzed, producing five latent constructs. All measurement models indicated good fit
measures and, with two exceptions, the standardized factor loadings were greater than 0.6.
The models, estimated using LISREL8.8 (Scientific Software International Inc.), describe:
(1) transport facilities with three indicators (travel time to work 0.794, proximity to transport
facilities 0.791, and proximity to all urban facilities 0.632) [ χ2 = 2.864, 1 degree of freedom
(df ); Bollen-Stine p = 0.075, root mean square error of approximation (RMSEA) = 0.051].
(2) car access with five indicators (distance to shops 0.708, school 0.535, medical services
0.733, natural environment 0.686, and rail station 0.631) ( χ2 = 5.929, 4 df , Bollen–Stine
p = 0.383, RMSEA = 0.026);
(3) walk and cycle access with five indicators (distance to shops 0.854, school 0.653, medical
services 0.847, natural environment 0.846, and school 0.748) ( χ2 = 21.779, 4 df, Bollen–
Stine p = 0.055, RMSEA = 0.079);
(4) dwelling and neighborhood with four indicators (property size 0.663, real estate value
0.705, house style 0.705, scenic attractive area 0.594) (c2 = 4.107, 2 df, Bollen–Stine
p = 0.169, RMSEA = 0.032);
(5) social dimension with five indicators (people with similar background 0.658, social
contact 0.655, familiarity with the area 0.622, safety 0.567, closer relations with family/
friends 0.526) (c2 = 6.346, 4 df, Bollen–Stine p = 0.358, RMSEA = 0.024).
Table 2 describes the lifecycle segments by their attitudes and location preferences,
expressed as standardized factor scores.
Table 1. Profile households by characteristics.
Household type NWeekly
Income
(AUD)
Number of
bedrooms
Car
ownership
Second
floor (%)
Owned
(%)
Single occupant <60 69 844.03 3.16 1.03 1 46.70
Single occupant 60+ 64 414.70 2.87 0.65 1 100.00
Sole parent 77 823.83 3.70 1.21 0 16.40
Couple only (HOH < 35) 72 1 391.84 3.47 1.98 6 6.80
Couple only (HOH 35+) 83 1 215.50 3.45 1.92 4 53.50
Nucleus family (HOH < 35) 81 1 232.75 3.75 2.01 3 10.10
Nucleus family (HOH 35+) 95 1 505.27 3.96 2.44 12 35.20
Pensioner couple (both partners) 143 602.19 3.39 1.40 8 91.80
Extended/shared household
(HOH < 35)
138 1 531.41 3.67 2.40 14 54.80
Extended/shared household
(HOH 35+)
212 1 389.75 4.13 2.89 18 34.60
Note: bold indicates the highest values; italics indicate the lowest. HOH—household head.
2504 B Smith, D Olaru
Having shops and parks within a five-minute walking and cycling radius offers the
highest latent score value, followed by transport facilities. In the comparison across lifecycle
segments, extended families/shared households appreciate car accessibility, whereas nucleus
families prefer walk and cycle access. Because their living standard appears strongly affected
by a relative lack of neighborhood resources, pensioner couples see both car-based and walk
and cycle access as environment enhancers. Sole parents indicate the highest preference for
transport facilities, along with young extended families and single occupants >60 years. Sole
parents do not seem to care much about car or walk/cycle access, which may be reflective
of their financial and mobility constraints. They favor good transport services, which enable
their children to travel independently to educational activities that may not be available
close to their residences. As expected, couples with children prefer large houses and high-
quality neighborhoods. The social dimension stands out as important primarily for senior
couples and citizens. The relative importance of various facilities to each household type also
highlights similar patterns between young couples and nucleus families, perhaps suggesting
their preparation for the next stage of their life course.
4 Modeling lifestyle latent classes
Choice models with the same attributes and class membership covariates were estimated with
one to four classes. In each model we included weekly household income and current tenancy
type (mortgage), along with the latent constructs expressing attitudes toward car access,
active modes, and public transport facilities. As we show in table 3, the four-class model
outperforms all others according to the Akaike information criterion (AIC), but the two-class
model is preferable based on the Bayesian information criterion (BIC). Models with more
than four classes exhibited parameter uncertainty. Ultimately, we determined that the four-
class model was most appropriate for further examination, on the basis of its fit statistics
and degree of interpretability. The number of class membership covariates and sample sizes
prevented us from testing five-class and six-class choice models. Subsequently, we identify
which household structures are associated with each of the four classes.
Table 4 provides the results for two models: the binomial logit model (BNL) that
assumes unified preferences for the sample and the latent class model (LCM) that identifies
classes of respondents with similar valuations of housing attributes. For the LCM model,
Table 2. Attitudes/preferences towards built environment across lifecycle stages.
Household type Transport
facilities
Car
access
Walk and
cycle access
Dwelling and
neighborhood
Social
dimension
Single occupant <60 −0.070 −0.058 −0.017 −0.249 −0.064
Single occupant 60+ 0.295 −0.031 −0.068 −0.241 0.451
Sole parent 0.307 −0.681 −0.458 −0.012 −0.223
Couple only (HOH < 35) 0.048 0.159 0.258 0.119 −0.137
Couple only (HOH 35+) −0.158 −0.118 −0.217 0.144 −0.121
Nucleus family (HOH < 35) 0.002 0.193 0.336 0.250 −0.037
Nucleus family (HOH 35+) 0.042 0.081 0.375 0.327 −0.028
Pensioner couple 0.170 0.234 0.334 0.138 0.285
Extended/shared household
(HOH < 35)
0.321 0.407 0.238 −0.056 0.153
Extended/shared household
(HOH 35+)
0.213 0.261 0.097 0.196 0.056
Note: The highest standardized values are highlighted in bold. Only the first three latent constructs
were significant in the analysis. HOH—household head.
Lifecycle stages and residential location choice 2505
the parameters listed in table 4 contain both the conditional marginal values of the attribute
[ie, c
b in equations (1) or (2) and the segment covariates [ c
i
in equation (3)]. The estimates
are based on standardized independent variables in the choice models, which allows for
comparisons of the importance of the attributes within a class, as well as the importance of an
attribute across classes. Alternatively, we might compare the marginal effects for each class
on the basis of the posterior parameter estimates (Train, 2003), using a segmentation strategy
that reflects respondents’ highest-class membership probability estimate. In the latter case the
values still would be subject to the scale of the independent variables (unless standardized
first). Elasticities, though dimensionless, depend on market shares and do not apply to an
unlabeled choice experiment.
The signs of the estimates generally are as we expected for both BNL and LCM: a
positive sign indicates that the feature adds worth to the house, whereas a negative sign
means detraction from the utility and housing choice likelihood. The positive parameter for
housing prices for class 3 (dwelling/school focused) indicates that these households may use
the real-estate value as a proxy for the quality of the neighborhood (or investment), over and
above the variables we have examined.
Dwelling characteristics and amenity are powerful indicators of household preferences,
after accounting for affordability. The quality of the residential location is enhanced by
proximity to local services. However, desirability for access is not universal; class 2 expresses
a (statistically significant) preference to be farther away from these services, perhaps due to
an aversion to density. The differences in the parameters for various lifestyle classes reflect
variations in the deep-rooted housing preferences across population. The parameter estimates
for the segment covariates indicate the degree of association between the class membership
and social-demographic variables (eg, income) or location (eg, Wellard).
Class-1 households (price/cost focused; 13% of the sample) are most sensitive to the
house price, as indicated by the dominant parameter estimate. They value proximity to shops
and a lower daily cost of living (transport costs). Along with class 4, this class is associated
with lower incomes. Still, they have the largest average number of children under 14 years
and the youngest HOH. These price-oriented households do not exhibit strong preferences
for a larger property or lot size and place little value on transport facilities (negative
parameter estimate for proximity to railway stations) or car accessibility. Instead, they prefer
accessibility by walking or cycling, which supports a cost-focused interpretation, as obtained
from the parameter estimate on travel costs.
Households in class 2 (space/isolation; 9%) value quiet locations; amenity is the
dominant parameter. They look for larger lot sizes and prefer isolation over access to schools
and shops. The price parameter is not significant, showing that they enjoy higher levels of
social status and wealth, which enable them to choose housing without considerable budget
Table 3. Model comparisons.
Information statistics Latent class models
no classes 2 classes 3 classes 4 classes
Parameters 9 24 41 57
Log likelihood −2 347.5 −2 209.4 −2 169.8 −2 119.9
Aikake information criterion 1.318 1.250 1.236 1.217
Bayesian information criterion 1.333 1.293 1.301 1.316
Adjusted R20.0462 0.102 0.114 0.145
Note: The highest values are highlighted in bold. N = 471 households, 3756 choice observations.
2506 B Smith, D Olaru
Table 4. Latent class model with four lifestyle classes.
Attribute BNL Class 1 (13%):
price/cost
Class 2 (9%):
space/isolation
Class 3 (41%):
dwelling/school
Class 4 (37%):
one-floor high
accessibility
PE t-stat PE t-stat PE t-stat PE t-stat PE t-stat
Price (AUD$ thousand) −0.421 −8.9 −3.444 −4.7 −0.250 −1.4 0.102 1.9 −0.864 −11.7
Additional floor −0.005 −0.2 0.026 0.1 −0.258 −2.9 0.282 8.1 −0.398 −8.5
Block size 0.249 8.4 0.091 0.5 0.723 6.5 0.216 6.7 0.390 8.7
Proximity to school 0.100 4.0 0.103 0.7 −0.500 −4.7 0.201 6.7 0.161 4.5
Proximity to shops 0.123 5.0 0.438 2.7 −0.461 −4.3 0.101 3.5 0.310 8.8
Proximity to train 0.087 3.5 −0.416 −2.9 −0.025 −0.3 0.006 0.2 0.351 9.4
Amenity 0.182 7.2 −0.031 −0.2 1.444 9.4 0.048 1.6 0.281 7.2
Travel cost −0.085 −3.4 −0.357 −2.3 −0.540 −0.1 −0.017 −0.6 −0.255 −6.9
Travel time −0.089 −3.7 0.023 0.2 −0.317 −3.4 −0.037 −1.3 −0.204 −5.7
Segment covariates
Constant −2.406 −3.7 −3.192 −4.2 −2.139 −4.2
Outer precinct: Wellard 1.959 1.9 2.409 2.3 2.921 2.9
Tenancy (mortgage) 0.457 1.6 −0.019 −0.1 0.073 0.3
Household income (AUD$/week) −0.990 −0.4 0.360 1.3 0.720 3.1
Transport facilities −2.091 −3.5 −1.725 −3.2 −1.035 −2.6
Car access −0.708 −1.2 −0.964 −1.7 −0.005 −0.2
Walk/cycle access 2.244 3.7 2.553 4.5 1.020 2.4
Note: The class proportions are the sums of the probability estimates from the class membership model. The parameters significant at 0.05 level are highlighted by
bold t-statistics. BNL—binomial logit model. PE—parameter estimate. Goodness-of-fit measures are provided in table 3.
Lifecycle stages and residential location choice 2507
constraints (they have the highest average income, 16% higher than class 3 and 33% higher
than the others). Classes 1 and 2 are fringe classes, representing only 22% of the sample; they
consider walk and cycle access very important, whereas public transport is not. They differ
mainly in their resources.
The parameter signs and magnitude for dwelling and block size suggest that households
in class 3 (dwelling/school focused; 41%) see them as the two most important attributes.
This class is the only one with a significant (positive) preference for an additional floor;
however, lot size is less important. Compared with the other classes, the proximity to
schools is extremely valuable for households in this group. Although the presence of public
transport is not as important as it is for class 4, walking/cycling access is more important.
Households in class 4 (one-floor high accessibility; 37%) weight the dwelling attributes
and locality features more evenly than other classes. They do not want an additional floor, but
they really value a larger lot size (backyard). Access to local facilities and transport services
contribute to their residential choices. These households rate access to public transport as
important and place a higher (though weakly significant) value on car accessibility, compared
with the two fringe groups.
Perhaps the most striking difference among the dominant classes is their current choice of
location. Class 3 is dwelling focused and far more likely to live in the outer station precinct,
Wellard. Class 4 instead likely has moved to Cockburn Central or Bull Creek.
Wellard Station was designed to mirror both TOD and ‘New Urbanism’ principles, with
the promise of a main street (4070 m2 of retail space) with the railway station at its center,
surrounded by higher density residential facilities. However, this development had not
been realized, even as neighboring suburban developments continued to grow. Perhaps at
37 km from Perth, with two satellite cities just outside their zone, households that desire the
connectivity of TOD find Wellard too isolated.
4.1 Association between lifecycle groups and lifestyle classes
Table 5 reveals the strength of the links between lifecycle and classes of households that are
relatively similar in their preferences for dwelling and neighborhood characteristics. A 2
|
test ( p < 0.001) indicated a significant level of association.
For example, sole parents and young nucleus families (HOH < 35) represent most of
class 1; they likely face substantial challenges in providing for the needs of their families.
The choice model reveals that households in this class are concerned about their proximity
to shops and transport facilities. In addition, sole occupants older than 60 years, extended
Table 5. Strength of association between lifecycle stages and lifestyle classes.
Class 1 Class 2 Class 3 Class 4
Single occupant <60 1.04 −1.65
Single occupant 60+ −1.55 1.12
Sole parent 3.52 −1.16 −1.54
Couple only (HOH < 35)
Couple only (HOH 35+) −2.02 1.15
Nucleus family (HOH < 35) 2.22 1.33 −1.69
Nucleus family (HOH 35+) 1.13 1.56
Pensioner couple −1.81 −1.13 −1.26 3.19
Extended/shared household (HOH < 35) 1.11
Extended/shared household (HOH 35+) −2.24 1.90
Note: Bold numbers indicate the strongest overlap between groups, italics refer to negative and
weaker associations, and white spaces refer to a lack of association. HOH—household head.
2508 B Smith, D Olaru
Table 6. Latent class models for single occupant, sole parent, and couple only.
Attribute Single occupant Sole parent Couple only
SO1 (55%) SO2 (45%) SP1 (42%) SP2 (58%) CO1 (16%) CO2 (84%)
PE t-stat PE t-stat PE t-stat PE t-stat PE t-stat PE t-stat
Price −0.0142 −5.8 0.0068 3.7 −0.1601 −1.9 −0.0012 −0.8 −0.0200 −2.2 −0.0030 −4.1
Additional floor 0.5064 1.4 −0.2369 −1.2 0.7665 0.5 0.3365 1.6 −2.6365 −1.8 0.5655 4.9
Block size 0.0045 2.5 0.0064 4.6 0.0149 1.0 0.0020 1.8 −0.0033 −0.7 0.0022 3.5
School 0.0581 2.0 0.0010 0.1 0.1293 1.0 0.0487 2.8 −0.0995 −1.0 0.0238 2.3
Shop 0.0959 3.4 −0.0524 −2.5 0.4832 1.7 0.0509 2.8 −0.0659 −0.7 0.0309 3.1
Train 0.0262 2.6 −0.0022 −0.3 0.0310 0.7 0.0076 0.9 −0.0759 −1.4 0.0117 2.9
Amenity 0.0556 0.3 0.8305 6.6 −0.2424 −0.3 0.1132 1.1 −1.3432 −2.1 0.3212 5.5
Travel cost −0.0274 −0.5 −0.0825 −1.9 −1.4041 −1.6 −0.0233 −0.5 0.0304 0.1 −0.0377 −1.6
Travel time −0.0268 −1.9 −0.0078 −0.7 0.0679 0.8 −0.0034 −0.3 −0.2278 −2.4 −0.0096 −1.7
Segment covariates
Constant 0.9109 2.0 0.2792 0.4 1.5216 0.8
HOH < 35 1.6717 1.4 −1.4736 −0.8
HOH < 60 −1.9300 −2.1
Income −0.3465 −1.2 −0.0009 −1.3
Employed −1.1320 −0.8 −1.6418 −0.9
Number of households/
number of scenarios
66/462 35/242 78/546
LL0 / LL model −320.23/−252.75 −167.74/−131.59 −378.46/−343.70
Adjusted R20.169 0.141 0.092
Note: The highest t-statistic values are highlighted in bold. PE—parameter estimate. HOH—household head.
Lifecycle stages and residential location choice 2509
households (HOH 35+; usually with at least one aged relative), and pensioner couples are
overrepresented (40% higher than expected) in class 4. This finding is relevant because
community characteristics relate to functional health and wellbeing—especially for older
people. As Gerstorf et al (2010, page 663) note,
“ inadequate public transportation may impose a high level of daily strain on residents and
detrimentally affect their daily routines. In contrast, improving walkability by adding and
maintaining barrier-free sidewalks and pedestrian amenities or clustering residential living
areas near retail stores can be expected to positively affect quality of life and well-being.”
The choice model also reveals that households with at least one aged occupant place a
relatively high emphasis on built environment accessibility.
Young couples without children are not associated with any specific class, though people
older than 35 years of age prefer accessibility to all facilities, as in class 3. This finding is
consistent with their incomes; younger people are at the beginning of their careers, whereas
the peak of professional activity comes later. Nucleus families span all classes, and younger
or newly created families show particular sensitivity to price and preferences for space and
green areas. Once children grow (and household age increases), dwelling and school take
prevalence, and preferences shift for all-around access. Living in extended families or shared
houses means budgetary restrictions are less severe (multiearner households), so they can
afford to locate in spacious properties or high-access areas that effectively meet multiple
requirements.
4.2 Latent class models of location choice by lifecycle segments
To examine the variation in lifestyle classifications across lifecycle segments, we applied
a series of LCMs for each lifecycle group identified by the cluster analysis in section 3.
The results (tables 6 and 7) show that lifecycle and lifestyles should be examined jointly
to understand preference heterogeneity. Each model includes two classes, mainly due to
sample sizes. The smallest segment—sole parents—includes 33 respondents and 242 choice
observations. All models provide a statistically significant fit to the data, and the latent class
structure significantly improves the fit compared with a corresponding BNL. The inclusion
of household covariates improves the AIC, though not every covariate is significant at the
5% level. In summarizing the results for the six lifecycle segments, we compare lifecycle
segments and their associations against the four classes estimated from our pooled data
choice model (table 4). For comprehensibility and clarity, we refer to the class numbering
together with the segment’s label, such that SO1 refers to single occupant class 1 in tables 6
and 7. The class numbers 1–4 (eg, class 1) refer to the pooled data model in table 4.
4.2.1 Single occupant
Class SO1 (55%) is most similar to the price/cost sensitive class 1 in the pooled data model.
However, it also wants accessibility to shops and transport. Class SO2 instead appears to match
with the space/isolation class 2, in accordance with the association in table 5. The 35–60-year
age bracket tends to be associated with class SO2, whereas single occupants who are older
or younger reflect SO1. Thus, the middle age group appears less sensitive to housing prices.
An alternative, two-class model, run with income as the only class membership covariate
( p = 0.24), revealed that class SO2 is (weakly) linked to higher incomes.
4.2.2 Sole parent
Class SP1 (42%) is highly price/cost sensitive. There is little evidence to indicate that it
factors in neighborhood characteristics, except for proximity to shops. These households
consider transport cost more than travel time when choosing a residential location. Another
class (SP2; 58%) resembles class 3, dwelling/school focused. The households seem to prefer
larger houses, on a larger block, with access to schools and shops. The HOH’s age does not
2510 B Smith, D Olaru
Table 7. Latent class (LC) models for pensioner couple, nucleus family, and extended/shared household.
Attribute Pensioner couple Nucleus family Extended family/shared
P1 (76%) P2 (24%) NF1 (37%) NF2 (63%) EFS1 (22%) EFS2 (78%)
PE t-stat PE t-stat PE t-stat PE t-stat PE t-stat PE t-stat
Price −0.0040 −3.0 −0.0015 −0.9 -0.0176 −5.4 −0.0002 −0.3 −0.0034 −2.4 0.0116 2.8
Additional floor −0.9881 −4.5 1.3702 4.7 -0.5818 −2.1 0.1764 1.7 −0.2906 −1.4 2.1051 3.2
Block size 0.0009 0.8 0.0092 5.1 0.0022 1.3 0.0045 8.1 0.0026 2.3 0.0250 4.6
School −0.0294 −1.8 0.0002 0.0 0.0423 1.8 0.0375 4.6 −0.0115 −0.7 −0.0119 −0.3
Shop −0.0631 −3.9 0.0396 2.0 0.0396 1.7 0.0288 3.6 −0.0204 −1.3 0.3379 4.3
Train −0.0112 −1.9 0.0137 1.5 0.0021 0.2 0.0052 1.6 −0.0200 −3.0 0.1385 4.2
Amenity 0.2032 2.0 0.1420 1.2 0.4073 2.3 0.3058 6.0 0.2252 2.2 0.6700 2.3
Travel cost −0.1327 −3.6 0.0360 0.7 −0.0401 −0.8 −0.0310 −1.6 −0.0342 −0.9 −0.0959 −1.0
Travel time −0.0161 −1.7 −0.0083 -0.7 −0.0189 −1.5 −0.0066 −1.4 −0.0183 −1.9 −0.0425 −1.7
Segment covariates
Constant −7.1407 −1.8 0.9463 1.4 .1.2346 0.7
Income 0.0012 2.1 −0.0008 2.4
HOH < 35 0.5875 1.9 1.1812 2.1
HOH 35–60 1.4563 1.3
Age of senior 60+ −0.2187 −2.1
Household size 0.5477 1.1
Children <10 −0.4343 −1.5
Work hours −0.3519 −1.1
Number of households/
number of scenarios
99/691 167/1169 67/466
LL0/LL model −478.96/−407.05 −810.26/−711.79 −323.01/−277.98
Adjusted R20.124 0.104 0.097
Note: The highest t-statistic values are highlighted in bold. PE—parameter estimate. HOH—household head.
Lifecycle stages and residential location choice 2511
distinguish these classes though. Finally, (weak) evidence indicates that sole parents who are
not in paid employment are more likely to belong to SP1
(1
4.8).
2
2
df
\=
4.2.3 Couple only
Table 5 suggests differences between younger and older couples. A small proportion of the
couple group, CO1 (16%), is difficult to categorize into any class from table 4. For example,
its demand for a single-floor dwelling and negative coefficients for proximity to facilities
suggest class 2, but the price coefficient and negative amenity coefficient make this group
appear most similar to class 1. The dominant class, CO2 (84%), seems to fit well with class 4;
empty nesters and older couples prefer accessibility. The main difference between CO2 and
class 4 is that the couples tend to prefer a second floor.
4.2.4 Pensioner couple
Generally associated with class 4, this group includes some older couples (P2, 24%) who
prefer a second floor. Income plays an important role in distinguishing P1 from P2. Less
well-off pensioner couples focus on price, with a moderate consideration of accessibility.
The more financially secure group favors a bigger house and proximity to facilities. In many
ways, P1 and P2 appear closely related to class 4, which explains their strong association in
table 5.
4.2.5 Nucleus family
Class NF1 (37%) is a blend of classes 1 and 4, with a strong price focus but also valuing
amenity and (weakly) considering access to local schools and shops. These households prefer
single-floor houses. They are associated with lower income, a young HOH, and fewer children
under the age of 10 years. Class NF2 (63%) instead bears some similarity to classes 3 and
4: an inclination for larger houses and amenity, though the coefficients for the accessibility
measures are stronger indicators of preference.
4.2.6 Extended family/shared household
The dominant class (EFS2; 78%) for this segment is closest to classes 3 and 4. It shows some
degree of preference for neighborhood facilities but with an emphasis on block size and
access. Households are more likely to have a senior member residing in the house (who may
not be the HOH), which may explain the strong preference for separated living areas. Class
EFS1 (22%) is closely related to class 2. Travel time is more important than transport cost.
The HOH also is likely to be in the 35–60-year age range, and these households have more
occupants.
Thus our findings show strong associations between lifecycles and lifestyles, as well as
evidence that lifecycle groups exhibit within-segment lifestyle preference heterogeneity. We
challenge the effectiveness of sociodemographics and life-course analyses that fail to account
for lifestyle preferences. Our results also signal the potential pitfalls of analyses conducted the
‘easy way’ (ie, the BNL model in table 4 is unable to capture the variability of preferences),
especially considering their substantial potential implications for policy valuation exercises.
5 Summary and conclusion
5.1 Choice modeling results
The latent choice model for the pooled data revealed two minority groups, one of which
exhibited preferences based on resources (price focused, lowest income, class 1) and another
that self-selected to live in a quiet location with a pleasant setting (high amenity, high income,
class 2). The other two classes (78% of households) exhibited preference variation in terms
of the relative importance of dwelling and neighborhood characteristics. Class 3 focused
primarily on the size of the household and local access; access by walking and cycling
was important. Class 4 preferred one-floor houses and rated access by public transport as
2512 B Smith, D Olaru
important. Two important attributes of these households are noted. The class oriented toward
the dwelling and school had higher incomes (significant at 5%) and was more likely to move
to the outer suburb, Wellard, with its continuing new development. The lower land value
there meant that these households could afford larger dwellings.
Having identified four latent classes, we turned our attention to the household lifecycles
and reestimated the LCM with two classes (tables 6 and 7). The association between lifestyles
and lifecycle stages demonstrated certain similarities, though the lifestyle latent classes
ultimately explained more preference heterogeneity than household composition. They
should not be confounded. Consistent with Myers and Gearin (2001) and Howie et al (2010),
older families, older single occupants, and pensioner couples were associated with latent
Class 4 and the desire for more urban amenities. Income helped explain whether a household
would be associated with class 1 or class 4, but for pensioners, physical limitations (age) also
influenced their demand for one floor (class 4) or two floors (class 3).
A similar finding emerged from the preference differences among nucleus, sole parent,
and extended families/shared households. Younger families with lower incomes were more
like class 1, but had a stronger preference for accessibility than similarly resourced couples,
sole parents, and single occupants. Older families favored the dwelling quality. Extended
families split along all classes, though more so in classes 2 and 3. The demand for a second
floor related to the absence of a senior resident.
5.2 Implications for modeling
Lifestyle (latent class) differences dominate the segmentation of housing preferences, which
highlights the significance of incorporating preferences in robust behavioral assessments.
The results support findings by Walker and Li (2007). Yet we also note that lifecycle stage
has a determinant role when it comes to residential locations. Our modeling strategy reveals
the relative importance of demographic variables and circumstances. Adding these variables
to a BNL model through taste modifiers or demographic translations may not reveal the
underlying preference heterogeneity if modelers do not first consider the latent preference
structure. Income and HOH discriminated between classes, but not to the same extent.
5.3 Implications for planning practice
From a policy perspective, our analysis revealed that not all households are affected
by neighborhood characteristics to the same degree. Dominant preference structures
are associated with car accessibility, but older and smaller households (couples, single
occupants, pensioner couples) value public transport access. Families concentrate primarily
on the dwelling and larger block size, but they also prefer walking access to local schools
and shops. Housing affordability is an important factor in determining choice; one segment
(class 1), dominated by young families and couples, less wealthy single occupants, and sole
parents, is restricted in housing choice and very price/cost conscious. For those with limited
resources (classes 1 and 4), housing and location features may be more influential than they
would be for households with the capability to access missing components from outside their
neighborhood boundaries (classes 2 and 3). To capture and match the “residential mosaic”
(Clark and Dieleman, 1996), planners and land developers should avoid standard packages in
residential areas and instead offer a more substantial account of behavioral aspects to develop
tailored strategies that address various lifestyle groups effectively.
5.4 Limitations and further research
This study suffers from several limitations: It has focused on households who moved into
the three precincts prior to the railway opening, without including movers to other areas
or households who might have liked to move to a TOD precinct but were unable to do so.
Considering the objective of TOD principles (namely, to create sustainable opportunity
Lifecycle stages and residential location choice 2513
spaces for urban residents), relying exclusively on responses from movers may create the
potential for bias in the parameter estimates. Caution thus is necessary when informing praxis.
Unfortunately, with a standalone stated preference survey, we can only make claims only
about the differences between lifecycle groups. The data do not support an examination of
transitions through lifecycle stages (housing trajectories), nor do they permit the investigation
of the transition from renting to ownership. These items should appear in further research.
These limitations notwithstanding, our analysis highlights the dominant role of lifestyle and
its complex interactions with lifecycle stages in shaping residential location decisions to
move to TODs.
Acknowledgements. This research received funding from the ARC LP0562422 project, which
included eleven partner organizations: Department for Planning and Infrastructure WA, Public
Transport Authority of WA, LandCorp, The Village at Wellard Joint Venture, Subiaco Redevelopment
Authority, City of Melville, East Perth Redevelopment Authority, City of Cockburn, Midland
Redevelopment Authority, Town of Kwinana, and the City of Rockingham.
References
Aassve A, Betti G, Mazzuco S, Mencarini L, 2007, “Marital disruption and economic well-being:
a comparative analysis” Journal of the Royal Statistical Society A 170 781–799
ABS, 2006, “Household Expenditure Survey, Australia: Summary of Results 2003–04”, ABS cat. no.
6530.0 Re-issue, Australian Bureau of Statistics, Canberra
ABS, 2010, “Household and Family Projections, Australia, 2006 to 2031”, Australian Bureau of
Statistics, http://www.abs.gov.au/ausstats/abs@.nsf/mf/3236.0
Æro T, 2006, “Residential choice from a lifestyle perspective” Housing, Theory and Society 23
109–130
Bailey N, Livingston M, 2008, “Selective migration and neighbourhood deprivation: evidence from
2001 census migration data for England and Scotland” Urban Studies 45 943–961
Blaauboer M, 2010, “Family background, individual resources and the homeownership of couples
and singles” Housing Studies 25 441–461
Brown T, King P, 2005, “The power to choose: effective choice and housing policy” European
Journal of Housing Policy 5(1) 59–75
Cahill D J, 2006 Lifestyle Market Segmentation (Haworth Press, New York)
Cao X(J), Mokhtarian P L, Handy SL, 2009, “Examining the impacts of residential self-selection on
travel behaviour: a focus on empirical findings Transport Reviews 29 359–395
Clark W A V, Deurloo M C, Dielemann F M, 2006, “Residential mobility and neighborhood
outcomes” Housing Studies 21 323–342
Clark W A V, Dielemann F M, 1996 Households and Housing. Choice and Outcomes in the Housing
Market (Center for Urban Policy Research, New Brunswick, NJ)
Clark W A V, Onaka J L, 1983, “Life cycle and housing adjustment as explanations for residential
mobility” Urban Studies 20 47–57
Clark W A V, Deurloo M C, 2006, “Aging in place and over-consumption” Journal of Housing and
Built Environment 21 257–270
De Groot C, Mulder C H, Manting D. 2011, “Intentions to move and actual moving behaviour in
The Netherlands” Housing Studies 26 307–328
Dieleman F M, 2001, “Modeling residential mobility: a review of recent trends in research” Journal
of Housing and the Built Environment 16 249–265
Ermisch J F, Jenkins S P, 1999, “Retirement and housing in later life: evidence from the British
Household Panel Survey” Labour Economics 6 311–333
Evans A, Unsworth R, 2012, “Housing densities and consumer choice” Urban Studies 49 1163–1177
Feijten P, 2005, “Union dissolution, unemployment and moving out of homeownership” European
Sociological Review 21 59–71
Galster G, 2001, “On the nature of neighbourhood” Urban Studies 38 2111–2124
Gerstorf D, Ram N, Goebel J, Schupp J, Lindenberger U, Wagner G G, 2010, “Where people live
and die makes a difference: individual and geographic disparities in well-being progression at
the end of life” Psychology and Aging 25 661–676
2514 B Smith, D Olaru
Hendershott P H, Ong R, Wood G A, Flatau P, 2009, “Marital history and home ownership: evidence
from Australia” Journal of Housing Economics 18 13–24
Howie P, Murphy S M, Wicks J, 2010, “An application of a stated preference method to value urban
amenities” Urban Studies 42 235–256
Li S-M, Wu F, 2004, “Contextualizing residential mobility and housing choice: evidence from urban
China” Environment and Planning A 36 1–6
Lund H, 2006, “Reasons for living in a transit-oriented development, and associated transit use”
Journal of the American Planning Association 72 357–366
Mokhtarian P L, Cao X, 2008, Examining the impacts of residential self-selection on travel behavior:
a focus on methodologies Transportation Research B 42 204–228
Mulder C H, Lauster N T, 2010, “Housing and family: an introduction” Housing Studies 25 433–440
Myers D, Gearin E, 2001, “Current preferences and future demand for denser residential environments”
Housing Policy Debate 12(4) 633–659
Olaru D, Smith B, Taplin J H E, 2011, “Residential location and transit-oriented development in a
new rail corridor” Transportation Research A 45 219–237
Rabe B, Taylor M, 2010, “Residential mobility, quality of the neighbourhood, and life course events”
Journal of the Royal Statistical Society 173 531–555
Rahman M M, 2010, “The Australian housing market—understanding the causes and effects of
rising prices” Policy Studies 31 577–590
Richards A, 2008, “Some observations on the cost of housing in Australia”, Address to 2008
Economic and Social Outlook Conference, Reserve Bank of Australia,
http://www.rba.gov.au/publications/bulletin/2008/apr/pdf/bu-0408-5.pdf
Rose J M, Bliemer M C J, 2009, “Constructing efficient stated choice experimental designs”
Transport Reviews 29 587–617
Rossi P H, 1955 Why Families Move: A Study of the Social Psychology of Urban Residential
Mobility (Free Press, Glencoe, IL)
Scheiner J, Kasper B, 2003, “Lifestyles, choice of housing location and daily mobility: the lifestyle
approach in the context of spatial mobility and planning” International Social Science Journal
55(176) 319–332
Schwanen T, Mokhtarian P L, 2005, “What if you live in the wrong neighborhood? The impact of
residential neighborhood type dissonance on distance travelled” Transportation Research D
10 127–151
Ström S, 2010, “Housing and first births in Sweden, 1972–2005” Housing Studies 25 509–526
Swait J, 1994, “A structural equation model of latent segmentation and product choice for cross-
sectional revealed preference data” Journal of Retailing and Consumer Services 1 77–89
Train K, 2003 Discrete Choice Methods with Simulation (Cambridge University Press, Cambridge)
Van Wee B, 2009, “Self-selection: a key to a better understanding of location choices, travel
behaviour and transport externalities” Transport Reviews: A Transnational Transdisciplinary
Journal 29 279–292
Veal A J, 2001, “Leisure, culture, and lifestyle” Society and Leisure 24 359–376
Wadell P, Bhat C, Eluru N, Wang L, Pendyala R M, 2007, “Modeling interdependence in household
residence and workplace choices” Transportation Research Record number 2003, 84–92
Walker J L, Li J, 2007, “Latent lifestyle preferences and household location decisions” Journal of
Geographical Systems 9 77–101
Yates J, 2001, “The rhetoric and reality of housing choice: the role of urban consolidation” Urban
Policy and Research 19 491–527