Content uploaded by Maksym Polyakov
Author content
All content in this area was uploaded by Maksym Polyakov on May 09, 2018
Content may be subject to copyright.
Preprint. Landscape and Urban Planning Volume 110, February 2013, Pages 134-142,
https://doi.org/10.1016/j.landurbplan.2012.11.001
The effect of street trees on property value in Perth, Western Australia
Ram Pandit, Maksym, Polyakov, Sorada Tapsuwan, and Timothy Moran
Abstract
Trees provide a variety of benefits to urban residents that are implicitly captured in the value of
residential properties. We apply a spatial hedonic model to estimate the value of urban trees in
23 suburbs of Perth Metropolitan Area in Western Australia. Results show that a broad-leaved
tree on the street verge increases the median property price by about AU$16,889, suggesting a
positive neighbourhood externality of broad-leaved trees. However, neither broad-leaved trees on
the property or on neighbouring properties nor palm trees irrespective of the locations
contributed significantly to sale price. Our result has potential implications on planting and
maintaining broad-leaved trees on street verges for neighbourhood development and urban
planning to generate public and private benefits of street trees.
Keywords: Broad-leaved trees; Spatial hedonic model; Property value; Street trees.
1
1. Introduction
Urban residents value a variety of amenities (e.g. wetlands, open spaces, parks and recreational
facilities) that offer environmental, cultural, aesthetic, and health benefits. Urban trees are one of
the amenities that are found on residential properties, street verges, parks and reserves in urban
areas. Environmental benefits of urban trees include protection of land from soil erosion,
reductions in storm-water run-off, habitats for wildlife, filtration of air pollutants, improvements
in local air quality, reductions in the urban “heat island” effect, and energy savings by providing
shading and insulation (Brack, 2002; Dwyer, Schroeder, & Gobster, 1991; McPherson, Simpson,
Peper, Maco, & Xiao, 2005; Nowak, Crane, & Stevens, 2006; Pandit & Laband, 2010; Sander,
Polasky, & Haight, 2010; Simpson, 1998). Urban trees also provide cultural and health benefits
that improve the quality of urban life, as trees may make a city neighbourhood more scenic,
provide privacy, shelter residents from negative effects of undesirable land uses, and improve
retail areas by creating environments that are more attractive to consumers (Dwyer et al., 1991;
Ellis, Lee, & Kweon, 2006; Sheets & Manzer, 1991; Wolf, 2005). These environmental, cultural,
and health benefits of urban trees are often difficult to translate into monetary terms (Anderson
& Cordell, 1988) as the market for most of these benefits are absent due to their public good
characteristics.
The hedonic pricing method (HPM) has been widely used to estimate the values of different
environmental and recreational amenities (or disamenities), which are bundled in property values
or sale prices (e.g. Dombrow, Rodriguez, & Sirmans, 2000; Geoghegan, Wainger, & Bockstael,
1997; Hatton MacDonald et al., 2010; Kim & Goldsmith, 2009; Loomis, Rameker, & Seidl,
2004; Mansfield, Pattanayak, McDow, McDonald, & Halpin, 2005; Payton, Lindsey, Wilson,
2
Ottensmann, & Man, 2008; Samarasinghe & Sharp, 2010; Tapsuwan, Ingram, Burton, &
Brennan, 2009; Zhang, Meng, & Polyakov, 2013). One of the several applications of HPM is to
estimate the values of urban trees, open space, and forest cover (for example see, Anderson &
Cordell, 1988; Cho, Poudyal, & Roberts, 2008; Dombrow et al., 2000; Donovan & Butry, 2010;
Kong, Yin, & Nakagoshi, 2007; Netusil, Chattopadhyay, & Kovacs, 2010; Sander et al., 2010;
Tyrväinen & Miettinen, 2000).
Recently, Donovan and Butry (2010) pointed out that the earlier studies that examined the effect
of trees on property value have not differentiated the relative impact of different types, sizes, and
species of trees on property value. It is reasonable to expect that prospective home buyers may
place different values to different types of trees as they provide different services. For instance,
conifers provide year round greenery to urban residents, even though they might be less
preferred than broad-leaved trees. Unlike conifers, broad-leaved trees provide different services
at different times of the year, such as shade in the summer whilst allowing warmth and light to
come through during the winter due to their deciduous nature (Pandit & Laband, 2010b).
Similarly, prospective buyers may place different values to the same type of tree depending on
its location – within one’s property (private space) or on the next-door neighbours’ property or
on the street verge adjacent to the property (public space).
There have been a few studies that examined location specific contribution of trees on property
values, mostly in Europe and North America, using tree cover variables such as proportional area
covered by trees within and adjacent to the property without differentiating the types of trees
(Netusil et al., 2010; Sander et al., 2010; Tyrväinen & Miettinen, 2000). As Mansfield et al.
(2005) pointed out, “each type of forest cover provides different amenities to the homeowner and
3
to society”, it is thus important to differentiate tree cover by tree types to estimate their relative
contributions on property values.
The missing link between type and the location of trees in urban areas and their effect on
property values is important to examine in an Australian context for multiple reasons. First, the
environmental and economic values of trees in residential settings have been widely recognized
and strongly held among Australian residents and city planners (City of Perth, 2010; Powell,
1976), yet the effect of trees on property price by their types and location had not been studied.
Second, the property market is a major industry in Australian cities and understanding the link
between tree location/types and sale price would be useful in property appraisal processes and/or
developing new residential neighbourhoods to accommodate cities’ growing populations.
Thirdly, it could lend empirical support to the existing effort of city councils to develop and
maintain urban greenery in residential areas.
The choice between traditional, i.e. non-spatial ordinary least squares model, and spatial hedonic
models is of empirical interest partly because spatial hedonic models capture unobserved spatial
effects among observations if such effects (or correlations) exist in the data. Non-spatial model
results are biased if a spatial lag process is significant, i.e. neighbouring property price affects the
price of a property; and they are inefficient if a spatial error process is present, i.e. errors among
observations are spatially correlated (Anselin, 1988). In addition, results are biased if omitted
variables in the model are spatially correlated with the dependent variable (LeSage & Pace,
2009). In a wildfire risk study using both non-spatial and spatial models, (Mueller & Loomis,
2008) found that the model results are identical and suggested that traditional hedonic models
might be preferable for policy purposes despite inefficient parameter estimates. However, their
4
data exhibited only spatial error process and ignoring it did not cause substantial bias in
coefficients estimates. On the other hand, in the presence of spatial effects spatial hedonic
models produce robust parameter estimates (Anselin, 1988). We therefore compare both types of
model results in our study.
The aim of this paper is to examine the effect of trees on property value using spatial hedonic
models with the emphasis on two types of trees, broad-leaved and palm trees. These two types of
trees are generally found in the northern suburbs of the Perth metropolitan region. Tree locations
are differentiated according to whether they are within a property boundary (private space) or in
an adjacent street verge (public space). The paper is organized into four sections. Following this
introduction, we discuss the methods of this study including a brief description of the study area,
data and variables, and the modelling process. Then we present and discuss the results, and make
some concluding remarks along with potential policy implications of the findings.
2. Materials and Methods
2.1 Study area
This study area covers 23 northern suburbs of the Perth metropolitan region in Western Australia
within four city councils: City of Bayswater, Town of Vincent, City of Stirling, and City of
Wanneroo. The study area extends approximately 22 km north-south and 2-4 km east-west,
covering an area of approximately 92 km2 with Wanneroo and Highgate as northern and southern
most suburbs respectively. The socio-economic setting of the area ranges from affluent and
established suburbs close to the city centre to middle-class newly developed suburbs farther
5
away on the northern outskirts. Figure 1 displays the extent of the study area, physical location of
the single-family houses sold in 2006, and the main roads within the study area.
The land use of the study area is dominated by residential housing, although a mixture of
industrial, recreational and commercial land uses is also present. Some significant environmental
amenities within or surrounding the area include the Swan River, Bold, Hyde and Kings Parks,
Warwick and Koondoola Bush Reserves, Lakes Monger, Joondalup and Goollelal and several
golf courses.
The study area is well connected by road networks with the Mitchell freeway near its western
border. The area runs parallel to the Indian Ocean in north-south direction with 2-3 km average
distance to the beach from the closest boundary of the area (Figure 1).
[Figure 1 about here]
2.2 Data and variables
Following a general practice of hedonic modelling and the insights from an earlier hedonic study
in Perth (Tapsuwan et al., 2009), we collected data on three groups of key variables: property
related, location specific and environmental characteristics. Using multiple sources, data on
property sales, geographic location and extent of the property, neighbourhood characteristics
including urban trees were collected. We used 2006 data on property sales that were acquired
from Landgate, a government agency which is the custodian of property data in Western
Australia. Property sale data contained parcel number, sale price, date of sale, and structural
characteristics of a property. Sale price was deflated to the 1 January 2006 price using Housing
6
Price Index (HPI) obtained from the Australian Bureau of Statistics (www.abs.gov.au). HPI is
available on a quarterly basis, so it was linearly interpolated for the dates between last days of
the quarters. Housing price index had increased 42% between the last quarter of 2005 and the
last quarter of 2006. For geospatial referencing and delineation of the boundaries of each
property, we retrieved cadastral map data from Shared Land Information Platform (SLIP) of
Landgate. The cadastral map data was also used to locate small neighbourhood parks in the
study area. These neighbourhood reserves are often small in size and covered with grass and
individual trees, generally used as play grounds. The locations of large parks, which are mostly
covered by natural vegetation with hiking and bush-walking facilities, were identified using the
‘parks and recreation’ layer of the land-use and zoning map. The location of golf courses, sport
reserves (i.e. parks with soccer/cricket pitches) and parks with lakes were identified using either
a cadastral, or land-use zoning maps. A high resolution (30 cm) aerial image of the study area
was accessed from SLIP using Web Map Service (WMS) to identify and count trees on and
around sample properties. To characterize suburb specific criminal activities in the study area,
we retrieved data on burglaries and assaults by suburb for 2005 from the State Police Service of
Western Australia website (http://www.police.wa.gov.au).
From all sale records for 2006, we selected single (detached) family homes with a property area
of less than 2 ha and sale price greater than $100,000, which resulted in a total of 2149
observations for analysis. Spatial analysis to link property location with other spatially explicit
variables was conducted using ArcGIS 10. The property boundary layer extracted from the
cadastral map was superimposed on the aerial image of the study area. Broad-leaved and palm
trees within the boundaries of each property, on the neighbouring properties, and on the verges
7
of the streets adjacent to each property were counted manually. Euclidian distances between each
property and the Central Business District (CBD) of Perth city, the nearest main road, and
different types of parks and reserves were computed in ArcGIS 10. The distance was measured
from each property to the central location of Perth city, the nearest main road edge, and to the
edges of the parks and reserves. A brief description of variables and their descriptive statistics
are presented in Table 1.
[Table 1 about here]
The number of trees on the property ranged from 0 to 11 for broad-leaved and 0 to 10 for palms
with a slightly higher mean for broad-leaved trees (i.e. 1.013 and 0.786 trees per property).
Compared to the average number of trees on residential properties, fewer numbers of broad-
leaved and palm trees were found on adjacent street verges (0.4 and 0.03) and on neighbouring
properties (0.52 and 0.16). This reduction in average number of trees on street verges is primarily
due to the fact that for most of the properties in our sample only one side was facing the street.
2.3 Models specification
The hedonic pricing model has been widely used to estimate the implicit price of different
structural, neighbourhood and environmental attributes of a house or property from its sale price
(Champ, Boyle, & Brown, 2003; Freeman, 1979). The model assumes that a house is a
differentiated good and the differences in house prices are due to different attributes of the house
including its neighbourhood and accessibility or location specific characteristics (Rosen, 1974).
Following Rosen’s seminal work we propose a traditional hedonic pricing model that represents
the price of a differentiated good – a house in this case – which reflects the value of structural,
8
neighbourhood, and location specific or accessibility characteristics associated with it (Equation
1). In this study we consider the number of trees within the property boundary (private space)
and on adjacent street verges (public space) as environmental characteristics of a house which
are parts of both structural and neighbourhood characteristics in the model.
(1)
where is the house sale price; is a j×1 vector of structural characteristics of the house and the
characteristics of the property (= j) that includes property area, house age, indicator for presence
of swimming pool, number of structural features of the house, including bathrooms, bedrooms,
dining and meal rooms, study rooms, parking spaces, and number of broad-leaved and palm trees
on the property; is a k×1 vector of neighbourhood and location characteristics (=k) that
includes the number of broad-leaved and palm trees on street verges adjacent to the property,
number of burglaries per 1000 houses and assaults per 1000 population by suburb, and distances
to the CBD, the nearest main road, large parks (i.e. large parks with hiking and/or bush walking
facilities), parks with lakes, sport reserves (i.e. large sporting fields for soccer/cricket/rugby
games), small neighbourhood reserves (i.e. playgrounds), and golf courses; α is the intercept,
and are parameter vectors to be estimated; and is the error term.
The hedonic model in equation 1 does not account for any spatial relationships. However, spatial
data such as house sale prices could exhibit spatial relationships in two-ways: spatial correlation
9
among observations of the dependent variable (sale prices) and model errors (Anselin, 1988).
Spatial relationships are rooted in the fundamental law of geography referred to as the “first law
of geography” that captures the essence of spatial or geographical influence among observational
units and states that “Everything is related to everything else, but near things are more related
than distant things” (Tobler, 1970). The relationship is referred to as spatial lag when the sale
price of a house is affected by the sale prices of neighbouring houses. When omitted variable
bias exists due to unobserved variables related to the location of a property, the errors of the
model may be spatially correlated. The model that takes into account of both spatial lag and
spatial error could be specified as (Equation 2):
(2)
where is the spatial lag parameter to be estimated and is a n×1 vector from the spatial
weight matrix, is the spatial error coefficient and is an uncorrelated error term, i.e.
.
The spatial weight matrix W defines the way in which observational units are believed to be
neighbours and determines the influence of neighbouring observations (see Anselin, 1988 ;
Conway, Li, Wolch, Kahle, & Jerrett, 2010; Taylor, 2003 for theory and applied examples).
Most of the observations in the data set are not immediate neighbours. In such cases, common
10
approaches to define the spatial weight matrix are the inclusion of N nearest neighbours or
observations within a certain cut-off distance. It is common to assume that the strength of the
spatial relationship declines as the distance between the two observations increases. Among
assumptions of weakening spatial relationship with distance, the most common is that the spatial
relationship decays proportionally to the inverse distance between the observations (Maddison,
2009). To determine the threshold of the inverse distance weight matrix, we examined the
semivariogram of the Ordinary Least Square (OLS) residuals (Donovan & Butry, 2010). Row
standardisation was used to normalise the total weight assigned to an observation to 1, which
makes the interpretation of the spatial error or spatial lag coefficients more intuitive. In this
study, we compared 8-nearest neighbour row-standardised weight matrix and inverse distance
row-standardised weight matrix.
We applied the Box-Cox transformations of the dependent variable (sales price) using SAS®
v.9.2 TRANSREG procedure to identify the most appropriate functional form for the hedonic
price function. Results indicate that the natural log transformed dependent variable is the most
appropriate functional form. Among the explanatory variables, all of the distance related
variables were also transformed to natural log form. Given the structural differences and other
variations on house attributes, it is possible that older houses may have some heritage premium
attached which could be reflected in house price. To examine potential nonlinearities associated
with cultural or heritage value of older houses, we included a squared term for house age in the
model.
11
3. Results and discussion
We estimated an OLS based traditional hedonic model (Table 2). To address a potential concern
that the number of trees on the verge of the street could be endogenous in the model as trees on
the street are more likely to be planted in rich communities, we performed a Hausman test of
endogeneity by using property frontage as an instrumental variable. This variable is uncorrelated
with the error term but correlated with the variable in question. The test failed to reject the
hypothesis of endogeneity (F-statistic =0.0586, p-value=0.8087), therefore we conclude that the
number of trees on the street verge is not endogenous.
We then explored the presence of spatial dependence using a semivariogram of the residuals of
the OLS model (Figure 2). It presents the semivariance as a function of distance between
observations. If the residual semivariance of closely located observations is smaller than the
residual semivariance of observations located further apart, spatial dependence is likely to be
present (Donovan & Butry, 2010). Analysis of the semivariogram in Figure 2 suggests the
presence of spatial dependence, which disappears after approximately 2000 m distance. We use
this distance as a threshold for the inverse distance row normalised spatial weight matrix. This
matrix was used to calculate Moran’s I statistic, which confirms the presence of spatial
dependence in the residuals (Table 3). For comparison, we also calculated Moran’s I statistic
using the 8-nearest neighbour spatial weight matrix. It also confirms the presence of spatial
dependence in the residuals, but the test using the 2000 m inverse distance spatial weight matrix
provided slightly stronger evidence of spatial dependency.
[Figure 2 about here]
12
[Table 2 about here]
[Table 3 about here]
In order to determine the type and magnitude of spatial dependence, we conducted a series of
Lagrange multiplier (LM) and Robust LM tests for presence of spatial lag and spatial error
(Anselin, Bera, Florax, & Yoon, 1996) using both types of spatial weight matrices (Table 3). The
LM tests test the null hypothesis of no spatial lag or spatial error dependence, while the robust
LM tests test the null hypothesis of no spatial error (spatial lag) dependence in the presence of
spatial lag (spatial error) dependence. The test statistics in Table 3 indicate the presence of strong
spatial error dependence for both LM and robust LM tests based on both types of weight
matrices, but the lag dependence was apparent only with the inverse distance weight matrix.
Finally, LM-SARMA test was applied to test the null hypothesis of both spatial lag and spatial
error equal to zero. The test results for both types of spatial weight matrices are highly
significant suggesting the presence of both lag and error dependence processes. However, the
test based on 2000 m inverse distance spatial weight matrix suggests a stronger evidence of these
processes. Based on these test results, we decided to estimate a model that accounts for both
spatial lag and spatial error using the 2000 m inverse distance spatial weight matrix. We apply a
general spatial two-stage least squares (GS2SLS) procedure that produces spatial heteroskedastic
and autocorrelation consistent (HAC) estimators of the variance-covariance matrix of the model
coefficients (Kelejian & Prucha, 2010) using ‘glsthet’ package (Piras, 2010) for R statistical
software (R Development Core Team, 2011). The model results are presented in Table 2.
13
The magnitude and significance of spatial control variables, as well as the result of the Wald test
for joint significance of spatial lag and spatial error coefficients in Table 2, confirm the presence
of both spatial error and spatial lag processes. The results are consistent across the OLS and
spatial models in their direction of impact (i.e. sign) and significance for most of the structural
variables and variables describing the number of trees on and around the property. The
coefficient for the number of broadleaved trees on the street verge is slightly overestimated in the
OLS model and the coefficient for palm trees on the property became insignificant in the spatial
model. This indicates that there might be some spatially correlated unobserved factors impacting
the OLS model estimates, which are unlikely to be correlated with tree variables. On the other
hand, there is a substantial bias in the coefficients for most of the distance based and suburb
based variables estimated using OLS (the magnitude is greater). For distance based variables, the
distance to a feature (such as CBD, park, or road) is only a proxy for the effect of the feature.
Furthermore, the values of the variables are spatially correlated due to the way they were
calculated. For suburb based variables, such as crime, a measurement error is likely to be present
due to the fact that the value of crime is assumed to be same throughout the suburb. In reality,
crime might be concentrated in a certain part of a suburb and have an effect on the neighbouring
suburb. These spatial effects are the likely causes of spatial autocorrelation. In the following
sections we discuss the results of the spatial model that accommodates both spatial error and
spatial lag parameters.
Regression coefficients of the structural characteristics of the house have the expected signs and
significance (Table 2). The numbers of bathrooms, bedrooms, dining and meal rooms, study
rooms, and parking spaces in the garage have positive and significant impacts on sale price. Land
14
area and presence of swimming pools also have similar impacts on sale price. House age and sale
price have a non-linear relationship with 49.4 years as a threshold age. Up until this age house
price decreases as age increases but after this house price increases as age increases.
Table 4 presents the implicit price of statistically significant variables in the model based on
median sale price. The marginal implicit prices of significant structural variables in Table 4 are
similar and fall between the findings of earlier Australian studies by Tapsuwan et al. (2009) and
Hatton MacDonald et al. (2010). For example, we found that the median-based marginal implicit
price for land area is about AU$ 222/m2, which is lower than AU$ 401/m2 found by Tapsuwan et
al. (2009) in neighbouring Perth suburbs and higher than AU$ 126/m2 found by Hatton
MacDonald et al. (2010) in Adelaide. Similarly, the implicit price of an additional bathroom
AU$ 32,215 is between AU$ 61,152 (Tapsuwan et al., 2009) and AU$ 21,580 (Hatton
MacDonald et al., 2010) found in these earlier studies.
[Table 4 about here]
We found that both the distance to the CBD and the nearest main road have significant
influences on sale price (Table 2 and 4). While proximity to the Perth city centre has a positive
effect of $ 7.28 for each metre closer to the city, proximity to the main road constitutes a
disamenity associated with traffic noise and decreases house prices by $ 34.84 for each metre
being closer to the main road. These results support earlier findings by Tapsuwan et al. (2009)
who found a premium of AU$ 31.84 and AU$ 26.13 for a house being closer to the city centre
and it being away from main highways in Perth, while Hatton MacDonald et al. (2010) found
15
slightly different premiums associated with distance to main road (i.e. AU$ 30/m) and distance
to Adelaide city centre (AU$ 4/m).
Proximity to public open space such as parks and reserves can influence house sale price
depending on the type and size of parks and reserves. Taking insights from the findings of
Hatton MacDonald et al. (2010) and Tapsuwan et al. (2009), we differentiated parks and reserves
into five different categories, namely, large parks, parks with lakes, sport reserves, small
neighbourhood reserves, and golf courses, to accurately capture the amenity value associated
with different types of parks and reserves. Our results suggest that proximity to parks with lakes
and small neighbourhood reserves have positive and statistically significant impacts on sale price
(the coefficient for the distance is negative); while proximity to large parks and sport reserves
have negative impacts (Table 2). This latter result is contrary to the findings of earlier studies
(Acharya & Bennett, 2001; Frech & Lafferty, 1984), but is consistent with various recent
findings (Bark, Osgood, Colby, & Halper, 2011; Fierro, Fullerton, & Donjuan-Callejo, 2009;
Hatton MacDonald et al., 2010). The reasons why proximity to large parks negatively impacts
house price depend on the study site and other contexts. For example, in northern Mexico, large
parks were considered a disamenity because of crime and poor park maintenance (Fierro et al.,
2009). In South Australia, the lack of aesthetic appeal and the fear of wild fire and poisonous
snakes also led to parks becoming disamenities (Hatton MacDonald et al., 2010).Our findings for
the effect of proximity to parks with lakes (Table 2) agree with the findings of (Hatton
MacDonald et al., 2010), however the effect of small neighbourhood reserves becomes
insignificant in the spatial model, and results of this study differs on the effect of proximity to
large sporting reserves.
16
The focus of our study was on examining the effect of broad-leaved and palm trees on house
sale prices depending on their location i.e. trees located on private space (within the property
boundary and on neighbouring properties) versus trees located on public space (i.e. along the
street verges next to the property). Contrary to our expectations, we found no effect of palm trees
on house sale price regardless of their location – either on one’s property, neighbouring
properties or on street verges (Table 2). We found positive and sizable effects of broad-leaved
trees on sale price only when such trees are on street verges, while trees on the property and trees
on neighbouring properties did not have statistically significant effects. A broad-leaved tree on
the street verge increases the median value of the property by 1.9% which corresponds to the
marginal implicit price of about AU$ 7,675 in our study area.
Our results are consistent with the findings of other studies that analysed the effects of trees both
on and around the property on sales price. In a recent study by Donovan and Butry (2010) in
Oregon, they found that 0.558 street trees in front of the house combined with canopy cover of
84 m2 within 30.5 m adds US$ 8,870 to sales price of an average house. Similarly, Sander et al.
(2010) in Minnesota found that a 10% increase in coverage of trees that are within 250 m of the
house, including trees on street verges, increases sales price by about US$ 836 (0.29%).Similar
to this study, they found no statistically significant effect of tree cover within the property on
sales price.
We believe that both trees on the property and trees on the street verge benefit homeowners.
However, trees on the property are associated with the cost of establishment and maintenance as
well as opportunity cost (trees compete for valuable space with other land uses such as lawns,
garden beds, swimming pools), which might outweigh the benefit. At the same time,
17
homeowners do not bear any cost associated with planting or maintaining trees on street verges
because they are maintained by public agencies (such as city councils). Therefore, the private net
benefit of street trees could be higher to residents while the opportunity costs associated with
these trees are lower compared to trees on the property. Furthermore, since tree cover is spatially
correlated, models that do not control for neighbourhood tree cover and spatial error might yield
inflated value of the coefficient of trees on the property (Sander et al., 2010).
We found that controlling for spatial autocorrelation has an impact on the estimated regression
coefficients and the marginal implicit price of the variables. Among two sets of model results
(Table 2), we found that the coefficient estimates for some of the highly significant variables are
substantially different between the models but the estimates for the number of broad-leaved trees
is not. For example, the 95% confidence interval of the OLS estimates of distance to the CBD (-
0.4617 to -0.4103), distance to bush-walking parks (0.0252 to 0.0462) and distance to sport
reserves (0.0176 to 0.0362) indicate statistical difference between the OLS and the spatial model
estimates. Consequently, the implicit prices in Table 4 associated with these variables would also
be substantially different; for example, the implicit price associated with proximity to the CDB is
AU$ 13.63/m and AU$ 7.28/m (Table 4) based on OLS and spatial model estimates,
respectively. The differences in implicit prices for variables between the models indicate an
omitted variable bias in the OLS model estimates, which signifies the importance and relevance
of controlling for spatial effects in hedonic modelling.
18
4. Conclusion
It has been shown by a number of hedonic studies that urban trees are valued by home owners
(Abbott & Klaiber, 2010; Anderson & Cordell, 1988; Dombrow et al., 2000; Tyrväinen &
Miettinen, 2000). Our study is consistent with these findings while improving our understanding
of how people value urban trees in two aspects. First, we found that different types of trees have
different effects on sale prices; broadleaved trees increase the sale prices while palm trees have
no effect. Second, broad-leaved trees are valued differently depending on where they are located,
whether within the property boundary (private space) or on the street verge adjacent to the
property (public space). A broad-leaved tree on the street verge, not on the property, increases
median property price of a house by about AU$7,675 (1.9%).
There might be several reasons why home-buyers value trees on public space more than trees on
private space. Trees on the property might have some disamenities such as blocking views,
dropping leaves, and damaging pavements (Donovan & Butry, 2010), despite their amenity
benefits. Broad-leaved trees found in our study area are mostly eucalyptus; they are fast growing
and often damage pavements and other infrastructure (i.e. drainage) due to their root system, in
addition to occupying valuable space that could be used for other purposes. Moreover,
maintaining these trees imposes costs on homeowners. In contrast, broad-leaved trees on public
space are highly valued by residents as they provide amenity services without incurring
significant private costs. The management costs associated with pruning, thinning, and removals
of street trees (public space) are borne by city councils while the benefits are shared among local
residents at a modest involvement at their will primarily for watering trees in the early stages of
19
plantings. Growing trees on street verges allows residents to enjoy benefits from street trees
without substantial involvement in their management.
One implication of our findings is that it is economical from the residents’ point of view to
promote broad-leaved trees along street verges compared to palm trees because broad-leaved
trees have other benefits to residents, including ameliorating micro-climate. For example, shade
casts by broad-leaved trees during hot summer months would help to reduce the temperature
underneath the tree thereby ameliorating the micro climate. This view may particularly be
relevant given the intense dry heat in Perth that could be in the low to mid 40 degree Celsius
during summer months. Secondly, residents have some stake in managing urban trees on the
street as it adds value to their property. Urban forestry programs targeted to develop greenery
along the streets would be positively viewed by local residents and therefore city councils can
use a public-private partnership approach in managing street trees, particularly for watering trees
in the early stages of establishment.
Methodologically, our study highlights the importance of incorporating spatial effects into
hedonic models. Spatial hedonic models accommodate the influence of neighbouring
observations and unobserved spatial correlation, thus the estimates are, arguably, robust,
consistent, and efficient. Ignoring spatial dependencies yields inflated regression coefficients
(Table 2) and thus imprecise implicit prices. For example, the implicit price of an additional
bathroom reduced from AU$ 45,781 to AU$ 32,216, while the price of a bedroom increased
from AU$ 9670 to AU$ 11,589 based on OLS and spatial model estimates, respectively. Such
differences could be attributed to omitted variables in the OLS model, such as size or type of
bedrooms and bathrooms, age of suburbs, different designs between old and new houses.
20
Similarly, the difference in estimated implicit price of proximity to sport reserve (AU$ 26.94/m
without spatial control versus AU$ 13.45/m with spatial controls) also revealed the importance
of controlling for spatial effects in hedonic studies.
We differentiated tree types into two broad categories by appearance but not by their sizes or
origins (such as exotic versus native). As Donovan and Butry (2011) suggested, it is important to
differentiate between sizes of the trees as residents could have size based preferences. In
differentiating trees by size, canopy cover can be used as a proxy, i.e. bigger trees have bigger
canopies. This is an avenue for further research which can be accomplished using remote sensing
data to delineate tree cover in private and public spaces to find amenity values associated with
different types and sizes of green spaces in Perth. In addition, valuing trees based on their origin
would provide useful information to city planners to make choices regarding tree species
whether to opt for native species that are more suitable for the dry Perth climate or non-native
species that could be more aesthetically pleasing.
21
References
Abbott, J. K., & Klaiber, H. A. (2010). Is all space created equal? Uncovering the relationship
between competing land uses in subdivisions. Ecological Economics, 70(2), 296-307.
Acharya, G., & Bennett, L. L. (2001). Valuing Open Space and Land-Use Patterns in Urban
Watersheds. Journal of Real Estate Finance and Economics, 22(2-3), 221-237.
Anderson, L., & Cordell, H. (1988). Influence of trees on residential property values in Athens,
Georgia (USA): a survey based on actual sales prices. Landscape and Urban Planning,
15(1-2), 153-164.
Anselin, L. (1988). Spatial Econometrics: Methods and Models. Dordrecht: Kluwer Academic
Publishers.
Anselin, L., Bera, A. K., Florax, R., & Yoon, M. J. (1996). Simple diagnostic tests for spatial
dependence. Regional Science and Urban Economics, 26(1), 77-104.
Bark, R. H., Osgood, D. E., Colby, B. G., & Halper, E. B. (2011). How Do Homebuyers Value
Different Types of Green Space? Journal of Agricultural and Resource Economics,
36(2), 395–415.
Brack, C. L. (2002). Pollution mitigation and carbon sequestration by an urban forest.
Environmental Pollution, 116(SUPPL. 1), S195-S200.
Champ, P. A., Boyle, K. J., & Brown, T. C. (2003). A primer on nonmarket valuation. Boston:
Kluwer Academic Publishers.
Cho, S. H., Poudyal, N. C., & Roberts, R. K. (2008). Spatial analysis of the amenity value of
green open space. Ecological Economics, 66(2-3), 403-416.
City of Perth. (2010). Visitor information available at: http://www.cityofperth.wa.gov.au/
web/Visiting/Attractions/Parks-and-Gardens/ [accessed on June 4, 2010]
22
Conway, D., Li, C. Q., Wolch, J., Kahle, C., & Jerrett, M. (2010). A spatial autocorrelation
approach for examining the effects of urban greenspace on residential property values.
Journal of Real Estate Finance and Economics, 41(2), 150-169.
Dombrow, J., Rodriguez, M., & Sirmans, C. (2000). The market value of mature trees in single-
family housing markets. Appraisal Journal, 68(1), 39-43.
Donovan, G. H., & Butry, D. T. (2010). Trees in the city: Valuing street trees in Portland,
Oregon. Landscape and Urban Planning, 94(2), 77-83.
Donovan, G. H., & Butry, D. T. (2011). The effect of urban trees on the rental price of single-
family homes in Portland, Oregon. Urban Forestry and Urban Greening, 10(3), 163-168.
Dwyer, J. F., Schroeder, H. W., & Gobster, P. H. (1991). The significance of urban trees and
forests: toward a deeper understanding of values. Journal of Arboriculture, 17(10), 276-
284.
Ellis, C. D., Lee, S. W., & Kweon, B. S. (2006). Retail land use, neighborhood satisfaction and
the urban forest: An investigation into the moderating and mediating effects of trees and
shrubs. Landscape and Urban Planning, 74(1), 70-78.
Fierro, K. P., Fullerton, T. M., & Donjuan-Callejo, K. E. (2009). Housing attribute preferences in
a Northern Mexico metropolitan economy. Atlantic Economic Journal, 37(2), 159-172.
Frech, H. E., & Lafferty, R. N. (1984). The effect of the California Coastal Commission on
housing prices. Journal of Urban Economics, 16(1), 105-123.
Freeman, A. M. (1979). Hedonic prices, property values and measuring environmental benefits:
a survey of the issues. Scandinavian Journal of Economics, 81(2), 154-173.
Geoghegan, J., Wainger, L. A., & Bockstael, N. E. (1997). Spatial landscape indices in a hedonic
framework: An ecological economics analysis using GIS. Ecological Economics, 23(3),
251-264.
Hatton MacDonald, D., Crossman, N. D., Mahmoudi, P., Taylor, L. O., Summers, D. M., &
Boxall, P. C. (2010). The value of public and private green spaces under water
restrictions. Landscape and Urban Planning, 95(4), 192-200.
23
Kelejian, H. H., & Prucha, I. R. (2010). Specification and estimation of spatial autoregressive
models with autoregressive and heteroskedastic disturbances. Journal of Econometrics,
157(1), 53-67. doi: 10.1016/j.jeconom.2009.10.025
Kim, J., & Goldsmith, P. (2009). A spatial hedonic approach to assess the impact of swine
production on residential property values. Environmental and Resource Economics,
42(4), 509-534.
Kong, F., Yin, H., & Nakagoshi, N. (2007). Using GIS and landscape metrics in the hedonic
price modeling of the amenity value of urban green space: A case study in Jinan City,
China. Landscape and Urban Planning, 79(3-4), 240-252.
LeSage, J. P., & Pace, R. K. (2009). Introduction to spatial econometrics (Vol. 196): Chapman
& Hall/CRC.
Loomis, J., Rameker, V., & Seidl, A. (2004). A hedonic model of public market transactions for
open space protection. Journal of Environmental Planning and Management, 47(1), 83-
96.
Maddison, D. (2009). A Spatio-temporal Model of Farmland Values. [Article]. Journal of
Agricultural Economics, 60(1), 171-189. doi: 10.1111/j.1477-9552.2008.00182.x
Mansfield, C., Pattanayak, S. K., McDow, W., McDonald, R., & Halpin, P. (2005). Shades of
Green: Measuring the value of urban forests in the housing market. Journal of Forest
Economics, 11(3), 177-199.
McPherson, G., Simpson, J. R., Peper, P. J., Maco, S. E., & Xiao, Q. F. (2005). Municipal forest
benefits and costs in five US cities. Journal of Forestry, 103(8), 411-416.
Mueller, J. M., & Loomis, J. B. (2008). Spatial dependence in hedonic property models: Do
different corrections for spatial dependence result in economically significant differences
in estimated implicit prices? Journal of Agricultural and Resource Economics, 33(2),
212-231.
Netusil, N. R., Chattopadhyay, S., & Kovacs, K. F. (2010). Estimating the demand for tree
canopy: A second-stage hedonic price analysis in Portland, Oregon. Land Economics,
86(2), 281-293.
24
Nowak, D. J., Crane, D. E., & Stevens, J. C. (2006). Air pollution removal by urban trees and
shrubs in the United States. Urban Forestry & Urban Greening, 4(3-4), 115-123.
Pandit, R., & Laband, D. N. (2010). Energy savings from tree shade. Ecological Economics,
69(6), 1324-1329. doi: 10.1016/j.ecolecon.2010.01.009
Pandit, R., & Laband, D. N. (2010b). A hedonic analysis of the impact of tree shade on
summertime residential energy consumption. Arboriculture and Urban Forestry, 36(3),
73-80.
Payton, S., Lindsey, G., Wilson, J., Ottensmann, J. R., & Man, J. (2008). Valuing the benefits of
the urban forest: A spatial hedonic approach. Journal of Environmental Planning and
Management, 51(6), 717-736.
Piras, G. (2010). sphet: Spatial models with heteroskedastic innovations in R. Journal of
Statistical Software, 35(1), 1-21.
Powell, J. M. (1976). Environmental Management in Australia, 1788-1914. Melbourne: Oxford
University Press.
R Development Core Team. (2011). R: A Language and Environment for Statistical Computing.
Vienna, Austria: R Foundation for Statistical Computing.
Rosen, S. (1974). Hedonic prices and implicit markets: product differentiation in pure
competition. The Journal of Political Economy, 82(1), 34-55.
Samarasinghe, O., & Sharp, B. (2010). Flood prone risk and amenity values: A spatial hedonic
analysis. Australian Journal of Agricultural and Resource Economics, 54(4), 457-475.
Sander, H., Polasky, S., & Haight, R. G. (2010). The value of urban tree cover: A hedonic
property price model in Ramsey and Dakota Counties, Minnesota, USA. Ecological
Economics, 69(8), 1646-1656.
Sheets, V. L., & Manzer, C. D. (1991). Affect, cognition, and urban vegetation: some effects of
adding trees along city streets. Environment & Behavior, 23(3), 285-304.
Simpson, J. R. (1998). Urban forest impacts on regional cooling and heating energy use:
Sacramento County case study. Journal of Arboriculture, 24(4), 201-209.
25
Tapsuwan, S., Ingram, G., Burton, M., & Brennan, D. (2009). Capitalized amenity value of
urban wetlands: A hedonic property price approach to urban wetlands in perth, western
Australia. Australian Journal of Agricultural and Resource Economics, 53(4), 527-545.
Taylor, L. O. (2003). The Hedonic Method. In P. A. Champ, K. J. Boyle & T. C. Brown (Eds.),
A primer on nonmarket valuation (pp. 331-375). Boston: Kluwer Academic Publishers.
Tobler, W. (1970). A computer movie simulating urban growth in the Detroit region. Economic
Geography, 46(2), 234-240.
Tyrväinen, L., & Miettinen, A. (2000). Property prices and urban forest amenities. Journal of
Environmental Economics and Management, 39(2), 205-223.
Wolf, K. L. (2005). Business district streetscapes, trees, and consumer response. Journal of
Forestry, 103(8), 396-400.
Zhang, D., Meng, L., & Polyakov, M. (2013). Determinants of the Prices of Bare Forestland and
Pre-Merchantable Timber Stands: A Spatial Hedonic Study. Forest Science, (In press).
doi: 10.5849/forsci.12-014
26
List of tables and table captions
1. Table 1 Model variables and descriptive statistics
2. Table 2 Ordinary least-squares (OLS) and spatial hedonic regression results of factors
affecting property values (dependent variable Log Property price, AU$)
3. Table 3 Tests of dependencies in the OLS model
4. Table 4 Marginal implicit price for significant variables based on median house price
(AU$395,000)
27
Table 1 Model variables and descriptive statistics
Variable
Median
Mean
STD
Minimum
Maximum
Dependent variable
Sale price (AU$)
$395,000
$448,000
$199,208
$102,500
$2,150,000
Explanatory variables
Land area (m2)
692
677
143.23
253
1657
House age (years)
23
26.32
19.47
1
102
Number of bathrooms
1
1.485
0.559
1
5
Number of bedrooms
3
3.344
0.772
1
6
Number of dining and meal rooms
1
0.954
0.649
0
2
Number of study rooms
0
0.17
0.386
0
3
Number of parking spaces in the
garage
0
0.737
0.894
0
4
Presence of swimming pool
0
0.162
0.368
0
1
Number of broad-leaved trees on
the property
0
1.013
1.381
0
11
Number of palm trees on the
property
0
0.786
1.441
0
10
Number of broad-leaved trees on
the neighbouring properties
0
0.518
0.761
0
5
Number of palm trees on the
neighbouring properties
0
0.164
0.526
0
5
Number of broad-leaved trees on
street verge adjacent to the property
0
0.403
0.756
0
9
Number of palm trees on street
verge adjacent to the property
0
0.03
0.245
0
6
Distance to the CBD (km)
12.63
12.83
5.619
1.575
25.39
Distance to the nearest main road
(m)
241
321
280
20
1541
Distance to the large park
(hiking/bushwalking) (m)
1121
1499
1413
10
6881
Distance to the park with lake (m)
3436
3291
1880
11
7460
Distance to the small
neighbourhood reserve (m)
152
180
121
8
674
Distance to the sport reserve (m)
409
540
456
9
2890
Number of burglaries in suburb per
1000 houses
30
36
18
11
70
Number of assaults in suburb per
1000 residents
6
11
8
0
29
28
Table 2 Ordinary least-squares (OLS) and spatial hedonic regression results of factors affecting property
values (dependent variable Log Property price, AU$)
Variable
OLS model
Spatial model
Intercept
16.14000***
(0.17630)
7.35970***
(0.92201)
Land area
0.00059***
(0.00003)
0.00056***
(0.00004)
House age
-0.00743***
(0.00090)
-0.00779***
(0.00111)
House age2
0.00008***
(0.00001)
0.00008***
(0.00001)
Number of bathrooms
0.11590***
(0.01098)
0.08156***
(0.01150)
Number of bedrooms
0.02448***
(0.00742)
0.02934***
(0.00736)
Number of dining and meal rooms
0.01813**
(0.00757)
0.01667**
(0.00665)
Number of study rooms
0.09611***
(0.01149)
0.07576***
(0.01067)
Number of garages spaces
0.06775***
(0.00514)
0.04205***
(0.00479)
Pool
0.04561***
(0.01112)
0.04754***
(0.01128)
Number of broad-leaved trees on the property
-0.00299
(0.00312)
-0.00243
(0.00314)
Number of palm trees on the property
-0.00525*
(0.00279)
-0.00055
(0.00253)
Number of broad-leaved trees on the
neighbouring properties
-0.00634
(0.00532)
-0.00418
(0.00483)
Number of palm trees on the neighbouring
properties
-0.00768
(0.00733)
-0.00405
(0.00621)
Number of broad-leaved trees on street verge
0.02013***
(0.00556)
0.01943***
(0.00585)
Number of palm trees on street verge
0.00652
(0.01566)
0.01529
(0.01133)
Log distance to the CBD
-0.43600***
(0.01311)
-0.23289***
(0.02785)
Log distance to the main road
0.02236***
(0.00437)
0.02124***
(0.00482)
Log distance to the park (bushwalking)
0.03566***
(0.00525)
0.01238*
(0.00717)
Log distance to the park with lake
-0.03308***
(0.00536)
-0.02094***
(0.00744)
Log distance to the small reserve
-0.01481***
(0.00497)
-0.00617
(0.00468)
Log distance to the sport reserve
0.02687***
(0.00474)
0.01393**
(0.00691)
Burglaries per 1000 houses
-0.00008
(0.00057)
0.00044
(0.00095)
Assaults per 1000 residents
-0.00822***
(0.00117)
-0.00214
(0.00192)
Spatial lag
0.54557***
(0.05684)
Spatial error
0.66025***
(0.08313)
R2
0.769
Adjusted R2
0.767
Wald test that spatial lag and spatial error are
both zero
167.31***
Note: standard errors are in parentheses.
* Significant at 10% level; ** significant at 5% level; *** significant at 1% level.
29
Table 3 Tests of dependencies in the OLS model
Test
Spatial weight matrix
8-nearest neighbours
2000 m Inverse distance
Spatial error dependence
Moran's I statistics standard deviate
15.30***
21.23***
Lagrange multiplier test
230.70***
434.49***
Robust Lagrange multiplier test
230.40***
215.14***
Spatial lag dependence
Lagrange multiplier test
0.30
300.16***
Robust Lagrange multiplier test
0.01
80.82***
Spatial error and lag dependence (SARMA)
Lagrange multiplier test
230.70***
515.31***
* Significant at 10% level; ** significant at 5% level; *** significant at 1% level.
30
Table 4 Marginal implicit price for significant variables based on median house price
(AU$ 395,000)
Variable
Marginal Implicit Price
Land area, m2
$488.55
House age, yearψ
-$3,603.69
Number of bathrooms
$70,891.91
Number of bedrooms
$25,502.94
Number of dining and meal rooms
$14,489.91
Number of study rooms
$65,847.82
Number of garage spaces
$36,551.60
Presence of swimming pool
$41,321.02
Broad-leaved trees on street verge
$16,888.96
Distance to the CBD (m)
-$16.02
Distance to the main road (m)
$76.68
Distance to the large park (m)
$9.60
Distance to the park with lake (m)
-$5.30
Distance to the sport reserve (m)
$29.59
. ψ The implicit prices for house age were calculated based on median age (23 year), however the age is non-linearly
related to sale price with a threshold age of 49.4 year.
31
Figures
Figure 1 Map of the study area and locations of the observations
32
Figure 2 Semivariogram of the residuals from the OLS hedonic price model
